Transformers give Clojurists some of the benefits of "Object Orientation" without many of the downsides Clojurists dislike about objects.

1. Objects couple behaviors required from multiple callers into a single class, while transformers do not change existing behaviors for existing callers by default
2. Objects push inheritance first design, whereas transformer inheritance is a function of shared structure between Clojure data structures derived from one another and design is driven by concrete implementation needs, like regular Clojure
3. Objects couple state and methods in spaghetti ways and transformers are just immutable maps. And just like how Clojure lets you stash stateful things like atoms in functions, transformers allow you to build stateful transformers, but like Clojure the default convention is to do everything immutably
4. Objects try to provide data hiding as a function of encapsulation whereas transformers are doing the opposite, exposing data otherwise hidden by a closure

There are many strategies for reusing code in the software industry. In Clojure, we use what some call a "lego" method of building small, single purpose functions that just can be used in a million different contexts, because of a tasteful use of simplicity in the right places. This works tremendously well for 95% of use cases. In certain use-cases, like for building hierarchies of functions that are highly self-similar, like with UI toolkits, transformers provide a better alternative.Transformers allow you to build a UI toolkit with 25% the code of normal function composition and 25% of the code required for evolution over time for the widgets in that hierarchy. The lego method is great for vertically composing things together, but when you want to make lateral changes for only certain callers in the tree, you have to defensively copy code between duplicative implementation trees and just call them "grandpa-function-1" and "grandpa-function-2" and then make versions 1 and 2 for all functions that wrapped the grandpa-functions afterwards. Transformers provide a solution for that situation, in the rare cases we end up in them in Clojure, without the downsides of a traditional object system.

Je ne comprends pas bien dans la solution ce bout de code : .lien-conteneur-photo:hover.photo-hover { display: flex; } Pourquoi la class .photo-hover se trouve après la pseudo classe hover ?

Reviewer #2 (Public Review):

In the presented manuscript, the authors first use structured microfluidic devices with gliding filamentous cyanobacteria inside in combination with micropipette force measurements to measure the bending rigidity of the filaments. The distribution of bending rigidities is very broad.

Next, they use triangular structures to trap the bacteria with the front against an obstacle. Depending on the length and rigidity, the filaments buckle under the propulsive force of the cells. The authors use theoretical expressions for the buckling threshold to infer propulsive force, given the measured length and (mean-) stiffnesses. They find nearly identical values for both species, 𝑓 ∼ (1.0 {plus minus} 0.6) nN∕µm, nearly independent of the velocity. These measurements have to be taken with additional care, as then inferred forces depend strongly on the bending rigidity, which already shows a broad distribution.

Finally, they measure the shape of the filament dynamically to infer friction coefficients via Kirchhoff theory. In this section they report a strong correlation with velocity and report propulsive forces that vary over two orders of magnitude.

From a theoretical perspective, not many new results are presented. The authors repeat the the well-known calculation for filaments buckling under propulsive load and arrive at the literature result of buckling when the dimensionless number (f L^3/B) is larger than 30.6 as previously derived by Sekimoto et al in 1995. In my humble opinion, the "buckling theory" section belongs to methods.<br /> Finally, the Authors use molecular dynamics type simulations similar to other models to reproduce the buckling dynamics from the experiments.

Data and source code are available via trusted institutional or third-party repositories that adhere to policies that make data discoverable, accessible and usable.

Author response:

Reviewer 1:

The paper “Quantifying gliding forces of filamentous cyanobacteria by self-buckling” combines experiments on freely gliding cyanobacteria, buckling experiments using two-dimensional V-shaped corners, and micropipette force measurements with theoretical models to study gliding forces in these organisms. The aim is to quantify these forces and use the results to perhaps discriminate between competing mechanisms by which these cells move. A large data set of possible collision events are analyzed, bucking events evaluated, and critical buckling lengths estimated. A line elasticity model is used to analyze the onset of buckling and estimate the effective (viscous type) friction/drag that controls the dynamics of the rotation that ensues post-buckling. This value of the friction/drag is compared to a second estimate obtained by consideration of the active forces and speeds in freely gliding filaments. The authors find that these two independent estimates of friction/drag correlate with each other and are comparable in magnitude. The experiments are conducted carefully, the device fabrication is novel, the data set is interesting, and the analysis is solid. The authors conclude that the experiments are consistent with the propulsion being generated by adhesion forces rather than slime extrusion. While consistent with the data, this conclusion is inferred.

We thank the reviewer for the positive evaluation of our work.

Summary:

The paper addresses important questions on the mechanisms driving the gliding motility of filamentous cyanobacteria. The authors aim to understand these by estimating the elastic properties of the filaments, and by comparing the resistance to gliding under a) freely gliding conditions, and b) in post-buckled rotational states. Experiments are used to estimate the propulsion force density on freely gliding filaments (assuming over-damped conditions). Experiments are combined with a theoretical model based on Euler beam theory to extract friction (viscous) coefficients for filaments that buckle and begin to rotate about the pinned end. The main results are estimates for the bending stiffness of the bacteria, the propulsive tangential force density, the buckling threshold in terms of the length, and estimates of the resistive friction (viscous drag) providing the dissipation in the system and balancing the active force. It is found that experiments on the two bacterial species yield nearly identical values of f (albeit with rather large variations). The authors conclude that the experiments are consistent with the propulsion being generated by adhesion forces rather than slime extrusion.

We appreciate this comprehensive summary of our work.

Strengths of the paper:

The strengths of the paper lie in the novel experimental setup and measurements that allow for the estimation of the propulsive force density, critical buckling length, and effective viscous drag forces for movement of the filament along its contour – the axial (parallel) drag coefficient, and the normal (perpendicular) drag coefficient (I assume this is the case, since the post-buckling analysis assumes the bent filament rotates at a constant frequency). These direct measurements are important for serious analysis and discrimination between motility mechanisms.

We thank the reviewer for this positive assessment of our work.

Weaknesses:

There are aspects of the analysis and discussion that may be improved. I suggest that the authors take the following comments into consideration while revising their manuscript.

The conclusion that adhesion via focal adhesions is the cause for propulsion rather than slime protrusion is consistent with the experimental results that the frictional drag correlates with propulsion force. At the same time, it is hard to rule out other factors that may result in this (friction) viscous drag - (active) force relationship while still being consistent with slime production. More detailed analysis aiming to discriminate between adhesion vs slime protrusion may be outside the scope of the study, but the authors may still want to elaborate on their inference. It would help if there was a detailed discussion on the differences in terms of the active force term for the focal adhesion-based motility vs the slime motility.

We appreciate this critical assessment of our conclusions. Of course we are aware that many different mechanisms may lead to similar force/friction characteristics, and that a definitive conclusion on the mechanism would require the combination of various techniques, which is beyond the scope of this work. Therefore, we were very careful in formulating the discussion of our findings, refraining, in particular, from a singular conclusion on the mechanism but instead indicating “support” for one hypothesis over another, and emphasizing “that many other possibilities exist”.

The most common concurrent hypotheses for bacterial gliding suggest that either slime extrusion at the junctional pore complex [A1], rhythmic contraction of fibrillar arrays at the cell wall [A2], focal adhesion sites connected to intracellular motor-microtubule complexes [A3], or modified type-IV pilus apparati [A4] provide the propulsion forces. For the slime extrusion hypothesis, which is still abundant today, one would rather expect an anticorrelation of force and friction: more slime extrusion would generate more force, but also enhance lubrication. The other hypotheses are more conformal to the trend we observed in our experiments, because both pili and focal adhesion require direct contact with a substrate. How contraction of fibrilar arrays would micromechanically couple to the environment is not clear to us, but direct contact might still facilitate force transduction. Please note that these hypotheses were all postulated without any mechanical measurements, solely based on ultra-structural electron microscopy and/or genetic or proteomic experiments. We see our work as complementary to that, providing a mechanical basis for evaluating these hypotheses.

We agree with the referee that narrowing down this discussion to focal adhesion should have been avoided. We rewrote the concluding paragraph (page 8):

“…it indicates that friction and propulsion forces, despite being quite vari able, correlate strongly. Thus, generating more force comes, inevitably, at the expense of added friction. For lubricated contacts, the friction coefficient is proportional to the thickness of the lubricating layer (Snoeijer et al., 2013 ), and we conjecture active force and drag both increase due to a more intimate contact with the substrate. This supports mechanisms like focal adhesion (Mignot et al., 2007 ) or a modified type-IV pilus (Khayatan et al., 2015 ), which generate forces through contact with extracellular surfaces, as the underlying mechanism of the gliding apparatus of filamentous cyanobacteria: more contacts generate more force, but also closer contact with the substrate, thereby increasing friction to the same extent. Force generation by slime extrusion (Hoiczyk and Baumeister, 1998 ), in contrast, would lead to the opposite behavior: More slime generates more propulsion, but also reduces friction. Besides fundamental fluid-mechanical considerations (Snoeijer et al., 2013 ), this is rationalized by two experimental observations: i. gliding velocity correlates positively with slime layer thickness (Dhahri et al., 2013 ) and ii. motility in slime-secretion deficient mutants is restored upon exogenous addition of polysaccharide slime. Still we emphasize that many other possibilities exist. One could, for instance, postulate a regulation of the generated forces to the experienced friction, to maintain some preferred or saturated velocity.”

Can the authors comment on possible mechanisms (perhaps from the literature) that indicate how isotropic friction may be generated in settings where focal adhesions drive motility? A key aspect here would probably be estimating the extent of this adhesion patch and comparing it to a characteristic contact area. Can lubrication theory be used to estimate characteristic areas of contact (knowing the radius of the filament, and assuming a height above the substrate)? If the focal adhesions typically cover areas smaller than this lubrication area, it may suggest the possibility that bacteria essentially present a flat surface insofar as adhesion is concerned, leading to a transversely isotropic response in terms of the drag. Of course, we will still require the effective propulsive force to act along the tangent.

We thank the referee for suggesting to estimate the dimensions of the contact region. Both pili and focal adhesion sites would be of sizes below one micron [A3, A4], much smaller than the typical contact region in the lubricated contact, which is on the order of the filament radius (few microns). So indeed, isotropic friction may be expected in this situation [A5] and is assumed frequently in theoretical work [A6–A8]. Anisotropy may then indeed be induced by active forces [A9], but we are not aware of measurements of the anisotropy of friction in bacterial gliding.

For a more precise estimate using lubrication theory, rheology and extrusion rate of the secreted polysaccharides would have to be known, but we are not aware of detailed experimental characterizations.

We extended the paragraph in the buckling theory on page 5 regarding the assumption of isotropic friction:

“We use classical Kirchhoff theory for a uniform beam of length L and bending modulus B, subject to a force density ⃗b = −f ⃗t− η ⃗v, with an effective active force density f along the tangent ⃗t, and an effective friction proportional to the local velocity ⃗v, analog to existing literature (Fily et al., 2020; Chelakkot et al., 2014; Sekimoto et al., 1995 ). Presumably, this friction is dominated by the lubrication drag from the contact with the substrate, filled by a thin layer of secreted polysaccharide slime which is much more viscous than the surrounding bulk fluid. Speculatively, the motility mechanism might also comprise adhering elements like pili (Khayatan et al., 2015 ) or foci (Mignot et al., 2007 ) that increase the overall friction (Pompe et al., 2015 ). Thus, the drag due to the surrounding bulk fluid can be neglected (Man and Kanso, 2019 ), and friction is assumed to be isotropic, a common assumption in motility models (Fei et al., 2020; Tchoufag et al., 2019; Wada et al., 2013 ). We assume…”

We also extended the discussion regarding the outcome of isotropic friction (page 7):

“…Thus we plot f/v over η in Figure 4 D, finding nearly identical values over about two decades. Since f and η are not correlated with v0, this is due to a correlation between f and η. This relation is remarkable in two aspects: On the one hand, it indicates that friction is mainly isotropic. This suggests that friction is governed by an isotropic process like bond friction or lubrication from the slime layer in the contact with the substrate, the latter being consistent with the observation that mutations deficient of slime secretion do not glide but exogenous addition of slime restores motility (Khayatan et al., 2015 ). In contrast, hydrodynamic drag from the surrounding bulk fluid (Man and Kanso, 2019 ), or the internal friction of the gliding apparatus would be expected to generate strongly anisotropic friction. If the latter was dominant, a snapping-like transition into the buckling state would be expected, rather than the continuously growing amplitude that is observed in experiments. On the other hand, it indicates that friction and propulsion forces…”

I am not sure why the authors mention that the power of the gliding apparatus is not rate-limiting. The only way to verify this would be to put these in highly viscous fluids where the drag of the external fluid comes into the picture as well (if focal adhesions are on the substrate-facing side, and the upper side is subject to ambient fluid drag). Also, the friction referred to here has the form of a viscous drag (no memory effect, and thus not viscoelastic or gel-like), and it is not clear if forces generated by adhesion involve other forms of drag such as chemical friction via temporary bonds forming and breaking. In quasi-static settings and under certain conditions such as the separation of chemical and elastic time scales, bond friction may yield overall force proportional to local sliding velocities.

We agree with the referee that the origin of the friction is not easily resolved. Lubrication yields an isotropic force density that is proportional to the velocity, and the same could be generated by bond friction. Importantly, both types of friction would be assumed to be predominantly isotropic. We explicitly referred to lubrication drag because it has been shown that mutations deficient of slime extrusion do not glide [A4].

Assuming, in contrast, that in free gliding, friction with the environment is not rate limiting, but rather the internal friction of the gliding apparatus, i.e., the available power, we would expect a rather different behavior during early-buckling evolution. During early buckling, the tangential motion is stalled, and the dynamics is dominated by the growing buckling amplitude of filament regions near the front end, which move mainly transversely. For geometric reasons, in this stage the (transverse) buckling amplitude grows much faster than the rear part of the filament advances longitudinally. Thus that motion should not be impeded much by the internal friction of the gliding apparatus, but by external friction between the buckling parts of the filament and the ambient. The rate at which the buckling amplitude initially grows should be limited by the accumulated compressive stress in the filament and the transverse friction with the substrate. If the latter were much smaller than the (logitudinal) internal friction of the gliding apparatus, we would expect a snapping-like transition into the buckled state, which we did not observe.

In our paper, we do not intend to evaluate the exact origin of the friction, quantifying the gliding force is the main objective. A linear force-velocity relation agrees with our observations. A detailed analysis of friction in cyanobacterial gliding would be an interesting direction for future work.

To make these considerations more clear, we rephrased the corresponding paragraph on page 7 & 8:

“…Thus we plot f/v over η in Figure 4 D, finding nearly identical values over about two decades. Since f and η are not correlated with v0, this is due to a correlation between f and η. This relation is remarkable in two aspects: On the one hand, it indicates that friction is mainly isotropic. This suggests that friction is governed by an isotropic process like bond friction or lubrication from the slime layer in the contact with the substrate, the latter being consistent with the observation that mutations deficient of slime secretion do not glide but exogenous addition of slime restores motility (Khayatan et al., 2015 ). In contrast, hydrodynamic drag from the surrounding bulk fluid (Man and Kanso, 2019 ), or the internal friction of the gliding apparatus would be expected to generate strongly anisotropic friction. If the latter was dominant, a snapping-like transition into the buckling state would be expected, rather than the continuously growing amplitude that is observed in experiments. On the other hand, it indicates that friction and propulsion forces…”

For readers from a non-fluids background, some additional discussion of the drag forces, and the forms of friction would help. For a freely gliding filament if f is the force density (per unit length), then steady gliding with a viscous frictional drag would suggest (as mentioned in the paper) f ∼ v! L η||. The critical buckling length is then dependent on f and on B the bending modulus. Here the effective drag is defined per length. I can see from this that if the active force is fixed, and the viscous component resulting from the frictional mechanism is fixed, the critical buckling length will not depend on the velocity (unless I am missing something in their argument), since the velocity is not a primitive variable, and is itself an emergent quantity.

We are not sure what “f ∼ v! L η||” means, possibly the spelling was corrupted in the forwarding of the comments.

We assumed an overdamped motion in which the friction force density ff (per unit length of the filament) is proportional to the velocity v0, i.e. ff ∼ η v0, with a friction coefficient η. Overdamped means that the friction force density is equal and opposite to the propulsion force density, so the propulsion force density is f ∼ ff ∼ η v0. The total friction and propulsion forces can be obtained by multiplication with the filament length

L, which is not required here. In this picture, v0 is an emergent quantity and f and η are assumed as given and constant. Thus, by observing v0, f can be inferred up to the friction coefficient η. Therefore, by using two descriptive variables, L and v0, with known B, the primitive variable η can be inferred by logistic regression, and f then follows from the overdamped equation of motion.

To clarify this, we revised the corresponding section on page 5 of the paper:

“The substrate contact requires lubrication from polysaccharide slime to enable bacteria to glide (Khayatan et al., 2015 ). Thus we assume an over- damped motion with co-linear friction, for which the propulsion force f and the free gliding velocity v0 of a filament are related by f = η v0, with a friction coefficient η. In this scenario, f can be inferred both from the observed Lc ∼ (f/B)−1/3 and, up to the proportionality coefficient η, from the observed free gliding velocity. Thus, by combining the two relations, one may expect also a strong correlation between Lc and v0. In order to test this relation for consistency with our data, we include v0 as a second regressor, by setting x = (L−Lc(v0))/∆Lc in Equation 1, with Lc(v0) = (η v0/(30.5722 B))−1/3, to reflect our expectation from theory (see below). Now, η rather than f is the only unknown, and its ensemble distribution will be determined in the regression. Figure 3 E,F show the buckling behavior…”

Reviewer 2:

In the presented manuscript, the authors first use structured microfluidic devices with gliding filamentous cyanobacteria inside in combination with micropipette force measurements to measure the bending rigidity of the filaments.

Next, they use triangular structures to trap the bacteria with the front against an obstacle. Depending on the length and rigidity, the filaments buckle under the propulsive force of the cells. The authors use theoretical expressions for the buckling threshold to infer propulsive force, given the measured length and stiffnesses. They find nearly identical values for both species, f ∼ (1.0 ± 0.6) nN/µm, nearly independent of the velocity.

Finally, they measure the shape of the filament dynamically to infer friction coefficients via Kirchhoff theory. This last part seems a bit inconsistent with the previous inference of propulsive force. Before, they assumed the same propulsive force for all bacteria and showed only a very weak correlation between buckling and propulsive velocity. In this section, they report a strong correlation with velocity, and report propulsive forces that vary over two orders of magnitude. I might be misunderstanding something, but I think this discrepancy should have been discussed or explained.

We regret the misunderstanding of the reviewer regarding the velocity dependence, which indicates that the manuscript should be improved to convey these relations correctly.

First, in the Buckling Measurements section, we did not assume the same propulsion force for all bacteria. The logistic regression yields an ensemble median for Lc (and thus an ensemble median for f ), along with the width ∆Lc of the distribution (and thus also the width of the distribution of f ). Our result f ∼ (1.0 ± 0.6) nN/µm indicates the median and the width of the distribution of the propulsion force densities across the ensemble of several hundred filaments used in the buckling measurements. The large variability of the forces found in the second part is consistently reflected by this very wide distribution of active forces detected in the logistic regression in the first part.

We did small modifications to the buckling theory paragraph to clarify that in the first part, a distribution of forces rather than a constant value is inferred (page 6)

“Inserting the population median and quartiles of the distributions of bending modulus and critical length, we can now quantify the distribution of the active force density for the filaments in the ensemble from the buckling measurements. We obtain nearly identical values for both species, f ∼ (1.0±0.6) nN/µm, where the uncertainty represents a wide distribution of f across the ensemble rather than a measurement error.”

The same holds, of course, when inferring the distribution of the friction coefficients (page 5):

“The substrate contact requires lubrication from polysaccharide slime to enable bacteria to glide (Khayatan et al., 2015 ). Thus we assume an over- damped motion with co-linear friction, for which the propulsion force f and the free gliding velocity v0 of a filament are related by f = η v0, with a friction coefficient η. In this scenario, f can be inferred both from the observed Lc ∼ (f/B)−1/3 and, up to the proportionality coefficient η, from the observed free gliding velocity. Thus, by combining the two relations, one may expect also a strong correlation between Lc and v0. In order to test this relation for consistency with our data, we include v0 as a second regressor, by setting x = (L−Lc(v0))/∆Lc in Equation 1, with Lc(v0) = (η v0/(30.5722 B))−1/3, to reflect our expectation from theory (see below). Now, η rather than f is the only unknown, and its ensemble distribution will be determined in the regression. Figure 3 E,F show the buckling behavior…”

The (naturally) wide distribution of force (and friction) leads to a distribution of Lc as well. However, due to the small exponent of 1/3 in the buckling threshold Lc ∼ f 1/3, the distribution of Lc is not as wide as the distributions of the individually inferred f or η. This is visualized in panel G of Figure 3, plotting Lc as a function of v0 (v0 is equivalent to f , up to a proportionality coefficient η). The natural length distribution, in contrast, is very wide. Therefore, the buckling propensity of a filament is most strongly characterized by its length, while force variability, which alters Lc of the individual, plays a secondary role.

In order to clarify this, we edited the last paragraph of the Buckling Measurements section on page 5 of the manuscript:

“…Within the characteristic range of observed velocities (1 − 3 µm/s), the median Lc depends only mildly on v0, as compared to its rather broad distribution, indicated by the bands in Figure 3 G. Thus a possible correlation between f and v0 would only mildly alter Lc. The natural length distribution (cf. Appendix 1—figure 1 ), however, is very broad, and we conclude that growth rather than velocity or force distributions most strongly impacts the buckling propensity of cyanobacterial colonies. Also, we hardly observed short and fast filaments of K. animale, which might be caused by physiological limitations (Burkholder, 1934 ).”

Second, in the Profile analysis section, we did not report a correlation between force and velocity. As can be seen in Figure 4—figure Supplement 1, neither the active force nor the friction coefficient, as determined from the analysis of individual filaments, show any significant correlation with the velocity. This is also written in the discussion (page 7):

We see no significant correlation between L or v0 and f or η, but the observed values of f and η cover a wide range (Figure 4 B, C and Figure 4—figure Supplement 1 ).

Note that this is indeed consistent with the logistic regression: Using v0 as a second regressor did not significantly reduce the width of the distribution of Lc as compared to the simple logistic regression, indicating that force and velocity are not strongly correlated.

In order to clarify this in the manuscript, we modified that part (page 7):

“…We see no significant correlation between L or v0 and f or η, but the observed values of f and η cover a wide range (Figure 4 B,C and Figure 4— figure Supplement 1 ). This is consistent with the logistic regression, where using v0 as a second regressor did not significantly reduce the width of the distribution of critical lengths or active forces. The two estimates of the friction coefficient, from logistic regression and individual profile fits, are measured in (predominantly) orthogonal directions: tangentially for the logistic regression where the free gliding velocity was used, and transversely for the evolution of the buckling profiles. Thus we plot f/v over η in Figure 4 D, finding nearly identical values over about two decades. Since f and η are not correlated with v0, this is due to a correlation between f and η. This relation is remarkable in two aspects: On the one hand, it indicates that friction is mainly isotropic…”

From a theoretical perspective, not many new results are presented. The authors repeat the well-known calculation for filaments buckling under propulsive load and arrive at the literature result of buckling when the dimensionless number (f L3/B) is larger than 30.6 as previously derived by Sekimoto et al in 1995 [1] (see [2] for a clamped boundary condition and simulations). Other theoretical predictions for pushed semi-flexible filaments [1–4] are not discussed or compared with the experiments. Finally, the Authors use molecular dynamics type simulations similar to [2–4] to reproduce the buckling dynamics from the experiments. Unfortunately, no systematic comparison is performed.

[1]        Ken Sekimoto, Naoki Mori, Katsuhisa Tawada, and Yoko Y Toyoshima. Symmetry breaking instabilities of an in vitro biological system. Physical review letters, 75(1):172, 1995.

[2]       Raghunath Chelakkot, Arvind Gopinath, Lakshminarayanan Mahadevan, and Michael F Hagan. Flagellar dynamics of a connected chain of active, polar, brownian particles. Journal of The Royal Society Interface, 11(92):20130884, 2014.

[3]       Rolf E Isele-Holder, Jens Elgeti, and Gerhard Gompper. Self-propelled worm-like filaments: spontaneous spiral formation, structure, and dynamics. Soft matter, 11(36):7181–7190, 2015.

[4]       Rolf E Isele-Holder, Julia J¨ager, Guglielmo Saggiorato, Jens Elgeti, and Gerhard Gompper. Dynamics of self-propelled filaments pushing a load. Soft Matter, 12(41):8495–8505, 2016.

We thank the reviewer for pointing us to these publications, in particular the work by Sekimoto we were not aware of. We agree with the referee that the calculation is straight forward (basically known since Euler, up to modified boundary conditions). Our paper focuses on experimental work, the molecular dynamics simulations were included mainly as a consistency check and not intended to generate the beautiful post-buckling patterns observed in references [2-4]. However, such shapes do emerge in filamentous cyanobacteria, and with the data provided in our manuscript, simulations can be quantitatively matched to our experiments, which will be covered by future work.

We included the references in the revision of our manuscript, and a statement that we do not claim priority on these classical theoretical results.

Introduction, page 2:

“…Self-Buckling is an important instability for self-propelling rod-like micro-organisms to change the orientation of their motion, enabling aggregation or the escape from traps (Fily et al., 2020; Man and Kanso, 2019; Isele-Holder et al., 2015; Isele-Holder et al., 2016 ). The notion of self-buckling goes back to work of Leonhard Euler in 1780, who described elastic columns subject to gravity (Elishakoff, 2000 ). Here, the principle is adapted to the self-propelling, flexible filaments (Fily et al., 2020; Man and Kanso, 2019; Sekimoto et al., 1995 ) that glide onto an obstacle. Filaments buckle if they exceed a certain critical length Lc ∼ (B/f)1/3, where B is the bending modulus and f the propulsion force density…”

Buckling theory, page 5:

“…The buckling of gliding filaments differs in two aspects: the propulsion forces are oriented tangentially instead of vertically, and the front end is supported instead of clamped. Therefore, with L < Lc all initial orientations are indifferently stable, while for L > Lc, buckling induces curvature and a resultant torque on the head, leading to rotation (Fily et al., 2020; Chelakkot et al., 2014; Sekimoto et al., 1995 ). Buckling under concentrated tangential end-loads has also been investigated in literature (de Canio et al., 2017; Wolgemuth et al., 2005 ), but leads to substantially different shapes of buckled filaments. We use classical Kirchhoff theory for a uniform beam of length L and bending modulus B, subject to a force density ⃗b = −f ⃗t − η ⃗v, with an effective active force density f along the tangent ⃗t, and an effective friction proportional to the local velocity ⃗v, analog to existing literature (Fily et al., 2020; Chelakkot et al., 2014; Sekimoto et al., 1995 )…”

Further on page 6:

“To derive the critical self-buckling length, Equation 5 can be linearized for two scenarios that lead to the same Lc: early-time small amplitude buckling and late-time stationary rotation at small and constant curvature (Fily et al., 2020; Chelakkot et al., 2014 ; Sekimoto et al., 1995 ). […] Thus, in physical units, the critical length is given by Lc = (30.5722 B/f)1/3, which is reproduced in particle based simulations (Appendix Figure 2 ) analogous to those in Isele-Holder et al. (2015, 2016).”

Discussion, page 7 & 8:

“…This, in turn, has dramatic consequences on the exploration behavior and the emerging patterns (Isele-Holder et al., 2015, 2016; Abbaspour et al., 2021; Duman et al., 2018; Prathyusha et al., 2018; Jung et al., 2020 ): (L/Lc)3 is, up to a numerical prefactor, identical to the flexure number (Isele-Holder et al., 2015, 2016; Duman et al., 2018; Winkler et al., 2017 ), the ratio of the Peclet number and the persistence length of active polymer melts. Thus, the ample variety of non-equilibrium phases in such materials (Isele-Holder et al., 2015, 2016; Prathyusha et al., 2018; Abbaspour et al., 2021 ) may well have contributed to the evolutionary success of filamentous cyanobacteria.”

Reviewer 3:

Summary:

This paper presents novel and innovative force measurements of the biophysics of gliding cyanobacteria filaments. These measurements allow for estimates of the resistive force between the cell and substrate and provide potential insight into the motility mechanism of these cells, which remains unknown.

We thank the reviewer for the positive evaluation of our work. We have revised the manuscript according to their comments and detail our replies and modifications next to the individual points below.

Strengths:

The authors used well-designed microfabricated devices to measure the bending modulus of these cells and to determine the critical length at which the cells buckle. I especially appreciated the way the authors constructed an array of pillars and used it to do 3-point bending measurements and the arrangement the authors used to direct cells into a V-shaped corner in order to examine at what length the cells buckled at. By examining the gliding speed of the cells before buckling events, the authors were able to determine how strongly the buckling length depends on the gliding speed, which could be an indicator of how the force exerted by the cells depends on cell length; however, the authors did not comment on this directly.

We thank the referee for the positive assessment of our work. Importantly, we do not see a significant correlation between buckling length and gliding speeds, and we also do not see a correlation with filament length, consistent with the assumption of a propulsion force density that is more or less homogeneously distributed along the filament. Note that each filament consists of many metabolically independent cells, which renders cyanobacterial gliding a collective effort of many cells, in contrast to gliding of, e.g., myxobacteria.

In response also to the other referees’ comments, we modified the manuscript to reflect more on the absence of a strong correlation between velocity and force/critical length. We modified the Buckling measurements section on page 5 of the paper:

“The substrate contact requires lubrication from polysaccharide slime to enable bacteria to glide (Khayatan et al., 2015 ). Thus we assume an over-damped motion with co-linear friction, for which the propulsion force f and the free gliding velocity v0 of a filament are related by f = η v0, with a friction coefficient η. In this scenario, f can be inferred both from the observed Lc ∼ (f/B)−1/3 and, up to the proportionality coefficient η, from the observed free gliding velocity. Thus, by combining the two relations, one may expect also a strong correlation between Lc and v0. In order to test this relation for consistency with our data, we include v0 as a second regressor, by setting x = (L−Lc(v0))/∆Lc in Equation 1, with Lc(v0) = (η v0/(30.5722 B))−1/3, to reflect our expectation from theory (see below). Now, η rather than f is the only unknown, and its ensemble distribution will be determined in the regression. Figure 3 E, F show the buckling behavior…”

Further, we edited the last paragraph of the Buckling measurements section on page 5 of the manuscript:

“Within the characteristic range of observed velocities (1 − 3 µm/s), the median Lc depends only mildly on v0, as compared to its rather broad distribution, indicated by the bands in Figure 3 G. Thus a possible correlation between f and v0 would only mildly alter Lc. The natural length distribution (cf. Appendix 1—figure 1 ), however, is very broad, and we conclude that growth rather than velocity or force distributions most strongly impacts the buckling propensity of cyanobacterial colonies. Also, we hardly observed short and fast filaments of K. animale, which might be caused by physiological limitations (Burkholder, 1934 ).”

We also rephrased the corresponding discussion paragraph on page 7:

“…Thus we plot f/v over η in Figure 4 D, finding nearly identical values over about two decades. Since f and η are not correlated with v0, this is due to a correlation between f and η. This relation is remarkable in two aspects: On the one hand, it indicates that friction is mainly isotropic. This suggests that friction is governed by an isotropic process like bond friction or lubrication from the slime layer in the contact with the substrate, the latter being consistent with the observation that mutations deficient of slime secretion do not glide but exogenous addition of slime restores motility (Khayatan et al., 2015 ). In contrast, hydrodynamic drag from the surrounding bulk fluid (Man and Kanso, 2019 ), or the internal friction of the gliding apparatus would be expected to generate strongly anisotropic friction. If the latter was dominant, a snapping-like transition into the buckling state would be expected, rather than the continuously growing amplitude that is observed in experiments. On the other hand, it indicates that friction and propulsion forces…”

Weaknesses:

There were two minor weaknesses in the paper.

First, the authors investigate the buckling of these gliding cells using an Euler beam model. A similar mathematical analysis was used to estimate the bending modulus and gliding force for Myxobacteria (C.W. Wolgemuth, Biophys. J. 89: 945-950 (2005)). A similar mathematical model was also examined in G. De Canio, E. Lauga, and R.E Goldstein, J. Roy. Soc. Interface, 14: 20170491 (2017). The authors should have cited these previous works and pointed out any differences between what they did and what was done before.

We thank the reviewer for pointing us to these references. The paper by Wolgemuth is theoretical work, describing A-motility in myxobacteria by a concentrated propulsion force at the rear end of the bacterium, possibly stemming from slime extrusion. This model was a little later refuted by [A3], who demonstrated that focal adhesion along the bacterial body and thus a distributed force powers A-motility, a mechanism that has by now been investigated in great detail (see [A10]). The paper by Canio et al. contains a thorough theoretical analysis of a filament that is clamped at one end and subject to a concentrated tangential load on the other. Since both models comprise a concentrated end-load rather than a distributed propulsion force density, they describe a substantially different motility mechanism, leading also to substantially different buckling profiles. Consequentially, these models cannot be applied to cyanobacterial gliding.

We included both citations in the revision and pointed out the differences to our work in the introduction (page 2):

“…A few species appear to employ a type-IV-pilus related mechanism (Khayatan et al., 2015; Wilde and Mullineaux, 2015 ), similar to the better- studied myxobacteria (Godwin et al., 1989; Mignot et al., 2007; Nan et al., 2014; Copenhagen et al., 2021; Godwin et al., 1989 ), which are short, rod-shaped single cells that exhibit two types of motility: S (social) motility based on pilus extension and retraction, and A (adventurous) motility based on focal adhesion (Chen and Nan, 2022 ) for which also slime extrusion at the trailing cell pole was earlier postulated as mechanism (Wolgemuth et al., 2005 ). Yet, most gliding filamentous cyanobacteria do not exhibit pili and their gliding mechanism appears to be distinct from myxobacteria (Khayatan et al., 2015 ).”

And in Buckling theory, page 5:

“….The buckling of gliding filaments differs in two aspects: the propulsion forces are oriented tangentially instead of vertically, and the front end is supported instead of clamped. Therefore, with L < Lc all initial orientations are indifferently stable, while for L > Lc, buckling induces curvature and a resultant torque on the head, leading to rotation (Fily et al., 2020; Chelakkot et al., 2014; Sekimoto et al., 1995 ). Buckling under concentrated tangential end-loads has also been investigated in literature (de Canio et al., 2017; Wolgemuth et al., 2005 ), but leads to substantially different shapes of buckled filaments.”

The second weakness is that the authors claim that their results favor a focal adhesion-based mechanism for cyanobacterial gliding motility. This is based on their result that friction and adhesion forces correlate strongly. They then conjecture that this is due to more intimate contact with the surface, with more contacts producing more force and pulling the filaments closer to the substrate, which produces more friction. They then claim that a slime-extrusion mechanism would necessarily involve more force and lower friction. Is it necessarily true that this latter statement is correct? (I admit that it could be, but is it a requirement?)

We thank the referee for raising this interesting question. Our claim regarding slime extrusion is based on three facts: i. mutations deficient of slime extrusion do not glide, but start gliding as soon as slime is provided externally [A4]. ii. A positive correlation between speed and slime layer thickness was observed in Nostoc [A11]. iii. The fluid mechanics of lubricated sliding contacts is very well understood and predicts a decreasing resistance with increasing layer thickness.

We included these considerations in the revision of our manuscript (page 8):

“…it indicates that friction and propulsion forces, despite being quite variable, correlate strongly. Thus, generating more force comes, inevitably, at the expense of added friction. For lubricated contacts, the friction coefficient is proportional to the thickness of the lubricating layer (Snoeijer et al., 2013 ), and we conjecture active force and drag both increase due to a more intimate contact with the substrate. This supports mechanisms like focal adhesion (Mignot et al., 2007 ) or a modified type-IV pilus (Khayatan et al., 2015 ), which generate forces through contact with extracellular surfaces, as the underlying mechanism of the gliding apparatus of filamentous cyanobacteria: more contacts generate more force, but also closer contact with the substrate, thereby increasing friction to the same extent. Force generation by slime extrusion (Hoiczyk and Baumeister, 1998 ), in contrast, would lead to the opposite behavior: More slime generates more propulsion, but also reduces friction. Besides fundamental fluid-mechanical considerations (Snoeijer et al., 2013 ), this is rationalized by two experimental observations: i. gliding velocity correlates positively with slime layer thickness (Dhahri et al., 2013 ) and ii. motility in slime-secretion deficient mutants is restored upon exogenous addition of polysaccharide slime. Still we emphasize that many other possibilities exist. One could, for instance, postulate a regulation of the generated forces to the experienced friction, to maintain some preferred or saturated velocity.”

Related to this, the authors use a model with isotropic friction. They claim that this is justified because they are able to fit the cell shapes well with this assumption. How would assuming a non-isotropic drag coefficient affect the shapes? It may be that it does equally well, in which case, the quality of the fits would not be informative about whether or not the drag was isotropic or not.

The referee raises another very interesting point. Given the typical variability and uncertainty in experimental measurements (cf. error Figure 4 A), a model with a sightly anisotropic friction could be fitted to the observed buckling profiles as well, without significant increase of the mismatch. Yet, strongly anisotropic friction would not be consistent with our observations.

Importantly, however, we did not conclude on isotropic friction based on the fit quality, but based on a comparison between free gliding and early buckling (Figure 4 D). In early buckling, the dominant motion is in transverse direction, while longitudinal motion is insignificant, due to geometric reasons. Thus, independent of the underlying model, mostly the transverse friction coefficiont is inferred. In contrast, free gliding is a purely longitudinal motion, and thus only the friction coefficient for longitudinal motion can be inferred. These two friction coefficients are compared in Figure 4 D. Still, the scatter of that data would allow to fit a certain anisotropy within the error margins. What we can exclude based on out observation is the case of a strongly anisotropic friction. If there is no ab-initio reason for anisotropy, nor a measurement that indicates it, we prefer to stick with the simplest

assumption. We carefully chose our wording in the Discussion as “mainly isotropic” rather

than “isotropic” or “fully isotropic”.

We added a small statement to the Discussion on page 7 & 8:

“... Thus we plot f/v over η in Figure 4 D, finding nearly identical values over about two decades. Since f and η are not correlated with v0, this is due to a correlation between f and η. This relation is remarkable in two aspects: On the one hand, it indicates that friction is mainly isotropic. This suggests that friction is governed by an isotropic process like bond friction or lubrication from the slime layer in the contact with the substrate, the latter being consistent with the observation that mutations deficient of slime secretion do not glide but exogenous addition of slime restores motility (Khayatan et al., 2015 ). In contrast, hydrodynamic drag from the surrounding bulk fluid (Man and Kanso, 2019 ), or the internal friction of the gliding apparatus would be expected to generate strongly anisotropic friction. If the latter was dominant, a snapping-like transition into the buckling state would be expected, rather than the continuously growing amplitude that is observed in experiments. On the other hand, it indicates that friction and propulsion forces ...”

Recommendations for the authors

The discussion regarding how the findings of this paper imply that cyanobacteria filaments are propelled by adhesion forces rather than slime extrusion should be improved, as this conclusion seems questionable. There appears to be an inconsistency with a buckling force said to be only weakly dependent on the gliding velocity, while its ratio with the velocity correlates with a friction coefficient. Finally, data and source code should be made publicly available.

In the revised version, we have modified the discussion of the force generating mechanism according to the reviewer suggestions. The perception of inconsistency in the velocity dependence of the buckling force was based on a misunderstanding, as we detailed in our reply to the referee. We revised the corresponding section to make it more clear. Data and source code have been uploaded to a public data repository.

Reviewer #2 (recommendations for the authors)

Despite eLife policy, the authors do not provide a Data Availability Statement. For the presented manuscript, data and source code should be provided “via trusted institutional or third-party repositories that adhere to policies that make data discoverable, accessible and usable.” https://elifesciences.org/inside-elife/51839f0a/for-authors-updates- to-elife-s-data-sharing-policies

Most of the issues in this reviewer’s public review should be easy to correct, so I would strongly support the authors to provide an amended manuscript.

We added the Data Availability Statement in the amended manuscript.

References

[A1] E. Hoiczyk and W. Baumeister. “The junctional pore complex, a prokaryotic secretion organelle, is the molecular motor underlying gliding motility in cyanobacteria”. In: Curr. Biol. 8.21 (1998), pp. 1161–1168. doi: 10.1016/s0960-9822(07)00487-3.

[A2] N. Read, S. Connell, and D. G. Adams. “Nanoscale Visualization of a Fibrillar Array in the Cell Wall of Filamentous Cyanobacteria and Its Implications for Gliding Motility”. In: J. Bacteriol. 189.20 (2007), pp. 7361–7366. doi: 10.1128/jb.00706- 07.

[A3] T. Mignot, J. W. Shaevitz, P. L. Hartzell, and D. R. Zusman. “Evidence That Focal Adhesion Complexes Power Bacterial Gliding Motility”. In: Science 315.5813 (2007), pp. 853–856. doi: 10.1126/science.1137223.

[A4] Behzad Khayatan, John C. Meeks, and Douglas D. Risser. “Evidence that a modified type IV pilus-like system powers gliding motility and polysaccharide secretion in filamentous cyanobacteria”. In: Mol. Microbiol. 98.6 (2015), pp. 1021–1036. doi: 10.1111/mmi.13205.

[A5] Tilo Pompe, Martin Kaufmann, Maria Kasimir, Stephanie Johne, Stefan Glorius, Lars Renner, Manfred Bobeth, Wolfgang Pompe, and Carsten Werner. “Friction- controlled traction force in cell adhesion”. In: Biophysical journal 101.8 (2011), pp. 1863–1870.

[A6] Hirofumi Wada, Daisuke Nakane, and Hsuan-Yi Chen. “Bidirectional bacterial gliding motility powered by the collective transport of cell surface proteins”. In: Physical Review Letters 111.24 (2013), p. 248102.

[A7] Jo¨el Tchoufag, Pushpita Ghosh, Connor B Pogue, Beiyan Nan, and Kranthi K Mandadapu. “Mechanisms for bacterial gliding motility on soft substrates”. In: Proceedings of the National Academy of Sciences 116.50 (2019), pp. 25087–25096.

[A8] Chenyi Fei, Sheng Mao, Jing Yan, Ricard Alert, Howard A Stone, Bonnie L Bassler, Ned S Wingreen, and Andrej Kosmrlj. “Nonuniform growth and surface friction determine bacterial biofilm morphology on soft substrates”. In: Proceedings of the National Academy of Sciences 117.14 (2020), pp. 7622–7632.

[A9] Arja Ray, Oscar Lee, Zaw Win, Rachel M Edwards, Patrick W Alford, Deok-Ho Kim, and Paolo P Provenzano. “Anisotropic forces from spatially constrained focal adhesions mediate contact guidance directed cell migration”. In: Nature communications 8.1 (2017), p. 14923.

[A10] Jing Chen and Beiyan Nan. “Flagellar motor transformed: biophysical perspectives of the Myxococcus xanthus gliding mechanism”. In: Frontiers in Microbiology 13 (2022), p. 891694.

[A11] Samia Dhahri, Michel Ramonda, and Christian Marliere. “In-situ determination of the mechanical properties of gliding or non-motile bacteria by atomic force microscopy under physiological conditions without immobilization”. In: PLoS One 8.4 (2013), e61663.

Reviewer #2 (Public Review):

Pyoverdines, siderophores produced by many Pseudomonads, are one of the most diverse groups of specialized metabolites and are frequently used as model systems. Thousands of Pseudomonas genomes are available, but large-scale analyses of pyoverdines are hampered by the biosynthetic gene clusters (BGCs) being spread across multiple genomic loci and existing tools' inability to accurately predict amino acid substrates of the biosynthetic adenylation (A) domains. The authors present a bioinformatics pipeline that identifies pyoverdine BGCs and predicts the A domain substrates with high accuracy. They tackled a second challenging problem by developing an algorithm to differentiate between outer membrane receptor selectivity for pyoverdines versus other siderophores and substrates. The authors applied their dataset to thousands of Pseudomonas strains, producing the first comprehensive overview of pyoverdines and their receptors and predicting many new structural variants.

The A domain substrate prediction is impressive, including the correction of entries in the MIBiG database. Their high accuracy came from a relatively small training dataset of A domains from 13 pyoverdine BGCs. The authors acknowledge that this small dataset does not include all substrates, and correctly point out that new sequence/structure pairs can be added to the training set to refine the prediction algorithm. The authors could have been more comprehensive in finding their training set data. For instance, the authors claim that histidine "had not been previously documented in pyoverdines", but the sequenced strain P. entomophila L48, incorporates His (10.1007/s10534-009-9247-y). The workflow cannot differentiate between different variants of Asp and OHOrn, and it's not clear if this is a limitation of the workflow, the training data, or both. The prediction workflow holds up well in Burkholderiales A domains, however, they fail to mention in the main text that they achieved these numbers by adding more A domains to their training set.

To validate their predictions, they elucidated structures of several new pyoverdines, and their predictions performed well. However, the authors did not include their MS/MS data, making it impossible to validate their structures. In general, the biggest limitation of the submitted manuscript is the near-empty methods section, which does not include any experimental details for the 20 strains or details of the annotation pipeline (such as "Phydist" and "Syndist"). The source code also does not contain the requisite information to replicate the results or re-use the pipeline, such as the antiSMASH version and required flags. That said, skimming through the source code and data (kindly provided upon request) suggests that the workflow itself is sound and a clear improvement over existing tools for pyoverdine BGC annotation.

Predicting outer membrane receptor specificity is likewise a challenging problem and the authors have made a promising achievement by finding specific gene regions that differentiate the pyoverdine receptor FpvA from FpvB and other receptor families. Their predictions were not tested experimentally, but the finding that only predicted FpvA receptors were proximate to the biosynthesis genes lends credence to the predictive power of the workflow. The authors find predicted pyoverdine receptors across an impressive 468 genera, an exciting finding for expanding the role of pyoverdines as public goods beyond Pseudomonas. However, whether or not these receptors can recognize pyoverdines (and if so, which structures!) remains to be investigated.

In all, the authors have assembled a rich dataset that will enable large-scale comparative genomic analyses. This dataset could be used by a variety of researchers, including those studying natural product evolution, public good eco/evo dynamics, and NRPS engineering.

Summary of the Talk on the Future of Web Frameworks by Ryan Carniado

• Introduction and Background:

  • Ryan Carniado, creator of SolidJS, has extensive experience in web development spanning 25 years, having worked with various technologies including ASP.NET, Rails, and jQuery.
  • SolidJS was started in 2016 and reflects a shift towards new paradigms in web frameworks, particularly in the front-end JavaScript ecosystem.
  • Quote: "I've been doing web development now for like 25 years... it wasn't really until the 2010s that my passion reignited for front-end JavaScript."

• Core Themes and Concepts:

  • Modern front-end development heavily relies on components (e.g., class components, function components, web components) which serve as fundamental building blocks for creating modular and composable applications.
  • Components have runtime implications due to their update models and life cycles, influencing the performance and design of web applications.
  • Traditional component models use either a top-down diffing approach (like virtual DOM) or rely on compilation optimizations to enhance performance.
  • Quote: "Modern front-end development for years has been about components... however, in almost every JavaScript framework components have runtime implications."

• Reactive Programming and Fine-Grained Reactivity:

  • Ryan advocates for a shift towards reactive programming to manage state changes more efficiently. This approach is likened to how spreadsheets work, where changes in input immediately affect outputs without re-execution of all logic.
  • Fine-grained reactivity involves three primitives: signals (atomic atoms), derived state (computeds or memos), and side effects (effects). These primitives help manage state and side effects without heavy reliance on the component architecture or compilation.
  • Quote: "What if the relationship held instead? What if whenever we changed B and C, A also immediately updated? That's basically what reactive programming is."

• Practical Demonstration and Code Examples:

  • Ryan demonstrated the implementation of fine-grained reactivity using SolidJS, showing how state management and updates can be handled more efficiently compared to traditional methods that rely heavily on component re-renders and hooks.
  • The examples provided emphasized how reactive programming can simplify state management and improve performance by only updating components that need to change, reducing unnecessary re-renders.
  • Quote: "The problem is that if any state in this component changes, the whole thing reruns again... what if we didn't? What if components didn't dictate the boundary of our performance?"

• Performance Implications and Advantages:

  • The "reactive advantage" in SolidJS and similar frameworks lies in their ability to run components minimally, avoiding stale closures and excessive dependencies that can degrade performance.
  • Ryan highlighted that in reactive frameworks, component boundaries do not dictate performance; instead, performance optimization is achieved through smarter state management and reactive updates.
  • Quote: "Components run once... state is independent of components. Component boundaries are for your sake, how you want to organize your code, not for performance."

• Future Directions and Framework Evolution:

  • The talk touched on the broader impact of reactive programming and fine-grained reactivity on the evolution of web frameworks. This includes the potential integration with AI and compilers to further optimize performance and developer experience.
  • Ryan suggested that the future of web development might see more frameworks adopting similar reactive principles, possibly leading to a "reactive renaissance" in the industry.
  • Quote: "A revolution is not in the cards, maybe just a reactive Renaissance."

• Q&A and Additional Insights:

  • During the Q&A, Ryan discussed the potential application of SolidJS principles in environments like React Native and native code development, indicating the flexibility and broad applicability of reactive programming principles across different platforms and technologies.
  • Quote: "The custom renderer and stuff is not something you need a virtual DOM to... the reactive tree as it turns out is completely independent."

Summary of Raph Levien's Blog: "Towards principled reactive UI"

Introduction

• Diversity of Reactive UI Systems: The blog notes the diversity in reactive UI systems primarily sourced from open-source projects. Levien highlights a lack of comprehensive literature but acknowledges existing sources offer insights into better practices. His previous post aimed to organize these diverse patterns.
  • "There is an astonishing diversity of 'literature' on reactive UI systems."

Goals of the Inquiry

• Clarifying Inquiry Goals: Levien sets goals not to review but to guide inquiry into promising avenues of reactive UI in Rust, likening it to mining for rich veins of ore rather than stamp collecting.
  • "I want to do mining, not stamp collecting."

Main Principles Explored

• Observable Objects vs. Future-like Polling: Discusses the importance of how systems manage observable objects or utilize future-like polling for efficient UI updates.
• Tree Mutations: How to express mutation in the render object tree is crucial, focusing on maintaining stable node identities within the tree.
  • "Then I will go into deeper into three principles, which I feel are critically important in any reactive UI framework."

Crochet: A Research Prototype

• Introduction of Crochet: Introduces 'Crochet', a prototype exploring these principles, acknowledging its current limitations and potential for development.
  • "Finally, I will introduce Crochet, a research prototype built for the purpose of exploring these ideas."

Goals for Reactive UI

• Concise Application Logic: Emphasizes the need for concise, clear application logic that drives UI efficiently, with reactive UI allowing declarative state expressions of the view tree.
  • "The main point of a reactive UI architecture is so that the app can express its logic clearly and concisely."
• Incremental Updates: Advocates for incremental updates in UI rendering to avoid performance issues related to full re-renders, highlighting the limitations of systems like imgui and the potential of systems like Conrod, despite its shortcomings.
  • "While imgui can express UI concisely, it cheats somewhat by not being incremental."

Evaluation of Existing Systems

• Comparison with Other Systems: Mentions SwiftUI, imgui, React, and Svelte, discussing their approaches to handling reactive UI and their adaptability to Rust.
  • "SwiftUI has gained considerable attention due to its excellent ergonomics in this regard."

Technical Challenges and Proposals

• Challenges in Tree Mutation and Stable Identity: Discusses the challenges in tree mutation techniques and the importance of stable identity in UI components to preserve user interaction states.
  • "Mutation of the DOM is expressed through a well-specified and reasonably ergonomic, if inefficient, interface."

Conclusion and Future Work

• Future Directions and Experiments: Encourages experimentation with the Crochet prototype and discusses the ongoing development and research in making reactive UIs more efficient and user-friendly.
  • "I encourage people to experiment with the Crochet code."

This blog post encapsulates Levien's ongoing exploration into developing a principled approach to reactive UI in Rust, highlighting the complexity of the task and his experimental prototype, Crochet, as a step towards solving these challenges.

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Strengths:

This work (almost didactically) demonstrates how to develop, calibrate, validate and analyze a comprehensive, spatially resolved, dynamical, multicellular model. Testable model predictions of (also non-monotonic) emergent behaviors are derived and discussed. The computational model is based on a widely-used simulation platform and shared openly such that it can be further analyzed and refined by the community.

Weaknesses:

While the parameter estimation approach is sophisticated, this work does not address issues of structural and practical non-identifiability (Wieland et al., 2021, DOI:10.1016/j.coisb.2021.03.005) of parameter values, given just tissue-scale summary statistics, and does not address how model predictions might change if alternative parameter combinations were used. Here, the calibrated model represents one point estimate (column "Value" in Suppl. Table 1) but there is specific uncertainty of each individual parameter value and such uncertainties need to be propagated (which is computationally expensive) to the model predictions for treatment scenarios.

We thank the reviewer for the excellent suggestions and observations. The CaliPro parameterization technique applied puts an emphasis on finding a robust parameter space instead of a global optimum. To address structural non-identifiability, we utilized partial rank correlation coefficient with each iteration of the calibration process to ensure that the sensitivity of each parameter was relevant to model outputs. We also found that there were ranges of parameter values that would achieve passing criteria but when testing the ranges in replicate resulted in inconsistent outcomes. This led us to further narrow the parameters into a single parameter set that still had stochastic variability but did not have such large variability between replicate runs that it would be unreliable. Additional discussion on this point has been added to lines 623-628. We acknowledge that there are likely other parameter sets or model rules that would produce similar outcomes but the main purpose of the model was to utilize it to better understand the system and make new predictions, which our calibration scheme allowed us to accomplish.

Regarding practical non-identifiability, we acknowledge that there are some behaviors that are not captured in the model because those behaviors were not specifically captured in the calibration data. To ensure that the behaviors necessary to answer the aims of our paper were included, we used multiple different datasets and calibrated with multiple different output metrics. We believe we have identified the appropriate parameters to recapitulate the dominating mechanisms underlying muscle regeneration. We have added additional discussion on practical non-identifiability to lines 621-623.

Suggested treatments (e.g. lines 484-486) are modeled as parameter changes of the endogenous cytokines (corresponding to genetic mutations!) whereas the administration of modified cytokines with changed parameter values would require a duplication of model components and interactions in the model such that cells interact with the superposition of endogenous and administered cytokine fields. Specifically, as the authors also aim at 'injections of exogenously delivered cytokines' (lines 578, 579) and propose altering decay rates or diffusion coefficients (Fig. 7), there needs to be a duplication of variables in the model to account for the coexistence of cytokine subtypes. One set of equations would have unaltered (endogenous) and another one have altered (exogenous or drugged) parameter values. Cells would interact with both of them.

Our perturbations did not include delivery of exogenously delivered cytokines and instead were focused on microenvironmental changes in cytokine diffusion and decay rates or specific cytokine concentration levels. For example, the purpose of the VEGF delivery perturbation was to test how an increase in VEGF concentrations would alter regeneration outcome metrics with the assumption that the delivered VEGF would act in the same manner as the endogenous VEGF. We have clarified the purpose of the simulations on line 410. We agree that exploring if model predictions would be altered if endogenous and exogenous were represented separately; however, we did not explore this type of scenario.

This work shows interesting emergent behavior from nonlinear cytokine interactions but the analysis does not provide insights into the underlying causes, e.g. which of the feedback loops dominates early versus late during a time course.

Indeed, analyzing the model to fully understand the time-varying interactions between the multiple feedback loops is a challenge in and of itself, and we appreciate the opportunity to elaborate on our approach to addressing this challenge. First: the crosstalk/feedback between cytokines and the temporal nature was analyzed in the heatmap (Fig. 6) and lines 474-482. Second: the sensitivity of cytokine parameters to specific outputs was included in Table 9 and full-time course sensitivity is included in Supplemental Figure 2. Further correlation analysis was also included to demonstrate how cytokine concentrations influenced specific output metrics at various timepoints (Supplemental Fig. 3). We agree that further elaboration of these findings is required; therefore, we added lines 504-509 to discuss the specific mechanisms at play with the combined cytokine interactions. We also added more discussion (lines 637-638) regarding future work that could develop more analysis methods to further investigate the complex behaviors in the model.

Reviewer #2 (Public Review):

Strengths:

The manuscript identified relevant model parameters from a long list of biological studies. This collation of a large amount of literature into one framework has the potential to be very useful to other authors. The mathematical methods used for parameterization and validation are transparent.

Weaknesses:>

I have a few concerns which I believe need to be addressed fully.

My main concerns are the following:

(1) The model is compared to experimental data in multiple results figures. However, the actual experiments used in these figures are not described. To me as a reviewer, that makes it impossible to judge whether appropriate data was chosen, or whether the model is a suitable descriptor of the chosen experiments. Enough detail needs to be provided so that these judgements can be made.

Thank you for raising this point. We created a new table (Supplemental table 6) that describes the techniques used for each experimental measurement.

(2) Do I understand it correctly that all simulations are done using the same initial simulation geometry? Would it be possible to test the sensitivity of the paper results to this geometry? Perhaps another histological image could be chosen as the initial condition, or alternative initial conditions could be generated in silico? If changing initial conditions is an unreasonably large request, could the authors discuss this issue in the manuscript?

We appreciate your insightful question regarding the initial simulation geometry in our model. The initial configuration of the fibers/ECM/microvascular structures was kept consistent but the location of the necrosis was randomly placed for each simulation. Future work will include an in-depth analysis of altered histology configuration on model predictions which has been added to lines 618-621. We did a preliminary example analysis by inputting a different initial simulation geometry, which predicted similar regeneration outcomes. We have added Supplemental Figure 5 that provides the results of that example analysis.

(3) Cytokine knockdowns are simulated by 'adjusting the diffusion and decay parameters' (line 372). Is that the correct simulation of a knockdown? How are these knockdowns achieved experimentally? Wouldn't the correct implementation of a knockdown be that the production or secretion of the cytokine is reduced? I am not sure whether it's possible to design an experimental perturbation which affects both parameters.

We appreciate that this important question has been posed. Yes, in order to simulate the knockout conditions, the cytokine secretion was reduced/eliminated. The diffusion and decay parameters were also adjusted to ensure that the concentration within the system was reduced. Lines 391-394 were added to clarify this assumption.

(4) The premise of the model is to identify optimal treatment strategies for muscle injury (as per the first sentence of the abstract). I am a bit surprised that the implemented experimental perturbations don't seem to address this aim. In Figure 7 of the manuscript, cytokine alterations are explored which affect muscle recovery after injury. This is great, but I don't believe the chosen alterations can be done in experimental or clinical settings. Are there drugs that affect cytokine diffusion? If not, wouldn't it be better to select perturbations that are clinically or experimentally feasible for this analysis? A strength of the model is its versatility, so it seems counterintuitive to me to not use that versatility in a way that has practical relevance. - I may well misunderstand this though, maybe the investigated parameters are indeed possible drug targets.

Thank you for your thoughtful feedback. The first sentence (lines 32-34) of the abstract was revised to focus on beneficial microenvironmental conditions to best reflect the purpose of the model. The clinical relevance of the cytokine modifications is included in the discussion (lines 547-558) with additional information added to lines 524-526. For example, two methods to alter diffusion experimentally are: antibodies that bind directly to the cytokine to prevent it from binding to its receptor on the cell surface and plasmins that induce the release of bound cytokines.

(5) A similar comment applies to Figure 5 and 6: Should I think of these results as experimentally testable predictions? Are any of the results surprising or new, for example in the sense that one would not have expected other cytokines to be affected as described in Figure 6?

We appreciate the opportunity to clarify the basis for these perturbations. The perturbations included in Figure 5 were designed to mimic the conditions of a published experiment that delivered VEGF in vivo (Arsic et al. 2004, DOI:10.1016/J.YMTHE.2004.08.007). The perturbation input conditions and experimental results are included in Table 8 and Supplemental Table 6 has been added to include experimental data and method description of the perturbation. The results of this analysis provide both validation and new predictions, because some the outputs were measured in the experiments while others were not measured. The additional output metrics and timepoints that were not collected in the experiment allow for a deeper understanding of the dynamics and mechanisms leading to the changes in muscle recovery (lines 437-454). These model outputs can provide the basis for future experiments; for example, they highlight which time points would be more important to measure and even provide predicted effect sizes that could be the basis for a power analysis (lines 639-640).

Regarding Figure 6, the published experimental outcomes of cytokine KOs are included in Table 8. The model allowed comparison of different cytokine concentrations at various timepoints when other cytokines were removed from the system due to the KO condition. The experimental results did not provide data on the impact on other cytokine concentrations but by using the model we were able to predict temporally based feedback between cytokines (lines 474-482). These cytokine values could be collected experimentally but would be time consuming and expensive. The results of these perturbations revealed the complex nature of the relationship between cytokines and how removal of one cytokine from the system has a cascading temporal impact. Lines 533-534 have been added to incorporate this into the discussion.

(6) In figure 4, there were differences between the experiments and the model in two of the rows. Are these differences discussed anywhere in the manuscript?

We appreciate your keen observation and the opportunity to address these differences. The model did not match experimental results for CSA output in the TNF KO and antiinflammatory nanoparticle perturbation or TGF levels with the macrophage depletion. While it did align with the other experimental metrics from those studies, it is likely that there are other mechanisms at play in the experimental conditions that were not captured by simulating the downstream effects of the experimental perturbations. We have added discussion of the differences to lines 445-454.

(7) The variation between experimental results is much higher than the variation of results in the model. For example, in Figure 3 the error bars around experimental results are an order of magnitude larger than the simulated confidence interval. Do the authors have any insights into why the model is less variable than the experimental data? Does this have to do with the chosen initial condition, i.e. do you think that the experimental variability is due to variation in the geometries of the measured samples?

Thank you for your insightful observations and questions. The lower model variability is attributed to the larger sample size of model simulations compared to experimental subjects. By running 100 simulations it narrows in the confidence interval (average 2.4 and max 3.3) compared to the experiments that typically had a sample size of less than 15. If the number of simulations had been reduced to 15 the stochasticity within the model results in a larger confidence interval (average 7.1 and max 10). There are also several possible confounding variables in the experimental protocols (i.e. variations in injury, different animal subjects for each timepoint, etc.) that are kept constant in the model simulation. We have added discussion of this point to the manuscript (lines 517519). Future work with the model will examine how variations in conditions, such as initial muscle geometry, injury, etc, alter regeneration outcomes and overall variability. This discussion has been incorporated into lines 640-643.

(8) Is figure 2B described anywhere in the text? I could not find its description.

Thank you for pointing that out. We have added a reference for Fig. 2B on line 190.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) The model code seems to be available from https://simtk.org/projects/muscle_regen but that website requests member status ("This is a private project. You must be a member to view its contents.") and applying for membership could violate eLife's blind review process. So, this reviewer liked to but couldn't run the model her/himself. To eLife: Can the authors upload their model to a neutral server that reviewers and editors can access anonymously?

The code has been made publicly available on the following sites:

SimTK: https://simtk.org/docman/?group_id=2635

Zendo: https://zenodo.org/records/10403014

GitHub: https://github.com/mh2uk/ABM-of-Muscle-Regeneration-with-MicrovascularRemodeling

Line 121 has been updated with the new link and the additional resources were added to lines 654-657.

(2) The muscle regeneration field typically studies 2D cross-sections and the present model can be well compared to these other 2D models but cells as stochastic and localized sources of diffusible cytokines may yield different cytokine fields in 3D vs. 2D. I would expect more broadened and smoothened cytokine fields (from sources in neighboring cross-sections) than what the 2D model predicts based on sources just within the focus cross-section. Such relations of 2D to 3D should be discussed.

We thank the reviewer for the excellent suggestions and observations. It has been reported in other Compucell3D models (Sego et al. 2017, DOI:10.1088/17585090/aa6ed4) that the convergence of diffusion solutions between 2D and 3D model configurations had similar outcomes, with the 3D simulations presenting excessive computational cost without contributing any noticeable additional accuracy. Similarly, other cell-based ABMs that incorporate diffusion mechanisms (Marino et al. 2018, DOI:10.3390/computation6040058) have found that 2D and 3D versions of the model both predict the same mechanisms and that the 2D resolution was sufficient for determining outcomes. Lines 615-618 were added to elaborate on this topic.

(3) Since the model (and title) focuses on "nonlinear" cytokine interactions, what would change if cytokine decay would not be linear (as modeled here) but saturated (with nonlinear Michaelis-Menten kinetics as ligand binding and endocytosis mechanisms would call for)?

Thank you for raising an intriguing point. The model includes a combination of cytokine decay as well as ligand binding and endocytosis mechanisms that can be saturated. For a cytokine-dependent model behavior to occur the cytokines necessary to induce that action had to reach a minimum threshold. Once that threshold was reached, that amount of the cytokine would be removed at that location to simulate ligand-receptor binding and endocytosis. These ligand binding and endocytosis mechanisms behave in a saturated way, removing a set amount when above a certain threshold or a defined ratio when under the threshold. Lines 313-315 was revised to clarify this point. There were certain concentrations of cytokines where we saw a plateau in outputs likely as a result of reaching a saturation threshold (Supplemental Fig. 3). In future work, more robust mathematical simulation of binding kinetics of cytokines (e.g., using ODEs) could be included.

(4) Limitations of the model should be discussed together with an outlook for model refinement. For example, fiber alignment and ECM ultrastructure may require anisotropic diffusion. Many of the rate equations could be considered with saturation parameters etc. There are so many model assumptions. Please discuss which would be the most urgent model refinements and, to achieve these, which would be the most informative next experiments to perform.

We appreciate your thoughtful consideration of the model's limitations and the need for a comprehensive discussion on model refinements and potential future experiments. The future direction section was expanded to discuss additional possible model refinements (lines 635-643) and additional possible experiments for model validation (lines 630-634).

(5) It is not clear how the single spatial arrangement that is used affects the model predictions. E.g. now the damaged area surrounds the lymphatic vessel but what if the opposite corner was damaged and the lymphatic vessel is deep inside the healthy area?

Thank you for highlighting the importance of considering different spatial arrangements in the model and its potential impact on predictions. We previously tested model perturbations that included specifying the injury surrounding the lymphatic vessel versus on the side opposite the vessel. Since this paper focuses more on cytokine dynamics, we plan to include this perturbation, along with other injury alterations, in a follow-on paper. We added more context about this in the future efforts section lines 640-643.

(6) It seems that not only parameter values but also the initial values of most of the model components are unknown. The parameter estimation strategy does not seem to include the initial (spatial) distributions of collagen and cytokines and other model components. Please discuss how other (reasonable) initial values or spatial arrangements will affect model predictions.

We appreciate your thoughtful consideration of unknown initial values/spatial arrangements and their potential influence on predictions. Initial cytokine levels prior to injury had a low relative concentration compared to levels post injury and were assumed to be negligible. Initial spatial distribution of cytokines was not defined as initial spatial inputs (except in knockout simulations) but are secreted from cells (with baseline resident cell counts defined from the literature). The distribution of cytokines is an emergent behavior that results from the cell behaviors within the model. The collagen distribution is altered in response to clearance of necrosis by the immune cells (decreased collagen with necrosis removal) and subsequent secretion of collagen by fibroblasts. The secretion of collagen from fibroblast was included in the parameter estimation sweep (Supplemental Table 1).

We are working on further exploring the model sensitivity to altered spatial arrangements and have added this to the future directions section (lines 618-621), as well as provided Supplemental Figure 5 to demonstrate that model outcomes are similar with altered initial spatial arrangements.

(7) Many details of the CC3D implementation are missing: overall lattice size, interaction neighborhood order, and "temperature" of the Metropolis algorithm. Are the typical adhesion energy terms used in the CPM Hamiltonian and if so, then how are these parameter values estimated?

Thank you for bringing attention to the missing details regarding the CC3D implementation in our manuscript. We have included supplemental information providing greater detail for CPM implementation (Lines 808-854). We also added two additional supplemental tables for describing the requested CC3D implementation details (Supplemental Table 4) and adhesion energy terms (Supplemental Table 5).

(8) Extending the model analysis of combinations of altered cytokine properties, which temporal schedules of administration would be of interest, and how could the timing of multiple interventions improve outcomes? Such a discussion or even analysis would further underscore the usefulness of the model.

In response to your valuable suggestion, lines 558-562 were added to discuss the potential of using the model as a tool to perturb different cytokine combinations at varying timepoints throughout regeneration. In addition, this is also included in future work in lines 636-637.

(9) The CPM is only weakly motivated, just one sentence on lines 142-145 which mentions diffusion in a misleading way as the CPM just provides cells with a shape and mechanical interactions. The diffusion part is a feature of the hybrid CompuCell3D framework, not the CPM.

Thank you for bringing up this distinction. We removed the statement regarding diffusion and updated lines 143-146 to focus on CPM representation of cellular behavior and interactions. We also added a reference to supplemental text that includes additional details on CPM.

(10) On lines 258-261 it does not become clear how the described springs can direct fibroblasts towards areas of low-density collagen ECM. Are the lambdas dependent on collagen density?

Thank you for highlighting this area for clarification. The fibroblasts form links with low collagen density ECM and then are pulled towards those areas based on a constant lambda value. The links between the fibroblast and the ECM will only be made if the collagen is below a certain threshold. We added additional clarification to lines 260-264.

(11) On line 281, what does the last part in "Fibers...were regenerating but not fully apoptotic cells" mean? Maybe rephrase this.

The last of part of that line indicates that there were some fibers surrounding the main injury site that were damaged but still had healthy portions, indicating that they were impacted by the injury and are regenerating but did not become fully apoptotic like the fiber cells at the main site of injury. We rephrased this line to indicate that the nearby fibers were damaged but not fully apoptotic.

(12) Lines 290-293 describe interactions of cells and fields with localized structures (capillaries and lymphatic vessel). Please explain in more detail how "capillary agents...transport neutrophiles and monocytes" in the CPM model formalism. Are new cells added following rules? How is spatial crowding of the lattice around capillaries affecting these rules? Moreover, how can "lymphatic vessel...drain the nearby cytokines and cells"? How is this implemented in the CPM and how is "nearby" calculated? We appreciate your detailed inquiry into the interactions of cells and fields with localized structures. The neutrophils and monocytes are added to the simulation at the lattice sites above capillaries (within the cell layer Fig. 2B) and undergo chemotaxis up their respective gradients. The recruitment of the neutrophils and monocytes are randomly distributed among the healthy capillaries that do not have an immune cell at the capillary location (a modeling artifact that is a byproduct of only having one cell per lattice site). This approach helped to prevent an abundance of crowding at certain capillaries. Because immune cells in the simulation are sufficiently small, chemotactic gradients are sufficiently large, and the simulation space is sufficiently large, we do not see aggregation of recruited immune cells in the CPM.

The lymphatic vessel uptakes cytokines at lattice locations corresponding to the lymphatic vessel and will remove cells located in lattice sites neighboring the lymphatic vessel. In addition, we have included a rule in our ABM to encourage cells to migrate towards the lymphatic vessel utilizing CompuCell3D External Potential Plugin. The influence of this rule is inversely proportional to the distance of the cells to the lymphatic vessel.

We have updated lines 294-298 and 305-309 to include the above explanation.

(13) Tables 1-4 define migration speeds as agent rules but in the typical CPM, migration speed emerges from random displacements biased by chemotaxis and other effects (like the slope of the cytokine field). How was the speed implemented as a rule while it is typically observable in the model?

We appreciate your inquiry regarding the implementation of migration speeds. To determine the lambda parameters (Table 7) for each cell type, we tested each in a simplified control simulation with a concentration gradient for the cell to move towards. We tuned the lambda parameters within this simulation until the model outputted cell velocity aligned with the literature reported cell velocity for each cell type (Tables 1-4). We have incorporated clarification on this to lines 177-180.

(14) Line 312 shows the first equation with number (5), either add eqn. (1-4) or renumber.

We have revised the equation number.

(15) Typos: Line 456, "expect M1 cell" should read "except M1 cell".

Line 452, "thresholds above that diminish fibroblast response (Supplemental Fig 3)." remains unclear, please rephrase.

Line 473, "at 28." should read "at 28 days.".

Line 474, is "additive" correct? Was the sum of the individual effects calculated and did that match?

Line 534, "complexity our model" should read "complexity in our model".

We have corrected the typos and clarified line 452 (updated line 594) to indicate that the TNF-α concentration threshold results in diminished fibroblast response. We updated terminology line 474 (updated line 512) to indicate that there was a synergistic effect with the combined perturbation.

(16) Table 7 defines cell target volumes with the same value as their diameter. This enforces a strange cell shape. Should there be brackets to square the value of the cell diameter, e.g. Value=(12µm)^2 ?

The target volume parameter values were selected to reflect the relative differences in average cell diameter as reported in the literature; however, there are no parameters that directly enforce a diameter for the cells in the CPM formalism separate from the volume. We have observed that these relative cell sizes allow the ABM to effectively reproduce cell behaviors described in the literature. Single cells that are too large in the ABM would be unable to migrate far enough per time step to carry out cell behaviors, and cells that are too small in the CPM would be unstable in the simulation environment and not persist in the simulation when they should. We removed the units for the cell shape values in Table 7 since the target volume is a relative parameter and does not directly represent µm.

(17) Table 7 gives estimated diffusion constants but they appear to be too high. Please compare them to measured values in the literature, especially for MCP-1, TNF-alpha and IL-10, or relate these to their molecular mass and compare to other molecules like FGF8 (Yu et al. 2009, DOI:10.1038/nature08391).

We utilized a previously published estimation method (Filion et al. 2004, DOI:10.1152/ajpheart.00205.2004) for estimating cytokine diffusivity within the ECM. This method incorporates the molecular masses and accounts for the combined effects of the collagen fibers and glycosaminoglycans. The paper acknowledged that the estimated value is faster than experimentally determined values, but that this was a result of the less-dense matrix composition which is more reflective of the tissue environment we are simulating in contrast to other reported measurements which were done in different environments. Using this estimation method also allowed us to more consistently define diffusion constants versus using values from the literature (which were often not recorded) that had varied experimental conditions and techniques (such as being in zebrafish embryo Yu et al. 2009, DOI:10.1038/nature08391 as opposed to muscle tissue). This also allowed for recalculation of the diffusivity throughout the simulation as the collagen density changed within the model. Lines 318-326 were updated to help clarify the estimation method.

(18) Many DOIs in the bibliography (Refs. 7,17,20,31,40,47...153) are wrong and do not resolve because the appended directory names are not allowed in the DOI, just with a journal's URL after resolution.

Thank you for bringing this to our attention. The incorrect DOIs have been corrected.

Reviewer #2 (Recommendations For The Authors):

Minor comments:

(9) On line 174, the authors say "We used the CC3D feature Flip2DimRatio to control the number of times the Cellular-Potts algorithm runs per mcs." What does this mean? Isn't one monte carlo timestep one iteration of the Cellular Potts model? How does this relate to physical timescales?

We appreciate your attention to detail and thoughtful question regarding the statement about the use of the CC3D feature Flip2DimRatio. Lines 175-177 were revised to simplify the meaning of Flip2DimRatio. That parameter alters the number of times the Cellular-Potts algorithm is run, which is the limiting factor for cell movement. The physical timescale is kept to a 15-minute timestep but a high Flip2DimRatio allows more flexibility and stability to allow the cells to move faster in one timestep.

(10) Has the costum matlab script to process histology images into initial conditions been made available?

The Matlab script along with CC3D code for histology initialization with documentation has been made available with the source code on the following sites:

SimTK: https://simtk.org/docman/?group_id=2635

Zendo: https://zenodo.org/records/10403014

GitHub: https://github.com/mh2uk/ABM-of-Muscle-Regeneration-with-MicrovascularRemodeling

(11) Equation 5 is provided without a reference or derivation. Where does it come from and what does it mean?

Thank you for highlighting the diffusion equation and seeking clarification on its origin and significance. Lines 318-326 were revised to clarify where the equation comes from. This is a previously published estimation method that we applied to calculate the diffusivity of the cytokines considering both collagen and glycosaminoglycans.

(12) Line 326: "For CSA, experimental fold-change from pre-injury was compared with fold-change in model-simulated CSA". Does this step rely on the assumption that the fold change will not depend on the CSA? If so, is this something that is experimentally known, or otherwise, can it be confirmed by simulations?

We appreciate the opportunity to clarify our rationale. The fold change was used as a method to normalize the model and experiment so that they could be compared on the same scale. Yes, this step relies on the assumption that fold change does not depend on pre-injury CSA. Experimentally it is difficult to determine the impact of initial fiber morphology on altered regeneration time course. This fold-change allows us to compare percent recovery which is a common metric utilized to assess muscle regeneration outcomes experimentally. Line 340-343 was revised to clarify.

(13) Line 355: "The final passing criteria were set to be within 1 SD for CSA recovery and 2.5 SD for SSC and fibroblast count" Does this refer to the experimental or the simulated SD?

The model had to fit within those experimental SD. Lines 371-372 was edited to specify that we are referring the experimental SD.

(14) "Following 8 iterations of narrowing the parameter space with CaliPro, we reached a set that had fewer passing runs than the previous iteration". Wouldn't one expect fewer passing runs with any narrowing of the parameter space? Why was this chosen as the stopping criterion for further narrowing?

We appreciate your observation regarding the statement about narrowing the parameter space with CaliPro. We started with a wide parameter space, expecting that certain parameters would give outputs that fall outside of the comparable data. So, when the parameter space was narrowed to enrich parts that give passing output, initially the number of passing simulations increased.

Once we have narrowed the set of possible parameters into an ideal parameter space, further narrowing will cut out viable parameters resulting in fewer passing runs. Therefore, we stopped narrowing once any fewer simulations passed the criteria that they had previously passed with the wider parameter set. Lines 375-379 have been updated to clarify this point.

(15) Line 516: 'Our model could test and optimize combinations of cytokines, guiding future experiments and treatments." It is my understanding that this is communicated as a main strength of the model. Would it be possible to demonstrate that the sentence is true by using the model to make actual predictions for experiments or treatments?

This is demonstrated by the combined cytokine alterations in Figure 7 and discussed in lines 509-513. We have also added in a suggested experiment to test the model prediction in lines 691-695.

(16) Line 456, typo: I think 'expect' should be 'except'.

Thank you for pointing that out. The typo has been corrected.

Résumé de la vidéo [00:14:07][^1^][1] - [00:37:04][^2^][2]:

Cette vidéo présente une conférence sur le rôle de l'école en tant que territoire vivant au cœur des valeurs de la République. Les intervenants discutent de l'importance de l'école dans la transmission des valeurs républicaines, surtout dans le contexte des événements récents qui ont touché l'académie.

Points forts: + [00:14:07][^3^][3] Introduction de la conférence * Accueil par Sébastien Jaibovski, directeur de l'INSP de l'académie de Lille * Présentation du programme et des intervenants * Contextualisation de la conférence dans les événements actuels + [00:17:01][^4^][4] Allocution d'Alain Frugère * Importance de partager et transmettre les valeurs de liberté, égalité et fraternité * Rappel historique des valeurs républicaines depuis la Révolution française * Nécessité de défendre ces valeurs face aux défis contemporains + [00:23:02][^5^][5] Intervention de Madame Lower * L'école comme vecteur de l'égalité des chances et de la lutte contre les inégalités * Rôle de l'école dans la reconnaissance de la capacité de tous les enfants à apprendre * Importance de la liberté d'éducation pour l'émancipation individuelle + [00:30:03][^6^][6] Contextualisation par Sébastien Jaibovski * L'école face au poids des attentes sociales et symboliques * Réflexion sur le rôle des institutions publiques et les moyens alloués * L'école comme lieu de partage et de construction de la citoyenneté Résumé de la vidéo [00:37:07][^1^][1] - [01:02:26][^2^][2]:

Cette partie de la vidéo aborde le rôle de l'école dans la transmission des valeurs républicaines, en mettant l'accent sur une approche citoyenne. Elle souligne l'importance de l'éducation aux valeurs comme fondement de la République et la nécessité d'une pédagogie qui favorise la pensée critique et la participation active des élèves.

Points forts: + [00:37:07][^3^][3] L'importance de l'éducation aux valeurs * L'école comme lieu de découverte et de compréhension des valeurs républicaines * La transmission des valeurs comme mission centrale de l'institution scolaire * Les valeurs de liberté, égalité, fraternité, laïcité et refus des discriminations + [00:47:00][^4^][4] Approche citoyenne des valeurs * Refus de l'inculcation, éducation à la liberté * Appel à la pensée critique et à l'interrogation des valeurs * Importance de l'expérience vécue des valeurs dans l'établissement scolaire + [00:57:00][^5^][5] Engagement dans la République * La valeur comme refus du réel et espace pour l'engagement * L'écart entre les valeurs et le réel comme opportunité d'action * L'importance de l'engagement citoyen de chacun dans l'éducation aux valeurs Résumé de la vidéo 01:02:28 - 01:25:00 : La vidéo explore l'évolution de la notion des valeurs de la République dans le discours public, les médias, et le droit français depuis les années 80. Elle examine l'augmentation significative de l'utilisation de cette notion dans les années 80 et 90, sa stabilisation dans les années 2010, et son absence de définition constitutionnelle. La vidéo souligne également l'importance de l'éducation à la laïcité dans les écoles françaises et comment elle est abordée dans les programmes scolaires.

Points saillants : + [01:02:28][^1^][1] L'évolution de la notion des valeurs de la République * Augmentation dans les publications et les médias depuis les années 80 * Stabilisation dans les années 2010 * Absence de définition constitutionnelle + [01:06:02][^2^][2] L'impact sur le droit de l'éducation et le droit des étrangers * Contribution significative du code de l'éducation et du droit des étrangers * Importance de l'apprentissage et de l'intégration des valeurs + [01:10:01][^3^][3] La définition et l'enseignement de la laïcité * Présence accrue dans les programmes scolaires depuis les années 80 * Nécessité d'expliquer les règles aux élèves * Approche pédagogique pour déconstruire les oppositions + [01:18:34][^4^][4] La perception et la compréhension des élèves sur la laïcité * Bonne maîtrise de la notion par les élèves * Importance de l'éducation à la laïcité pour une société inclusive Résumé de la vidéo [01:25:02][^1^][1] - [01:38:50][^2^][2] :

Cette partie de la vidéo aborde l'importance de l'éducation à la laïcité et aux valeurs de la République dans les écoles françaises, y compris celles à l'étranger. Elle souligne les défis culturels et les différences dans l'approche de l'enseignement des valeurs universelles.

Points forts : + [01:25:02][^3^][3] La laïcité dans l'éducation * La laïcité n'est pas une hostilité envers les croyances religieuses * Importance de partager une vision positive de la laïcité * Nécessité d'adapter l'enseignement aux contextes culturels variés + [01:30:10][^4^][4] L'approche citoyenne à l'école * Utilisation de la pensée critique et de la liberté * Transformation des valeurs en réalité concrète pour les élèves * Engagement des élèves envers les valeurs de la République + [01:35:00][^5^][5] L'école, un territoire vivant * L'école est un lieu d'échange et d'apprentissage actif des valeurs républicaines * L'actualité influence l'enseignement et la perception des valeurs * L'importance de l'équilibre entre idéal et réalité dans l'éducation

author mentions this as the locking in of rotting google search.

The start of Google search decreasing effectiveness

After reading this article I was reminded of assignment 3 where we did bot trolling and I thought of how difficult it would be to code to catch for people not obeying user policies. For instance the comments weren't blatantly explicit, commenting sexually inappropriate comments, but they would contain a string of sexually suggestive emoji or insinuate some form of sexual abuse. At this rate, making sure your platform is safe for children would be extremely difficult. The way we coded the automatic response in assignment 3, we had it recognize a specific sentence but harassment is a spectrum and it's intepreted a lot of the time (making it difficult to detect).

Software development as a conversation

Scherer, Glenn. “Brazil’s New Forest Code Puts Vast Areas of Protected Amazon Forest at Risk.” Mongabay Environmental News, July 18, 2019. https://news.mongabay.com/2019/03/brazils-new-forest-code-puts-vast-areas-of-protected-amazon-forest-at-risk/.

The Nature Conservancy. “Brazil’s Forest Code,” July 1, 2018. https://www.nature.org/en-us/about-us/where-we-work/latin-america/brazil/stories-in-brazil/brazils-forest-code/.

So you do not think physical quantities exist out of our mind?

And remove the code because it should link to the discounted checkout cart, right?

a way of writing kind of like morse code

Author response:

The following is the authors’ response to the original reviews.

Editor’s summary:

This paper by Castello-Serrano et al. addresses the role of lipid rafts in trafficking in the secretory pathway. By performing carefully controlled experiments with synthetic membrane proteins derived from the transmembrane region of LAT, the authors describe, model and quantify the importance of transmembrane domains in the kinetics of trafficking of a protein through the cell. Their data suggest affinity for ordered domains influences the kinetics of exit from the Golgi. Additional microscopy data suggest that lipid-driven partitioning might segregate Golgi membranes into domains. However, the relationship between the partitioning of the synthetic membrane proteins into ordered domains visualised ex vivo in GPMVs, and the domains in the TGN, remain at best correlative. Additional experiments that relate to the existence and nature of domains at the TGN are necessary to provide a direct connection between the phase partitioning capability of the transmembrane regions of membrane proteins and the sorting potential of this phenomenon.

The authors have used the RUSH system to study the traffic of model secretory proteins containing single-pass transmembrane domains that confer defined affinities for liquid ordered (lo) phases in Giant Plasma Membrane derived Vesicles (GPMVs), out of the ER and Golgi. A native protein termed LAT partitioned into these lo-domains, unlike a synthetic model protein termed LAT-allL, which had a substituted transmembrane domain. The authors experiments provide support for the idea that ER exit relies on motifs in the cytosolic tails, but that accelerated Golgi exit is correlated with lo domain partitioning.

Additional experiments provided evidence for segregation of Golgi membranes into coexisting lipid-driven domains that potentially concentrate different proteins. Their inference is that lipid rafts play an important role in Golgi exit. While this is an attractive idea, the experiments described in this manuscript do not provide a convincing argument one way or the other. It does however revive the discussion about the relationship between the potential for phase partitioning and its influence on membrane traffic.

We thank the editors and scientific reviewers for thorough evaluation of our manuscript and for positive feedback. While we agree that our experimental findings present a correlation between trafficking rates and raft affinity, in our view, the synthetic, minimal nature of the transmembrane protein constructs in question makes a strong argument for involvement of membrane domains in their trafficking. These constructs have no known sorting determinants and are unlikely to interact directly with trafficking proteins in cells, since they contain almost no extramembrane amino acids. Yet, the LATTMD traffics through Golgi similarly to the full-length LAT protein, but quite different from mutants with lower raft phase affinity. We suggest that these observations can be best rationalized by involvement of raft domains in the trafficking fates and rates of these constructs, providing strong evidence (beyond a simple correlation) for the existence and relevance of such domains.

We have substantially revised the manuscript to address all reviewer comments, including several new experiments and analyses. These revisions have substantially improved the manuscript without changing any of the core conclusions and we are pleased to have this version considered as the “version of record” in eLife.

Below is our point-by-point response to all reviewer comments.

ER exit:

The experiments conducted to identify an ER exit motif in the C-terminal domain of LAT are straightforward and convincing. This is also consistent with available literature. The authors should comment on whether the conservation of the putative COPII association motif (detailed in Fig. 2A) is significantly higher than that of other parts of the C-terminal domain.

Thank you for this suggestion, this information has now been included as Supp Fig 2B. While there are other wellconserved residues of the LAT C-terminus, many regions have relatively low conservation. In contrast, the essential residues of the COPII association motif (P148 and A150) are completely conserved across in LAT across all species analyzed.

One cause of concern is that addition of a short cytoplasmic domain from LAT is sufficient to drive ER exit, and in its absence the synthetic constructs are all very slow. However, the argument presented that specific lo phase partitioning behaviour of the TMDs do not have a significant effect on exit from the ER is a little confusing. This is related to the choice of the allL-TMD as the 'non-lo domain' partitioning comparator. Previous data has shown that longer TMDs (23+) promote ER export (eg. Munro 91, Munro 95, Sharpe 2005). The mechanism for this is not, to my knowledge, known. One could postulate that it has something to do with the very subject of this manuscript- lipid phase partitioning. If this is the case, then a TMD length of 22 might be a poor choice of comparison. A TMD 17 Ls' long would be a more appropriate 'non-raft' cargo. It would be interesting to see a couple of experiments with a cargo like this.

The basis for the claim that raft affinity has relatively minor influence on ER exit kinetics, especially in comparison to the effect of the putative COPII interaction motif, is in Fig 1G. We do observe some differences between constructs and they may be related to raft affinity, however we considered these relatively minor compared to the nearly 4-fold increase in ER efflux induced by COPII motifs.

We have modified the wording in the manuscript to avoid the impression that we have ruled out an effect of raft affinity of ER exit.

We believe that our observations are broadly consistent with those of Munro and colleagues. In both their work and ours, long TMDs were able to exit the ER. In our experiments, this was true for several proteins with long TMDs, either as fulllength or as TMD-only versions (see Fig 1G). We intentionally did not measure shorter synthetic TMDs because these would not have been comparable with the raft-preferring variants, which all require relatively long TMDs, as demonstrated in our previous work1,2. Thus, because our manuscript does not make any claims about the influence of TMD length on trafficking, we did not feel that experiments with shorter non-raft constructs would substantively influence our conclusions.

However, to address reviewer interest, we did complete one set of experiments to test the effect of shortening the TMD on ER exit. We truncated the native LAT TMD by removing 6 residues from the C-terminal end of the TMD (LAT-TMDd6aa). This construct exited the ER similarly to all others we measured, revealing that for this set of constructs, short TMDs did not accumulate in the ER. ER exit of the truncated variant was slightly slower than the full-length LAT-TMD, but somewhat faster than the allL-TMD. These effects are consistent with our previous measurements with showed that this shortened construct has slightly lower raft phase partitioning than the LAT-TMD but higher than allL2. While these are interesting observations, a more thorough exploration of the effect of TMD length would be required to make any strong conclusion, so we did not include these data in the final manuscript.

Author response image 1.

Golgi exit:

For the LAT constructs, the kinetics of Golgi exit as shown in Fig. 3B are surprisingly slow. About half of the protein Remains in the Golgi at 1 h after biotin addition. Most secretory cargo proteins would have almost completely exited the Golgi by that time, as illustrated by VSVG in Fig. S3. There is a concern that LAT may have some tendency to linger in the Golgi, presumably due to a factor independent of the transmembrane domain, and therefore cannot be viewed as a good model protein. For kinetic modeling in particular, the existence of such an additional factor would be far from ideal. A valuable control would be to examine the Golgi exit kinetics of at least one additional secretory cargo.

We disagree that LAT is an unusual protein with respect to Golgi efflux kinetics. In our experiments, Golgi efflux of VSVG was similar to full-length LAT (t1/2 ~ 45 min), and both of these were similar to previously reported values3. Especially for the truncated (i.e. TMD) constructs, it is very unlikely that some factor independent of their TMDs affects Golgi exit, as they contain almost no amino acids outside the membrane-embedded TMD.

Practically, it has proven somewhat challenging to produce functional RUSH-Golgi constructs. We attempted the experiment suggested by the reviewer by constructing SBP-tagged versions of several model cargo proteins, but all failed to trap in the Golgi. We speculate that the Golgin84 hook is much more sensitive to the location of the SBP on the cargo, being an integral membrane protein rather than the lumenal KDEL-streptavidin hook. This limitation can likely be overcome by engineering the cargo, but we did not feel that another control cargo protein was essential for the conclusions we presented, thus we did not pursue this direction further.

Comments about the trafficking model

(1) In Figure 1E, the export of LAT-TMD from the ER is fitted to a single-exponential fit that the authors say is "well described". This is unclear and there is perhaps something more complex going on. It appears that there is an initial lag phase and then similar kinetics after that - perhaps the authors can comment on this?

This is a good observation. This effect is explainable by the mechanics of the measurement: in Figs 1 and 2, we measure not ‘fraction of protein in ER’ but ‘fraction of cells positive for ER fluorescence’. This is because the very slow ER exit of the TMD-only constructs present a major challenge for live-cell imaging, so ER exit was quantified on a population level, by fixing cells at various time points after biotin addition and quantifying the fraction of cells with observable ER localization (rather than tracking a single cell over time).

For fitting to the kinetic model (which attempts to describe ‘fraction in ER/Golgi’) we re-measured all constructs by livecell imaging (see Supp Fig 5) to directly quantify relative construct abundance in the ER or Golgi. These data did not have the plateau in Fig 1E, suggesting that this is an artifact of counting “ER positive cells” which would be expected to have a longer lag than “fraction of protein in ER”. Notably however, t1/2 measured by both methods was similar, suggesting that the population measurement agrees well with single-cell live imaging.

We have included all these explanations and caveats in the manuscript. We have also changed the wording from “well described” to “reasonably approximated”.

(2) The model for Golgi sorting is also complicated and controversial, and while the authors' intention to not overinterpreting their data in this regard must be respected, this data is in support of the two-phase Golgi export model (Patterson et al PMID:18555781).

The reviewers are correct, our observations and model are consistent with Patterson et al and it was a major oversight that a reference to this foundational work was not included. We have now added a discussion regarding the “two phase model” of Patterson and Lippincott-Schwartz.

Furthermore contrary to the statement in lines 200-202, the kinetics of VSVG exit from the Golgi (Fig. S3) are roughly linear and so are NOT consistent with the previous report by Hirschberg et al.

Regarding kinetics of VSVG, our intention was to claim that the timescale of VSVG efflux from the Golgi was similar to previously reported in Hirschberg, i.e. t1/2 roughly between 30-60 minutes. We have clarified this in the text. Minor differences in the details between our observations and Hirschberg are likely attributable to temperature, as those measurements were done at 32°C for the tsVSVG mutant.

Moreover, the kinetics of LAT export from the Golgi (Fig. 3B) appear quite different, more closely approximating exponential decay of the signal. These points should be described accurately and discussed.

Regarding linear versus exponential fits, we agree that the reality of Golgi sorting and efflux is far more complicated than accounted for by either the phenomenological curve fitting in Figs 1-3 or the modeling in Fig 4. In addition to the possibility of lateral domains within Golgi stacks, there is transport between stacks, retrograde traffic, etc. The fits in Figs 1-3 are not intended to model specifics of transport, but rather to be phenomenological descriptors that allowed us to describe efflux kinetics with one parameter (i.e. t1/2). In contrast, the more refined kinetic modeling presented in Figure 4 is designed to test a mechanistic hypothesis (i.e. coexisting membrane domains in Golgi) and describes well the key features of the trafficking data.

Relationship between membrane traffic and domain partitioning:

(1) Phase segregation in the GPMV is dictated by thermodynamics given its composition and the measurement temperature (at low temperatures 4degC). However at physiological temperatures (32-37degC) at which membrane trafficking is taking place these GPMVs are not phase separated. Hence it is difficult to argue that a sorting mechanism based solely on the partitioning of the synthetic LAT-TMD constructs into lo domains detected at low temperatures in GPMVs provide a basis (or its lack) for the differential kinetics of traffic of out of the Golgi (or ER). The mechanism in a living cell to form any lipid based sorting platforms naturally requires further elaboration, and by definition cannot resemble the lo domains generated in GPMVs at low temperatures.

We thank the reviewers for bringing up this important point. GPMVs are a useful tool because they allow direct, quantitative measurements of protein partitioning between coexisting ordered and disordered phases in complex, cell-derived membranes. However, we entirely agree, that GPMVs do not fully represent the native organization of the living cell plasma membrane and we have previously discussed some of the relevant differences4,5. Despite these caveats, many studies have supported the cellular relevance of phase separation in GPMVs and the partitioning of proteins to raft domains therein 6-9. Most notably, elegant experiments from several independent labs have shown that fluorescent lipid analogs that partition to Lo domains in GPMVs also show distinct diffusive behaviors in live cells 6,7, strongly suggesting the presence of nanoscopic Lo domains in live cells. Similarly, our recent collaborative work with the lab of Sarah Veatch showed excellent agreement between raft preference in GPMVs and protein organization in living immune cells imaged by super-resolution microscopy10. Further, several labs6,7, including ours11, have reported nice correlations between raft partitioning in GPMVs and detergent resistance, which is a classical (though controversial) assay for raft association.

Based on these points, we feel that GPMVs are a useful tool for quantifying protein preference for ordered (raft) membrane domains and that this preference is a useful proxy for the raft-associated behavior of these probes in living cells. We propose that this approach allows us to overcome a major reason for the historical controversy surrounding the raft field: nonquantitative and unreliable methodologies that prevented consistent definition of which proteins are supposed to be present in lipid rafts and why. Our work directly addresses this limitation by relating quantitative raft affinity measurements in a biological membrane with a relevant and measurable cellular outcome, specifically inter-organelle trafficking rates.

Addressing the point about phase transition temperatures in GPMVs: this is the temperature at which macroscopic domains are observed. Based on physical models of phase separation, it has been proposed that macroscopic phase separation at lower temperatures is consistent sub-microscopic, nanoscale domains at higher temperatures8,12. These smaller domains can potentially be stabilized / functionalized by protein-protein interactions in cells13 that may not be present in GPMVs (e.g. because of lack of ATP).

(2) The lipid compositions of each of these membranes - PM, ER and Golgi are drastically different. Each is likely to phase separate at different phase transition temperatures (if at all). The transition temperature is probably even lower for Golgi and the ER membranes compared to the PM. Hence, if the reported compositions of these compartments are to be taken at face value, the propensity to form phase separated domains at a physiological temperature will be very low. Are ordered domains even formed at the Golgi at physiological temperatures?

It is a good point that the membrane compositions and the resulting physical properties (including any potential phase behavior) will be very different in the PM, ER, and Golgi. Whether ordered domains are present in any of these membranes in living cells remains difficult to directly visualize, especially for non-PM membranes which are not easily accessible by probes, are nanoscopic, and have complex morphologies. However, the fact that raft-preferring probes / proteins share some trafficking characteristics, while very similar non-raft mutants behave differently argues that raft affinity plays a role in subcellular traffic.

(3) The hypothesis of 'lipid rafts' is a very specific idea, related to functional segregation, and the underlying basis for domain formation has been also hotly debated. In this article the authors conflate thermodynamic phase separation mechanisms with the potential formation of functional sorting domains, further adding to the confusion in the literature. To conclude that this segregation is indeed based on lipid environments of varying degrees of lipid order, it would probably be best to look at the heterogeneity of the various membranes directly using probes designed to measure lipid packing, and then look for colocalization of domains of different cargo with these domains.

This is a very good suggestion, and a direction we are currently following. Unfortunately, due to the dynamic nature and small size of putative lateral membrane domains, combined with the interior of a cell being filled with lipophilic environments that overlay each other, directly imaging domains in organellar membranes with lipid packing probes remains extremely difficult with current technology (or at least available to us). We argue that the TMD probes used in this manuscript are a reasonable alternative, as they are fluorescent probes with validated selectivity for membrane compartments with different physical properties.

Ultimately, the features of membrane domains suggested by a variety of techniques – i.e. nanometric, dynamic, relatively similar in composition to the surrounding membrane, potentially diverse/heterogeneous – make them inherently difficult to microscopically visualize. This is one reason why we believe studies like ours, which use a natural model system to directly quantify raft-associated behaviors and relate them to cellular effects (in our case, protein sorting), are a useful direction for this field.

We believe we have been careful in our manuscript to avoid confusing language surrounding lipid rafts, phase separation, etc. Our experiments clearly show that mammalian membranes have the capacity to phase separate, that some proteins preferentially interact with more ordered domains, and that this preference is related to the subcellular trafficking fates and rates of these proteins. We have edited the manuscript to emphasize these claims and avoid the historical controversies and confusions.

(4) In the super-resolution experiments (by SIM- where the enhancement of resolution is around two fold or less compared to optical), the authors are able to discern a segregation of the two types of Golgi-resident cargo that have different preferences for the lo-domains in GPMVs. It should be noted that TMD-allL and the LATallL end up in the late endosome after exit of the Golgi. Previous work from the Bonafacino laboratory (PMID: 28978644) has shown that proteins (such as M6PR) destined to go to the late endosome bud from a different part of the Golgi in vesicular carriers, while those that are destined for the cell surface first (including TfR) bud with tubular vesicular carriers. Thus at the resolution depicted in Fig 5, the segregation seen by the authors could be due to an alternative explanation, that these molecules are present in different areas of the Golgi for reasons different from phase partitioning. The relatively high colocalization of TfR with the GPI probe in Fig 5E is consistent with this explanation. TfR and GPI prefer different domains in the GPMV assays yet they show a high degree of colocalization and also traffic to the cell surface.

This is a good point. Even at microscopic resolutions beyond the optical diffraction limit, we cannot make any strong claims that the segregation we observe is due to lateral lipid domains and not several reasonable alternatives, including separation between cisternae (rather than within), cargo vesicles moving between cisternae, or lateral domains that are mediated by protein assemblies rather than lipids. We have explicitly included this point in the Discussion: “Our SIM imaging suggests segregation of raft from nonraft cargo in the Golgi shortly (5 min) after RUSH release (Fig 5B), but at this level of resolution, we can only report reduced colocalization, not intra-Golgi protein distributions. Moreover, segregation within a Golgi cisterna would be very difficult to distinguish from cargo moving between cisternae at different rates or exiting via Golgi-proximal vesicles.”

We have also added a similar caveat in the Results section of the manuscript: “These observations support the hypothesis that proteins can segregate in Golgi based on their affinity for distinct membrane domains; however, it is important to emphasize that this segregation does not necessarily imply lateral lipid-driven domains within a Golgi cisterna. Reasonable alternative possibilities include separation between cisternae (rather than within), cargo vesicles moving between cisternae, or lateral domains that are mediated by protein assemblies rather than lipids.”

Finally, while probes with allL TMD do eventually end up in late endosomes (consistent with the Bonifacino lab’s findings which we include), they do so while initially transiting the PM2,11.

Minor concerns:

(1) Generally, the quantitation is high quality from difficult experimental data. Although a lot appears to be manual, it appears appropriately performed and interpreted. There are some claims that are made based on this quantitation, however, where there are no statistics performed. For example, figure 1B. Any quantitation with an accompanying conclusion should be subject to a statistical test. I think the quality of the model fits- this is particularly important.

We appreciate the thoughtful feedback, the quantifications and fits were not trivial, but we believe important. We have added statistical significance to Figure 1B and others where it was missing.

(2) Modulation of lipid levels in Fig 4E shows a significant change for the trafficking rate for the LAT-TMD construct and a not so significant change for all-TMD construct. However, these data are not convincing and appear to depend on a singular data point that appears to lower the mean value. In general, the experiment with the MZA inhibitor (Fig. 4D-F) is hard to interpret because cells will likely be sick after inhibition of sphingolipid and cholesterol synthesis. Moreover, the difference in effects for LAT-TMD and allL-TMD is marginal.

We disagree with this interpretation. Fig 4E shows the average of three experiments and demonstrates clearly that the inhibitors change the Golgi efflux rate of LAT-TMD but not allL-TMD. This is summarized in the t1/2 quantifications of Fig 4F, which show a statistically significant change for LAT-TMD but not allL-TMD. This is not an effect of a singular data point, but rather the trend across the dataset.

Further, the inhibitor conditions were tuned carefully to avoid cells becoming “sick”: at higher concentrations, cells did adopt unusual morphologies and began to detach from the plates. We pursued only lower concentrations, which cells survived for at least 48 hrs and without major morphological changes.

(3) Line 173: 146-AAPSA-152 should read either 146-AAPSA-150 or 146-AAPSAPA-152, depending on what the authors intended.

Thanks for the careful reading, we intended the former and it has been fixed.

(4) What is the actual statistical significance in Fig. 3C and Fig. 3E? There is a single asterisk in each panel of the figure but two asterisks in the legend.

Apologies, a single asterisk representing p<0.05 was intended. It has been fixed.

(5) The code used to calculate the model. is not accessible. It is standard practice to host well-annotated code on Github or similar, and it would be good to have this publicly available.

We have deposited the code on a public repository (doi: 10.5281/zenodo. 10478607) and added a note to the Methods.

(1) Lorent, J. H. et al. Structural determinants and func7onal consequences of protein affinity for membrane ra=s. Nature communica/ons 8, 1219 (2017).PMC5663905

(2) Diaz-Rohrer, B. B., Levental, K. R., Simons, K. & Levental, I. Membrane ra= associa7on is a determinant of plasma membrane localiza7on. Proc Natl Acad Sci U S A 111, 8500-8505 (2014).PMC4060687

(3) Hirschberg, K. et al. Kine7c analysis of secretory protein traffic and characteriza7on of golgi to plasma membrane transport intermediates in living cells. J Cell Biol 143, 1485-1503 (1998).PMC2132993

(4) Levental, K. R. & Levental, I. Giant plasma membrane vesicles: models for understanding membrane organiza7on. Current topics in membranes 75, 25-57 (2015)

(5) Sezgin, E. et al. Elucida7ng membrane structure and protein behavior using giant plasma membrane vesicles. Nat Protoc 7, 1042-1051 (2012)

(6) Komura, N. et al. Ra=-based interac7ons of gangliosides with a GPI-anchored receptor. Nat Chem Biol 12, 402-410 (2016)

(7) Kinoshita, M. et al. Ra=-based sphingomyelin interac7ons revealed by new fluorescent sphingomyelin analogs. J Cell Biol 216, 1183-1204 (2017).PMC5379944

(8) Stone, M. B., Shelby, S. A., Nunez, M. F., Wisser, K. & Veatch, S. L. Protein sor7ng by lipid phase-like domains supports emergent signaling func7on in B lymphocyte plasma membranes. eLife 6 (2017).PMC5373823

(9) Machta, B. B. et al. Condi7ons that Stabilize Membrane Domains Also Antagonize n-Alcohol Anesthesia. Biophys J 111, 537-545 (2016)

(10) Shelby, S. A., Castello-Serrano, I., Wisser, I., Levental, I. & S., V. Membrane phase separa7on drives protein organiza7on at BCR clusters. Nat Chem Biol in press (2023)

(11) Diaz-Rohrer, B. et al. Rab3 mediates a pathway for endocy7c sor7ng and plasma membrane recycling of ordered microdomains Proc Natl Acad Sci U S A 120, e2207461120 (2023)

(12) Veatch, S. L. et al. Cri7cal fluctua7ons in plasma membrane vesicles. ACS Chem Biol 3, 287-293 (2008)

(13) Wang, H. Y. et al. Coupling of protein condensates to ordered lipid domains determines func7onal membrane organiza7on. Science advances 9, eadf6205 (2023).PMC10132753

the answers to exercise 7.4a and b should also be done with Python code

Reviewer #1 (Public Review):

Summary:

Li and colleagues describe an experiment whereby sequences of dots in different locations were presented to participants while electroencephalography (EEG) was recorded. By presenting fixed sequences of dots in different locations repeatedly to participants, the authors assumed that participants had learned the sequences during the experiment. The authors also trained classifiers using event-related potential (ERP) data recorded from separate experimental blocks of dots presented in a random (i.e., unpredictable) order. Using these trained classifiers, the authors then assessed whether patterns of brain activity could be detected that resembled the neural response to a dot location that was expected, but not presented. They did this by presenting an additional set of sequences whereby only one of the dots in the learned sequence appeared, but not the other dots. They report that, in these sequences with omitted stimuli, patterns of EEG data resembled the visual response evoked by a dot location for stimuli that could be expected, but were not presented. Importantly, this only occurred for an omitted dot stimulus that would be expected to appear immediately after the dot that was presented in these partial sequences.

This exciting finding complements previous demonstrations of the ability to decode expected (but not presented) stimuli in Blom et al. (2020) and Robinson et al. (2020) that are cited in this manuscript. It suggests that the visual system is able to generate patterns of activity that resemble expected sensory events, approximately at times at which an observer would expect them.

Strengths:

The experiment was carefully designed and care was taken to rule out some confounding factors. For example, gaze location was tracked over time, and deviations from fixation were marked, in order to minimise the contributions of saccades to above-chance decoding of dot position. The use of a separate block of dots (with unpredictable locations) to train the classifiers was also useful in isolating visual responses evoked by each dot location independently of any expectations that might be formed during the experiment. A large amount of data was also collected from each participant, which is important when using classifiers to decode stimulus features from EEG data. This careful approach is commendable and draws on best practices from existing work.

Weaknesses:

While there was clear evidence of careful experiment design, there are some aspects of the data analysis and results that significantly limit the inferences that can be drawn from the data. Both issues raised here relate to the use of pre-stimulus baselines and associated problems. As these issues are somewhat technical and may not be familiar to many readers, I will try to unpack each line of reasoning below. Here, it should be noted that these problems are complex, and similar issues often go undetected even by highly experienced EEG researchers.

Relevant to both issues, the authors derived segments of EEG data relative to the time at which each dot was presented in the sequences (or would have appeared when the stimuli were omitted in the partial sequences). Segments were derived that spanned -100ms to 300ms relative to the actual or expected onset of the dot stimulus. The 300ms post-stimulus time period corresponds to the duration of each dot in the sequence (100ms) plus the inter-stimulus interval (ISI) that was 200ms in duration before the next dot appeared (or would be expected to appear in the partial sequences). Importantly, a pre-stimulus baseline was applied to each of these segments of data, meaning that the average amplitude at each electrode between -100ms and 0ms relative to (actual or expected) stimulus onset was subtracted from each segment of data (i.e., each epoch in common EEG terminology). While this type of baseline subtraction procedure is commonplace in EEG studies, in this study design it is likely to cause problematic effects that could plausibly lead to the patterns of results reported in this manuscript.

First of all, the authors compare event-related potentials (ERPs) evoked by dots in the full as compared to partial sequences, to test a hypothesis relating to attentional tuning. They reported ERP amplitude differences across these conditions, for epochs corresponding to when a dot was presented to a participant (i.e., excluding epochs time-locked to omitted dots). However, these ERP comparisons are complicated by the fact that, in the full sequences, dot presentations are preceded by the presentation of other dots in the sequence. This means that ERPs evoked by the preceding dots in the full sequences will overlap in time with the ERPs corresponding to the dots presented at the zero point in the derived epochs. Importantly, this overlap would not occur in the partial sequence conditions, where only one dot was presented in the sequence. This essentially makes any ERP comparisons between full and partial sequences very difficult to interpret, because it is unclear if ERP differences are simply a product of overlapping ERPs from previously presented dots in the full sequence conditions. For example, there are statistically significant differences observed even in the pre-stimulus baseline period for this ERP analysis, which likely reflects the contributions ERPs evoked by the preceding dots in the full sequences, which are absent in the partial sequences.

The problems with interpreting this data are also compounded by the use of pre-stimulus baselines as described above. Importantly, the use of pre-stimulus baselines relies on the assumption that the ERPs in the baseline period (here, the pre-stimulus period) do not systematically differ across the conditions that are compared (here, the full vs. partial sequences). This assumption is violated due to the overlapping ERPs issue described just above. Accordingly, the use of the pre-stimulus baseline subtraction can produce spurious effects in the time period after stimulus onset (for examples see Feuerriegel & Bode, 2022, Neuroimage). This also makes it very difficult to meaningfully compare the ERPs following dot stimulus onset in these analyses.

The second issue relates to the use of pre-stimulus baselines and concerns the key finding reported in the paper: that EEG patterns corresponding to expected but omitted events can be decoded in the partial sequences. In the partial sequences, there are two critical epochs that were derived: One time-locked to the presentation of the dot, and another that was time-locked to 300ms after the dot was presented (i.e. when the next dot would be expected to appear). The latter epoch was used to test for representations of expected, but omitted, stimulus locations.

For the epochs in which the dots were presented, above-chance decoding can be observed spanning a training time range from around 100-300ms and a testing time range of a similar duration (see the plot in Figure 4b). This plot indicates that, during the time window of around 200-300ms following dot stimulus onset, the position of the dot can be decoded not only from trained classifiers using the same time windows spanning 200-300ms, but also using classifiers trained using earlier time windows of around 100-200ms.

This is important because the 200-300ms time period after dot onset in the partial sequences is the window used for pre-stimulus baseline subtraction when deriving epochs corresponding to the first successor representation (i.e., the first stimulus that might be expected to follow from the presented dot, but did not actually appear). In other words, the 200-300ms time window from dot onset corresponds to the -100 to 0 ms time window in the first successor epochs. Accordingly, the pattern that is indicative of the preceding, actually presented dot position would be subtracted from the EEG data used to test for the successor representation. Notably, the first successor condition would always be in another visual field quadrant (90-degree rotated or the opposite quadrant) as stated in the methods. In other words, the omitted stimulus would be expected to appear in the opposite vertical and/or horizontal visual hemifield as compared to the previously presented dot in these partial sequences.

This is relevant because ERPs tend to show reversed polarity across hemifields. For example, a stimulus presented in the right hemifield will have reversed polarity patterns at the same electrode as compared to an equivalent stimulus presented in the left hemifield (e.g., Supplementary Figure 3 in the comparable study of Blom et al., 2020). By subtracting the ERP patterns evoked by the presented dot in the partial sequences during the time period of 200-300ms (corresponding to the -100 to 0ms baseline window), this would be expected to bias patterns of EEG data in the first successor epochs to resemble stimulus positions in opposite hemifields. This could plausibly produce above-chance decoding accuracy in the time windows identified in Figure 5a, where the training time windows broadly correspond to the periods of above-chance decoding during 200-300ms from dot stimulus onset in Figure 4b.

In other words, the above-chance decoding of the first successor representation may plausibly be an artefact of the pre-stimulus baseline subtraction procedure used when deriving the epochs. This casts some doubt as to whether genuine successor representations were actually detected in the study. Additional tests for successor representations using ERP baselines prior to the presented dot in the partial sequences may be able to get around this, but such analyses were not presented, and the code and data were not accessible at the time of this review.

Although the study is designed well and a great amount of care was taken during the analysis stage, these issues with ERP overlap and baseline subtraction raise some doubts regarding the interpretability of the findings in relation to the analyses currently presented.

Reviewer #2 (Public Review):

Summary:

The authors investigate axonal and synapse development in two distinct visual feature-encoding neurons (VPN), LC4 and LPLC2. They first show that they occupy distinct regions on the GF dendrites, and likely arrive sequentially. Analysis of the VPNs' morphology throughout development, and synaptic gene and protein expression data reveals the temporal order of maturation. Functional analysis then shows that LPLC2 occupancy of the GF dendrites is constrained by LC4 presence.

Strengths:

The authors investigate an interesting and very timely topic, which will help to understand how neurons coordinate their development. The manuscript is very well written, and data are of high quality, that generally support the conclusions drawn (but see some comments for Fig. 2 below). A thorough descriptive analysis of the LC4/LPLC2 to GF connectivity is followed by some functional assessment showing that one neuron's occupancy of the GF dendrite depend on another.<br /> The manuscripts uses versatile methods to look at membrane contact, gene and protein expression (using scRNAseq data and state-of-the art genetic tools) and functional neuronal properties. I find it especially interesting and elegant how the authors combine their findings to highlight the temporal trajectory of development in this system.

Weaknesses:

After reading the summary, I was expecting a more comprehensive analysis of many VPNs, and their developmental relationships. For a better reflection of the data, the summary could state that the authors investigate *two* visual projection neurons (VPNs) and that ablation *of one cell type of VPNs* results in the expansion of the remaining VPN territory.

The manuscript is falling a bit short of putting the results into the context of what is known about synaptic partner choice/competition between different neurons during neuronal or even visual system development. Lots of work has been done in the peripheral the visual system, from the Hiesinger lab and others. Both the introduction and the discussion section should elaborate on this.

The one thing that the manuscript does not unambiguously show is when the connections between LC4 and LPLC2 become functional.

Figure 2:<br /> Figure 2A-C: I found the text related to that figure hard to follow, especially when talking about filopodia. Overall, life imaging would probably clarify at which time point there really are dynamic filopodia. For this study, high magnification images of what the authors define as filopodia would certainly help.<br /> L137ff: This section talks about filopodia between 24-48 hAPF, but only 36h APF is shown in A, where one could see filopodia. The other time points are shown in B and C, but number of filopodia is not quantified.<br /> L143: "filopodia were still present, but visibly shorter": This is hard to see, and again, not quantified.<br /> L144f: "from 72h APF to eclosion, the volume of GF dendrites significantly decreased": this is not actually quantified, comparisons are only done to 24, 36 and 48 h APF.<br /> Furthermore, 72h APF is not shown here, but in Figure 2D, so either show here, or call this figure panel already?

Figure 2D/E: to strengthen the point that LC4 and LPLC2 arrive sequentially, it would help to show all time points analyzed in Figure D/E.

L208: "significant increase ... from 60h APF to 72h APF": according to the figure caption, this comparison is marked by "+" but there is no + in the figure itself.

Figure 3:<br /> A key point of the manuscript is the sequential arrival of different VPN classes. So then why is the scRNAseq analysis in Figure 3 shown pooled across VPNs? Certainly, the reader at this point is interested in temporal differences in gene expression. The class-specific data are somewhat hidden in Supp. Fig. 9, and actually do not show temporal differences. This finding should be presented in the main data.

L438: "silencing LC4 by expressing Kir2.1... reduced the GF response": Is this claim backed by some quantification?

Figure 4K: Do the control data have error bars, which are just too small to see? And what is tested against what? Is blue vs. black quantified as well? What do red, blue, and black asterisks indicate? Please clarify in figure caption.

Optogenetics is mentioned in methods (in "fly rearing", in the genotypes, and there is an extra "Optogenetics" section in methods), but no such data are shown in the manuscripts. (If the authors have those data, it would be great to know when the VPN>GF connections become functional!)

Methods:

Antibody concentrations are not given anywhere and will be useful information for the reader

Could the authors please give more details on the re-analysis of the scRNAseq dataset? How did you identify cell type clusters in there, for example?

L785 and L794: I am curious. Why is it informative to mention what was *not* done?

Custom-written analysis code is mentioned in a few places. Is this code publicly available?

Reviewer #1 (Public Review):

Summary:

The manuscript gives a broad overview of how to write NeuroML, and a brief description of how to use it with different simulators and for different purposes - cells to networks, simulation, optimization, and analysis. From this perspective, it can be an extremely useful document to introduce new users to NeuroML.

However, the manuscript itself seems to lose sight of this goal in many places, and instead, the description at times seems to target software developers. For example, there is a long paragraph on the board and user community. The discussion on simulator tools seems more for developers, not users. All the information presented at the level of a developer is likely to be distracting to readers..

Strengths:

The modularity of NeuroML is indeed a great advantage. For example, the ability to specify the channel file allows different channels to be used with different morphologies without redundancy. The hierarchical nature of NeuroML also is commendable, and well illustrated in Figures 2a through c.

The number of tools available to work with NeuroML is impressive.

The abstract, beginning, and end of the manuscript present and discuss incorporating NeuroML into research workflows to support FAIR principles.

Having a Python API and providing examples using this API is fantastic. Exporting to NeuroML from Python is also a great feature.

Weaknesses:

Though modularity is a strength, it is unclear to me why the cell morphology isn't also treated similarly, i.e., specify the morphology of a multi-compartmental model in a separate file, and then allow the cell file to specify not only the files containing channels, but also the file containing the multi-compartmental morphology, and then specify the conductance for different segment groups. Also, after pynml_write_neuroml2_file, you would not have a super long neuroML file for each variation of conductances, since there would be no need to rewrite the multi-compartmental morphology for each conductance variation.

This would be especially important for optimizations, if each trial optimization wrote out the neuroML file, then including the full morphology of a realistic cell would take up excessive disk space, as opposed to just writing out the conductance densities. As long as cell morphology must be included in every cell file, then NeuroML is not sufficiently modular, and the authors should moderate their claim of modularity (line 419) and building blocks (551). In addition, this is very important for downloading NeuroML-compliant reconstructions from NeuroMorpho.org. If the cell morphology cannot be imported, then the user has to edit the file downloaded from NeuroMorpho.org, and provenance can be lost. Also, Figure 2d loses the hierarchical nature by showing ion channels, synapses, and networks as separate main branches of NeuroML.

In Figure 5, the difference between the core and native simulator is unclear. What is involved in helper scripts? I thought neurons could read NeuroML? If so, why do you need the export simulator-specific scripts? In addition, it seems strange to call something the "core" simulation engine, when it cannot support multi-compartmental models. It is unclear why "other simulators" that natively support NeuroML cannot be called the core. It might be more helpful to replace this sort of classification with a user-targeted description. The authors already state which simulators support NeuroML and which ones need code to be exported. In contrast, lines 369-370 mention that not all NeuroML models are supported by each simulator. I recommend expanding this to explain which features are supported in each simulator. Then, the unhelpful separation between core and native could be eliminated.

The body of the manuscript has so much other detail that I lose sight of how NeuroML supports FAIR. It is also unclear who is the intended audience. When I get to lines 336-344, it seems that this description is too much detail for the audience. The paragraph beginning on line 691 is a great example of being unclear about who is the audience. Does someone wanting to develop NeuroML models need to understand XSD schema? If so, the explanation is not clear. XSD schema is not defined and instead explains NeuroML-specific aspects of XSD. Lines 734-735 are another example of explaining to code developers (not model developers).

Reviewer #2 (Public Review):

Summary:

Developing neuronal models that are shareable, reproducible, and interoperable allows the neuroscience community to make better use of published models and to collaborate more effectively. In this manuscript, the authors present a consolidated overview of the NeuroML model description system along with its associated tools and workflows. They describe where different components of this ecosystem lay along the model development pathway and highlight resources, including documentation and tutorials, to help users employ this system.

Strengths:

The manuscript is well-organized and clearly written. It effectively uses the delineated model development life cycle steps, presented in Figure 1, to organize its descriptions of the different components and tools relating to NeuroML. It uses this framework to cover the breadth of the software ecosystem and categorize its various elements. The NeuroML format is clearly described, and the authors outline the different benefits of its particular construction. As primarily a means of describing models, NeuroML also depends on many other software components to be of high utility to computational neuroscientists; these include simulators (ones that both pre-date NeuroML and those developed afterwards), visualization tools, and model databases.

Overall, the rationale for the approach NeuroML has taken is convincing and well-described. The pointers to existing documentation, guides, and the example usages presented within the manuscript are useful starting points for potential new users. This manuscript can also serve to inform potential users of features or aspects of the ecosystem that they may have been unaware of, which could lower obstacles to adoption. While much of what is presented is not new to this manuscript, it still serves as a useful resource for the community looking for information about an established, but perhaps daunting, set of computational tools.

Weaknesses:

The manuscript in large part catalogs the different tools and functionalities that have been produced through the long development cycle of NeuroML. As discussed above, this is quite useful, but it can still be somewhat overwhelming for a potential new user of these tools. There are new user guides (e.g., Table 1) and example code (e.g. Box 1), but it is not clear if those resources employ elements of the ecosystem chosen primarily for their didactic advantages, rather than general-purpose utility. I feel like the manuscript would be strengthened by the addition of clearer recommendations for users (or a range of recommendations for users in different scenarios).

For example, is the intention that most users should primarily use the core NeuroML tools and expand into the wider ecosystem only under particular circumstances? What are the criteria to keep in mind when making that decision to use alternative tools (scale/complexity of model, prior familiarity with other tools, etc.)? The place where it seems most ambiguous is in the choice of simulator (in part because there seem to be the most options there) - are there particular scenarios where the authors may recommend using simulators other than the core jNeuroML software?

The interoperability of NeuroML is a major strength, but it does increase the complexity of choices facing users entering into the ecosystem. Some clearer guidance in this manuscript could enable computational neuroscientists with particular goals in mind to make better strategic decisions about which tools to employ at the outset of their work.

Author response:

The following is the authors’ response to the original reviews.

eLife assessment

Connelly and colleagues provide convincing genetic evidence that importation from mainland Tanzania is a major source of Plasmodium falciparum lineages currently circulating in Zanzibar. This study also reveals ongoing local malaria transmission and occasional near-clonal outbreaks in Zanzibar. Overall, this research highlights the role of human movements in maintaining residual malaria transmission in an area targeted for intensive control interventions over the past decades and provides valuable information for epidemiologists and public health professionals.

Reviewer #1 (Public Review):

Zanzibar archipelago is close to achieving malaria elimination, but despite the implementation of effective control measures, there is still a low-level seasonal malaria transmission. This could be due to the frequent importation of malaria from mainland Tanzania and Kenya, reservoirs of asymptomatic infections, and competent vectors. To investigate population structure and gene flow of P. falciparum in Zanzibar and mainland Tanzania, they used 178 samples from mainland Tanzania and 213 from Zanzibar that were previously sequenced using molecular inversion probes (MIPs) panels targeting single nucleotide polymorphisms (SNPs). They performed Principal Component Analysis (PCA) and identity by descent (IBD) analysis to assess genetic relatedness between isolates. Parasites from coastal mainland Tanzania contribute to the genetic diversity in the parasite population in Zanzibar. Despite this, there is a pattern of isolation by distance and microstructure within the archipelago, and evidence of local sharing of highly related strains sustaining malaria transmission in Zanzibar that are important targets for interventions such as mass drug administration and vector control, in addition to measures against imported malaria.

Strengths:

This study presents important samples to understand population structure and gene flow between mainland Tanzania and Zanzibar, especially from the rural Bagamoyo District, where malaria transmission persists and there is a major port of entry to Zanzibar. In addition, this study includes a larger set of SNPs, providing more robustness for analyses such as PCA and IBD. Therefore, the conclusions of this paper are well supported by data.

Weaknesses:

Some points need to be clarified:

(1) SNPs in linkage disequilibrium (LD) can introduce bias in PCA and IBD analysis. Were SNPs in LD filtered out prior to these analyses?

Thank you for this point. We did not filter SNPs in LD prior to this analysis. In the PCA analysis in Figure 1, we did restrict to a single isolate among those that were clonal (high IBD values) to prevent bias in the PCA. In general, disequilibrium is minimal only over small distances <5-10kb without selective forces at play. This is much less than the average spacing of the markers in the panel. If there is minimal LD, the conclusions drawn on relative levels and connections at high IBD are unlikely to be confounded by any effects of disequilibrium.

( 2) Many IBD algorithms do not handle polyclonal infections well, despite an increasing number of algorithms that are able to handle polyclonal infections and multiallelic SNPs. How polyclonal samples were handled for IBD analysis?

Thank you for this point. We added lines 157-161 to clarify. This section now reads:

“To investigate genetic relatedness of parasites across regions, identity by descent (IBD) estimates were assessed using the within sample major alleles (coercing samples to monoclonal by calling the dominant allele at each locus) and estimated utilizing a maximum likelihood approach using the inbreeding_mle function from the MIPanalyzer package (Verity et al., 2020). This approach has previously been validated as a conservative estimate of IBD (Verity et al., 2020).”

Please see the supplement in (Verity et al., 2020) for an extensive simulation study that validates this approach.

Reviewer #1 (Recommendations For The Authors):

(3) I think Supplementary Figures 8 and 9 are more visually informative than Figure 2.

Thank you for your response. We performed the analysis in Figure 2 to show how IBD varies between different regions and is higher within a region than between.

Reviewer #2 (Public Review):

This manuscript describes P. falciparum population structure in Zanzibar and mainland Tanzania. 282 samples were typed using molecular inversion probes. The manuscript is overall well-written and shows a clear population structure. It follows a similar manuscript published earlier this year, which typed a similar number of samples collected mostly in the same sites around the same time. The current manuscript extends this work by including a large number of samples from coastal Tanzania, and by including clinical samples, allowing for a comparison with asymptomatic samples.

The two studies made overall very similar findings, including strong small-scale population structure, related infections on Zanzibar and the mainland, near-clonal expansion on Pemba, and frequency of markers of drug resistance. Despite these similarities, the previous study is mentioned a single time in the discussion (in contrast, the previous research from the authors of the current study is more thoroughly discussed). The authors missed an opportunity here to highlight the similar findings of the two studies.

Thank you for your insights. We appreciated the level of detail of your review and it strengthened our work. We have input additional sentences on lines 292-295, which now reads:

“A recent study investigating population structure in Zanzibar also found local population microstructure in Pemba (Holzschuh et al., 2023). Further, both studies found near-clonal parasites within the same district, Micheweni, and found population microstructure over Zanzibar.”

Strengths:

The overall results show a clear pattern of population structure. The finding of highly related infections detected in close proximity shows local transmission and can possibly be leveraged for targeted control.

Weaknesses:

A number of points need clarification:

(1) It is overall quite challenging to keep track of the number of samples analyzed. I believe the number of samples used to study population structure was 282 (line 141), thus this number should be included in the abstract rather than 391. It is unclear where the number 232 on line 205 comes from, I failed to deduct this number from supplementary table 1.

Thank you for this point. We have included 282 instead of 391 in the abstract. We added a statement in the results at lines 203-205 to clarify this point, which now reads:

“PCA analysis of 232 coastal Tanzanian and Zanzibari isolates, after pruning 51 samples with an IBD of greater than 0.9 to one representative sample, demonstrates little population differentiation (Figure 1A).”

(2) Also, Table 1 and Supplementary Table 1 should be swapped. It is more important for the reader to know the number of samples included in the analysis (as given in Supplementary Table 1) than the number collected. Possibly, the two tables could be combined in a clever way.

Thank you for this advice. Rather than switch to another table altogether, we appended two columns to the original table to better portray the information (see Table 1).

Methods

(3) The authors took the somewhat unusual decision to apply K-means clustering to GPS coordinates to determine how to combine their data into a cluster. There is an obvious cluster on Pemba islands and three clusters on Unguja. Based on the map, I assume that one of these three clusters is mostly urban, while the other two are more rural. It would be helpful to have a bit more information about that in the methods. See also comments on maps in Figures 1 and 2 below.

Cluster 3 is a mix of rural/urban while the clusters 2, 4 and 5 are mostly rural. This analysis was performed to see how IBD changes in relation to local context within different regions in Zanzibar, showing that there is higher IBD within locale than between locale.

(4) Following this point, in Supplemental Figure 5 I fail to see an inflection point at K=4. If there is one, it will be so weak that it is hardly informative. I think selecting 4 clusters in Zanzibar is fine, but the justification based on this figure is unclear.

The K-means clustering experiment was used to cluster a continuous space of geographic coordinates in order to compare genetic relatedness in different regions. We selected this inflection point based on the elbow plot and based the number to obtain sufficient subsections of Zanzibar to compare genetic relatedness. This point is added to the methods at lines 174-178, which now reads:

“The K-means clustering experiment was used to cluster a continuous space of geographic coordinates in order to compare genetic relatedness in different regions. We selected K = 4 as the inflection point based on the elbow plot (Supplemental Figure 5) and based the number to obtain sufficient subsections of Zanzibar to compare genetic relatedness.”

(5) For the drug resistance loci, it is stated that "we further removed SNPs with less than 0.005 population frequency." Was the denominator for this analysis the entire population, or were Zanzibar and mainland samples assessed separately? If the latter, as for all markers <200 samples were typed per site, there could not be a meaningful way of applying this threshold. Given data were available for 200-300 samples for each marker, does this simply mean that each SNP needed to be present twice?

Population frequency is calculated based on the average within sample allele frequency of each individual in the population, which is an unbiased estimator. Within sample allele frequency can range from 0 to 1. Thus, if only one sample has an allele and it is at 0.1 within sample frequency, the population allele frequency would be 0.1/100 = 0.001. This allele is removed even though this would have resulted in a prevalence of 0.01. This filtering is prior to any final summary frequency or prevalence calculations (see MIP variant Calling and Filtering section in the methods). This protects against errors occurring only at low frequency.

Discussion:

(6) I was a bit surprised to read the following statement, given Zanzibar is one of the few places that has an effective reactive case detection program in place: "Thus, directly targeting local malaria transmission, including the asymptomatic reservoir which contributes to sustained transmission (Barry et al., 2021; Sumner et al., 2021), may be an important focus for ultimately achieving malaria control in the archipelago (Björkman & Morris, 2020)." I think the current RACD program should be mentioned and referenced. A number of studies have investigated this program.

Thank you for this point. We have added additional context and clarification on lines 275-280, which now reads:

“Thus, directly targeting local malaria transmission, including the asymptomatic reservoir which contributes to sustained transmission (Barry et al., 2021; Sumner et al., 2021), may be an important focus for ultimately achieving malaria control in the archipelago (Björkman & Morris, 2020). Currently, a reactive case detection program within index case households is being implemented, but local transmission continues and further investigation into how best to control this is warranted (Mkali et al. 2023).”

(7) The discussion states that "In Zanzibar, we see this both within and between shehias, suggesting that parasite gene flow occurs over both short and long distances." I think the term 'long distances' should be better defined. Figure 4 shows that highly related infections rarely span beyond 20-30 km. In many epidemiological studies, this would still be considered short distances.

Thank you for this point. We have edited the text at lines 287-288 to indicate that highly related parasites mainly occur at the range of 20-30km, which now reads:

“In Zanzibar, highly related parasites mainly occur at the range of 20-30km.”

(8) Lines 330-331: "Polymorphisms associated with artemisinin resistance did not appear in this population." Do you refer to background mutations here? Otherwise, the sentence seems to repeat lines 324. Please clarify.

We are referring to the list of Pfk13 polymorphisms stated in the Methods from lines 146-148. We added clarifying text on lines 326-329:

“Although polymorphisms associated with artemisinin resistance did not appear in this population, continued surveillance is warranted given emergence of these mutations in East Africa and reports of rare resistance mutations on the coast consistent with spread of emerging Pfk13 mutations (Moser et al., 2021). “

(9) Line 344: The opinion paper by Bousema et al. in 2012 was followed by a field trial in Kenya (Bousema et al, 2016) that found that targeting hotspots did NOT have an impact beyond the actual hotspot. This (and other) more recent finding needs to be considered when arguing for hotspot-targeted interventions in Zanzibar.

We added a clarification on this point on lines 335-345, which now reads:

“A recent study identified “hotspot” shehias, defined as areas with comparatively higher malaria transmission than other shehias, near the port of Zanzibar town and in northern Pemba (Bisanzio et al., 2023). These regions overlapped with shehias in this study with high levels of IBD, especially in northern Pemba (Figure 4). These areas of substructure represent parasites that differentiated in relative isolation and are thus important locales to target intervention to interrupt local transmission (Bousema et al., 2012). While a field cluster-randomized control trial in Kenya targeting these hotspots did not confer much reduction of malaria outside of the hotspot (Bousema et al. 2016), if areas are isolated pockets, which genetic differentiation can help determine, targeted interventions in these areas are likely needed, potentially through both mass drug administration and vector control (Morris et al., 2018; Okell et al., 2011). Such strategies and measures preventing imported malaria could accelerate progress towards zero malaria in Zanzibar.”

Figures and Tables:

(10) Table 2: Why not enter '0' if a mutation was not detected? 'ND' is somewhat confusing, as the prevalence is indeed 0%.

Thank you for this point. We have put zero and also given CI to provide better detail.

(11) Figure 1: Panel A is very hard to read. I don't think there is a meaningful way to display a 3D-panel in 2D. Two panels showing PC1 vs. PC2 and PC1 vs. PC3 would be better. I also believe the legend 'PC2' is placed in the wrong position (along the Y-axis of panel 2).

Supplementary Figure 2B suffers from the same issue.

Thank you for your comment. A revised Figure 1 and Supplemental Figure 2 are included, where there are separate plots for PC1 vs. PC2 and PC1 vs. PC3.

(12) The maps for Figures 1 and 2 don't correspond. Assuming Kati represents cluster 4 in Figure 2, the name is put in the wrong position. If the grouping of shehias is different between the Figures, please add an explanation of why this is.

Thank you for this point. The districts with at least 5 samples present are plotted in the map in Figure 1B. In Figure 2, a totally separate analysis was performed, where all shehias were clustered into separate groups with k-means and the IBD values were compared between these clusters. These maps are not supposed to match, as they are separate analyses. Figure 1B is at the district level and Figure 2 is clustering shehias throughout Zanzibar.

The figure legend of Figure 1B on lines 410-414 now reads:

“B) A Discriminant Analysis of Principal Components (DAPC) was performed utilizing isolates with unique pseudohaplotypes, pruning highly related isolates to a single representative infection. Districts were included with at least 5 isolates remaining to have sufficient samples for the DAPC. For plotting the inset map, the district coordinates (e.g. Mainland, Kati, etc.) are calculated from the averages of the shehia centroids within each district.”

The figure legend of Figure 2 on lines 417-425 now reads:

“Figure 2. Coastal Tanzania and Zanzibari parasites have more highly related pairs within their given region than between regions. K-means clustering of shehia coordinates was performed using geographic coordinates all shehias present from the sample population to generate 5 clusters (colored boxes). All shehias were included to assay pairwise IBD between differences throughout Zanzibar. Pairwise comparisons of within cluster IBD (column 1 of IBD distribution plots) and between cluster IBD (column 2-5 of IBD distribution plots) was done for all clusters. In general, within cluster IBD had more pairwise comparisons containing high IBD identity.”

(13) Figure 2: In the main panel, please clarify what the lines indicate (median and quartiles?). It is very difficult to see anything except the outliers. I wonder whether another way of displaying these data would be clearer. Maybe a table with medians and confidence intervals would be better (or that data could be added to the plots). The current plots might be misleading as they are dominated by outliers.

Thank you for this point and it greatly improved this figure. We changed the plotting mechanisms through using a beeswarm plot, which plots all pairwise IBD values within each comparison group.

(14) In the insert, the cluster number should not only be given as a color code but also added to the map. The current version will be impossible to read for people with color vision impairment, and it is confusing for any reader as the numbers don't appear to follow any logic (e.g. north to south).

Thank you very much for these considerations. We changed the color coding to a color blind friendly palette and renamed the clusters to more informative names; Pemba, Unguja North (Unguja_N), Unguja Central (Unguja_C), Unguja South (Unguja_S) and mainland Tanzania (Mainland).

(15) The legend for Figure 3 is difficult to follow. I do not understand what the difference in binning was in panels A and B compared to C.

Thank you for this point. We have edited the legend to reflect these changes. The legend for Figure 3 on lines 427-433 now reads:

“Figure 3. Isolation by distance is shown between all Zanzibari parasites (A), only Unguja parasites (B) and only Pemba parasites (C). Samples were analyzed based on geographic location, Zanzibar (N=136) (A), Unguja (N=105) (B) or Pemba (N=31) (C) and greater circle (GC) distances between pairs of parasite isolates were calculated based on shehia centroid coordinates. These distances were binned at 4km increments out to 12 km. IBD beyond 12km is shown in Supplemental Figure 8. The maximum GC distance for all of Zanzibar was 135km, 58km on Unguja and 12km on Pemba. The mean IBD and 95% CI is plotted for each bin.”

(16) Font sizes for panel C differ, and it is not aligned with the other panels.

Thank you for pointing this out. Figure 3 and Supplemental Figure 10 are adjusted with matching formatting for each plot.

(17) Why is Kusini included in Supplemental Figure 4, but not in Figure 1?

In Supplemental Figure 4, all isolates were used in this analysis and isolates with unique pseudohaplotypes were not pruned to a single representative infection. That is why there are additional isolates in Kusini. The legend for Supplemental Figure 4 now reads:

“Supplemental Figure 4. PCA with highly related samples shows population stratification radiating from coastal Mainland to Zanzibar. PCA of 282 total samples was performed using whole sample allele frequency (A) and DAPC was performed after retaining samples with unique pseudohaplotypes in districts that had 5 or more samples present (B). As opposed to Figure 1, all isolates were used in this analysis and isolates with unique pseudohaplotypes were not pruned to a single representative infection.”

(18) Supplemental Figures 6 and 7: What does the width of the line indicate?

The sentence below was added to the figure legends of Supplemental Figures 6 and 7 and the legends of each network plot were increased in size:

“The width of each line represents higher magnitudes of IBD between pairs.”

(19) What was the motivation not to put these lines on the map, as in Figure 4A? This might make it easier to interpret the data.

Thank you for this comment. For Supplemental Figure 8 and 9, we did not put these lines that represent lower pairwise IBD to draw the reader's attention to the highly related pairs between and within shehias.

Reviewer #2 (Recommendations For The Authors):

(1) There is a rather long paragraph (lines 300-323) on COI of asymptomatic infections and their genetic structure. Given that the current study did not investigate most of the hypotheses raised there (e.g. immunity, expression of variant genes), and the overall limited number of asymptomatic samples typed, this part of the discussion feels long and often speculative.

Thank you for your perspective. The key sections highlighted in this comment, regarding immunity and expression of variant genes, were shortened. This section on lines 300-303 now reads:

“Asymptomatic parasitemia has been shown to be common in falciparum malaria around the globe and has been shown to have increasing importance in Zanzibar (Lindblade et al., 2013; Morris et al., 2015). What underlies the biology and prevalence of asymptomatic parasitemia in very low transmission settings where anti-parasite immunity is not expected to be prevalent remains unclear (Björkman & Morris, 2020).”

(2) As a detail, line 304 mentions "few previous studies" but only one is cited. Are there studies that investigated this and found opposite results?

Thank you for this comment. We added additional studies that did not find an association between clinical disease and COI. These changes are on lines 303-308, which now reads:

“Similar to a few previous studies, we found that asymptomatic infections had a higher COI than symptomatic infections across both the coastal mainland and Zanzibar parasite populations (Collins et al., 2022; Kimenyi et al., 2022; Sarah-Matio et al., 2022). Other studies have found lower COI in severe vs. mild malaria cases (Robert et al., 1996) or no significant difference between COI based on clinical status (Earland et al. 2019; Lagnika et al. 2022; Conway et al. 1991; Kun et al. 1998; Tanabe et al. 2015)”

(3) Table 2: Percentages need to be checked. To take one of several examples, for Pfk13-K189N a frequency of 0.019 for the mutant allele is given among 137 samples. 2/137 equals to 0.015, and 3/137 to 0.022. 0.019 cannot be achieved. The same is true for several other markers. Possibly, it can be explained by the presence of polyclonal infections. If so, it should be clarified what the total of clones sequenced was, and whether the prevalence is calculated with the number of samples or number of clones as the denominator.

Thank you for this point. We mistakenly reported allele frequency instead of prevalence. An updated Table 2 is now in the manuscript. The method for calculating the prevalence is now at lines 148-151:

“Prevalence was calculated separately in Zanzibar or mainland Tanzania for each polymorphism by the number of samples with alternative genotype calls for this polymorphism over the total number of samples genotyped and an exact 95% confidence interval was calculated using the Pearson-Klopper method for each prevalence.”

START OF THE HEADER!

Throughout Jaime Cortez’s short story Gordro, we find instances of where Gordo is assimilating to the mainstream image of a boy with masculine interests. This need of being masculine is pushed by his father and Cesar. The passage highlighted how Gordo quickly code switched from singing a girly song to doing something masculine hence counting. Although his father scolded him on counting, Gordo could’ve gone back to singing but he did not. As his first instinct to sing a song about being a princess. But Gordo wants to become more like or have his fathers approval, Gordo wants to please his fathers so he crosses a boundary to be accepted.

sneaking suspicion

lot code in my way

no bang per line of code

// We do not get much out of higher-level languages because people do not appreciate that "that advantage of high-level languages is notational rather then computatioal" John Allen - Anatomy of Lisp

I learn this 40 years ago.

I only realise now that the problem with Programming Languages and with Software "Enginnering#2 which is nothing of the sort

The lesson from the first NATO conference on the Software Crisis

ended up identifying that it is not angineering what we doe

that that was what was needed but clearly out of sight

yet the pretence persisted and we were kidding ourselves that what we do is ingineering

The reason is that programming languages constituted in terms of the means of primitives, means of abstractions and means of combinations,

where what is needed is to raise th level of expressive power of our notations by building everything that is needed into a coherent complex self-orgamzing system that supports such complex way of arrticulatinon that the task calls for

Articulating intent to the point where it is amenable to the pun to actuallyr un on a machine

Author response:

Reviewer #3 (Public Review):

Software UX design is not a trivial task and a point-and-click interface may become difficult to use or misleading when such design is not very well crafted. While Phantasus is a laudable effort to bring some of the out-of-the box transcriptomics workflows closer to the broader community of point-and-click users, there are a number of shortcomings that the authors may want to consider improving.

Thank you for such an in-depth review. We really appreciate this feedback and have tried to address all of the concerns in the new version of Phantasus.

Here I list the ones I found running Phantasus locally through the available Bioconductor package:

(1) The feature of loading in one click one of the thousands of available GEO datasets is great. However, one important use of any such interfaces is the possibility for the users to analyze his/her own data. One of the standard formats for storing tables of RNA-seq counts are CSV files. However, if we try to upload from the computer a CSV file with expression data, such as the counts stored in the file GSE120660_PCamerge_hg38.csv.gz from https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE120660, a first problem is that the system does not recognize that the CSV file is compressed. A second problem is that it does not recognize that values are separated by commas, the very original CSV format, giving a cryptic error "columnVector is undefined". If we transform the CSV format into tab-separated values (TSV) format, then it works, but this constitutes already a first barrier for the target user of Phantasus.

Thank you for highlighting this issue of file formats support. We acknowledge the commonality of CSV and CSV.gz files in gene expression analysis. As a response, we have updated our data loading procedure to support these file formats. Moreover, the most recent version of our web application is able to recognize gzip-archived file in any of supported table formats: GCT, TSV, CSV and XLSX.

(2) Many RNA-seq processing pipelines use Ensembl annotations, which for the purpose of downstream interpretation of the analysis, need to be translated into HUGO gene symbols. When I try to annotate the rows to translate the Ensembl gene identifiers, I get the error

"There is no AnnotationDB on server. Ask administrator to put AnnotationDB sqlite databases in cacheDir/annotationdb folder"

Thank you for revealing this issue. Indeed, locally installed instances of the Phantatus might lose some functionality in absence of some auxiliary files. For example, gene annotation mapping is unavailable without annotation databases. Previously, the user had to perform additional setup steps to unlock a few features, which might be confusing and unclear. In order to overcome this we have revised significantly the installation procedure. Newly added ‘setupPhantasus’ function is able to create all necessary configuration files and provides an interactive dialog with the user that helps to load all necessary data files from our official cache mirror (https://alserglab.wsutl.edu/files/phantasus/minimal-cache/). Docker-based installation follows the same approach, however it is configured to install everything by default. Thus, with help of the new installation procedure locally installed Phantasus now has the whole functionality available at the official mirrors. The comprehensive installation description is now available at https://ctlab.github.io/phantasus-doc/installation.

(3) When trying to normalize the RNA-seq counts, there are no standard options such as within-library (RPKM, FPKM) or between-library (TMM) normalization procedures.

Appreciating your feedback, we've expanded the available normalization options in the updated version of Phantasus. We added support for TMM normalization as suggested by the edgeR package and voom normalization from the limma package. However, certain strategies like RPKM/FPKM or TPM rely on gene-specific effective lengths, which are challenging to infer without protocol and alignment details. As Phantasus operates on gene expression matrices and doesn't execute alignment steps, the implementation of these normalization seems infeasible. On the other hand, if the user has the matrix with FPKM or TPM gene values (for example from a core facility), such a matrix can be loaded into Phantasus and used for the analysis.

If I take log2(1+x) a new tab is created with the normalized data, but it's not easy to realize what happened because the tab has the same name as the previous one and while the colors of the heatmap changed to reflect the new scale of the data, this is quite subtle. This may cause that an unexperienced user to apply the same normalization step again on the normalized data. Ideally, the interface should lead the user through a pipeline, reducing unnecessary degrees of freedom associated with each step.

Thank you for your comment. Indeed our approach to create a new tab for each alteration to the expression values preserving the name might be the source of confusion for a user. On the other hand, generating informative tab names without overwhelming users with too much detail is also challenging. As a compromise we have an option for the user to manually rename the tab. Still, we agree that this remains an area for improvement. We also consider it to be a part of a larger issue: for example, the loaded data can already be log-scaled, so that even one round of log-scale transformation in Phantasus would be incorrect. Accordingly, we are exploring ways to address this issue in the future by adding automated checks for the tools or, as you suggested, implementing stricter pipelines.

(4.4) Phantasus allows one to filter out lowly-expressed genes by averaging expression of genes across samples and discarding/selecting genes using some cutoff value on that average. This strategy is fine, but to make an informed decision on that cutoff it would be useful to see a density plot of those averages that would allow one to identify the modes of low and high expression and decide the cutoff value that separates them.

Thank you for the suggestion. Indeed a density plot might help users to make informed decisions during gene filtration. We have added such a plot into the ‘Plot/Chart’ tool as a ‘histogram’ chart type.

It would be also nice to have an interface to the filterByExpr() function from the edgeR package, which provides more control on how to filter out lowly-expressed genes.

Thank you for proposing the inclusion of an interface for the filterByExpr() function from the edgeR package. In the recent update we have incorporated filterByExpr() as part of the voom normalization tool. For now, for simplicity, we have decided to keep only the default parameter values. However, we will explore the addition of the dedicated filtering tool in the future.

(5) When attempting a differential expression (DE) analysis, a popup window appears saying:

"Your dataset is filtered. Limma will apply to unfiltered dataset. Consider using New Heat Map tool."

One of the main purposes of filtering lowly-expressed genes is mainly to conduct a DE analysis afterwards, so it does not make sense that the tool says that such an analysis will be done on the unfiltered dataset. The reference to the "New Heat Map tool" is vague and unclear where should the user look for that other tool, without any further information or link.

Thank you for highlighting this issue. We agree that the message in the popup window and the default action were confusing. In response to your feedback, we've updated the default behavior of our DE tools to automatically use the filtered data in a new tab. Additionally, we've clarified the warning message to ensure a better understanding of this process.

(6) The DE analysis only allows for a two-sample group comparison, which is an important limitation in the question we may want to address. The construction of more complex designs could be graphically aided by using the ExploreModelMatrix Bioconductor package (Soneson et al, F1000Research, 2020).

Indeed, the ability to create complex designs and various comparisons is important for many applications for gene expression analysis. Accordingly, in the latest Phantasus version, we've introduced an advanced design feature for the DE analysis, enabling the utilization of multiple column annotations for the design matrix. Combined with the existing ability to create new annotations, this update facilitates the setup of diverse design matrices. While at the moment we do not allow setting a complex contrast, we hope that the current interface will cover most of the differential expression use cases.

(7) When trying to perform a pathway analysis with FGSEA, I get the following error:

"Couldn't load FGSEA meta information. Please try again in a moment. Error: cannot open the connection In call: file(file, "rt")

We hope that this issue should be resolved after we have implemented a more streamlined setup process. Among others, the new approach aims to eliminate the unexpected absence of metafiles in local installations. The latest Phantasus package version explicitly prompts the user to load necessary additional files automatically during the initial run, reducing options for an invalid setup.

Finally, there have been already some efforts to approach R and Bioconductor transcriptomics pipelines to point-and-click users, such as iSEE (Rue-Albrecht et al, 2018) and GeneTonic (Marini et al, 2021) but they are not compared or at least cited in the present work.

Indeed, our comparison was focused toward tools that offer non-programmatic functionalities for gene expression data analysis. While tools like iSEE and GeneTonic are adept at visualizing data and hold their own in providing extensive abilities, they do necessitate additional data preparation using R, distinguishing them from the specific scope of tools we assessed.

One nice features of these two tools that I missed in Phantasus is the possibility of generating the R code that produces the analysis performed through the interface. This is important to provide a way to ensure the reproducibility of the analyses performed.

The ability to generate R code within tools like these indeed aids in ensuring analysis reproducibility. Moreover, we have previously attempted implementing this functionality in Phantasus, however it proved to be hard to do in a useful fashion due to potential complex interactions between user and the client-side part of Phantasus. Nevertheless, we acknowledge the significance of such a feature and aim to introduce it in the future.

Sure! Imagine you're building a structure, like a house. Before you start building, you need a plan, right? That's what a schema is in programming – it's like your blueprint.

So, in this code, we're using a tool called Joi to make our blueprint. We want our structure to have a specific type, like a string, and maybe some rules, like a minimum length.

Here's a simple explanation:

1. Constructing the Schema: First, we make our blueprint using Joi. In this case, we're saying we want something called a to be a string. Think of it like saying, "In my house blueprint, I want a room called a, and it should be a string."

javascript const schema = Joi.object({ a: Joi.string() });

1. Adding Rules (Constraints): Now, let's say we want to add a rule to our blueprint, like saying that our room a must be at least 5 characters long. When we add rules, Joi gives us back a new blueprint with that rule added. It's like updating our original blueprint with extra details.

javascript const schemaWithRule = schema.keys({ a: Joi.string().min(5) });

So, in simple terms, we're creating a plan for our data, and then we can add rules to that plan to make sure our data follows certain conditions.

Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

Learn more at Review Commons

---

Reply to the reviewers

Reviewer #1

Evidence, reproducibility and clarity

Seleit and colleagues set out to explore the genetics of developmental timing and tissue size by mapping natural genetic variation associated with segmentation clock period and presomitic mesoderm (PSM) size in different species of Medaka fish. They first establish the extent of variation between five different Medaka species of in terms of organismal size, segmentation rate, segment size and presomitic mesoderm size, among other traits. They find that these traits are species-specific but strongly correlated. In a massive undertaking, they then perform developmental QTL mapping for segmentation clock period and PSM size in a set of ~600 F2 fish resulting from the cross of Orizyas sakaizumii (Kaga) and Orizyas latipes (Cab). Correlation between segmentation period and segment size was lost among the F2s, indicating that distinct genetic modules control these traits. Although the researchers fail to identify causal variants driving these traits, they perform proof of concept perturbations by analyzing F0 Crispants in which candidate genes were knocked out. Overall, the study introduces a completely new methodology (QTL mapping) to the field of segmentation and developmental tempo, and therefore provides multiple valuable insights into the forces driving evolution of these traits.

Major comments: - The first sentence in the abstract reads "How the timing of development is linked to organismal size is a longstanding question". It is therefore disappointing that organismal size is not reported for the F2 hybrids. Was larval length measured in the F2s? If so, it should be reported. It is critical to understand whether the correlation between larval size and segmentation clock period is preserved in F2s or not, therefore determining if they represent a single or separate developmental modules. If larval length data were not collected, the authors need to be more careful with their wording.

The question the reviewer raises here is indeed a very relevant one, and a question that we also were curious about ourselves. While it was not possible (logistically) to grow the 600 F2 fish to adulthood, we did measure larval length in a subset of F2 hatchling (n=72) to ask precisely the question the reviewer raises here. Our results (new Supplementary Figure 5) show that the correlation between larval length and segmentation timing (which we report across the Oryzias species) is absent in the F2s. This indeed argues that the traits represent separate developmental modules.

In the current version of the paper, organismal size is often incorrectly equated to tissue size (e.g. PSM size, segment size). For example, in page 3 lines 33-34, the authors state that faster segmentation occurred in embryos of smaller size (Fig. 1D). However, Fig. 1D shows correlation between segmentation rate and unsegmented PSM area. The appropriate data to show would be segmentation rate vs. larval or adult length.

The reviewer is correct. We have now linked the data more clearly to data we show in Supplementary Figure 1, which shows that adult length and adult mass are strongly correlated (S1A) and that adult mass is in turn strongly correlated with segmentation rate in the different Oryzias species (S1B). Additionally main Figure 1B shows that larval length is correlated with PSM length. We have corrected the main text to reflect these relationships more clearly.

• Is my understanding correct in that the her7-venus reporter is carried by the Cab F0 but not the Kaga F0? Presumably only F2s which carried the reporter were selected for phenotyping. I would expect the location of the reporter in the genome to be obvious in Figure 3J as a region that is only Cab or het but never Kaga. Can the authors please point to the location of the reporter?

The reviewer is correct. Indeed the location of our her7-venus KI is on chromosome 16 and the recombination patterns on this chromosome overwhelmingly show either Hom Cab (green) or Het Cab/Kaga (Black). This is expected as we selected fish carrying the her7-venus KI for phenotyping.

• devQTL mapping in this study seems like a wasted opportunity. The authors perform mapping only to then hand pick their targets based on GO annotations. This biases the study towards genes known to be involved in PSM development, when part of the appeal of QTL mapping is precisely its unbiased nature and the potential to discover new functionally relevant genes. The authors need to better justify their rationale for candidate prioritization from devQTL peaks. The GO analysis should be shown as supplemental data. What criteria were used to select genes based on GO annotations?

We have now commented on these valid points and outlined our rationale in more detail in the text (page 4, lines 20-30). Our rationale now also includes selection of differentially expressed genes (n=5 genes) that fall within segmentation timing devQTL hits (for more details see below). Essentially, while we indeed finally focused on the proof of principle using known genes, these genes were previously not known to play a role in either setting the timing of segmentation or controlling the size of the PSM. Hence, we do think our strategy demonstrates the "the potential to discover new functionally relevant genes", even though the genes themselves had been involved overall in somitogenesis. We added the GO analysis as supplemental data as requested (new Supplementary Figure 7E).

• Analysis of the predicted functional consequence of divergent SNPs (Fig. S6B, F) is superficial. Among missense variants, which genes harbor the most deleterious mutations? Which missense variants are located in highly conserved residues? Which genes carry variants in splice donors/acceptors? Carefully assessing the predicted effect of SNPs in coding regions would provide an alternative, less biased approach to prioritize candidate genes.

We now included our analysis of SNPs based on the Variant effect predictor (VEP) tool from ensembl. This analysis does rank the predicted severity of the SNP on protein structure and function (Impact: low, moderate, high) and does annotate which variants can affect splice donors/acceptors. The VEP analysis for both phenotypes is now added to the manuscript as supplemental data (new Supplementary Data S2, S5).

• Another potential way to prioritize candidate genes within devQTL peaks would be to use the RNA seq data. The authors should perform differential expression analysis between Kaga and Cab RNA-seq datasets. Do any of the differentially expressed genes fall within the devQTL peaks?

As suggested we have performed this additional experiment and report the RNAseq differential analysis in new Supplement Figure 7C-D. The analysis revealed 2606 differentially expressed genes in the PSM between Kaga and Cab, five of which were candidate genes from the devQTL analysis. We now tested all of these (5 in total, 4 new and 1 previously targeted adgrg1) for segmentation timing by CRISPR/Cas9 KO in the her7-venus background, none of which showed a timing phenotype (new Supplementary Figure 7F-F'). We provide the complete set of results in new Supplementary Figure 7 , Supplementary Data file 3 (DE-genes), all data were deposited on publicly available repository Biostudies under accession number: E-MTAB-13927.

• The use of crispants to functionally test candidate genes is inappropriate. Crispants do not mimic the effect of divergent SNPs and therefore completely fail to prove causality. While it is completely understandable that Medaka fish are not amenable to the creation of multiple knock-in lines where divergent SNPs are interconverted between species, better justification is needed. For instance, is there enough data to suggest that the divergent alleles for the candidate genes tested are loss of function? Why was a knockout approach chosen as opposed to overexpression?

We agree with the reviewer that we do not address the causality of SNPs with the CRISPR/Cas9 KO approach we followed. And medaka does offer the genome editing capabilities to create tailored sequence modifications. So in principle, this can be done. In practice, however, we reasoned that any given SNP will contribute only partially to the observed phenotypes and combinatorial sequence edits are simply very laborious given the current state of the art in genome editing technologies. We therefore opted for an alternative proof of principle approach that aims to "to discover new functionally relevant genes", not SNPs.

-Along the same line, now that two candidate genes have been shown to modulate the clock period in crispants (mespb and pcdh10b), the authors should at least attempt to knock in the respective divergent SNPs for one of the genes. This is of course optional because it would imply several months of work, but it would significantly increase the impact of the study.

As above, this is in principle the correct rationale to follow though very time, cost and labour intensive. It is for the later practical consideration that we decided not to follow this option.

Minor Comments - It would be highly beneficial to describe the ecological differences between the two Medaka species. For example, do the northern O. sakaizumii inhabit a colder climate than the southern O. latipes? Is food more abundant or easily accessible for one species compared to the other? What, if anything, has been described about each species' ecology?

There are indeed differences in the ecology of both species, with the northern O.sakaizumii inhabiting a colder climate than the southern O. latipes. In addition, it is known that the breeding season is shorter in the north than the south, and also there is the fact that northern species have been shown to have a faster juvenile growth rate than southern species. While it would be premature to link those ecological factors to the timing differences we observe, we can certainly speculate. A line to this effect has been added to the main text (Page 5, line 28-30).

• The authors describe two different methods for quantifying segmentation clock period (mean vs. intercept). It is still unclear what is the difference between Figs. 3A (clock period), S4A (mean period) and S4B (intercept period). Is clock period just mean period? Are the data then shown twice? How do Fig. 3A and S4A differ?

The clock period shown in all the main figures is the intercept period, which was also used for the devQTL analysis. Both measurements (mean and intercept) are indeed highly correlated and we include both in supplement for completeness.

• devQTL as shorthand for developmental QTL should be defined in page 4 line 1 (where the term first appears), not later in line 12 of the same page.

Noted and corrected, we thank the reviewer for spotting this error.

• Python code for period quantification should be uploaded to Github and shared with reviewers.

All period quantification code that was used in this study was obtained from the publicly available tool Pyboat (https://www.biorxiv.org/content/10.1101/2020.04.29.067744v3). All code that is used in PyBoat is available from the Github page of the creator of the tool (https://github.com/tensionhead/pyBOAT). Both are linked in the references and materials and methods sections.

• RNA-seq data should be uploaded to a publicly accessible repository and the reviewer token shared with reviewers.

We have uploaded all RNA-sequencing Data to public repository BioStudies under accession numbers : E-MTAB-13927, E-MTAB-13928. This information is now also added to material and methods in the manuscript text.

Why are the maintenance (27-28C) vs. imaging (30C) temperatures different?

Medaka fish have a wide range of temperatures they can physiologically tolerate, i.e. 17-33. The temperature 30C was chosen for practical reasons, i.e. a slightly faster developmental rate enables higher sample throughput in overnight real-time imaging experiments.

• For Crispants, control injections should have included a non-targeting sgRNA control instead of simply omitting the sgRNA.

We agree a non-targeting sgRNA control can be included, though we choose a different approach. For clarity, we now also include a control targeting Oca2, a gene involved in the pigmentation of the eye to probe for any injection related effect on timing and PSM size. As expected, 3 sgRNAs + Cas9 against Oca2 had no impact on timing or PSM size. This data is now shown in new Supplementary Figure 9 F-G'.

It is difficult to keep track of the species and strains. It would be most helpful if Fig. S1 appeared instead in main figure 1.

We agree and included an overview of the phylogenetic relationship of all species and their geographical locales in new Figure 1 A-B.

Significance

• The study introduces a new way of thinking about segmentation timing and size scaling by considering natural variation in the context of selection. This new framing will have an important impact on the field.
• Perhaps the most significant finding is that the correlation between segment timing and size in wild populations is driven not by developmental constraints but rather selection pressure, whereas segment size scaling does form a single developmental module. This finding should be of interest to a broad audience and will influence how researchers in the field approach future studies.
• It would be helpful to add to the conclusion the author's opinion on whether segmentation timing is a quantitative trait based on the number of QTL peaks identified.
• The authors should be careful not to assign any causality to the candidate genes that they test in crispants.
• The data and results are generally well-presented, and the research is highly rigorous.
• Please note I do have the expertise to evaluate the statistical/bioinformatic methods used for devQTL mapping.

Reviewer #2

Evidence, reproducibility and clarity

Seleit et al. investigate the correlation between segment size, presomitic mesoderm and the rhythm of periodic oscilations in the segmentation clock of developing medaka fish. Specifically, they aim to identify the genetic determinants for said traits. To do so, they employ a common garden approach and measure such traits in separate strains (F0) and in interbreedings across two generations (F1 and F2). They find that whereas presomitic mesoderm and segment size are genetically coupled, the tempo of her7 oscilations it is not. Genetic mapping of the F0 and F2 progeny allows them to identify regions associated to said traits. They go on an perturb 7 loci associated to the segmentation clock and X related to segment size. They show that 2/7 have a tempo defect, and 2/ affect size.

Major comments: The conclusions are convincing and well supported by the data. I think the work could be published as is in its current state, and no additional experiments that I can think of are needed to support the claims in the paper.

Minor comments: - The authors could provide a more detailed characterization of the identified SNPs associated to the clock and to PSM size. For the segmentation clock, the authors identify 46872 SNPs, most of which correspond to non-coding regions and are associated to 57 genes. They narrow down their approach to those expressed in the PSM of Cab Kaga. Was the RNA selected from F1 hybrids? I wonder if this would impact the analysis for tempo and or size in any way, as F2 are derived from these, and they show broader variability in the clock period than the F0 and F1 fishes.

The RNA was obtained from the pure F0 strains and we have now extended this analysis by deep bulk-RNA sequencing and differential gene expression analysis. As indicated also to reviewer 1, this revealed 2606 differentially expressed genes in the unsegmented tails of Kaga and Cab embryos, some of which occurred in devQTL peaks. Based on this information we expanded our list of CRISPR/Cas9 KOs by targeting all differentially expressed genes (5 in total, 4 new and 1 previously targeted) for segmentation timing, none of which showed a timing phenotype (new Supplementary figure 7C-D). We provide the complete set of results in new Supplementary Figure 7, Supplementary Data file 3 (DE-genes). All data were deposited on publicly available repository Biostudies under accession number: E-MTAB-13927.

It would be good if the authors could discuss if there were any associated categories or overall functional relationships between the SNPs/genes associated to size. And what about in the case of timing?

In the case of PSM size there were no clear GO terms or functional relationships between the genes that passed the significance threshold on chromosome 3.

For the 35 genes related to segmentation timing, there were a number of GO enrichment terms directly related to somitogenesis. We have included the GO analysis in the new Supplementary Figure 7E.

• Have any of the candidate genes or regulatory loci been associated to clock defects (57) or segment size (204) previously in the literature?

To the best of our knowledge none of the genes have been associated with clock or PSM size defects so far. It might be worthwhile using our results to probe their function in other systems enabling higher throughput functional analysis, such as newly developed organoid models.

• When the authors narrow down the candidate list, it is not clear if the genes selected as expressed in the PSM are tissue specific. If they are, I wonder if genes with ubiquitous expression would be more informative to investigate tempo of development more broadly. It would be good if the authors could specifically discuss this point in the manuscript.

We have not addressed the spatial expression pattern of the 35 identified PSM genes in this study, so we cannot speculate further. But the reviewer raises an important point, how timing of individual processes (body axis segmentation) are linked at organismal scale is indeed a fundamental, additional, question that will be addressed in future studies, indeed the in-vivo context we follow here would be ideal for such investigations.

Can the authors speculate mechanistically why mespb or pchd10b accelerates the period of her7 oscillations?

While we do not have a mechanistic explanation yet, an additional experiment we performed, i.e. bulk-RNAsequencing on WT and mespb mutant tails, provided additional insight, we now added this data to the manuscript . This analysis revealed 808 differentially expressed genes between wt and mespb mutants. Interestingly, many of these affected genes are known to be expressed outside of the mespb domain, i.e. in the most posterior PSM (i.e. tbxt, foxb1,msgn1, axin2, fgf8, amongst others). This indicates that the effect of mespb downregulation is widespread and possibly occurs at an earlier developmental stage. This requires more follow up studies. This data is now shown in new Supplementary figure 9A, Supplementary Data file S4. We now comment on this point in the revised manuscript.

• Are there any size difference associated to the functionally validated clock mutants?

We addressed this point directly and added this analysis as supplementary Figure 9H-H'. While pcdh10b mutants do not show any detectable difference in PSM size, we find a small, statistically significant reduction in PSM size (area but not length) in mespb mutants. All this data is now included in the revised manuscript.

-Ref 27 shows a lack of correlation between body size and the segmentation period in various species of mammals. The work supports their findings, and it would be good to see this discussed in the text.

We are not certain how best to compare our in-vivo results in externally developing fish embryos to in-vitro mammalian 2-D cell cultures. In our view, the correlation of embryo size, larval and adult size that we find in Oryzias might not necessarily hold in mammalian species, which would make a comparison more difficult. We do cite the work mentioned so the reader is pointed towards this interesting, complementary literature.

Significance

The work is quite remarkable in terms of the multigenerational genetic analysis performed. The authors have analysed >600 embryos from three separate generations to obtain quantitative data to answer their question (herculean task!). Moreover, they have associated this characterization to specific SNPs. Then, to go beyond the association, they have generated mutant lines and identified specific genes associated to the traits they set out to decipher.

To my knowledge, this is the first project that aims to identify the genetic determinants for developmental timing. Recent work on developmental timing in mammals has focused on interspecies comparisons and does not provide genetic evidence or insight into how tempo is regulated in the genome. As for vertebrates, recent work from zebrafish has profiled temperature effects on cell proportions and developmental timing. However, the genetic approach of this work is quite elegant and neat.

Conceptually, it is quite important and unexpected that overall size and tempo are not related. Body size, lifespan, basal metabolic rates and gestational period correlate positively and we tend to think that mechanistically they would all be connected to one another. This paper and Lazaro et al. 2023 (ref 27) are one of the first in which this preconception is challenged in a very methodical and conclusive manner. I believe the work is a breakthrough for the field and this work would be interesting for the field of biological timing, for the segmentation clock community and more broadly for all developmental biologists.

My field is quantitative stem cell biology and I work on developmental timing myself, so I acknowledge that I am biased in the enthusiasm for the work. It should be noted that as an expert on the field, I have identified instances where other work hasn't been as insightful or well developed in comparison to this piece. It is also worth noting that I am not an expert in fish development, phylogenetic studies or GWAS analyses, so I am not capable to asses any pitfalls in that respect.

__Reviewer #3 (Evidence, reproducibility and clarity (Required)): __

__Summary: __

This manuscript explores the temporal and spatial regulation of vertebrate body axis development and patterning. In the early stages of vertebrate embryo development, the axial mesoderm (presomitic mesoderm - PSM) undergoes segmentation, forming structures known as somites. The exact genetic regulation governing somite and PSM size, and their relationship to the periodicity of somite formation remains unclear.

To address this, the authors used two evolutionarily closely related Medaka species, Oryzias sakaizumii and Oryzias latipes, which, although having distinct characteristics, can produce viable offspring. Through analysis spanning parental (generation F0) and offspring (generations F1 and F2) generations, the authors observed a correlation between PSM and somite size. However, they found that size scaling does not correlate with the timing of somitogenesis.

Furthermore, employing developmental quantitative trait loci (devQTL) mapping, the authors identified several new candidate loci that may play a role during somitogenesis, influencing timing of segment formation or segment size. The significance of these loci was confirmed through an innovative CRISPR-Cas9 gene editing approach.

This study highlights that the spatial and temporal aspects of vertebrate segmentation are independently controlled by distinct genetic modular mechanisms.

__Major comments: __

1) In the main text page 3, lines 11 and 12, the authors state that the periodicity of the embryo clock of the F1 generation is the intermediate between the parental F0 lineages. However, the authors look only at the periodicity of the Cab strain (Oryzias latipes) segmentation clock. The authors should have a reporter fish line for the Kaga strain (Oryzias sakaizumii) to compare the segmentation clock of both parental strains and their offspring. Since it could be time consuming and laborious, I advise to alternatively rephrase the text of the manuscript.

We agree a careful distinction between segment forming rate (measured based on morphology) and clock period (measured using the novel reporter we generated) is essential. We show that both measures correlate very well in Cab, in both F0 and F1 and F2 carrying the Cab allele. For Kaga F0, we indeed can only provide the rate of somite formation, which nevertheless allows comparison due to the strong correlation to the clock period we have found. We have rephrased the text accordingly.

2) It is evident that only a few F0 and F1 animals were analyzed in comparison with the F2 generation. Could the authors kindly explain whether and how this could bias or skew the observed results?

We provide statistical evidence through the F-test of equality that the variances between the F0, F1 and F2 samples are equal. Additionally if we sub-sample and separate the F2 data into groups of 100 embryos (instead of all 638) we get the same distribution of the F2s. We therefore believe that this is sufficient evidence against a bias or skew in the results.

3) It would be interesting to create fish lines with the validated CRISPR-Cas9 gene manipulations in different genetic contexts (Cab or Kaga) to analyze the true impact on the segmentation clock and/or PSM & somite sizes.

We agree with the reviewer this would in principle be of interest indeed, please see our response to reviewer 1 earlier.

4) Please add the results of the Go Analysis as supplementary material.

We have added the GO analysis in new Supplementary Figure 7E.

__Minor comments: __

1) In the main text, page 2, line 29, Supplementary Figure 1D should be referenced.

We have added a clearer phylogeny and geographical location of the different species in new Figure 1 A-B. And reference it at the requested location.

2) In the main text, page 2, line 32, the authors refer to Figure 1B, but it should be 1C.

We have corrected the information.

3) Regarding the topic "Correlation of segmentation timing and size in the Oryzias genus" the authors should also give information on the total time of development of the different Oryzias species, as well as the total number of formed somites.

We follow this recommendation and have added this information in new Supplementary Figure 5. We also now include segment number measured in F2 embryos. We indeed view segmentation rate as a proxy for developmental rate, which however needs to be distinguished from total developmental time. The latter can be measured for instance by quantifying hatching time, which we did. These measurements show that Kaga, Cab and O.hubbsi embryos kept at constant 28 degrees started hatching on the same day while O.minutillus and O.mekongensis embryos started hatching one day earlier. We have not included this data in the manuscript because we think a distinction should be made between rate of development and total development time.

4) In Figures 3A and B, please add info on the F1 lines for comparison.

The information on F1 lines is provided in Supplementary Figure 3

5) Supplementary Figures 2F shows that the generation F1 PSM is similar to Cab F0, and not an intermediate between Kaga F0 and Cab F0. This is interesting and should be discussed.

We show that the F1 PSM is indeed closer to the PSM of Cab than it is to the Kaga PSM. This is indeed intriguing and we have now commented on this point directly in the text.

6) Supplementary Figures 6C to H are not mentioned either in the main text or in the extended information. Please add/mention accordingly.

We have added references to both in the text

7) The order of Supplementary Figure 8 E to H and A to D appears to be not correct and not following the flow of the text. Please update/correct accordingly.

We have updated the text accordingly.

8) The authors should choose between "Fig.", "Fig", "fig.", "fig" or "Figure". All 'variants' can be found in the text.

Noted, and updated. Fig. is used for main figures and fig. is used for supplementary figures.

9) The color scheme of several figures (graphs with colored dots) should be revised. Several appear to be difficult to discern and analyze.

We have enhanced the colours and increased the font on the figure panels. The colour panel was chosen to be colour-blind friendly.

10) Please address/discuss following questions: What are the known somitogenesis regulating genes in Medaka? How do they correlate with the new candidates?

The candidates we found and tested had not been implicated in regulating the tempo of segmentation or PSM size, while for some a role in somite formation had been previously established, hence the enrichment in GO analysis Somitogenesis.

Reviewer #3 (Significance (Required)):

General assessment:

This interesting manuscript describes a novel approach to study and find new players relevant to the regulation of vertebrate segmentation. By employing this innovative methodology, the authors could elegantly demonstrate that the segmentation clock periodicity is independent from the sizes of the PSM and forming somites. The authors were further able to find new genes that may be involved in the regulation of the segmentation clock periodicity and/or the size of the PSM & somites. A limitation of this study is the fact that the results mainly rely on differences between the two species. The integration of additional Medaka species would be beneficial and may help uncover relevant genes and genetic contexts.

Advance:

To my best knowledge this is the first time that such a methodology was employed to study the segmentation clock and axial development. Although the topic has been extensively studied in several model organisms, such as mice, chicken, and zebrafish, none of them correlated the size of the embryonic tissues and the periodicity of the embryo clock. This study brings novel technological and functional advances to the study of vertebrate axial development.

Audience:

This work is particularly interesting to basic researchers, especially in the field of developmental biology and represents a fresh new approach to study a core developmental process. This study further opens the exciting possibility of using a similar methodology to investigate other aspects of vertebrate development. It is a timely and important manuscript which could be of interest to a wider scientific audience and readership.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

---

Referee #1

Evidence, reproducibility and clarity

Seleit and colleagues set out to explore the genetics of developmental timing and tissue size by mapping natural genetic variation associated with segmentation clock period and presomitic mesoderm (PSM) size in different species of Medaka fish. They first establish the extent of variation between five different Medaka species of in terms of organismal size, segmentation rate, segment size and presomitic mesoderm size, among other traits. They find that these traits are species-specific but strongly correlated. In a massive undertaking, they then perform developmental QTL mapping for segmentation clock period and PSM size in a set of ~600 F2 fish resulting from the cross of Orizyas sakaizumii (Kaga) and Orizyas latipes (Cab). Correlation between segmentation period and segment size was lost among the F2s, indicating that distinct genetic modules control these traits. Although the researchers fail to identify causal variants driving these traits, they perform proof of concept perturbations by analyzing F0 Crispants in which candidate genes were knocked out. Overall, the study introduces a completely new methodology (QTL mapping) to the field of segmentation and developmental tempo, and therefore provides multiple valuable insights into the forces driving evolution of these traits.

Major comments:

• The first sentence in the abstract reads "How the timing of development is linked to organismal size is a longstanding question". It is therefore disappointing that organismal size is not reported for the F2 hybrids. Was larval length measured in the F2s? If so, it should be reported. It is critical to understand whether the correlation between larval size and segmentation clock period is preserved in F2s or not, therefore determining if they represent a single or separate developmental modules. If larval length data were not collected, the authors need to be more careful with their wording. In the current version of the paper, organismal size is often incorrectly equated to tissue size (e.g. PSM size, segment size). For example, in page 3 lines 33-34, the authors state that faster segmentation occurred in embryos of smaller size (Fig. 1D). However, Fig. 1D shows correlation between segmentation rate and unsegmented PSM area. The appropriate data to show would be segmentation rate vs. larval or adult length.
• Is my understanding correct in that the her7-venus reporter is carried by the Cab F0 but not the Kaga F0? Presumably only F2s which carried the reporter were selected for phenotyping. I would expect the location of the reporter in the genome to be obvious in Figure 3J as a region that is only Cab or het but never Kaga. Can the authors please point to the location of the reporter?
• devQTL mapping in this study seems like a wasted opportunity. The authors perform mapping only to then hand pick their targets based on GO annotations. This biases the study towards genes known to be involved in PSM development, when part of the appeal of QTL mapping is precisely its unbiased nature and the potential to discover new functionally relevant genes. The authors need to better justify their rationale for candidate prioritization from devQTL peaks. The GO analysis should be shown as supplemental data. What criteria were used to select genes based on GO annotations?
• Analysis of the predicted functional consequence of divergent SNPs (Fig. S6B, F) is superficial. Among missense variants, which genes harbor the most deleterious mutations? Which missense variants are located in highly conserved residues? Which genes carry variants in splice donors/acceptors? Carefully assessing the predicted effect of SNPs in coding regions would provide an alternative, less biased approach to prioritize candidate genes.
• Another potential way to prioritize candidate genes within devQTL peaks would be to use the RNA seq data. The authors should perform differential expression analysis between Kaga and Cab RNA-seq datasets. Do any of the differentially expressed genes fall within the devQTL peaks?
• The use of crispants to functionally test candidate genes is inappropriate. Crispants do not mimic the effect of divergent SNPs and therefore completely fail to prove causality. While it is completely understandable that Medaka fish are not amenable to the creation of multiple knock-in lines where divergent SNPs are interconverted between species, better justification is needed. For instance, is there enough data to suggest that the divergent alleles for the candidate genes tested are loss of function? Why was a knockout approach chosen as opposed to overexpression?
• Along the same line, now that two candidate genes have been shown to modulate the clock period in crispants (mespb and pcdh10b), the authors should at least attempt to knock in the respective divergent SNPs for one of the genes. This is of course optional because it would imply several months of work, but it would significantly increase the impact of the study.

Minor Comments

• It would be highly beneficial to describe the ecological differences between the two Medaka species. For example, do the northern O. sakaizumii inhabit a colder climate than the southern O. latipes? Is food more abundant or easily accessible for one species compared to the other? What, if anything, has been described about each species' ecology?
• The authors describe two different methods for quantifying segmentation clock period (mean vs. intercept). It is still unclear what is the difference between Figs. 3A (clock period), S4A (mean period) and S4B (intercept period). Is clock period just mean period? Are the data then shown twice? How do Fig. 3A and S4A differ?
• devQTL as shorthand for developmental QTL should be defined in page 4 line 1 (where the term first appears), not later in line 12 of the same page.
• Python code for period quantification should be uploaded to Github and shared with reviewers.
• RNA-seq data should be uploaded to a publicly accessible repository and the reviewer token shared with reviewers.
• Why are the maintenance (27-28C) vs. imaging (30C) temperatures different?
• For Crispants, control injections should have included a non-targeting sgRNA control instead of simply omitting the sgRNA.
• It is difficult to keep track of the species and strains. It would be most helpful if Fig. S1 appeared instead in main figure 1.

Significance

• The study introduces a new way of thinking about segmentation timing and size scaling by considering natural variation in the context of selection. This new framing will have an important impact on the field.
• Perhaps the most significant finding is that the correlation between segment timing and size in wild populations is driven not by developmental constraints but rather selection pressure, whereas segment size scaling does form a single developmental module. This finding should be of interest to a broad audience and will influence how researchers in the field approach future studies.
• It would be helpful to add to the conclusion the author's opinion on whether segmentation timing is a quantitative trait based on the number of QTL peaks identified.
• The authors should be careful not to assign any causality to the candidate genes that they test in crispants.
• The data and results are generally well-presented, and the research is highly rigorous.
• Please note I do have the expertise to evaluate the statistical/bioinformatic methods used for devQTL mapping.

Multifactor authentication is a system where a site will only allow access to the site when 2 or more pieces of authenticating evidence are presented. this may come in the form of a password along with a code sent through SMS or email. This allows for a site to be more secure when multiple factors of authentication are presented to avoid unwanted access.

Both are correct. This aspirational ideal - crafting a program with a small, tight, and beautiful core - is possible if a program is intended to be an artifact.

One definition of an artifact - a program designed to serve a specific use case in a specific point of time forever. It is crafted then left untouched.

By contrast, software to most businesses is a living, breathing beast - we have strict time constraints to implement, modify, adjust to, and tack on features or the business dies. This business of crafting a perfect, beautiful core would require a rewrite of the entire system every time you intended to add a new feature or reinvestigate the model.

Software engineering is, then, a process of compromising - continuously declaring that edge X is the one least likely to shoot yourself in the foot.

https://expressjs.com/en/5x/api.html#res.redirect:~:text=res.redirect(%5Bstatus,redirect%20vulnerabilities.

Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

Learn more at Review Commons

---

Reply to the reviewers

Manuscript number: RC-2023-02218R

Corresponding author(s): Steven, McMahon

1. General Statements [optional]

*We were pleased to receive the encouraging critiques and very much appreciate the Reviewer's specific comments and suggestions. In this revised version of our manuscript, we have made a number of substantive additions and modifications in response to these comments/suggestions. We hope you agree that the study is now improved to the point where it is suitable for publication. *

2. Point-by-point description of the revisions

Reviewer #1 (Evidence, reproducibility and clarity (Required)):

Summary This study describes efforts to characterize differences in the roles of the two related human decapping factors Dcp1a and Dcp1b by assessing mRNA decay and protein associations in knockdown and knockout cell lines. The authors conclude that these proteins are non-redundant based on the observations that loss of DCp1a versus Dcp1b impacts the decapping complex (interactome) and the transcriptome differentially.

Major comments • While the experiments appear to be well designed and executed and the data of generally high quality, the conclusions are drawn without sufficient consideration for the fact that these two proteins form a heterotrimeric complex. The authors assume that there are distinct homotrimeric complexes rather than a single complex with both proteins in. Homotrimers may have new/different functions not normally seen when both proteins are expressed. Thus while it is acceptable to infer that the functions of these two proteins within the decapping complex are distinct, it is not clear that they act separately, or that complexes naturally exist without one or the other. A careful evaluation of the relative ratios of Dcp1a and b overall and in decapping complexes would be informative if the authors want to make stronger statements about the roles of these two factors.

RESPONSE: Thank you for this valuable comment. We have substantially edited the manuscript to incorporate these points. Examples include a detailed analysis of iBAQ values for the DDX6, DCP1a, and DCP1b interactomes (which now allows us to estimate the ratios of DCP1a and DCP1b in these complexes) and cellular fractionation to interrogate complex integrity (using Superose 6).

• The concept of buffering is not adequately introduced and the interpretation of observations that RNAs with increased half life do not show increased protein abundance - that Dcp1a/b are involved in transcript buffering is nebulous. In order to support this interpretation, the mRNA abundances (NOT protein abundances) should be assessed, and even then, there is no way to rule out indirect effects. RESPONSE: Thank you for this comment. In the revised version of the manuscript, we introduced the concept of transcript buffering at an earlier stage as one of the potential explanations for our findings. We were also able to use a new algorithm (grandR) to estimate half-lives and synthesis rates from our data. These new data add strength to the argument that DCP1a and DCP1b are linked to transcript buffering pathways.

• It might be interesting to see what happens when both factors are depleted to get an idea of the overall importance of each one.

RESPONSE: In our work we tried to emphasize the differences between the two paralogs. We believe that doing double knockout or knockdown would mask the distinct impacts of the paralogs. In data not included in this study, we have shown that cells lacking both DCP1a and DCP1b are viable. We did check PARP cleavage in the CRISPR generated cell pools of DCP1a KO, DCP1b KO, and the double KO. The WB measuring the PARP cleavage is shown in the supplemental material (Supplementary Material: Replicates)

• The algorithms etc used for data analysis should be included at the time of publication. Version number and settings used for SMART to define protein domains, and webgestalt should be indicated

RESPONSE: We apologize for this oversight. Version number and settings used for the webtools (SMART, Webgestalt) are now included. The analysis pipeline for half-lives and synthesis rates estimation as well as all the files and the code needed to generate the figures in the paper are available on zenodo (https://zenodo.org/records/10725429).

• Statistical analysis is not provided for the IP experiments, the number of replicates performed is not indicated and quantification of KD efficiency are not provided.

RESPONSE: The number of replicates performed in each experiment is now clearly indicated and quantifications of knockdown efficiency are provided (Supplemental Figure 3A and 3B, Figure 3A, Figure 3B).

• The possibility that the IP Antibody interferes with protein-protein interactions is not mentioned.

RESPONSE: Thank you for this comment. The revised manuscript includes a discussion of the antibody epitope location and the potential for impact on protein-protein interactions.

Minor comments • P4 - "This translational repression of mRNA associated with decapping can be reversed, providing another point at which gene expression can be regulated (21)" - implies that decapping can be reversed or that decapped RNAs are translated. I don't think this is technically true.

RESPONSE: There have been several studies that document the reversal of decapping. These findings are summarized in the following reviews.

Schoenberg, D. R., & Maquat, L. E. (2009). Re-capping the message. Trends in biochemical sciences, 34(9), 435-442.

Trotman, J. B., & Schoenberg, D. R. (2019). A recap of RNA recapping. Wiley Interdisciplinary Reviews: RNA, 10(1), e1504.

• P11 - how common is it for higher eukaryotes to have 2 DCP genes? *RESPONSE: Metazoans have 2 DCP1 genes. *

• Fig S1 - says "mammalian tissues" in the text but the data is all human. The statement that "expression analyses revealed that DCP1a and DCP1b have concordant rather than reciprocal expression patterns across different mammalian tissues (Supplemental Figure 1)" is a bit misleading as no evidence for correlation or anti-correlation is provided. Also co-expression is not strong support for the idea that these genes have non-redundant functions. Both genes are just expressed in all tissues - there's no evidence provided that they are concordantly expressed. In bone marrow it may be worth noting that one is high and the other low - i.e. reciprocal. *RESPONSE: We appreciate this comment. We have corrected the interpretation of the aforementioned dataset. We have also incorporated a more detailed discussion in the text of the paper. As the Reviewer pointed out, there are a subset of tissues where their expression appears to be reciprocal. *

• Fig 1A - it is not clear what the different colors mean. Does Sc DCP1 have 1 larger EVH or 2 distinct ones. Are the low complexity regions in Sc DCP2 the SLiMs. *RESPONSE: Thank you for this comment. We have corrected this ambiguity to reflect that Sc DCP1 has one EVH1 domain that is interconnected by a flexible hinge. The low-complexity regions typically contain short linear motifs (SLIMs), however, not all low-complexity regions have been verified to contain them. In the figure, only low-complexity regions are shown. The text of the paper refers only to verified SLIMs . *

• P11 - why were HCT116 cells selected? RESPONSE: HCT116 cells are an easily transfectable human cell line and have been widely used in biochemical and molecular studies, including studies of mRNA decapping (see references below). Since decapping is impacted by viral proteins we avoided the use of other commonly used cell models such as HEK293T or HeLa.

https://pubmed.ncbi.nlm.nih.gov/?term=decapping+hct116&sort=date&size=200

• Fig 1B - what are the asterisks by the RNA names? Might be worth noting that over-expression of DCP1b reduced IP of DCP1a. There's no quantification and no indication of the number of times this experiment was repeated. Data from replicates and quantification of the knockdown efficiency in each replicate would be nice to see. *RESPONSE: Thank you for this comment. Asterisks indicate that those bands were from a second gel, as DCP1a and DCP1b run at approximately the same molecular weight. We have now included a note in our figure legend to indicate this. The knockdown efficiency is provided (Figure 3 and Supplemental Figure 3). We also noted the number of replicas for each IP in figure 1. The replicas are provided as supplementary material (Supplementary Materials: Replicates). *

• Fig 1C/1D - why are there 3 bands in the DCP1a blot? Quantification of the IP bands is necessary to say whether there is an effect or not of over-expression/KO. RESPONSE: The additional bands in DCP1a blots are background. When we stained the whole blot for DCP1a, in cells which with complete DCP1a KO cells (clone A3), these bands still appear (Supplementary Material: Validation of the KO clones). Quantifications of the bands in the overexpression experiments is now provided.

• Fig 3 - is it possible that differences are due to epitope positions for the antibodies used for IP? RESPONSE: We do not believe so. DCP1a antibody binds roughly 300-400 residues on DCP1a, and DCP1b antibody binds around Val202. Antibodies therefore do not bind DCP1a or DCP1b low-complexity regions (which are largely responsible for interacting with the decapping complex interactome). Antibodies don't bind the EVH1 domains or the trimerization domain, which are needed for their interaction with DCP2 and each other.

• Fig 5A - the legend doesn't match the colors in the figure. It is not clear how the pRESPONSE: Thank you for this comment. We have corrected this issue in the revised version of the paper. High-confidence proteins are those with pRESPONSE: Thank you for this comment. We have corrected this issue in the revised version of the paper.*

• There are a few more recent studies on buffering that should be cited and more discussion of this in the introduction is necessary if conclusions are going to be drawn about buffering. *RESPONSE: We have included a discussion of transcript buffering in the introduction. *

• The heatmaps in figure 2 are hard to interpret. RESPONSE: To clarify the heatmaps, we included a more detailed description in the figure legends, have enlarged the heatmaps themselves, and have added more extensive labeling.

Reviewer #1 (Significance (Required)):

• Strengths: The experiments appear to be done well and the datasets should be useful for the field. • Limitations: The results are overinterpreted - different genes are affected by knocking down one or other of these two similar proteins but this does not really tell us all that much about how the two proteins are functioning in a cell where both are expressed. • Audience: This study will appeal most to a specialized audience consisting of those interested in the basic mechanisms of mRNA decay. Others may find the dataset useful. • This study might complement and/or be informed by another recent study in BioRXiv - https://doi.org/10.1101/2023.09.04.556219 • My field of expertise is mRNA decay - I am qualified to evaluate the findings within the context of this field. I do not have much experience of LC-MS-MS and therefore cannot evaluate the methods/analysis of this part of the study.

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

The authors provide evidence that Dcp1a and Dcp1b - two paralogous proteins of the mRNA decapping complex - may have divergent functions in a cancer cell line. In the first part, the authors show that interaction of Dcp2 with EDC4 is diminished upon depletion of Dcp1a but not affected by depletion of Dcp1b. The results have been controlled by overexpression of Dcp1b as it may be limiting factor (i.e. expression levels too low to compensate for depletion of Dcp1a reduced interaction with EDC3/4 while depletion of Dcp1b lead to opposite and increase interactions). They then defined the protein interactome of DDX6 in parental and Dcp1a or Dcp1b depleted cells. Here, the authors show some differential association with EDC4 again, which is along results shown in the first part. The authors further performed SLAM-seq and identified subsets of mRNA whose decay rates are common but also different upon depletion with Dcp1a and Dcp1b. Interestingly, it seems that Dcp1a preferentially targets mRNAs for proteins regulating lymphocyte differentiation. To further test whether changes in RNA decay rates are also reflected at the protein levels, they finally performed an MS analysis with Dcp1a/b depleted cells. However no significant overlap with mRNAs showing altered stability could be observed; and the authors suggested that the lack of congruence reflects translational repression.

Major comments: 1. While functional difference between Dcp1a and Dcp1b are interesting and likely true, there are overinterpretations that need correction or further evidence for support. Sentences like "DCP1a regulates RNA cap binding proteins association with the decapping complex and DCP1b controls translational initiation factors interactions (Figure 2E)" sound misleading. While differential association with proteins has been recognised with MS-data, it does not necessary implement an active process of control/regulation. To make the claim on 'control/regulation', and inducible system or introduction of mutants would be required.

RESPONSE: This set of comments were particularly useful in helping us refine the presentation of our findings. We have edited our manuscript to be more specific about the limits of our data.

1. The MS analysis is not clearly described in the text and it is unclear how authors selected high-confident proteins. The reader needs to consider the supplemental tables to find out what controls were used. Furthermore, the authors should show correlation plots of MS data between replicates. For instance, there seems to be limited correlation among some of the replicates (e.g. Dcp1b_ko3 sample, Fig. 2c). Any explanation in this variance?

*RESPONSE: We have now included a clear description of how all high-confidence proteins were selected in the Methods and Results sections. The revised manuscript also includes a more thorough description of the controls used and the number of replicates for individual experiments. The PCA plots have now been included where appropriate. The variance in this sample is likely technical. *

1. GO analysis for the proteome analysis should consider the proteome and not the genome as the background. The authors should also indicate the corrected P-values (multiple testing) FDRs.

*RESPONSE: Webgestalt uses a reference set of IDs to recognize the input IDs, and it does not use it for the background analysis in the classical sense. We repeated a subset of our proteome analyses using the 'genome-protein coding' as background and obtained the same result as in our original analysis. All ontology analyses now include raw p-values and/or FDRs when appropriate. *

1. Fig 2E. The figures display GO enrichments needs better explanation and additional data can be added. The enrichment ratio is not explained (is this normalised?) and p-values and FDRs, number of proteins in respective GO category should be added. *RESPONSE: More thorough explanations of the GO enrichments are now included. The supplemental data contains all p-values (raw and adjusted), as well as the number of proteins in each GO category. The Enrichment ratio is normalized and contains information about the number of proteins that are redundant in multiple groups. GO Ontology analyses are now displayed with p-values and/or FDR values, and in this case the enrichment ratio contains information regarding the number of proteins found in our input set and the number of expected proteins in the GO group. The network analysis shows the FDR values and the number of proteins found in the groups compared. *

Minor: 5. These studies were performed in a colorectal carcinoma cell line (HCT116). The authors should justify the choice of this specialised cell line. Furthermore, one wonders whether similar conclusions can be drawn with other cell lines or whether findings are specific to this cancer line.

RESPONSE: The study that is currently in pre-print in BioRxiv (https://doi.org/10.1101/2023.09.04.556219*) utilized HEK293Ts and found similar results to ours when examining the various relationships between the core decapping core members. *

1. Fig. 1B. It is unclear what DCP1b* refers to? There are bands of different size that are not mentioned by the authors - are those protein isoforms or what are those referring to? A molecular marker should be added to each Blots. Uncropped Western images and markers should be provided in the Supplement. *RESPONSE: The asterisk indicates that these images came from a second western blot gel (DCP1a and DCP1b have a similar molecular weight and cannot be probed on the same membrane). Uncropped western blot images and markers (as available) are provided in the supplement. *

2. MS data submitted to public repository with access. No. indicated in the manuscript.

RESPONSE: MS data is submitted as supplementary datasets to the paper. It contains the analyzed data as well as the LCMSMS output. We are in the process of submitting the raw LSMSMS data to a public repository.

Fig 3. A Venn Diagram displaying the overlap of identified proteins should be added. GO analysis should be done considering the proteome as background (as mentioned above).

*RESPONSE: A Venn diagram showing the overlap among the proteins identified is now included in the revised version. *

Reviewer #2 (Significance (Required)):

Overall, this is a large-scale integrative -omics study that suggest functional difference between Dcp1 paralogues. While it seems clear that both paralogous have some different functions and impact, there are overinterpretations in place and further evidence would to be provided to substantiate conclusions made in the paper. For instance, while the interactions with Dcp2/Ddx6 in the absence of Dcp1a,b with EDC4/3 may be altered (Fig. 1, 2), the functional implications of this changed associations remains unresolved and not further discussed. As such, it remains somehow disconnected with the following experiments and compromises the flow of the study. The observed differences in decay-rates for distinct functionally related sets of mRNAs is interesting; however, it remains unclear whether those are direct or rather indirect effects. This is further obscured by the absence of any correlation to changes in protein levels, which the authors interpreted as 'transcriptional buffering'. In this regard, it is puzzling how the authors can make a statement about transcriptional buffering? While this may be an interesting aspect and concept of the discussion, there is no primary data showing such a functional impact.

As such, the study is interesting as it claims functional differences between DCP1a/b paralogous in a cancer cell line. Nevertheless, I am not sure how trustful the MS analysis and decay measurements are as there is not further validation. It woudl be interesting if the authors could go a bit further and draw some hypothesis how the selectivty could be achieved i.e interaction with RNA-binding proteins that may add some specificity towards the target RNAs for differential decay. As such, the study remains unfortunately rather descriptive without further functional insight.

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

Review on "Non-redundant roles for the human mRNA decapping cofactor paralogs DCP1a and DCP1b" by Steven McMahon and co-workers mRNA decay is a critical step in the regulation of gene expression. In eukaryotes, mRNA turnover typically begins with the removal of the poly(A) tail, followed by either removal of the 5' cap structure or exonucleolytic 3'-5' decay catalyzed by the exosome. The decapping enzyme DCP2 forms a complex with its co-activator DCP1, which enhances decapping activity. Mammals are equipped with two DCP1 paralogs, namely DCP1a and DCP1b. Metazoans' decapping complexes feature additional components, such as enhancer of decapping 4 (EDC4), which supports the interaction between DCP1 and DCP2, thereby amplifying the efficiency of decapping. This work focuses on DCP1a and DCP1b and investigates their distinct functions. Using DCP1a- and DCP1a-specific knockdowns as well as K.O. cell lines, the authors find surprising differences between the DCP1 paralogs. While DCP1a is essential for the assembly of EDC4-containig decapping complexes and interactions with mRNA cap binding proteins, DCP1b mediates interactions with the translational machinery. Furthermore, DCP1a and DCP1b target different mRNAs for degradation, indicating that they execute non-overlapping functions. The findings reported here expand our understanding of mRNA decapping in human cells, shedding light on the unique contributions of DCP1a and DCP1b to mRNA metabolism. The manuscript tackles an interesting subject. Historically, the emphasis has been on studying DCP1a, while DCP1b has been deemed a functionally redundant homolog of DCP1a. Therefore, it is commendable that the authors have taken on this topic and, with the help of knockout cell lines, aimed to dissect the function of DCP1a and DCP1b. Despite recognizing the significance of the subject and approach, the manuscript falls short of persuading me. Following a promising start in Figure 1 (which still has room for improvement), there is a distinct decline in overall quality, with only relatively standard analyses being conducted. However, I do not want to give the authors a detailed advice on maximizing the potential of their data and presenting it convincingly. So, here are just a few key points for improvement: Figure 1C: Upon closer examination, a faint band is still visible at the size of DCP1a in the DCP1a knockout cells. Could this be leaky expression of DCP1a? The authors should provide an in-depth characterization of their cells (possibly as supplementary material), including identification of genomic changes (e.g. by sequencing of the locus) and Western blots with longer exposure, etc.

*RESPONSE: Thank you for this comment. The in-depth characterization of our cells is now included in the Supplementary Material. DCP1a KO cells and DCP1b KO cells indicated as single cell clones have been confirmed to have no DCP1a or DCP1b expression. In Figure 1D and Figure 3, polyclonal pool cells were used as indicated (only for DCP1a KO). *

Figure 2: It is great to see that the effects of the KOs are also visible in the DDX6 immunoprecipitation. However, I wonder if the IP clearly confirms that the KO cells indeed do not express DCP1a or DCP1b. In the heatmap in Figure 2B, it appears as if the proteins are only reduced by a log2-fold change of approximately 1.5? Additionally, Figure 2 shows a problem that persists in the subsequent figures. The visual presentation is not particularly appealing, and essential details, such as the scale of the heatmap in 2B (is it log2 fold?), are lacking.

*RESPONSE: The in-depth characterization of our cells is included in the Supplementary Materials and confirms the presence of single-cell clones where indicated. As noted above, only Figure 1D and Figure 3 used DCP1a KO pooled cells. The heatmap in Figure 2B is scaled by row using the pheatfunction in R studio. The actual data for the heatmap comes from protein intensities from the LC-MS/MS analysis. We have improved the visual presentation in the revised manuscript. *

Figure 3: I wonder why there are no primary data shown here, only processed GO analyses. Wouldn't one expect that DCP2 interacts mainly with DCP1a, but less with DCP1b? Is this visible in the data? Moreover, such analyses are rather uninformative (as reflected in the GO terms themselves, for instance, "oxoglutarate dehydrogenase complex" doesn't provide much meaningful insight). The authors should rather try to derive functional and mechanistic insights from their data.

RESPONSE: We have now revised this Figure to include primary data as well as the IP of DCP1a in DCP1b KO cells (single cell clones) and the IP of DCP1b in DCP1a KO cells (pooled cells). We identified EDC3 in the high-confidence protein pool. The EDC3:DCP1a interaction is enhanced in DCP1b KO cells. We also found that the EDC3:DCP1b interaction is less abundant in DCP1a KO cells. This is consistent with our data in Figures 1 and 2. DCP2 was not identified in the interactomes of either DCP1a or DCP1b. This is not unusual as DCP2 is highly flexible and the association between DCP1s with DCP2 is transient and facilitated by other proteins.

In Fig. 4 the potential of the approach is not fully exploited. Firstly, I would advocate for omitting the GO analyses, as, in my opinion, they offer little insight. Again, crucial information is missing to assess the results. While 75 nt reads are mentioned in the methods, the sequencing depth remains unspecified. Figure 4b should be included in the supplements. Furthermore, I strongly recommend concentrating on insights into the mechanisms of DCP1a and DCP1b-containing complexes. E.g. what characteristics distinguish DCP1a and DCP1b-dependent mRNAs? Are these targets inherently unstable? Why are they degraded? Are they known decapping substrates?

*RESPONSE: Thank you for this comment. We have now revised this figure and have included information about sequencing depth and other pertinent information. We have been able to use a newly available algorithm (grandR) and were able to estimate half-lives and synthesis rates. This is a significant addition to the paper. We were also able to compare significantly impacted mRNAs (by DCP1a or DCP1b loss) to the established DCP2 target list. *

In general, I suggest the authors revise the manuscript with a focus on the potential readers. Reduce Gene Ontology (GO) analyses and heatmaps, and instead, incorporate more analyses regarding the molecular processes associated with the different decapping complexes.

*RESPONSE: We removed selected GO analyses and heatmaps from the main body of the manuscript (included as Supplementary Figures instead). For our LC-MS/MS datasets, we added iBAQ analyses of the DDX6 IP, DCP1a IP, and DCP1b IP in the control conditions. Cellular fractionation studies (using Superose 6 chromatography) were also added to the paper and allow us to interrogate decapping complex composition in more detail. The revised version of the manuscript includes a new 4SU labeling experiment (pulse-chase) as well as estimation of half-lives and synthesis rates in our conditions. Also included is relevant information about DCP1b transcriptional regulation. *

Reviewer #3 (Significance (Required)):

The manuscript in its current form could benefit from substantial revisions for it to be considered impactful for researchers in the field.

Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

Learn more at Review Commons

---

Reply to the reviewers

*Reviewer #1 (Evidence, reproducibility and clarity (Required)):

I have trialled the package on my lab's data and it works as advertised. It was straightforward to use and did not require any special training. I am confident this is a tool that will be approachable even to users with limited computational experience. The use of artificial data to validate the approach - and to provide clear limits on applicability - is particularly helpful.

The main limitation of the tool is that it requires the user to manually select regions. This somewhat limits the generalisability and is also more subjective - users can easily choose "nice" regions that better match with their hypothesis, rather than quantifying the data in an unbiased manner. However, given the inherent challenges in quantifying biological data, such problems are not easily circumventable.

*

* I have some comments to clarify the manuscript:

1. A "straightforward installation" is mentioned. Given this is a Method paper, the means of installation should be clearly laid out.*

__This sentence is now modified. In the revised manuscript we now describe how to install the toolset and we give the link to the toolset website if further information is needed. __On this website, we provide a full video tutorial and a user manual. The user manual is provided as a supplementary material of the manuscript.

* It would be helpful if there was an option to generate an output with the regions analysed (i.e., a JPG image with the data and the drawn line(s) on top). There are two reasons for this: i) A major problem with user-driven quantification is accidental double counting of regions (e.g., a user quantifies a part of an image and then later quantifies the same region). ii) Allows other users to independently verify measurements at a later time.*

We agree that it is helpful to save the analyzed regions. To answer this comment and the other two reviewers' comments pointing at a similar feature, we have now included an automatic saving of the regions of interest. The user will be able to reopen saved regions of interest using a new function we included in the new version of PatternJ.

* 3. Related to the above point, it is highlighted that each time point would need to be analysed separately (line 361-362). It seems like it should be relatively straightforward to allow a function where the analysis line can be mapped onto the next time point. The user could then adjust slightly for changes in position, but still be starting from near the previous timepoint. Given how prevalent timelapse imaging is, this seems like (or something similar) a clear benefit to add to the software.*

We agree that the analysis of time series images can be a useful addition. We have added the analysis of time-lapse series in the new version of PatternJ. The principles behind the analysis of time-lapse series and an example of such analysis are provided in Figure 1 - figure supplement 3 and Figure 5, with accompanying text lines 140-153 and 360-372. The analysis includes a semi-automated selection of regions of interest, which will make the analysis of such sequences more straightforward than having to draw a selection on each image of the series. The user is required to draw at least two regions of interest in two different frames, and the algorithm will automatically generate regions of interest in frames in which selections were not drawn. The algorithm generates the analysis immediately after selections are drawn by the user, which includes the tracking of the reference channel.

* Line 134-135. The level of accuracy of the searching should be clarified here. This is discussed later in the manuscript, but it would be helpful to give readers an idea at this point what level of tolerance the software has to noise and aperiodicity.

*

We agree with the reviewer that a clarification of this part of the algorithm will help the user better understand the manuscript.__ We have modified the sentence to clarify the range of search used and the resulting limits in aperiodicity (now lines 176-181). __Regarding the tolerance to noise, it is difficult to estimate it a priori from the choice made at the algorithm stage, so we prefer to leave it to the validation part of the manuscript. We hope this solution satisfies the reviewer and future users.

*

**Referees cross-commenting**

I think the other reviewer comments are very pertinent. The authors have a fair bit to do, but they are reasonable requests. So, they should be encouraged to do the revisions fully so that the final software tool is as useful as possible.

Reviewer #1 (Significance (Required)):

Developing software tools for quantifying biological data that are approachable for a wide range of users remains a longstanding challenge. This challenge is due to: (1) the inherent problem of variability in biological systems; (2) the complexity of defining clearly quantifiable measurables; and (3) the broad spread of computational skills amongst likely users of such software.

In this work, Blin et al., develop a simple plugin for ImageJ designed to quickly and easily quantify regular repeating units within biological systems - e.g., muscle fibre structure. They clearly and fairly discuss existing tools, with their pros and cons. The motivation for PatternJ is properly justified (which is sadly not always the case with such software tools).

Overall, the paper is well written and accessible. The tool has limitations but it is clearly useful and easy to use. Therefore, this work is publishable with only minor corrections.

*We thank the reviewer for the positive evaluation of PatternJ and for pointing out its accessibility to the users.

*

Reviewer #2 (Evidence, reproducibility and clarity (Required)):

# Summary

The authors present an ImageJ Macro GUI tool set for the quantification of one-dimensional repeated patterns that are commonly occurring in microscopy images of muscles.

# Major comments

In our view the article and also software could be improved in terms of defining the scope of its applicability and user-ship. In many parts the article and software suggest that general biological patterns can be analysed, but then in other parts very specific muscle actin wordings are used. We are pointing this out in the "Minor comments" sections below. We feel that the authors could improve their work by making a clear choice here. One option would be to clearly limit the scope of the tool to the analysis of actin structures in muscles. In this case we would recommend to also rename the tool, e.g. MusclePatternJ. The other option would be to make the tool about the generic analysis of one-dimensional patterns, maybe calling the tool LinePatternJ. In the latter case we would recommend to remove all actin specific wordings from the macro tool set and also the article should be in parts slightly re-written.

*

We agree with the reviewer that our initial manuscript used a mix of general and muscle-oriented vocabulary, which could make the use of PatternJ confusing especially outside of the muscle field. To make PatternJ useful for the largest community, we corrected the manuscript and the PatternJ toolset to provide the general vocabulary needed to make it understandable for every biologist. We modified the manuscript accordingly.

* # Minor/detailed comments

# Software

We recommend considering the following suggestions for improving the software.

## File and folder selection dialogs

In general, clicking on many of the buttons just opens up a file-browser dialog without any further information. For novel users it is not clear what the tool expects one to select here. It would be very good if the software could be rewritten such that there are always clear instructions displayed about which file or folder one should open for the different buttons.*

We experienced with the current version of macOS that the file-browser dialog does not display any message; we suspect this is the issue raised by the reviewer. This is a known issue of Fiji on Mac and all applications on Mac since 2016. We provided guidelines in the user manual and on the tutorial video to correct this issue by changing a parameter in Fiji. Given the issues the reviewer had accessing the material on the PatternJ website, which we apologize for, we understand the issue raised. We added an extra warning on the PatternJ website to point at this problem and its solution. Additionally, we have limited the file-browser dialog appearance to what we thought was strictly necessary. Thus, the user will experience fewer prompts, speeding up the analysis.

*

## Extract button

The tool asks one to specify things like whether selections are drawn "M-line-to-M-line"; for users that are not experts in muscle morphology this is not understandable. It would be great to find more generally applicable formulations. *

We agree that this muscle-oriented vocabulary can make the use of PatternJ confusing. We have now corrected the user interface to provide both general and muscle-specific vocabulary ("center-to-center or edge-to-edge (M-line-to-M-line or Z-disc-to-Z-disc)").*

## Manual selection accuracy

The 1st step of the analysis is always to start from a user hand-drawn profile across intensity patterns in the image. However, this step can cause inaccuracy that varies with the shape and curve of the line profile drawn. If not strictly perpendicular to for example the M line patterns, the distance between intensity peaks will be different. This will be more problematic when dealing with non-straight and parallelly poised features in the image. If the structure is bended with a curve, the line drawn over it also needs to reproduce this curve, to precisely capture the intensity pattern. I found this limits the reproducibility and easy-usability of the software.*

We understand the concern of the reviewer. On curved selections this will be an issue that is difficult to solve, especially on "S" curved or more complex selections. The user will have to be very careful in these situations. On non-curved samples, the issue may be concerning at first sight, but the errors go with the inverse of cosine and are therefore rather low. For example, if the user creates a selection off by 5 degrees, which is visually obvious, lengths will be affected by an increase of only 0.38%. The point raised by the reviewer is important to discuss, and we therefore added a paragraph to comment on the choice of selection (lines 94-98) and a supplementary figure to help make it clear (Figure 1 - figure supplement 1).*

### Reproducibility

Since the line profile drawn on the image is the first step and very essential to the entire process, it should be considered to save together with the analysis result. For example, as ImageJ ROI or ROIset files that can be re-imported, correctly positioned, and visualized in the measured images. This would greatly improve the reproducibility of the proposed workflow. In the manuscript, only the extracted features are being saved (because the save button is also just asking for a folder containing images, so I cannot verify its functionality). *

We agree that this is a very useful and important feature. We have added ROI automatic saving. Additionally, we now provide a simplified import function of all ROIs generated with PatternJ and the automated extraction and analysis of the list of ROIs. This can be done from ROIs generated previously in PatternJ or with ROIs generated from other ImageJ/Fiji algorithms. These new features are described in the manuscript in lines 120-121 and 130-132.

*

## ? button

It would be great if that button would open up some usage instructions.

*

We agree with the reviewer that the "?" button can be used in a better way. We have replaced this button with a Help menu, including a simple tutorial showing a series of images detailing the steps to follow by the user, a link to the user website, and a link to our video tutorial.

* ## Easy improvement of workflow

I would suggest a reasonable expansion of the current workflow, by fitting and displaying 2D lines to the band or line structure in the image, that form the "patterns" the author aims to address. Thus, it extracts geometry models from the image, and the inter-line distance, and even the curve formed by these sets of lines can be further analyzed and studied. These fitted 2D lines can be also well integrated into ImageJ as Line ROI, and thus be saved, imported back, and checked or being further modified. I think this can largely increase the usefulness and reproducibility of the software.

*

We hope that we understood this comment correctly. We had sent a clarification request to the editor, but unfortunately did not receive an answer within the requested 4 weeks of this revision. We understood the following: instead of using our 1D approach, in which we extract positions from a profile, the reviewer suggests extracting the positions of features not as a single point, but as a series of coordinates defining its shape. If this is the case, this is a major modification of the tool that is beyond the scope of PatternJ. We believe that keeping our tool simple, makes it robust. This is the major strength of PatternJ. Local fitting will not use line average for instance, which would make the tool less reliable.

* # Manuscript

We recommend considering the following suggestions for improving the manuscript. Abstract: The abstract suggests that general patterns can be quantified, however the actual tool quantifies specific subtypes of one-dimensional patterns. We recommend adapting the abstract accordingly.

*

We modified the abstract to make this point clearer.

* Line 58: Gray-level co-occurrence matrix (GLCM) based feature extraction and analysis approach is not mentioned nor compared. At least there's a relatively recent study on Sarcomeres structure based on GLCM feature extraction: https://github.com/steinjm/SotaTool with publication: *https://doi.org/10.1002/cpz1.462

• *

We thank the reviewer for making us aware of this publication. We cite it now and have added it to our comparison of available approaches.

* Line 75: "...these simple geometrical features will address most quantitative needs..." We feel that this may be an overstatement, e.g. we can imagine that there should be many relevant two-dimensional patterns in biology?!*

We have modified this sentence to avoid potential confusion (lines 76-77).

• *

• Line 83: "After a straightforward installation by the user, ...". We think it would be convenient to add the installation steps at this place into the manuscript. *

__This sentence is now modified. We now mention how to install the toolset and we provide the link to the toolset website, if further information is needed (lines 86-88). __On the website, we provide a full video tutorial and a user manual.

* Line 87: "Multicolor images will give a graph with one profile per color." The 'Multicolor images' here should be more precisely stated as "multi-channel" images. Multi-color images could be confused with RGB images which will be treated as 8-bit gray value (type conversion first) images by profile plot in ImageJ. *

We agree with the reviewer that this could create some confusion. We modified "multicolor" to "multi-channel".

* Line 92: "...such as individual bands, blocks, or sarcomeric actin...". While bands and blocks are generic pattern terms, the biological term "sarcomeric actin" does not seem to fit in this list. Could a more generic wording be found, such as "block with spike"? *

We agree with the reviewer that "sarcomeric actin" alone will not be clear to all readers. We modified the text to "block with a central band, as often observed in the muscle field for sarcomeric actin" (lines 103-104). The toolset was modified accordingly.

* Line 95: "the algorithm defines one pattern by having the features of highest intensity in its centre". Could this be rephrased? We did not understand what that exactly means.*

We agree with the reviewer that this was not clear. We rewrote this paragraph (lines 101-114) and provided a supplementary figure to illustrate these definitions (Figure 1 - figure supplement 2).

* Line 124 - 147: This part the only description of the algorithm behind the feature extraction and analysis, but not clearly stated. Many details are missing or assumed known by the reader. For example, how it achieved sub-pixel resolution results is not clear. One can only assume that by fitting Gaussian to the band, the center position (peak) thus can be calculated from continuous curves other than pixels. *

Note that the two sentences introducing this description are "Automated feature extraction is the core of the tool. The algorithm takes multiple steps to achieve this (Fig. S2):". We were hoping this statement was clear, but the reviewer may refer to something else. We agree that the description of some of the details of the steps was too quick. We have now expanded the description where needed.

* Line 407: We think the availability of both the tool and the code could be improved. For Fiji tools it is common practice to create an Update Site and to make the code available on GitHub. In addition, downloading the example file (https://drive.google.com/file/d/1eMazyQJlisWPwmozvyb8VPVbfAgaH7Hz/view?usp=drive_link) required a Google login and access request, which is not very convenient; in fact, we asked for access but it was denied. It would be important for the download to be easier, e.g. from GitHub or Zenodo.

*

We are sorry for issues encountered when downloading the tool and additional material. We thank the reviewer for pointing out these issues that limited the accessibility of our tool. We simplified the downloading procedure on the website, which does not go through the google drive interface nor requires a google account. Additionally, for the coder community the code, user manual and examples are now available from GitHub at github.com/PierreMangeol/PatternJ, and are provided as supplementary material with the manuscript. To our knowledge, update sites work for plugins but not for macro toolsets. Having experience sharing our codes with non-specialists, a classical website with a tutorial video is more accessible than more coder-oriented websites, which deter many users.

* Reviewer #2 (Significance (Required)):

The strength of this study is that a tool for the analysis of one-dimensional repeated patterns occurring in muscle fibres is made available in the accessible open-source platform ImageJ/Fiji. In the introduction to the article the authors provide an extensive review of comparable existing tools. Their new tool fills a gap in terms of providing an easy-to-use software for users without computational skills that enables the analysis of muscle sarcomere patterns. We feel that if the below mentioned limitations could be addressed the tool could indeed be valuable to life scientists interested in muscle patterning without computational skills.

In our view there are a few limitations, including the accessibility of example data and tutorials at sites.google.com/view/patternj, which we had trouble to access. In addition, we think that the workflow in Fiji, which currently requires pressing several buttons in the correct order, could be further simplified and streamlined by adopting some "wizard" approach, where the user is guided through the steps.

*As answered above, the links on the PatternJ website are now corrected. Regarding the workflow, we now provide a Help menu with:

1. __a basic set of instructions to use the tool, __
2. a direct link to the tutorial video in the PatternJ toolset
3. a direct link to the website on which both the tutorial video and a detailed user manual can be found. We hope this addresses the issues raised by this reviewer.

*Another limitation is the reproducibility of the analysis; here we recommend enabling IJ Macro recording as well as saving of the drawn line ROIs. For more detailed suggestions for improvements please see the above sections of our review. *

We agree that saving ROIs is very useful. It is now implemented in PatternJ.

We are not sure what this reviewer means by "enabling IJ Macro recording". The ImageJ Macro Recorder is indeed very useful, but to our knowledge, it is limited to built-in functions. Our code is open and we hope this will be sufficient for advanced users to modify the code and make it fit their needs.*

Reviewer #3 (Evidence, reproducibility and clarity (Required)):

Summary In this manuscript, the authors present a new toolset for the analysis of repetitive patterns in biological images named PatternJ. One of the main advantages of this new tool over existing ones is that it is simple to install and run and does not require any coding skills whatsoever, since it runs on the ImageJ GUI. Another advantage is that it does not only provide the mean length of the pattern unit but also the subpixel localization of each unit and the distributions of lengths and that it does not require GPU processing to run, unlike other existing tools. The major disadvantage of the PatternJ is that it requires heavy, although very simple, user input in both the selection of the region to be analyzed and in the analysis steps. Another limitation is that, at least in its current version, PatternJ is not suitable for time-lapse imaging. The authors clearly explain the algorithm used by the tool to find the localization of pattern features and they thoroughly test the limits of their tool in conditions of varying SNR, periodicity and band intensity. Finally, they also show the performance of PatternJ across several biological models such as different kinds of muscle cells, neurons and fish embryonic somites, as well as different imaging modalities such as brightfield, fluorescence confocal microscopy, STORM and even electron microscopy.

This manuscript is clearly written, and both the section and the figures are well organized and tell a cohesive story. By testing PatternJ, I can attest to its ease of installation and use. Overall, I consider that PatternJ is a useful tool for the analysis of patterned microscopy images and this article is fit for publication. However, i do have some minor suggestions and questions that I would like the authors to address, as I consider they could improve this manuscript and the tool:

*We are grateful to this reviewer for this very positive assessment of PatternJ and of our manuscript.

* Minor Suggestions: In the methodology section is missing a more detailed description about how the metric plotted was obtained: as normalized intensity or precision in pixels. *

We agree with the reviewer that a more detailed description of the metric plotted was missing. We added this information in the method part and added information in the Figure captions where more details could help to clarify the value displayed.

* The validation is based mostly on the SNR and patterns. They should include a dataset of real data to validate the algorithm in three of the standard patterns tested. *

We validated our tool using computer-generated images, in which we know with certainty the localization of patterns. This allowed us to automatically analyze 30 000 images, and with varying settings, we sometimes analyzed 10 times the same image, leading to about 150 000 selections analyzed. From these analyses, we can provide with confidence an unbiased assessment of the tool precision and the tool capacity to extract patterns. We already provided examples of various biological data images in Figures 4-6, showing all possible features that can be extracted with PatternJ. In these examples, we can claim by eye that PatternJ extracts patterns efficiently, but we cannot know how precise these extractions are because of the nature of biological data: "real" positions of features are unknown in biological data. Such validation will be limited to assessing whether a pattern was found or not, which we believe we already provided with the examples in Figures 4-6.

* The video tutorial available in the PatternJ website is very useful, maybe it would be worth it to include it as supplemental material for this manuscript, if the journal allows it. *

As the video tutorial may have been missed by other reviewers, we agree it is important to make it more prominent to users. We have now added a Help menu in the toolset that opens the tutorial video. Having the video as supplementary material could indeed be a useful addition if the size of the video is compatible with the journal limits.

* An example image is provided to test the macro. However, it would be useful to provide further example images for each of the three possible standard patterns suggested: Block, actin sarcomere or individual band.*

We agree this can help users. We now provide another multi-channel example image on the PatternJ website including blocks and a pattern made of a linear intensity gradient that can be extracted with our simpler "single pattern" algorithm, which were missing in the first example. Additionally, we provide an example to be used with our new time-lapse analysis.

* Access to both the manual and the sample images in the PatternJ website should be made publicly available. Right now they both sit in a private Drive account. *

As mentioned above, we apologize for access issues that occurred during the review process. These files can now be downloaded directly on the website without any sort of authentication. Additionally, these files are now also available on GitHub.

* Some common errors are not properly handled by the macro and could be confusing for the user: When there is no selection and one tries to run a Check or Extraction: "Selection required in line 307 (called from line 14). profile=getProfile( ;". A simple "a line selection is required" message would be useful there. When "band" or "block" is selected for a channel in the "Set parameters" window, yet a 0 value is entered into the corresponding "Number of bands or blocks" section, one gets this error when trying to Extract: "Empty array in line 842 (called from line 113). if ( ( subloc . length == 1 ) & ( subloc [ 0 == 0) ) {". This error is not too rare, since the "Number of bands or blocks" section is populated with a 0 after choosing "sarcomeric actin" (after accepting the settings) and stays that way when one changes back to "blocks" or "bands".*

We thank the reviewer for pointing out these bugs. These bugs are now corrected in the revised version.

* The fact that every time one clicks on the most used buttons, the getDirectory window appears is not only quite annoying but also, ultimately a waste of time. Isn't it possible to choose the directory in which to store the files only once, from the "Set parameters" window?*

We have now found a solution to avoid this step. The user is only prompted to provide the image folder when pressing the "Set parameter" button. We kept the prompt for directory only when the user selects the time-lapse analysis or the analysis of multiple ROIs. The main reason is that it is very easy for the analysis to end up in the wrong folder otherwise.

* The authors state that the outputs of the workflow are "user friendly text files". However, some of them lack descriptive headers (like the localisations and profiles) or even file names (like colors.txt). If there is something lacking in the manuscript, it is a brief description of all the output files generated during the workflow.*

PatternJ generates multiple files, several of which are internal to the toolset. They are needed to keep track of which analyses were done, and which colors were used in the images, amongst others. From the user part, only the files obtained after the analysis All_localizations.channel_X.txt and sarcomere_lengths.txt are useful. To improve the user experience, we now moved all internal files to a folder named "internal", which we think will clarify which outputs are useful for further analysis, and which ones are not. We thank the reviewer for raising this point and we now mention it in our Tutorial.

I don't really see the point in saving the localizations from the "Extraction" step, they are even named "temp".

We thank the reviewer for this comment, this was indeed not necessary. We modified PatternJ to delete these files after they are used.

* In the same line, I DO see the point of saving the profiles and localizations from the "Extract & Save" step, but I think they should be deleted during the "Analysis" step, since all their information is then grouped in a single file, with descriptive headers. This deleting could be optional and set in the "Set parameters" window.*

We understand the point raised by the reviewer. However, the analysis depends on the reference channel picked, which is asked for when starting an analysis, and can be augmented with additional selections. If a user chooses to modify the reference channel or to add a new profile to the analysis, deleting all these files would mean that the user will have to start over again, which we believe will create frustration. An optional deletion at the analysis step is simple to implement, but it could create problems for users who do not understand what it means practically.

* Moreover, I think it would be useful to also save the linear roi used for the "Extract & Save" step, and eventually combine them during the "Analysis step" into a single roi set file so that future re-analysis could be made on the same regions. This could be an optional feature set from the "Set parameters" window. *

We agree with the reviewer that saving ROIs is very useful. ROIs are now saved into a single file each time the user extracts and saves positions from a selection. Additionally, the user can re-use previous ROIs and analyze an image or image series in a single step.

* In the "PatternJ workflow" section of the manuscript, the authors state that after the "Extract & Save" step "(...) steps 1, 2, 4, and 5 can be repeated on other selections (...)". However, technically, only steps 1 and 5 are really necessary (alternatively 1, 4 and 5 if the user is unsure of the quality of the patterning). If a user follows this to the letter, I think it can lead to wasted time.

*

We agree with the reviewer and have corrected the manuscript accordingly (line 119-120).

• *

*I believe that the "Version Information" button, although important, has potential to be more useful if used as a "Help" button for the toolset. There could be links to useful sources like the manuscript or the PatternJ website but also some tips like "whenever possible, use a higher linewidth for your line selection" *

We agree with the reviewer as pointed out in our previous answers to the other reviewers. This button is now replaced by a Help menu, including a simple tutorial in a series of images detailing the steps to follow, a link to the user website, and a link to our video tutorial.

* It would be interesting to mention to what extent does the orientation of the line selection in relation to the patterned structure (i.e. perfectly parallel vs more diagonal) affect pattern length variability?*

As answered to reviewer 1, we understand this concern, which needs to be clarified for readers. The issue may be concerning at first sight, but the errors grow only with the inverse of cosine and are therefore rather low. For example, if the user creates a selection off by 3 degrees, which is visually obvious, lengths will be affected by an increase of only 0.14%. The point raised by the reviewer is important to discuss, and we therefore have added a comment on the choice of selection (lines 94-98) as well as a supplementary figure (Figure 1 - figure supplement 1).

* When "the algorithm uses the peak of highest intensity as a starting point and then searches for peak intensity values one spatial period away on each side of this starting point" (line 133-135), does that search have a range? If so, what is the range? *

We agree that this information is useful to share with the reader. The range is one pattern size. We have modified the sentence to clarify the range of search used and the resulting limits in aperiodicity (now lines 176-181).

* Line 144 states that the parameters of the fit are saved and given to the user, yet I could not find such information in the outputs. *

The parameters of the fits are saved for blocks. We have now clarified this point by modifying the manuscript (lines 186-198) and modifying Figure 1 - figure supplement 5. We realized we made an error in the description of how edges of "block with middle band" are extracted. This is now corrected.

* In line 286, authors finish by saying "More complex patterns from electron microscopy images may also be used with PatternJ.". Since this statement is not backed by evidence in the manuscript, I suggest deleting it (or at the very least, providing some examples of what more complex patterns the authors refer to). *

This sentence is now deleted.

* In the TEM image of the fly wing muscle in fig. 4 there is a subtle but clearly visible white stripe pattern in the original image. Since that pattern consists of 'dips', rather than 'peaks' in the profile of the inverted image, they do not get analyzed. I think it is worth mentioning that if the image of interest contains both "bright" and "dark" patterns, then the analysis should be performed in both the original and the inverted images because the nature of the algorithm does not allow it to detect "dark" patterns. *

We agree with the reviewer's comment. We now mention this point in lines 337-339.

* In line 283, the authors mention using background correction. They should explicit what method of background correction they used. If they used ImageJ's "subtract background' tool, then specify the radius.*

We now describe this step in the method section.

*

Reviewer #3 (Significance (Required)):

• Describe the nature and significance of the advance (e.g. conceptual, technical, clinical) for the field. Being a software paper, the advance proposed by the authors is technical in nature. The novelty and significance of this tool is that it offers quick and simple pattern analysis at the single unit level to a broad audience, since it runs on the ImageJ GUI and does not require any programming knowledge. Moreover, all the modules and steps are well described in the paper, which allows easy going through the analysis.
• Place the work in the context of the existing literature (provide references, where appropriate). The authors themselves provide a good and thorough comparison of their tool with other existing ones, both in terms of ease of use and on the type of information extracted by each method. While PatternJ is not necessarily superior in all aspects, it succeeds at providing precise single pattern unit measurements in a user-friendly manner.
• State what audience might be interested in and influenced by the reported findings. Most researchers working with microscopy images of muscle cells or fibers or any other patterned sample and interested in analyzing changes in that pattern in response to perturbations, time, development, etc. could use this tool to obtain useful, and otherwise laborious, information. *

We thank the reviewer for these enthusiastic comments about how straightforward for biologists it is to use PatternJ and its broad applicability in the bio community.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

---

Referee #2

Evidence, reproducibility and clarity

Summary

The authors present an ImageJ Macro GUI tool set for the quantification of one-dimensional repeated patterns that are commonly occurring in microscopy images of muscles.

Major comments

In our view the article and also software could be improved in terms of defining the scope of its applicability and user-ship. In many parts the article and software suggest that general biological patterns can be analysed, but then in other parts very specific muscle actin wordings are used. We are pointing this out in the "Minor comments" sections below. We feel that the authors could improve their work by making a clear choice here. One option would be to clearly limit the scope of the tool to the analysis of actin structures in muscles. In this case we would recommend to also rename the tool, e.g. MusclePatternJ. The other option would be to make the tool about the generic analysis of one-dimensional patterns, maybe calling the tool LinePatternJ. In the latter case we would recommend to remove all actin specific wordings from the macro tool set and also the article should be in parts slightly re-written.

Minor/detailed comments

Software

We recommend considering the following suggestions for improving the software.

File and folder selection dialogs

In general, clicking on many of the buttons just opens up a file-browser dialog without any further information. For novel users it is not clear what the tool expects one to select here. It would be very good if the software could be rewritten such that there are always clear instructions displayed about which file or folder one should open for the different buttons.

Extract button

The tool asks one to specify things like whether selections are drawn "M-line-to-M-line"; for users that are not experts in muscle morphology this is not understandable. It would be great to find more generally applicable formulations.

Manual selection accuracy

The 1st step of the analysis is always to start from a user hand-drawn profile across intensity patterns in the image. However, this step can cause inaccuracy that varies with the shape and curve of the line profile drawn. If not strictly perpendicular to for example the M line patterns, the distance between intensity peaks will be different. This will be more problematic when dealing with non-straight and parallelly poised features in the image. If the structure is bended with a curve, the line drawn over it also needs to reproduce this curve, to precisely capture the intensity pattern. I found this limits the reproducibility and easy-usability of the software.

Reproducibility

Since the line profile drawn on the image is the first step and very essential to the entire process, it should be considered to save together with the analysis result. For example, as ImageJ ROI or ROIset files that can be re-imported, correctly positioned, and visualized in the measured images. This would greatly improve the reproducibility of the proposed workflow. In the manuscript, only the extracted features are being saved (because the save button is also just asking for a folder containing images, so I cannot verify its functionality).

? button

It would be great if that button would open up some usage instructions.

Easy improvement of workflow

I would suggest a reasonable expansion of the current workflow, by fitting and displaying 2D lines to the band or line structure in the image, that form the "patterns" the author aims to address. Thus, it extracts geometry models from the image, and the inter-line distance, and even the curve formed by these sets of lines can be further analyzed and studied. These fitted 2D lines can be also well integrated into ImageJ as Line ROI, and thus be saved, imported back, and checked or being further modified. I think this can largely increase the usefulness and reproducibility of the software.

Manuscript

We recommend considering the following suggestions for improving the manuscript. Abstract: The abstract suggests that general patterns can be quantified, however the actual tool quantifies specific subtypes of one-dimensional patterns. We recommend adapting the abstract accordingly.

Line 58: Gray-level co-occurrence matrix (GLCM) based feature extraction and analysis approach is not mentioned nor compared. At least there's a relatively recent study on Sarcomeres structure based on GLCM feature extraction: https://github.com/steinjm/SotaTool with publication: https://doi.org/10.1002/cpz1.462

Line 75: "...these simple geometrical features will address most quantitative needs..." We feel that this may be an overstatement, e.g. we can imagine that there should be many relevant two-dimensional patterns in biology?!

Line 83: "After a straightforward installation by the user, ...". We think it would be convenient to add the installation steps at this place into the manuscript.

Line 87: "Multicolor images will give a graph with one profile per color." The 'Multicolor images' here should be more precisely stated as "multi-channel" images. Multi-color images could be confused with RGB images which will be treated as 8-bit gray value (type conversion first) images by profile plot in ImageJ.

Line 92: "...such as individual bands, blocks, or sarcomeric actin...". While bands and blocks are generic pattern terms, the biological term "sarcomeric actin" does not seem to fit in this list. Could a more generic wording be found, such as "block with spike"?

Line 95: "the algorithm defines one pattern by having the features of highest intensity in its centre". Could this be rephrased? We did not understand what that exactly means.

Line 124 - 147: This part the only description of the algorithm behind the feature extraction and analysis, but not clearly stated. Many details are missing or assumed known by the reader. For example, how it achieved sub-pixel resolution results is not clear. One can only assume that by fitting Gaussian to the band, the center position (peak) thus can be calculated from continuous curves other than pixels.

Line 407: We think the availability of both the tool and the code could be improved. For Fiji tools it is common practice to create an Update Site and to make the code available on GitHub. In addition, downloading the example file (https://drive.google.com/file/d/1eMazyQJlisWPwmozvyb8VPVbfAgaH7Hz/view?usp=drive_link) required a Google login and access request, which is not very convenient; in fact, we asked for access but it was denied. It would be important for the download to be easier, e.g. from GitHub or Zenodo.

Significance

The strength of this study is that a tool for the analysis of one-dimensional repeated patterns occurring in muscle fibres is made available in the accessible open-source platform ImageJ/Fiji. In the introduction to the article the authors provide an extensive review of comparable existing tools. Their new tool fills a gap in terms of providing an easy-to-use software for users without computational skills that enables the analysis of muscle sarcomere patterns. We feel that if the below mentioned limitations could be addressed the tool could indeed be valuable to life scientists interested in muscle patterning without computational skills.

In our view there are a few limitations, including the accessibility of example data and tutorials at sites.google.com/view/patternj, which we had trouble to access. In addition, we think that the workflow in Fiji, which currently requires pressing several buttons in the correct order, could be further simplified and streamlined by adopting some "wizard" approach, where the user is guided through the steps. Another limitation is the reproducibility of the analysis; here we recommend enabling IJ Macro recording as well as saving of the drawn line ROIs. For more detailed suggestions for improvements please see the above sections of our review.

¿Cómo afectará esto a los clientes finales?

Microsoft seguramente comience a ofrecer como alternativa su propio software equivalente a Redis, https://github.com/microsoft/garnet

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this manuscript, Bell et al. provide an exhaustive and clear description of the diversity of a new class of predicted type IV restriction systems that the authors denote as CoCoNuTs, for their characteristic presence of coiled-coil segments and nuclease tandems. Along with a comprehensive analysis that includes phylogenetics, protein structure prediction, extensive protein domain annotations, and an in-depth investigation of encoding genomic contexts, they also provide detailed hypotheses about the biological activity and molecular functions of the members of this class of predicted systems. This work is highly relevant, it underscores the wide diversity of defence systems that are used by prokaryotes and demonstrates that there are still many systems to be discovered. The work is sound and backed-up by a clear and reasonable bioinformatics approach. I do not have any major issues with the manuscript, but only some minor comments.

Strengths:

The analysis provided by the authors is extensive and covers the three most important aspects that can be covered computationally when analysing a new family/superfamily: phylogenetics, genomic context analysis, and protein-structure-based domain content annotation. With this, one can directly have an idea about the superfamily of the predicted system and infer their biological role. The bioinformatics approach is sound and makes use of the most current advances in the fields of protein evolution and structural bioinformatics.

Weaknesses:

It is not clear how coiled-coil segments were assigned if only based on AF2-predicted models or also backed by sequence analysis, as no description is provided in the methods. The structure prediction quality assessment is based solely on the average pLDDT of the obtained models (with a threshold of 80 or better). However, this is not enough, particularly when multimeric models are used. The PAE matrix should be used to evaluate relative orientations, particularly in the case where there is a prediction that parts from 2 proteins are interacting. In the case of multimers, interface quality scores, such as the ipTM or pDockQ, should also be considered and, at minimum, reported.

A description of the coiled-coil predictions has been added to the Methods. For multimeric models, PAE matrices and ipTM+pTM scores have been included in Supplementary Data File S1.

Reviewer #2 (Public Review):

Summary:

In this work, using in-depth computational analysis, Bell et al. explore the diverse repertoire of type IV McrBC modification-dependent restriction systems. The prototypical two-component McrBC system has been structurally and functionally characterised and is known to act as a defence by restricting phage and foreign DNA containing methylated cytosines. Here, the authors find previously unanticipated complexity and versatility of these systems and focus on detailed analysis and classification of a distinct branch, the so-called CoCoNut, named after its composition of coiled-coil structures and tandem nucleases. These CoCoNut systems are predicted to target RNA as well as DNA and to utilise defence mechanisms with some similarity to type III CRISPR-Cas systems.

Strengths:

This work is enriched with a plethora of ideas and a myriad of compelling hypotheses that now await experimental verification. The study comes from the group that was amongst the first to describe, characterize, and classify CRISPR-Cas systems. By analogy, the findings described here can similarly promote ingenious experimental and conceptual research that could further drive technological advances. It could also instigate vigorous scientific debates that will ultimately benefit the community.

Weaknesses:

The multi-component systems described here function in the context of large oligomeric complexes. Some of the single chain AF2 predictions shown in this work are not compatible, for example, with homohexameric complex formation due to incompatible orientation of domains. The recent advances in protein structure prediction, in particular AlphaFold2 (AF2) multimer, now allow us to confidently probe potential protein-protein interactions and protein complex formation. This predictive power could be exploited here to produce a better glimpse of these multimeric protein systems. It can also provide a more sound explanation for some of the observed differences amongst different McrBC types.

Hexameric CnuB complexes with CnuC stimulatory monomers for Type I-A, I-B, I-C, II, and III-A CoCoNuT systems have been modeled with AF2 and included in Supplementary Data File S1, albeit without the domains fused to the GTPase N-terminus (with the exception of Type I-B, which lacks the long coiled-coil domain fused to the GTPase and was modeled with its entire sequence). Attempts to model the other full-length CnuB hexamers did not lead to convincing results.

Recommendations for the authors:

Reviewing Editor:

The detailed recommendations by the two reviewers will help the authors to further strengthen the manuscript, but two points seem particularly worth considering: 1. The methods are barely sketched in the manuscript, but it could be useful to detail them more closely. Particularly regarding the coiled-coil segments, which are currently just statists, useful mainly for the name of the family, more detail on their prediction, structural properties, and purpose would be very helpful. 2. Due to its encyclopedic nature, the wealth of material presented in the paper makes it hard to penetrate in one go. Any effort to make it more accessible would be very welcome. Reviewer 1 in particular has made a number of suggestions regarding the figures, which would make them provide more support for the findings described in the text.

A description of the techniques used to identify coiled-coil segments has been added to the Methods. Our predictions ranged from near certainty in the coiled-coils detected in CnuB homologs, to shorter helices at the limit of detection in other factors. We chose to report all probable coiled-coils, as the extensive coiled-coils fused to CnuB, which are often the only domain present other than the GTPase, imply involvement in mediating complex formation by interacting with coiled-coils in other factors, particularly the other CoCoNuT factors. The suggestions made by Reviewer 1 were thoughtful and we made an effort to incorporate them.

Reviewer #1 (Recommendations For The Authors):

I do not have any major issues with the manuscript. I have however some minor comments, as described below.

• The last sentence of the abstract at first reads as a fact and not a hypothesis resulting from the work described in the manuscript. After the second read, I noticed the nuances in the sentence. I would suggest a rephrasing to emphasize that the activity described is a theoretical hypothesis not backed-up by experiments.

This sentence has been rephrased to make explicit the hypothetical nature of the statement.

• In line 64, the authors rename DUF3578 as ADAM because indeed its function is not unknown. Did the authors consider reaching out to InterPro to add this designation to this DUF? A search in interpro with DUF3578 results in "MrcB-like, N-terminal domain" and if a name is suggested, it may be worthwhile to take it to the IntrePro team.

We will suggest this nomenclature to InterPro.

• I find Figure 1E hard to analyse and think it occupies too much space for the information it provides. The color scheme, the large amount of small slices, and the lack of numbers make its information content very small. I would suggest moving this to the supplementary and making it instead a bar plot. If removed from Figure 1, more space is made available for the other panels, particularly the structural superpositions, which in my opinion are much more important.

We have removed Figure 1E from the paper as it adds little information beyond the abundance and phyletic distribution of sequenced prokaryotes, in which McrBC systems are plentiful.

• In Figure 2, it is not clear due to the presence of many colorful "operon schemes" that the tree is for a single gene and not for the full operon segment. Highlighting the target gene in the operons or signalling it somehow would make the figure easy to understand even in the absence of the text and legend. The same applies to Supplementary Figure 1.

The legend has been modified to show more clearly that this is a tree of McrB-like GTPases.

• In line 146, the authors write "AlphaFold-predicted endonucelase fold" to say that a protein contains a region that AF2 predicts to fold like an endonuclease. This is a weird way of writing it and can be confusing to non-expert readers. I would suggest rephrasing for increased clarity.

This sentence has been rephrased for greater clarity.

• In line 167, there is a [47]. I believe this is probably due to a previous reference formatting.

Indeed, this was a reference formatting error and has been fixed.

• In most figures, the color palette and the use of very similar color palettes for taxonomy pie charts, genomic context composition schemes, and domain composition diagrams make it really hard to have a good understanding of the image at first. Legends are often close to each other, and it is not obvious at first which belong to what. I would suggest changing the layouts and maybe some color schemes to make it easier to extract the information that these figures want to convey.

It seemed that Figure 4 was the most glaring example of these issues, and it has been rearranged for easier comprehension.

• In the paragraph that starts at line 199, the authors mention an Ig-like domain that is often found at the N-terminus of Type I CoCoNuTs. Are they all related to each other? How conserved are these domains?

These domains are all predicted to adopt a similar beta-sandwich fold and are found at the N-terminus of most CoCoNuT CnuC homologs, suggesting they are part of the same family, but we did not undertake a more detailed sequenced-based analysis of these regions.

We also find comparable domains in the CnuC/McrC-like partners of the abundant McrB-like NxD motif GTPases that are not part of CoCoNuT systems, and given the similarity of some of their predicted structures to Rho GDP-dissociation inhibitor 1, we suspect that they have coevolved as regulators of the non-canonical NxD motif GTPase type. Our CnuBC multimer models showing consistent proximity between these domains in CnuC and CnuB GTPase domains suggest this could indeed be the case. We plan to explore these findings further in a forthcoming publication.

• In line 210, the authors write "suggesting a role in overcrowding-induced stress response". Why so? In >all other cases, the authors justify their hypothesis, which I really appreciated, but not here.

A supplementary note justifying this hypothesis has been added to Supplementary Data File S1.

• At the end of the paragraph that starts in line 264, the authors mention that they constructed AF2 multimeric models to predict if 2 proteins would interact. However, no quality scores were provided, particularly the PAE matrix. This would allow for a better judgement of this prediction, and I would suggest adding the PAE matrix as another panel in the figure where the 3D model of the complex is displayed.

The PAE matrix and ipTM+pTM scores for this and other multimer models have been added to Supplementary Data File S1. For this model in particular, the surface charge distribution of the model has been presented to support the role of the domains that have a higher PAE in RNA binding.

• In line 306, "(supplementary data)" refers to what part of the file?

This file has been renamed Supplementary Table S3 and referenced as such.

• In line 464, the authors suggest that ShdA could interact with CoCoNuTs. Why not model the complex as done for other cases? what would co-folding suggest?

As we were not able to convincingly model full-length CnuB hexamers with N-terminal coiled-coils, we did not attempt modeling of this hypothetical complex with another protein with a long coiled-coil, but it remains an interesting possibility.

• In line 528, why and how were some genes additionally analyzed with HHPred?

Justification for this analysis has been added to the Methods, but briefly, these genes were additionally analyzed if there were no BLAST hits or to confirm the hits that were obtained.

• In the first section of the methods, the first and second (particularly the second) paragraphs are extremely long. I would suggest breaking them to facilitate reading.

This change has been made.

• In line 545, what do the authors mean by "the alignment (...) were analyzed with HHPred"?

A more detailed description of this step has been added to the Methods.

• The authors provide the models they produced as well as extensive supplementary tables that make their data reusable, but they do not provide the code for the automated steps, as to excise target sequence sections out of multiple sequence alignments, for example.

The code used for these steps has been in use in our group at the NCBI for many years. It will be difficult to utilize outside of the NCBI software environment, but for full disclosure, we have included a zipped repository with the scripts and custom-code dependencies, although there are external dependencies as well such as FastTree and BLAST. In brief, it involves PSI-BLAST detection of regions with the most significant homology to one of a set of provided alignments (seals-2-master/bin/wrappers/cog_psicognitor). In this case, the reference alignments of McrB-like GTPases and DUF2357 were generated manually using HHpred to analyze alignments of clustered PSI-BLAST results. This step provided an output of coordinates defining domain footprints in each query sequence, which were then combined and/or extended using scripts based on manual analysis of many examples with HHpred (footprint_finders/get_GTPase_frags.py and footprint_finders/get_DUF2357_frags.py), then these coordinates were used to excise such regions from the query amino acid sequence with a final script (seals-2-master/bin/misc/fa2frag).

Reviewer #2 (Recommendations For The Authors):

(1) Page 4, line 77 - 'PUA superfamily domains' could be more appropriate to use instead of "EVE superfamily".

While this statement could perhaps be applied to PUA superfamily domains, our previous work we refer to, which strongly supports the assertion, was restricted to the EVE-like domains and we prefer to retain the original language.

(2) Page 5. lines 128-130 - AF2 multimer prediction model could provide a more sound explanation for these differences.

Our AF2 multimer predictions added in this revision indeed show that the NxD motif McrB-like CoCoNuT GTPases interact with their respective McrC-like partners such that an immunoglobulin-like beta-sandwich domain, fused to the N-termini of the McrC homologs and similar to Rho GDP-dissociation inhibitor 1, has the potential to physically interact with the GTPase variants. However, we did not probe this in greater detail, as it is beyond the scope of this already highly complex article, but we plan to study it in the future.

(3) Page 8, line 252 - The surface charge distribution of CnuH OB fold domain looks very different from SmpB (pdb3iyr). In fact, the regions that are in contact with RNA in SmpB are highly acidic in CoCoNut CnuH. Although it looks likely that this domain is involved in RNA binding, the mode of interaction should be very different.

We did not detect a strong similarity between the CnuH SmpB-like SPB domain and PDB 3IYR, but when we compare the surface charge distribution of PDB 1WJX and the SPB domain, while there is a significant area that is positively charged in 1WJX that is negatively charged in SPB, there is much that overlaps with the same charge in both domains.

The similarity between SmpB and the SPB domain is significant, but definitely not exact. An important question for future studies is: If the domains are indeed related due to an ancient fusion of SmpB to an ancestor of CnuH, would this degree of divergence be expected?

In other words, can we say anything about how the function of a stand-alone tmRNA-binding protein could evolve after being fused to a complex predicted RNA helicase with other predicted RNA binding domains already present? Experimental validation will ultimately be necessary to resolve these kinds of questions, but for now, it may be safe to say that the presence of this domain, especially in conjunction with the neighboring RelE-like RTL domain and UPF1-like helicase domain, signals a likely interaction with the A-site of the ribosome, and perhaps restriction of aberrant/viral mRNA.

Here is a detailed summary of the article "Super Charging Fine-Grained Reactive Performance" by Milo:

1. Introduction to Reactivity in JavaScript

   • Definition and Importance: "Reactivity allows you to write lazy variables that are efficiently cached and updated, making it easier to write clean and fast code."
   • Introduction to Reactively: "I've been working on a new fine grained reactivity library called Reactively inspired by my work on the SolidJS team."

2. Characteristics of Fine-Grained Reactivity Libraries

   • Library Examples and Usage: "Fine-grained reactivity libraries... Examples include new libraries like Preact Signals, µsignal, and now Reactively, as well as longer-standing libraries like Solid, S.js, and CellX."
   • Functionality and Advantages: "With a library like Reactively, you can easily add lazy variables, caching, and incremental recalculation to your typescript/javascript programs."

3. Core Concepts in Reactively

   • Dependency Graphs: "Reactive libraries work by maintaining a graph of dependencies between reactive elements."
   • Implementation Example: "import { reactive } from '@reactively/core'; const nthUser = reactive(10);"

4. Goals and Features of Reactive Libraries

   • Efficiency and State Consistency: "Efficient: Never overexecute reactive elements... Glitch free: Never allow user code to see intermediate state where only some reactive elements have updated."

5. Comparison Between Lazy and Eager Evaluation

   • Evaluation Strategies: "A lazy library... will first ask B then C to update, then update D after the B and C updates have been completed."
   • Algorithm Challenges: "The first challenge is what we call the diamond problem... The second challenge is the equality check problem."

6. Algorithm Descriptions

   • MobX: "MobX uses a two pass algorithm, with both passes proceeding from A down through its observers... MobX stores a count of the number of parents that need to be updated with each reactive element."
   • Preact Signals: "Preact checks whether the parents of any signal need to be updated before updating that signal... Preact also has two phases, and the first phase 'notifies' down from A."
   • Reactively: "Reactively uses one down phase and one up phase. Instead of version numbers, Reactively uses only graph coloring."

7. Benchmarking Results

   • Performance Observations: "In early experiments with the benchmarking tool, what we've discovered so far is that Reactively is the fastest."
   • Framework Comparisons: "The Solid algorithm performs best on wider graphs... The Preact Signal implementation is fast and very memory efficient."

This summary encapsulates the key concepts, methodologies, and findings presented in the article, focusing on the innovations and performance of various fine-grained reactivity libraries, especially the newly introduced Reactively.

Author Response

The following is the authors’ response to the original reviews.

eLife assessment

This work provides a valuable contribution and assessment of what it means to replicate a null study finding, and what are the appropriate methods for doing so (apart from a rote p-value assessment). Through a convincing re-analysis of results from the Reproducibility Project: Cancer Biology using frequentist equivalence testing and Bayes factors, the authors demonstrate that even when reducing 'replicability success' to a single criterion, how precisely replication is measured may yield differing results. Less focus is directed to appropriate replication of non-null findings.

Reviewer #1 (Public Review):

Summary:

The goal of Pawel et al. is to provide a more rigorous and quantitative approach for judging whether or not an initial null finding (conventionally with p ≥ 0.05) has been replicated by a second similarly null finding. They discuss important objections to relying on the qualitative significant/non-significant dichotomy to make this judgment. They present two complementary methods (one frequentist and the other Bayesian) which provide a superior quantitative framework for assessing the replicability of null findings.

Strengths:

Clear presentation; illuminating examples drawn from the well-known Reproducibility Project: Cancer Biology data set; R-code that implements suggested analyses. Using both methods as suggested provides a superior procedure for judging the replicability of null findings.

Weaknesses:

The proposed frequentist and the Bayesian methods both rely on binary assessments of an original finding and its replication. I'm not sure if this is a weakness or is inherent to making binary decisions based on continuous data.

For the frequentist method, a null finding is considered replicated if the original and replication 90% confidence intervals for the effects both fall within the equivalence range. According to this approach, a null finding would be considered replicated if p-values of both equivalences tests (original and replication) were, say, 0.049, whereas would not be considered replicated if, for example, the equivalence test of the original study had a p-value of 0.051 and the replication had a p-value of 0.001. Intuitively, the evidence for replication would seem to be stronger in the second instance. The recommended Bayesian approach similarly relies on a dichotomy (e.g., Bayes factor > 1).

Thanks for the suggestions, we now emphasize more strongly in the “Methods for assessing replicability of null results” and “Conclusions” sections that both TOST p-values and Bayes factors are quantitative measures of evidence that do not require dichotomization into “success” or “failure”.

Reviewer #2 (Public Review):

Summary:

The study demonstrates how inconclusive replications of studies initially with p > 0.05 can be and employs equivalence tests and Bayesian factor approaches to illustrate this concept. Interestingly, the study reveals that achieving a success rate of 11 out of 15, or 73%, as was accomplished with the non-significance criterion from the RPCB (Reproducibility Project: Cancer Biology), requires unrealistic margins of Δ > 2 for equivalence testing.

Strengths:

The study uses reliable and shareable/open data to demonstrate its findings, sharing as well the code for statistical analysis. The study provides sensitivity analysis for different scenarios of equivalence margin and alfa level, as well as for different scenarios of standard deviations for the prior of Bayes factors and different thresholds to consider. All analysis and code of the work is open and can be replicated. As well, the study demonstrates on a case-by-case basis how the different criteria can diverge, regarding one sample of a field of science: preclinical cancer biology. It also explains clearly what Bayes factors and equivalence tests are.

Weaknesses:

It would be interesting to investigate whether using Bayes factors and equivalence tests in addition to p-values results in a clearer scenario when applied to replication data from other fields. As mentioned by the authors, the Reproducibility Project: Experimental Philosophy (RPEP) and the Reproducibility Project: Psychology (RPP) have data attempting to replicate some original studies with null results. While the RPCB analysis yielded a similar picture when using both criteria, it is worth exploring whether this holds true for RPP and RPEP. Considerations for further research in this direction are suggested. Even if the original null results were excluded in the calculation of an overall replicability rate based on significance, sensitivity analyses considering them could have been conducted. The present authors can demonstrate replication success using the significance criteria in these two projects with initially p < 0.05 studies, both positive and non-positive.

Other comments:

• Introduction: The study demonstrates how inconclusive replications of studies initially with p > 0.05 can be and employs equivalence tests and Bayesian factor approaches to illustrate this concept. Interestingly, the study reveals that achieving a success rate of 11 out of 15, or 73%, as was accomplished with the non-significance criterion from the RPCB (Reproducibility Project: Cancer Biology), requires unrealistic margins of Δ > 2 for equivalence testing.

• Overall picture vs. case-by-case scenario: An interesting finding is that the authors observe that in most cases, there is no substantial evidence for either the absence or the presence of an effect, as evidenced by the equivalence tests. Thus, using both suggested criteria results in a picture similar to the one initially raised by the paper itself. The work done by the authors highlights additional criteria that can be used to further analyze replication success on a case-by-case basis, and I believe that this is where the paper's main contributions lie. Despite not changing the overall picture much, I agree that the p-value criterion by itself does not distinguish between (1) a situation where the original study had low statistical power, resulting in a highly inconclusive non-significant result that does not provide evidence for the absence of an effect and (2) a scenario where the original study was adequately powered, and a non-significant result may indeed provide some evidence for the absence of an effect when analyzed with appropriate methods. Equivalence testing and Bayesian factor approaches are valuable tools in both cases.

Regarding the 0.05 threshold, the choice of the prior distribution for the SMD under the alternative H1 is debatable, and this also applies to the equivalence margin. Sensitivity analyses, as highlighted by the authors, are helpful in these scenarios.

Thank you for the thorough review and constructive feedback. We have added an additional “Appendix C: Null results from the RPP and EPRP” that shows equivalence testing and Bayes factor analyses for the RPP and EPRP null results.

Reviewer #3 (Public Review):

Summary:

The paper points out that non-significance in both the original study and a replication does not ensure that the studies provide evidence for the absence of an effect. Also, it can not be considered a "replication success". The main point of the paper is rather obvious. It may be that both studies are underpowered, in which case their non-significance does not prove anything. The absence of evidence is not evidence of absence! On the other hand, statistical significance is a confusing concept for many, so some extra clarification is always welcome.

One might wonder if the problem that the paper addresses is really a big issue. The authors point to the "Reproducibility Project: Cancer Biology" (RPCB, Errington et al., 2021). They criticize Errington et al. because they "explicitly defined null results in both the original and the replication study as a criterion for replication success." This is true in a literal sense, but it is also a little bit uncharitable. Errington et al. assessed replication success of "null results" with respect to 5 criteria, just one of which was statistical (non-)significance.

It is very hard to decide if a replication was "successful" or not. After all, the original significant result could have been a false positive, and the original null-result a false negative. In light of these difficulties, I found the paper of Errington et al. quite balanced and thoughtful. Replication has been called "the cornerstone of science" but it turns out that it's actually very difficult to define "replication success". I find the paper of Pawel, Heyard, Micheloud, and Held to be a useful addition to the discussion.

Strengths:

This is a clearly written paper that is a useful addition to the important discussion of what constitutes a successful replication.

Weaknesses:

To me, it seems rather obvious that non-significance in both the original study and a replication does not ensure that the studies provide evidence for the absence of an effect. I'm not sure how often this mistake is made.

Thanks for the feedback. We do not have systematic data on how often the mistake of confusing absence of evidence with evidence of absence has been made in the replication context, but we do know that it has been made in at least three prominent large-scale replication projects (the RPP, RPEP, RPCB). We therefore believe that there is a need for our article.

Moreover, we agree that the RPCB provided a nuanced assessment of replication success using five different criteria for the original null results. We emphasize this now more in the “Introduction” section. However, we do not consider our article as “a little bit uncharitable” to the RPCB, as we discuss all other criteria used in the RPCB and note that our intent is not to diminish the important contributions of the RPCB, but rather to build on their work and provide constructive recommendations for future researchers. Furthermore, in response to comments made by Reviewer #2, we have added an additional “Appendix B: Null results from the RPP and EPRP” that shows equivalence testing and Bayes factor analyses for null results from two other replication projects, where the same issue arises.

Reviewer #1 (Recommendations For The Authors):

The authors may wish to address the dichotomy issue I raise above, either in the analysis or in the discussion.

Thank you, we now emphasize that Bayes factors and TOST p-values do not need to be dichotomized but can be interpreted as quantitative measures of evidence, both in the “Methods for assessing replicability of null results” and the “Conclusions” sections.

Reviewer #2 (Recommendations For The Authors):

Given that, here follow additional suggestions that the authors should consider in light of the manuscript's word count limit, to avoid confusing the paper's main idea:

2) Referencing: Could you reference the three interesting cases among the 15 RPCB null results (specifically, the three effects from the original paper #48) where the Bayes factor differs qualitatively from the equivalence test?

We now explicitly cite the original and replication study from paper #48.

3) Equivalence testing: As the authors state, only 4 out of the 15 study pairs are able to establish replication success at the 5% level, in the sense that both the original and the replication 90% confidence intervals fall within the equivalence range. Among these 4, two (Paper #48, Exp #2, Effect #5 and Paper #48, Exp #2, Effect #6) were initially positive with very low p-values, one (Paper #48, Exp #2, Effect #4) had an initial p of 0.06 and was very precisely estimated, and the only one in which equivalence testing provides a clearer picture of replication success is Paper #41, Exp #2, Effect #1, which had an initial p-value of 0.54 and a replication p-value of 0.05. In this latter case (or in all these ones), one might question whether the "liberal" equivalence range of Δ = 0.74 is the most appropriate. As the authors state, "The post-hoc specification of equivalence margins is controversial."

We agree that the post hoc choice of equivalence ranges is a controversial issue. The margins define an equivalence region where effect sizes are considered practically negligible, and we agree that in many contexts SMD = 0.74 is a large effect size that is not practically negligible. We therefore present sensitivity analyses for a wide range of margins. However, we do not think that the choice of this margin is more controversial for the mentioned studies with low p-values than for other studies with greater p-values, since the question of whether a margin plausibly encodes practically negligible effect sizes is not related to the observed p-value of a study. Nevertheless, for the new analyses of the RPP and EPRP data in Appendix B, we have added additional sensitivity analyses showing how the individual TOST p-values and Bayes factors vary as a function of the margin and the prior standard deviation. We think that these analyses provide readers with an even more transparent picture regarding the implications of the choice of these parameters than the “project-wise” sensitivity analyses in Appendix A.

4) Bayes factor suggestions: For the Bayes factor approach, it would be interesting to discuss examples where the BF differs slightly. This is likely to occur in scenarios where sample sizes differ significantly between the original study and replication. For example, in Paper #48, Exp #2 and Effect #4, the initial p is 0.06, but the BF is 8.1. In the replication, the BF dramatically drops to < 1/1000, as does the p-value. The initial evidence of 8.1 indicates some evidence for the absence of an effect, but not strong evidence ("strong evidence for H0"), whereas a p-value of 0.06 does not lead to such a conclusion; instead, it favors H1. It would be interesting if the authors discussed other similar cases in the paper. It's worth noting that in Paper #5, Exp #1, Effect #3, the replication p-value is 0.99, while the BF01 is 2.4, almost indicating "moderate" evidence for H0, even though the p-value is inconclusive.

We agree that some of the examples nicely illustrate conceptual differences between p-values and Bayes factors, e.g., how they take into account sample size and effect size. As methodologists, we find these aspects interesting ourselves, but we think that emphasizing them is beyond the scope of the paper and would distract eLife readers from the main messages.

Concerning the conceptual differences between Bayes factors and TOST p-values, we already discuss a case where there are qualitative differences in more detail (original paper #48). We added another discussion of this phenomenon in the Appendix C as it also occurs for the replication of Ranganath and Nosek (2008) that was part of the RPP.

5) p-values, magnitude and precision: It's noteworthy to emphasize, if the authors decide to discuss this, that the p-value is influenced by both the effect's magnitude and its precision, so in Paper #9, Exp #2, Effect #6, BF01 = 4.1 has a higher p-value than a BF01 = 2.3 in its replication. However, there are cases where both p-values and BF agree. For example, in Paper #15, Exp #2, Effect #2, both the original and replication studies have similar sample sizes, and as the p-value decreases from p = 0.95 to p = 0.23, BF01 decreases from 5.1 ("moderate evidence for H0") to 1.3 (region of "Absence of evidence"), moving away from H0 in both cases. This also occurs in Paper #24, Exp #3, Effect #6.

We appreciate the suggestions but, as explained before, think that the message of our paper is better understood without additional discussion of more general differences between p-values and Bayes factors.

6) The grey zone: Given the above topic, it is important to highlight that in the "Absence of evidence grey zone" for the null hypothesis, for example, in Paper #5, Exp #1, Effect #3 with a p = 0.99 and a BF01 = 2.4 in the replication, BF and p-values reach similar conclusions. It's interesting to note, as the authors emphasize, that Dawson et al. (2011), Exp #2, Effect #2 is an interesting example, as the p-value decreases, favoring H1, likely due to the effect's magnitude, even with a small sample size (n = 3 in both original and replications). Bayes factors are very close to one due to the small sample sizes, as discussed by the authors.

We appreciate the constructive comments. We think that the two examples from Dawson et al. (2011) and Goetz et al. (2011) already nicely illustrate absence of evidence and evidence of absence, respectively, and therefore decided not to discuss additional examples in detail, to avoid redundancy.

7) Using meta-analytical results (?): For papers from RPCB, comparing the initial study with the meta-analytical results using Bayes factor and equivalence testing approaches (thus, increasing the sample size of the analysis, but creating dependency of results since the initial study would affect the meta-analytical one) could change the conclusions. This would be interesting to explore in initial studies that are replicated by much larger ones, such as: Paper #9, Exp #2, Effect #6; Goetz et al. (2011), Exp #1, Effect #1; Paper #28, Exp #3, Effect #3; Paper #41, Exp #2, Effect #1; and Paper #47, Exp #1, Effect #5).

Thank you for the suggestion. We considered adding meta-analytic TOST p-values and Bayes factors before, but decided that Figure 3 and the results section are already quite technical, so adding more analyses may confuse more than help. Nevertheless, these meta-analytic approaches are discussed in the “Conclusions” section.

8) Other samples of fields of science: It would be interesting to investigate whether using Bayes factors and equivalence tests in addition to p-values results in a clearer scenario when applied to replication data from other fields. As mentioned by the authors, the Reproducibility Project: Experimental Philosophy (RPEP) and the Reproducibility Project: Psychology (RPP) have data attempting to replicate some original studies with null results. While the RPCB analysis yielded a similar picture when using both criteria, it is worth exploring whether this holds true for RPP and RPEP. Considerations for further research in this direction are suggested. Even if the original null results were excluded in the calculation of an overall replicability rate based on significance, sensitivity analyses considering them could have been conducted. The present authors can demonstrate replication success using the significance criteria in these two projects with initially p < 0.05 studies, both positive and non-positive.

Thank you for the excellent suggestion. We added an Appendix B where the null results from the RPP and EPRP are analyzed with our proposed approaches. The results are also discussed in the “Results” and “Conclusions” sections.

9) Other approaches: I am curious about the potential impact of using an approach based on equivalence testing (as described in https://arxiv.org/abs/2308.09112). It would be valuable if the authors could run such analyses or reference the mentioned work.

Thank you. We were unaware of this preprint. It seems related to the framework proposed by Stahel W. A. (2021) New relevance and significance measures to replace p-values. PLoS ONE 16(6): e0252991. https://doi.org/10.1371/journal.pone.0252991

We now cite both papers in the discussion.

10) Additional evidence: There is another study in which replications of initially p > 0.05 studies with p > 0.05 replications were also considered as replication successes. You can find it here: https://www.medrxiv.org/content/10.1101/2022.05.31.22275810v2. Although it involves a small sample of initially p > 0.05 studies with already large sample sizes, the work is currently under consideration for publication in PLOS ONE, and all data and materials can be accessed through OSF (links provided in the work).

Thank you for sharing this interesting study with us. We feel that it is beyond the scope of the paper to include further analyses as there are already analyses of the RPCB, RPP, and EPRP null results. However, we will keep this study in mind for future analysis, especially since all data are openly available.

11) Additional evidence 02: Ongoing replication projects, such as the Brazilian Reproducibility Initiative (BRI) and The Sports Replication Centre (https://ssreplicationcentre.com/), continue to generate valuable data. BRI is nearing completion of its results, and it promises interesting data for analyzing replication success using p-values, equivalence regions, and Bayes factor approaches.

We now cite these two initiatives as examples of ongoing replication projects in the introduction. Similarly as for your last point, we think that it is beyond the scope of the paper to include further analyses as there are already analyses of the RPCB, RPP, and EPRP null results.

Reviewer #3 (Recommendations For The Authors):

I have no specific recommendations for the authors.

Thank you for the constructive review.

Reviewing Editor (Recommendations For the Authors):

I recognize that it was suggested to the authors by the previous Reviewing Editor to reduce the amount of statistical material to be made more suitable for a non-statistical audience, and so what I am about to say contradicts advice you were given before. But, with this revised version, I actually found it difficult to understand the particulars of the construction of the Bayes Factors and would have appreciated a few more sentences on the underlying models that fed into the calculations. In my opinion, the provided citations (e.g., Dienes Z. 2014. Using Bayes to get the most out of non-significant results) did not provide sufficient background to warrant a lack of more technical presentation here.

Thank you for the feedback. We added a new “Appendix C: Technical details on Bayes factors” that provides technical details on the models, priors, and calculations underlying the Bayes factors.

The explanation provided about how the PRAW library functions as a mediator between Python applications and Reddit through the use of APIs is quite illuminating. APIs, as described, serve as the bridge that facilitates these interactions under a set of defined rules and protocols. This brings to mind the essential nature of understanding the limits and capabilities of any API when developing software that depends on external services. It would be interesting to explore further how robust the error handling capabilities of the PRAW library are. Specifically, how does PRAW manage or relay errors that arise from API limitations or disruptions in Reddit's service? This is crucial for developers to ensure their applications can gracefully handle such issues and maintain a good user experience.

I am not well informed on how API works but it sounds like it has a lot of connection to the internet and other kind of information systems. Going back to the sources of social media data, one of the things that platforms could record are what users click on, when they log on or off, etc. What arises from this I think ties back in with the ethical frameworks we were talking about in chapter 2. The discussion becomes what course of action is correct because even if the platform records information on the behavior of user interaction, it boils down to the idea that even if it's to mazimize user experience, should platforms be allowed to record information that could be personal?

yummy

Cubit和Bloc的区别：

Cubit只需要定义state和能改变state的function

Bloc要定义state,event和event handler

In the article it mentioned that many non facebook sites use JavaScript code that tells the mothership what kind of content you're looking at when you're not on Facebook's site and apps. Even if done legally, that doesn't make it the most ethical choice. I think there are better and more ethical ways to understand the target market of your audience. I think this is smart to engage and understand users better but it just doesn't feel right to be about in someone's own personal life when they don't know about it.

Object methods

The person has-a student and instructor rather than the student is-a person (inheritance)

delete instructions

课程代码仓库

suite à la création de la liste non ordonnée une erreur Prettier s'affiche dans la console et Prettier ne met plus en page le code comme auparavant. quelqu'un sait il pourquoi ?

L'utilisation de plusieurs balises

pour avoir le même résultat est-elle également considéré comme une mauvaise pratique ?

Author response:

The following is the authors’ response to the original reviews.

Reviewer 1

(1) Given the low trial numbers, and the point of sequential vs clustered reactivation mentioned in the public review, it would be reassuring to see an additional sanity check demonstrating that future items that are currently not on-screen can be decoded with confidence, and if so, when in time the peak reactivation occurs. For example, the authors could show separately the decoding accuracy for near and far items in Fig. 5A, instead of plotting only the difference between them.

We have now added the requested analysis showing the raw decoded probabilities for near and distant items separately in Figure 5A. We have also chosen to replace Figure 5B with the new figure as we think it provides more information than the previous Figure 5B. Instead, we have moved Figure 5B to the supplement. The median peak decoded accuracy for near and distant items is equivalent. We have added the following description to the figure:

“Decoded raw probabilities for off-screen items, that were up to two steps ahead of the current stimulus cue (‘near’,) vs. distant items that were more than two steps away on the graph, on trials with correct answers. The median peak decoded probability for near and distant items was at the same time point for both probability categories. Note that displayed lines reflect the average probability while, to eliminate influence of outliers, the peak displays the median.”

(2) The non-sequential reactivation analyses often use a time window of peak decodability, and it was not entirely clear to me what data this time window is determined on, e.g., was it determined based on all future reactivations irrespective of graph distance? This should be clarified in the methods.

Thank you for raising this. We now clarify this in the relevant section to read: “First, we calculated a time point of interest by computing the peak probability estimate of decoders across all trials, i.e., the average probability for each timepoint of all trials (except previous onscreen items) of all distances, which is equivalent to the peak of the differential reactivation analysis”

(3) Fig 4 shows evidence for forward and backward sequential reactivation, suggesting that both forward and backward replay peak at a lag of 40-50msec. It would be helpful if this counterintuitive finding could be picked up in the discussion, explaining how plausible it is, physiologically, to find forward and backward replay at the same lag, and whether this could be an artifact of the TDLM method.

This is an important point and we agree that it appears counterintuitive. However, we would highlight this exact time range has been reported in previous studies, though t never for both forward and backward replay. We now include a discussion of this finding. The section now reads:

“[… ] Even though we primarily focused on the mean sequenceness scores across time lags, there appears s to be a (non-significant) peak at 40-60 milliseconds. While simultaneous forward and backward replay is theoretically possible, we acknowledge that it is somewhat surprising and, given our paradigm, could relate to other factors such as autocorrelations (Liu, Dolan, et al., 2021).”

(4) It is reported that participants with below 30% decoding accuracy are excluded from the main analyses. It would be helpful if the manuscript included very specific information about this exclusion, e.g., was the criterion established based on the localizer cross-validated data, the temporal generalisation to the cued item (Fig. 2), or only based on peak decodability of the future sequence items? If the latter, is it applied based on near or far reactivations, or both?

We now clarify this point to include more specific information, which reads:

“[…] Therefore, we decided a priori that participants with a peak decoding accuracy of below 30% would be excluded from the analysis (nine participants in all) as obtained from the cross-validation of localizer trials”

(5) Regarding the low amount of data for the reactivation analysis, the manuscript should be explicit about the number of trials available for each participant. For example, Supplemental Fig. 1 could provide this information directly, rather than the proportion of excluded trials.

We have adapted the plot in the supplement to show the absolute number of rejected epochs per participant, in addition to the ratio.

(6) More generally, the supplements could include more detailed information in the legends.

We agree and have added more extensive explanation of the plots in the supplement legends.

(7) The choice of comparing the 2 nearest with all other future items in the clustered reactivation analysis should be better motivated, e.g., was this based on the Wimmer et al. (2020) study?

We have added our motivation for taking the two nearest items and contrasting them with the items further away. The paragraph reads:

“[…] We chose to combine the following two items for two reasons: First, this doubled the number of included trials; secondly, using this approach the number of trials for each category (“near” and “distant”) was more balanced. […]”

Reviewer 2

(1) Focus exclusively on retrieval data (and here just on the current image trials).

If I understand correctly, you focus all your analyses (behavioural as well as MEG analyses) on retrieval data only and here just on the current image trials. I am surprised by that since I see some shortcomings due to that. These shortcomings can likely be addressed by including the learning data (and predecessor image trials) in your analyses.

a) Number of trials: During each block, you presented each of the twelve edges once. During retrieval, participants then did one "single testing session block". Does that mean that all your results are based on max. 12 trials? Given that participants remembered, on average, 80% this means even fewer trials, i.e., 9-10 trials?

This is correct and a limitation of the paper. However, while we used only correct trials for the reactivation analysis, the sequential analysis was conducted using all trials disregarding the response behaviour. To retain comparability with previous studies we mainly focused on data from after a consolidation phase. Nevertheless, despite the trial limitation we consider the results are robust and worth reporting. Additionally, based on the suggestion of the referee, we now include results from learning blocks (see below).

b) Extend the behavioural and replay/reactivation analysis to predecessor images.

Why do you restrict your analyses to the current image trials? Especially given that you have such a low trial number for your analyses, I was wondering why you did not include the predecessor trials (except the non-deterministic trials, like the zebra and the foot according to Figure 2B) as well.

We agree it would be great to increase power by adding the predecessor images to the current image cue analysis, excluding the ambiguous trials, we did not do so as we considered the underlying retrieval processes of these trial types are not the same, i.e. cannot be simply combined. Nevertheless, we have performed the suggested analysis to check if it increases our power. We found, that the reactivation effect is robust and significant at the same time point of 220-230 ms. However, the effect size actually decreased: While before, peak differential reactivation was at 0.13, it is now at 0.07. This in fact makes conceptual sense. We suspect that the two processes that are elicited by showing a single cue and by showing a second, related, cue are distinct insofar as the predecessor image acts as a primer for the current image, potentially changing the time course/speed of retrieval. Given our concerns that the two processes are not actually the same we consider it important to avoid mixing these data.

We have added a statement to the manuscript discussing this point. The section reads:

“Note that we only included data from the current image cue, and not from the predecessor image cue, as we assume the retrieval processes differ and should not be concatenated.”

c) Extend the behavioural and replay/reactivation analysis to learning trials.

Similar to point 1b, why did you not include learning trials in your analyses?

The advantage of including (correct and incorrect) learning trials has the advantage that you do not have to exclude 7 participants due to ceiling performance (100%).

Further, you could actually test the hypothesis that you outline in your discussion: "This implies that there may be a switch from sequential replay to clustered reactivation corresponding to when learned material can be accessed simultaneously without interference." Accordingly, you would expect to see more replay (and less "clustered" reactivation) in the first learning blocks compared to retrieval (after the rest period).

To track reactivation and replay over the course of learning is a great idea. We have given a lot of thought as to how to integrate these findings but have not found a satisfying solution. Thus, analysis of the learning data turned out to be quite tricky: We decided that each participant should perform as many blocks as necessary to reach at least 80% (with a limit of six and lower bound of two, see Supplement figure 4). Indeed, some participant learned 100% of the sequence after one block (these were mostly medical students, learning things by hard is their daily task). With the benefit of hindsight, we realise our design means that different blocks are not directly comparable between participants. In theory, we would expect that replay emerges in parallel with learning and then gradually changes to clustered reactivation as memory traces become consolidated/stronger. However, it is unclear when replay should emerge and when precisely a switch to clustered reactivation would happen. For this reason, we initially decided not to include the learning trials into the paper.

Nevertheless, to provide some insight into the learning process, and to see how consolidation impacts differential reactivation and replay, we have split our data into pre and post resting state, aggregating all learning trials of each participant. While this does not allow us to track processes on a block basis, it does offer potential (albeit limited) insight into the hypothesis we outline in the discussion.

For reactivation, we see emergence of a clear increase, further strengthening the outlined hypothesis, however, for replay the evidence is less clear, as we do not know over how many learning blocks replay is expected.

We calculated individual trajectories of how reactivation and replay changes from learning to retrieval and related these to performance. Indeed, we see an increase of reactivation is nominally associated with higher learning performance, while an increase in replay strength is associated with lower performance (both non-significant). However, due to the above-mentioned reasons we think it would premature to add this weak evidence to the paper.

To mitigate problems of experiment design in relation to this question we are currently implementing a follow-study, where we aim to normalize the learning process across participants and index how replay/reactivation changes over the course of learning and after consolidation.

We have added plots showing clustered reactivation sequential replay measures during learning (Figure 5D and Supplement 8)

The added section(s) now read:

“To provide greater detail on how the 8-minute consolidation period affected reactivation we, post-hoc, looked at relevant measures across learning trials in contrast to retrieval trials. For all learning trials, for each participant, we calculated differential reactivation for the same time point we found significant in the previous analysis (220-260 milliseconds). On average, differential reactivation probability increased from pre to post resting state (Figure 5D). […]

Nevertheless, even though our results show a nominal increase in reactivation from learning to retrieval (see Figure 5D), due to experimental design features our data do not enable us to test for an hypothesized switch for sequential replay (see also “limitations” and Supplement 8).”

d) Introduction (last paragraph): "We examined the relationship of graph learning to reactivation and replay in a task where participants learned a ..." If all your behavioural analyses are based on retrieval performance, I think that you do not investigate graph learning (since you exclusively focus the analyses on retrieving the graph structure). However, relating the graph learning performance and replay/reactivation activity during learning trials (i.e., during graph learning) to retrieval trials might be interesting but beyond the scope of this paper.

We agree. We have changed the wording to be more accurate. Indeed, we do not examine graph learning but instead examine retrieval from a graph, after graph learning. The mentioned sentence now read

“[…] relationship of retrieval from a learned graph structure to reactivation [...]”

e) It is sometimes difficult to follow what phase of the experiment you refer to since you use the terms retrieval and test synonymously. Not a huge problem at all but maybe you want to stick to one term throughout the whole paper.

Thank you for pointing this out. We have now adapted the manuscript to exclusively refer to “retrieval” and not to “test”.

(2) Is your reactivation clustered?

In Figure 5A, you compare the reactivation strength of the two items following the cue image (i.e., current image trials) with items further away on the graph. I do not completely understand why your results are evidence for clustered reactivation in contrast to replay.

First, it would be interesting to see the reactivation of near vs. distant items before taking the difference (time course of item probabilities).

(copied answer from response to Reviewer 1, as the same remark was raised)

We have added the requested analysis showing the raw decoded probabilities for near and distant items separately in Figure 5A. We have chosen to replace Figure 5B with the new figure as we think that it offers more information than the previous Figure 5B. Instead, we have moved Figure 5B to the supplement. The median peak decoded accuracy for near and distant items is equivalent. We have added the following description to the figure:

“Decoded raw probabilities for off-screen items, that were up to two steps ahead of the current stimulus cue (‘near’,) vs. distant items that were more than two steps away on the graph, on trials with correct answers. The median peak decoded probability for near and distant items was at the same time point for both probability categories. Note that displayed lines reflect the average probability while, to eliminate influence of outliers, the peak displays the median. .”

Second, could it still be that the first item is reactivated before the second item? By averaging across both items, it becomes not apparent what the temporal courses of probabilities of both items look like (and whether they follow a sequential pattern). Additionally, the Gaussian smoothing kernel across the time dimension might diminish sequential reactivation and favour clustered reactivation. (In the manuscript, what does a Gaussian smoothing kernel of  = 1 refer to?). Could you please explain in more detail why you assume non-sequential clustered reactivation here and substantiate this with additional analyses?

We apologise for the unclear description. Note the Gaussian kernel is in fact only used for the reactivation analysis and not the replay analysis, so any small temporal successions would have been picked up by the sequential analysis. We now clarify this in the respective section of the sequential analysis and also explain the parameter of delta= 1 in the reactivation analysis section. The paragraph now reads

“[…] As input for the sequential analysis, we used the raw probabilities of the ten classifiers corresponding to the stimuli. [...]

[…] Therefore, to address this we applied a Gaussian smoothing kernel (using scipy.ndimage.gaussian_filter with the default parameter of σ=1 which corresponds approximately to taking the surrounding timesteps in both direction with the following weighting: current time step: 40%, ±1 step: 25%, ±2 step: 5%, ±3 step: 0.5%) [...]”

(3) Replay and/or clustered reactivation?

The relationship between the sequential forward replay, differential reactivation, and graph reactivation analysis is not really apparent. Wimmer et al. demonstrated that high performers show clustered reactivation rather than sequential reactivation. However, you did not differentiate in your differential reactivation analysis between high vs. low performers. (You point out in the discussion that this is due to a low number of low performers.)

We agree that a split into high vs low performers would have been preferably for our analysis. However, there is one major obstacle that made us opt for a correlational analysis instead: We employed criteria learning, rendering a categorical grouping conceptually biased. Even though not all participants reached the criteria of 80%, our sample did not naturally split between high and low performers but was biased towards higher performance, leaving the groups uneven. The median performance was 83% (mean ~81%), with six of our subjects (~1/4th of included participant) having this exact performance. This makes a median or mean split difficult, as either binning assignment choice would strongly affect the results. We have added a limitations section in which we extensively discuss this shortcoming and reasoning for not performing a median split as in Wimmer et al (2020). The section now reads:

“There are some limitations to our study, most of which originate from a suboptimal study design. [...], as we performed criteria learning, a sub-group analysis as in Wimmer et al., (2020) was not feasible, as median performance in our sample would have been 83% (mean 81%), with six participants exactly at that threshold. [...]”

It might be worth trying to bring the analysis together, for example by comparing sequential forward replay and differential reactivation at the beginning of graph learning (when performance is low) vs. retrieval (when performance is high).

Thank you for the suggestion to include the learning segments, which we think improves the paper quite substantially. However, analysis of the learning data turned out to be quite tricky> We had decided that each participant should perform as many blocks as necessary to reach at least 80% accuracy (with a limit of six and lower bound of two, see Supplement figure 4). Some participants learned 100% of the sequence after one block (these were mostly medical students, learning things by hard is their daily task). This in hindsight is an unfortunate design feature in relation to learning as it means different blocks are not directly comparable between participants.

In theory, we would expect that replay emerges in parallel with learning and then gradually change to clustered reactivation, as memory traces get consolidated/stronger. However, it is unclear when replay would emerge and when the switch to reactivation would happen. For this reason, we initially decided not to include the learning trials into the paper at all.

Nevertheless, to give some insight into the learning process and to see how consolidation effects differential reactivation and replay, we have split our data into pre and post resting state, aggregating all learning trials of each participant. While this does not allow us to track measures of interest on a block basis, it gives some (albeit limited) insight into the hypothesis outlined in our discussion.

For reactivation, we see a clear increase, further strengthening the outlined hypothesis, However, for replay the evidence is less obvious, potentially due to that fact that we do not know across how many learning blocks replay is to be expected.

The added section(s) now read:

“To examine how the 8-minute consolidation period affected reactivation we, post-hoc, looked at relevant measures during learning trials in contrast to retrieval trials. For all learning trial, for each participant, we calculated differential reactivation for the time point we found significant during the previous analysis (220-260 milliseconds). On average, differential reactivation probability increased from pre to post resting state (Figure 5D).

[…]

Nevertheless, even though our results show a nominal increase in reactivation from learning to retrieval (see Figure 5D), our data does not enable us to show an hypothesized switch for sequential replay (see also “limitations” and Supplement 8).”

Additionally, the main research question is not that clear to me. Based on the introduction, I thought the focus was on replay vs. clustered reactivation and high vs. low performance (which I think is really interesting). However, the title is more about reactivation strength and graph distance within cognitive maps. Are these two research questions related? And if so, how?

We agree we need to be clearer on this point. We have added two sentences to the introduction, which should address this point. The section now reads:

“[…] In particular, the question remains how the brain keeps track of graph distances for successful recall and whether the previously found difference between high and low performers also holds true within a more complex graph learning context.”

(4) Learning the graph structure.

I was wondering whether you have any behavioural measures to show that participants actually learn the graph structure (instead of just pairs or triplets of objects). For example, do you see that participants chose the distractor image that was closer to the target more frequently than the distractor image that was further away (close vs. distal target comparison)? It should be random at the beginning of learning but might become more biased towards the close target.

Thanks, this is an excellent suggestion. Our analysis indeed shows that people take the near lure more often than the far lure in later blocks, while it is random in the first block.

Nevertheless, we have decided to put these data into the supplement and reference it in the text. This is because analysis of the learning blocks is challenging and biased in general. Each participant had a different number of learning blocks based on their learning rate, and this makes it difficult to compare learning across participants. We have tried our best to accommodate and explain these difficulties in the figure legend. Nevertheless, we thank the referee for guidance here and this analysis indeed provides further evidence that participants learned the actual graph structure.

The added section reads

“Additionally, we have included an analysis showing how wrong answers participants provided were random in the first block and biased towards closer graph nodes in later blocks. This is consistent with participants actually learning the underlying graph structure as opposed to independent triplets (see figure and legend of Supplement 6 for details).”

(5) Minor comments

a) "Replay analysis relies on a successive detection of stimuli where the chance of detection exponentially decreases with each step (e.g., detecting two successive stimuli with a chance of 30% leaves a 9% chance of detecting the replay event). " Could you explain in more detail why 30% is a good threshold then?

Thank you. We have further clarified the section. As we are working mainly with probabilities, it is useful to keep in mind that accuracy is a class metric that only provides a rough estimate of classifier ability. Alternatively, something like a Top-3-Accuracy would be preferable, but also slightly silly in the context of 10 classes.

Nevertheless, subtle changes in probability estimates are present and can be picked up by the methods we employ. Therefore, the 30% is a rough lower bound and decided based on pilot data that showed that clean MEG data from attentive participants can usually reach this threshold. The section now reads:

“(e.g., detecting two successive stimuli with a chance of 30% leaves a 9% chance of detecting a replay event). However, one needs to bear in mind that accuracy is a “winnertakes-all” metric indicating whether the top choice also has the highest probability, disregarding subtle, relative changes in assigned probability. As the methods used in this analysis are performed on probability estimates and not class labels, one can expect that the 30% are a rough lower bound and that the actual sensitivity within the analysis will be higher. Additionally, based on pilot data, we found that attentive participants were able to reach 30% decodability, allowing us to use decodability as a data quality check. “

b) Could you make explicit how your decoders were designed? Especially given that you added null data, did you train individual decoders for one class vs. all other classes (n = 9 + null data) or one class vs. null data?

We added detail to the decoder training. The section now reads

“Decoders were trained using a one-vs-all approach, which means that for each class, a separate classifier was trained using positive examples (target class) and negative examples (all other classes) plus null examples (data from before stimulus presentation, see below). In detail, null data was.”

c) Why did you choose a ratio of 1:2 for your null data?

Our choice for using a higher ratio was based upon previous publications reporting better sensitivity of TDLM using higher ratios, as spatial sensor correlations are decreasing. Nevertheless, this choice was not well investigated beforehand. We have added more information to this to the manuscript

d) You could think about putting the questionnaire results into the supplement if they are sanity checks.

We have added the questionnaire results. However, due to the size of the tables, we have decided to add them as excel files into the supplementary files of the code repository. We have mentioned the existence file in the publication.

e) Figure 2. There is a typo in D: It says "Precessor Image" instead of "Predecessor Image".

Fixed typo in figure.

f) You write "Trials for the localizer task were created from -0.1 to 0.5 seconds relative to visual stimulus onset to train the decoders and for the retrieval task, from 0 to 1.5 seconds after onset of the second visual cue image." But the Figure legend 3D starts at -0.1 seconds for the retrieval test.

We have now clarified this. For the classifier cross-validation and transfer sanity check and clustered analysis we used trials from -0.1 to 0.5s, whereas for the sequenceness analysis of the retrieval, we used trials from 0 to 1.5 seconds

컴퓨터 소프트웨어의 제작에 사용되는 설계도.

The following code is not working getting the following error: julia> [xs i.(xs)] ERROR: UndefVarError: i not defined Stacktrace: [1] top-level scope @ REPL[41]:1

All running in the browser not on remote WMs

delete the session on your best fit occupations--that info goes into your Career Ready Portfolio, not the SLJ.

Author response:

The following is the authors’ response to the previous reviews.

eLife assessment

This important study advances our understanding of how past and future information is jointly considered in visual working memory by studying gaze biases in a memory task that dissociates the locations during encoding and memory tests. The evidence supporting the conclusions is convincing, with state-of-the-art gaze analyses that build on a recent series of experiments introduced by the authors. This work, with further improvements incorporating the existing literature, will be of broad interest to vision scientists interested in the interplay of vision, eye movements, and memory.

We thank the Editors and the Reviewers for their enthusiasm and appreciation of our task, our findings, and our article. We also wish to thank the Reviewers for their constructive comments that we have embraced to improve our article. Please find below our point-by-point responses to this valuable feedback, where we also state relevant revisions that we have made to our article.

In addition, please note that we have now also made our data and code publicly available.

Reviewer 1, Comments:

In this study, the authors offer a fresh perspective on how visual working memory operates. They delve into the link between anticipating future events and retaining previous visual information in memory. To achieve this, the authors build upon their recent series of experiments that investigated the interplay between gaze biases and visual working memory. In this study, they introduce an innovative twist to their fundamental task. Specifically, they disentangle the location where information is initially stored from the location where it will be tested in the future. Participants are tasked with learning a novel rule that dictates how the initial storage location relates to the eventual test location. The authors leverage participants' gaze patterns as an indicator of memory selection. Intriguingly, they observe that microsaccades are directed toward both the past encoding location and the anticipated future test location. This observation is noteworthy for several reasons. Firstly, participants' gaze is biased towards the past encoding location, even though that location lacks relevance to the memory test. Secondly, there's a simultaneous occurrence of an increased gaze bias towards both the past and future locations. To explore this temporal aspect further, the authors conduct a compelling analysis that reveals the joint consideration of past and future locations during memory maintenance. Notably, microsaccades biased towards the future test location also exhibit a bias towards the past encoding location. In summary, the authors present an innovative perspective on the adaptable nature of visual working memory. They illustrate how information relevant to the future is integrated with past information to guide behavior.

Thank you for your enthusiasm for our article and findings as well as for your constructive suggestions for additional analyses that we respond to in detail below.

This short manuscript presents one experiment with straightforward analyses, clear visualizations, and a convincing interpretation. For their analysis, the authors focus on a single time window in the experimental trial (i.e., 0-1000 ms after retro cue onset). While this time window is most straightforward for the purpose of their study, other time windows are similarly interesting for characterizing the joint consideration of past and future information in memory. First, assessing the gaze biases in the delay period following the cue offset would allow the authors to determine whether the gaze bias towards the future location is sustained throughout the entire interval before the memory test onset. Presumably, the gaze bias towards the past location may not resurface during this delay period, but it is unclear how the bias towards the future location develops in that time window. Also, the disappearance of the retro cue constitutes a visual transient that may leave traces on the gaze biases which speaks again for assessing gaze biases also in the delay period following the cue offset.

Thank you for raising this important point. We initially focused on the time window during the cue given that our central focus was on gaze-biases associated with mnemonic item selection. By zooming in on this window, we could best visualize our main effects of interest: the joint selection (in time) of past and future memory attributes.

At the same time, we fully agree that examining the gaze biases over a more extended time window yields a more comprehensive view of our data. To this end, we have now also extended our analysis to include a wider time range that includes the period between cue offset (1000 ms after cue onset) and test onset (1500 ms after cue onset). We present these data below. Because we believe our future readers are likely to be interested in this as well, we have now added this complementary visualization as Supplementary Figure 4 (while preserving the focus in our main figure on the critical mnemonic selection period of interest).

Author response image 1.

Supplementary Figure 4. Gaze biases in extended time window as a complement to Figure 1 and Supplementary Figure 2. This extended analysis reveals that while the gaze bias towards the past location disappears around 600 ms after cue onset, the gaze bias towards the future location persists (panel a) and that while the early (joint) future bias occurs predominantly in the microsaccade range below 1 degree visual angle, the later bias to the future location incorporates larger eye movement that likely involve preparing for optimally perceiving the anticipated test stimulus (panel b).

This extended analysis reveals that while the gaze bias towards the past location disappears around 600 ms after cue onset (consistent with our prior reports of this bias), the gaze bias towards the future location persists. Moreover, as revealed by the data in panel b above, while the early (joint) future bias occurs predominantly in the microsaccade range below 1 degree visual angle, the later bias to the future location incorporates larger eye movement that likely involve preparing for optimally perceiving the anticipated test stimulus.

We now also call out these additional findings and figure in our article:

Page 2 (Results): “Gaze biases in both axes were driven predominantly by microsaccades (Supplementary Fig. 2) and occurred similarly in horizontal-to-vertical and vertical-tohorizontal trials (Supplementary Fig. 3). Moreover, while the past bias was relatively transient, the future bias continued to increase in anticipation of the of the test stimulus and increasingly incorporated eye-movements beyond the microsaccade range (see Supplementary Fig. 4 for a more extended time range)”.

Moreover, assessing the gaze bias before retro-cue onset allows the authors to further characterize the observed gaze biases in their study. More specifically, the authors could determine whether the future location is considered already during memory encoding and the subsequent delay period (i.e., before the onset of the retro cue). In a trial, participants encode two oriented gratings presented at opposite locations. The future rule indicates the test locations relative to the encoding locations. In their example (Figure 1a), the test locations are shifted clockwise relative to the encoding location. Thus, there are two pairs of relevant locations (each pair consists of one stimulus location and one potential test location) facing each other at opposite locations and therefore forming an axis (in the illustration the axis would go from bottom left to top right). As the future rule is already known to the participants before trial onset it is possible that participants use that information already during encoding. This could be tested by assessing whether more microsaccades are directed along the relevant axis as compared to the orthogonal axis. The authors should assess whether such a gaze bias exists already before retro cue onset and discuss the theoretical consequences for their main conclusions (e.g., is the future location only jointly used if the test location is implicitly revealed by the retro cue).

Thank you – this is another interesting point. We fully agree that additional analysis looking at the period prior to retrocue onset may also prove informative. In accordance with the suggested analysis, we have therefore now also analysed the distribution of saccade directions (including in the period from encoding to retrocue) as a function of the future rule (presented below, and now also included as Supplementary Fig. 5). Complementary recent work from our lab has shown how microsaccade directions can align to the axis of memory contents during retention (see de Vries & van Ede, eNeuro, 2024). Based on this finding, one may predict that if participants retain the items in a remapped fashion, their microsaccades may align with the axis of the future rule, and this could potentially already happen prior to cue onset.

These complementary analyses show that saccade directions are predominantly influenced by the encoding locations rather than the test locations, as seen most clearly by the saccade distribution plots in the middle row of the figure below. To obtain time-courses, we categorized saccades as occurring along the axis of the future rule or along the orthogonal axis (bottom row of the figure below). Like the distribution plots, these time course plots also did not reveal any sign of a bias along the axis of the future rule itself.

Importantly, note how this does not argue against our main findings of joint selection of past and future memory attributes, as for that central analysis we focused on saccade biases that were specific to the selected memory item, whereas the analyses we present below focus on biases in the axes in which both memory items are defined; not only the cued/selected memory item.

Author response image 2.

Supplementary Figure 5. Distribution of saccade directions relative to the future rule from encoding onset. (Top panel) The spatial layouts in the four future rules. (Middle panel) Polar distributions of saccades during 0 to 1500 ms after encoding onset (i.e., the period between encoding onset and cue onset). The purple quadrants represent the axis of the future rule and the grey quadrants the orthogonal axis. (Bottom panel) Time courses of saccades along the above two axes. We did not observe any sign of a bias along the axis of the future rule itself.

We agree that these additional results are important to bring forward when we interpret our findings. Accordingly, we now mention these findings at the relevant section in our Discussion:

Page 5 (Discussion): “First, memory contents could have directly been remapped (cf. 4,24–26) to their future-relevant location. However, in this case, one may have expected to exclusively find a future-directed gaze bias, unlike what we observed. Moreover, using a complementary analysis of saccade directions along the axis of the future rule (cf. 24), we found no direct evidence for remapping in the period between encoding and cue (Supplementary Fig. 5)”.

Reviewer 2, Comments:

The manuscript by Liu et al. reports a task that is designed to examine the extent to which "past" and "future" information is encoded in working memory that combines a retro cue with rules that indicate the location of an upcoming test probe. An analysis of microsaccades on a fine temporal scale shows the extent to which shifts of attention track the location of the location of the encoded item (past) and the location of the future item (test probe). The location of the encoded grating of the test probe was always on orthogonal axes (horizontal, vertical) so that biases in microsaccades could be used to track shifts of attention to one or the other axis (or mixtures of the two). The overall goal here was then to (1) create a methodology that could tease apart memory for the past and future, respectively, (2) to look at the time-course attention to past/future, and (3) to test the extent to which microsaccades might jointly encode past and future memoranda. Finally, some remarks are made about the plausibility of various accounts of working memory encoding/maintenance based on the examination of these time courses.

Strengths:

This research has several notable strengths. It has a clear statement of its aims, is lucidly presented, and uses a clever experimental design that neatly orthogonalizes "past" and "future" as operationalized by the authors. Figure 1b-d shows fairly clearly that saccade directions have an early peak (around 300ms) for the past and a "ramping" up of saccades moving in the forward direction. This seems to be a nice demonstration the method can measure shifts of attention at a fine temporal resolution and differentiate past from future-oriented saccades due to the orthogonal cue approach. The second analysis shown in Figure 2, reveals a dependency in saccade direction such that saccades toward the probe future were more likely also to be toward the encoded location than away from the encoded direction. This suggests saccades are jointly biased by both locations "in memory".

Thank you for your overall appreciation of our work and for highlighting the above strengths. We also thank you for your constructive comments and call for clarifications that we respond to below.

Weaknesses:

(1) The "central contribution" (as the authors characterize it) is that "the brain simultaneously retains the copy of both past and future-relevant locations in working memory, and (re)activates each during mnemonic selection", and that: "... while it is not surprising that the future location is considered, it is far less trivial that both past and future attributes would be retained and (re)activated together. This is our central contribution." However, to succeed at the task, participants must retain the content (grating orientation, past) and probe location (future) in working memory during the delay period. It is true that the location of the grating is functionally irrelevant once the cue is shown, but if we assume that features of a visual object are bound in memory, it is not surprising that location information of the encoded object would bias processing as indicated by microsaccades. Here the authors claim that joint representation of past and future is "far less trivial", this needs to be evaluaed from the standpoint of prior empirical data on memory decay in such circumstances, or some reference to the time-course of the "unbinding" of features in an encoded object.

Thank you. We agree that our participants have to use the future rule – as otherwise they do not know to which test stimulus they should respond. This was a deliberate decision when designing the task. Critically, however, this does not require (nor imply) that participants have to incorporate and apply the rule to both memory items already prior to the selection cue. It is at least as conceivable that participants would initially retain the two items at their encoded (past) locations, then wait for the cue to select the target memory item, and only then consider the future location associated with the target memory item. After all, in every trial, there is only 1 relevant future location: the one associated with the cued memory item. The time-resolved nature of our gaze markers argues against such a scenario, by virtue of our observation of the joint (simultaneous) consideration of past and future memory attributes (as opposed to selection of past-before-future). These temporal dynamics are central to the insights provided by our study.

In our view, it is thus not obvious that the rule would be applied at encoding. In this sense, we do not assume that the future location is part of both memory objects from encoding, but rather ask whether this is the case – and, if so, whether the future location takes over the role of the past location, or whether past and future locations are retained jointly.

Our statements regarding what is “trivial” and what is “less trivial” regard exactly this point: it is trivial that the future is considered (after all, our task demanded it). However, it is less trivial that (1) the future location was already available at the time of initial item selection (as reflected in the simultaneous engagement of past and future locations), and (2) that in presence of the future location, the past location was still also present in the observed gaze biases.

Having said that, we agree that an interesting possibility is that participants remap both memory items to their future-relevant locations ahead of the cue, but that the past location is not yet fully “unbound” by the time of the cue. This may trigger a gaze bias not only to the new future location but also to the “sticky” (unbound) past location. We now acknowledge this possibility in our discussion (also in response to comment 3 below) where we also suggest how future work may be able to tap into this:

Page 6 (Discussion): “In our study, the past location of the memory items was technically irrelevant for the task and could thus, in principle, be dropped after encoding. One possibility is that participants remapped the two memory items to their future locations soon after encoding, and had started – but not finished – dropping the past location by the time the cue arrived. In such a scenario, the past signal is merely a residual trace of the memory items that serves no purpose but still pulls gaze. Alternatively, however, the past locations may be utilised by the brain to help individuate/separate the two memory items. Moreover, by storing items with regard to multiple spatial frames (cf. 37) – here with regard to both past and future visual locations – it is conceivable that memories may become more robust to decay and/or interference. Also, while in our task past locations were never probed, in everyday life it may be useful to remember where you last saw something before it disappeared behind an occluder. In future work, it will prove interesting to systematically vary to the delay between encoding and cue to assess whether the reliance on the past location gradually dissipates with time (consistent with dropping an irrelevant feature), or whether the past trace remains preserved despite longer delays (consistent with preserving utility for working memory).”

(2) The authors refer to "future" and "past" information in working memory and this makes sense at a surface level. However, once the retrocue is revealed, the "rule" is retrieved from long-term memory, and the feature (e.g. right/left, top/bottom) is maintained in memory like any other item representation. Consider the classic test of digit span. The digits are presented and then recalled. Are the digits of the past or future? The authors might say that one cannot know, because past and future are perfectly confounded. An alternative view is that some information in working memory is relevant and some is irrelevant. In the digit span task, all the digits are relevant. Relevant information is relevant precisely because it is thought be necessary in the future. Irrelevant information is irrelevant precisely because it is not thought to be needed in the immediate future. In the current study, the orientation of the grating is relevant, but its location is irrelevant; and the location of the test probe is also relevant.

Thank you for this stimulating reflection. We agree that in our set-up, past location is technically “task-irrelevant” while future location is certainly “task-relevant”. At the same time, the engagement of the past location suggests to us that the brain uses past location for the selection – presumably because the brain uses spatial location to help individuate/separate the items, even if encoded locations are never asked about. Therefore, whether something is relevant or irrelevant ultimately depends on how one defines relevance (past location may be relevant/useful for the brain even if technically irrelevant from the perspective of the task). In comparison, the use of “past” and “future” may be less ambiguous.

It is also worth noting how we interpret our findings in relation to demands on visual working memory, inspired by dynamic situations whereby visual stimuli may be last seen at one location but expected to re-appear at another, such as a bird disappearing behind a building (the example in our introduction). Thus, past for us does not refer to the memory item perse (like in the digit span analogue) but, rather, quite specifically to the past location of a dynamic visual stimulus in memory (which, in our experiment, was operationalised by the future rule, for convenience).

(3) It is not clear how the authors interpret the "joint representation" of past and future. Put aside "future" and "past" for a moment. If there are two elements in memory, both of which are associated with spatial bindings, the attentional focus might be a spatial average of the associated spatial indices. One might also view this as an interference effect, such that the location of the encoded location attracts spatial attention since it has not been fully deleted/removed from working memory. Again, for the impact of the encoded location to be exactly zero after the retrieval cue, requires zero interference or instantaneous decay of the bound location information. It would be helpful for the authors to expand their discussion to further explain how the results fit within a broader theoretical framework and how it fits with empirical data on how quickly an irrelevant feature of an object can be deleted from working memory.

Thank you also for this point (that is related to the two points above). As we stated in our reply to comment 1 above, we agree that one possibility is that the past location is merely “sticky” and pulls the task-relevant future bias toward the past location. If so, our time courses suggest that such “pulling” occurs only until approximately 600 ms after cue onset, as the past bias is only transient. An alternative interpretation is that the past location may not be merely a residual irrelevant trace, but actually be useful and used by the brain.

For example, the encoded (past) item locations provide a coordinate system in which to individuate/separate the two memory items. While the future locations also provide such a coordinate system, the brain may benefit from holding onto both coordinate systems at the same time, rendering our observation of joint selection in both frames. Indeed, in a recent VR experiment in which we had participants (rather than the items) rotate, we also found evidence for the joint use of two spatial frames, even if neither was technically required for the upcoming task (see Draschkow, Nobre, van Ede, Nature Human Behaviour, 2022). Though highly speculative at this stage, such reliance on multiple spatial frames may make our memories more robust to decay and/or interference. Moreover, while past location was never explicitly probed in our task, in daily life the past location may sometimes (unexpectedly) become relevant, hence it may be useful to hold onto it, just in case. Thus, considering the past location merely as an “irrelevant feature” (that takes time to delete) may not do sufficient justice to the potential roles of retaining past locations of dynamic visual objects held in working memory.

As also stated in response to comment 1 above, we now added these relevant considerations to our Discussion:

Page 5 (Discussion): “In our study, the past location of the memory items was technically irrelevant for the task and could thus, in principle, be dropped after encoding. One possibility is that participants remapped the two memory items to their future locations soon after encoding, and had started – but not finished – dropping the past location by the time the cue arrived. In such a scenario, the past signal is merely a residual trace of the memory items that serves no purpose but still pulls gaze. Alternatively, however, the past locations may be utilised by the brain to help individuate/separate the two memory items. Moreover, by storing items with regard to multiple spatial frames (cf. 37) – here with regard to both past and future visual locations – it is conceivable that memories may become more robust to decay and/or interference. Also, while in our task past locations were never probed, in everyday life it may be useful to remember where you last saw something before it disappeared behind an occluder. In future work, it will prove interesting to systematically vary to the delay between encoding and cue to assess whether the reliance on the past location gradually dissipates with time (consistent with dropping an irrelevant feature), or whether the past trace remains preserved despite longer delays (consistent with preserving utility for working memory).”

Reviewer 3, Comments:

This study utilizes saccade metrics to explore, what the authors term the "past and future" of working memory. The study features an original design: in each trial, two pairs of stimuli are presented, first a vertical pair and then a horizontal one. Between these two pairs comes the cue that points the participant to one target of the first pair and another of the second pair. The task is to compare the two cued targets. The design is novel and original but it can be split into two known tasks - the first is a classic working memory task (a post-cue informs participants which of two memorized items is the target), which the authors have used before; and the second is a classic spatial attention task (a pre-cue signal that attention should be oriented left or right), which was used by numerous other studies in the past. The combination of these two tasks in one design is novel and important, as it enables the examination of the dynamics and overlapping processes of these tasks, and this has a lot of merit. However, each task separately is not new. There are quite a few studies on working memory and microsaccades and many on spatial attention and microsaccades. I am concerned that the interpretation of "past vs. future" could mislead readers to think that this is a new field of research, when in fact it is the (nice) extension of an existing one. Since there are so many studies that examined pre-cues and post-cues relative to microsaccades, I expected the interpretation here to rely more heavily on the existing knowledge base in this field. I believe this would have provided a better context of these findings, which are not only on "past" vs. "future" but also on "working memory" vs. "spatial attention".

Thank you for considering our findings novel and important, while at the same time reminding us of the parallels to prior tasks studying spatial attention in perception and working memory. We fully agree that our task likely engages both attention to the (past) memory item as well as spatial attention to the upcoming (future) test stimulus. At the same time, there is a critical difference in spatial attention for the future in our task compared with ample prior tasks engaging spatial cueing of attention for perception. In our task, the cue never directly cues the future location. Rather, it exclusively cues the relevant memory item. It is the memory item that is associated with the relevant future location, according to the future rule. This integration of the rule-based future location into the memory representation is distinct from classical spatial-attention tasks in which attention is cued directly to a specific location via, for example, a spatial cue such as an arrow.

Thus, if we wish to think about our task as engaging cueing of spatial attention for perception, we have to at least also invoke the process of cueing the relevant location via the appropriate memory item. We feel it is more parsimonious to think of this as attending to both the past and future location of a dynamic visual object in working memory.

If we return to our opening example, when we see a bird disappear behind a building, we can keep in working memory where we last saw it, while anticipating where it will re-appear to guide our external spatial attention. Here too, spatial attention is fully dependent on working-memory content (the bird itself) – mirroring the dynamic semng in our study. Thus, we believe our findings contribute a fresh perspective, while of course also extending established fields. We now contextualize our finding within the literature and clarify our unique contribution in our revised manuscript:

Page 5 (Discussion): “Building on the above, at face value, our task may appear like a study that simply combines two established tasks: tasks using retro-cues to study attention in working memory (e.g.,2,31-33) and tasks using pre-cues to study orienting of spatial attention to an upcoming external stimulus (e.g., 31,32,34–36). A critical difference with common pre-cue studies, however, is that the cue in our task never directly informed the relevant future location. Rather, as also stressed above, the future location was a feature of the cued memory item (according to the future rule), and not of the cue itself. Note how this type of scenario may not be uncommon in everyday life, such as in our opening example of a bird flying behind a building. Here too, the future relevant location is determined by the bird – i.e. the memory content – itself.”

Reviewer 2, Recommendations:

It would be helpful to set up predictions based on existing working memory models. Otherwise, the claim that the joint coding of past/future is "not trivial" is simply asserted, rather than contradicting an existing model or prior empirical results. If the non-trivial aspect is simply the ability to demonstrate the joint coding empirical through a good experimental design, make it clear that this is the contribution. For example, it may be that prevailing models predict exactly this finding, but nobody has been able to demonstrate it cleanly, as the authors do here. So the non-triviality is not that the result contradicts working memory models, but rather relates to the methodological difficulty of revealing such an effect.

Thank you for your recommendation. First, please see our point-by-point responses to the individual comments above, where we also state relevant changes that we have made to our article, and where we clarify what we meant with “non trivial”. As we currently also state in our introduction, our work took as a starting point the framework that working memory is inherently about the past while being for the future (cf. van Ede & Nobre, Annual Review of Psychology, 2023). By virtue of our unique task design, we were able to empirically demonstrate that visual contents in working memory are selected via both their past and their future-relevant locations – with past and future memory attributes being engaged together in time. With “not trivial” we merely intend to make clear that there are viable alternatives than the findings we observed. For example, past could have been replaced by the future, or it could have been that item selection (through its past location) was required before its future-relevant location could be considered (i.e. past-before-future, rather than joint selection as we reported). We outline these alternatives in the second paragraph of our Discussion:

Page 5 (Discussion): “Our finding of joint utilisation of past and future memory attributes emerged from at least two alternative scenarios of how the brain may deal with dynamic everyday working memory demands in which memory content is encoded at one location but needed at another.

First, [….]”

Our work was not motivated from a particular theoretical debate and did not aim to challenge ongoing debates in the working-memory literature, such as: slot vs. resource, active vs. silent coding, decay vs. interference, and so on. To our knowledge, none of these debates makes specific claims about the retention and selection of past and future visual memory attributes – despite this being an important question for understanding working memory in dynamics everyday semngs, as we hoped to make clear by our opening example.

Reviewer 3, Recommendations:

I recommend that the present findings be more clearly interpreted in the context of previous findings on working memory and attention. The task design includes two components - the first (post-cue) is a classic working memory task and the second (the pre-cue) is a classic spatial attention design. Both components were thoroughly studied in the past and this previous knowledge should be better integrated into the present conclusions. I specifically feel uncomfortable with the interpretation of past vs. future. I find this framework to be misleading because it reads like this paper is on a topic that is completely new and never studied before, when in fact this is a study on the interaction between working memory and spatial attention. I recommend the authors minimize this past-future framing or be more explicit in explaining how this new framework relates to the more common terminology in the field and make sure that the findings are not presented in a vacuum, as another contribution to the vibrant field that they are part of.

Thank you for these recommendations. Please also see our point-by-point responses to the individual comments above. Here, we explained our logic behind using the terminology of past vs. future (in addition, see also our response to point 2 or reviewer 2). Here, we also stated relevant changes that we have made to our manuscript to explain how our findings complement – but are also distinct from – prior tasks that used pre-cues to direct spatial attention to an upcoming stimulus. As we explained above, in our task, the cue itself never contained information about the upcoming test location. Rather, the upcoming test location was a property of the memory item (given the future rule). Hence, we referred to this as a “future attribute” of the cued memory item, rather than as the “cued location” for external spatial attention. Still, we agree the future bias likely (also) reflects spatial allocation to the upcoming test array, and we explicitly acknowledge this in our discussion. For example:

Page 5 (Discussion): “This signal may reflect either of two situations: the selection of a future-copy of the cued memory content or anticipatory attention to its the anticipated location of its associated test-stimulus. Either way, by the nature of our experimental design, this future signal should be considered a content-specific memory attribute for two reasons. First, the two memory contents were always associated with opposite testing locations, hence the observed bias to the relevant future location must be attributed specifically to the cued memory content. Second, we cued which memory item would become tested based on its colour, but the to-be-tested location was dependent on the item’s encoding location, regardless of its colour. Hence, consideration of the item’s future-relevant location must have been mediated by selecting the memory item itself, as it could not have proceeded via cue colour directly.”

Page 6 (Discussion): “Building on the above, at face value, our task may appear like a study that simply combines two established tasks: tasks using retro-cues to study attention in working memory (e.g.,2,31-33) and tasks using pre-cues to study orienting of spatial attention to an upcoming external stimulus (e.g., 31,32,34–36). A critical difference with common pre-cue studies, however, is that the cue in our task never directly informed the relevant future location. Rather, as also stressed above, the future location was a feature of the cued memory item (according to the future rule), and not of the cue itself. Note how this type of scenario may not be uncommon in everyday life, such as in our opening example of a bird flying behind a building. Here too, the future relevant location is determined by the bird – i.e. the memory content – itself.”

Author response:

The following is the authors’ response to the previous reviews.

Public Reviews:

Reviewer #1 (Public Review):

Gap junction channels establish gated intercellular conduits that allow the diffusion of solutes between two cells. Hexameric connexin26 (Cx26) hemichannels are closed under basal conditions and open in response to CO2. In contrast, when forming a dodecameric gapjunction, channels are open under basal conditions and close with increased CO2 levels. Previous experiments have implicated Cx26 residue K125 in the gating mechanism by CO2, which is thought to become carbamylated by CO2. Carbamylation is a labile post-translational modification that confers negative charge to the K125 side chain. How the introduction of a negative charge at K125 causes a change in gating is unclear, but it has been proposed that carbamylated K125 forms a salt bridge with the side chain at R104, causing a conformational change in the channel. It is also unclear how overall gating is controlled by changes in CO2, since there is significant variability between structures of gap-junction channels and the cytoplasmic domain is generally poorly resolved. Structures of WT Cx26 gap-junction channels determined in the presence of various concentrations of CO2 have suggested that the cytoplasmatic N-terminus changes conformation depending on the concentration of the gas, occluding the pore when CO2 levels are high.

In the present manuscript, Deborah H. Brotherton and collaborators use an intercellular dyetransfer assay to show that Cx26 gap-junction channels containing the K125E mutation, which mimics carbamylation caused by CO2, is constitutively closed even at CO2 concentrations where WT channels are open. Several cryo-EM structures of WT and mutant Cx26 gap junction channels were determined at various conditions and using classification procedures that extracted more than one structural class from some of the datasets. Together, the features on each of the different structures are generally consistent with previously obtained structures at different CO2 concentrations and support the mechanism that is proposed in the manuscript. The most populated class for K125E channels determined at high CO2 shows a pore that is constricted by the N-terminus, and a cytoplasmic region that was better resolved than in WT channels, suggesting increased stability. The K125E structure closely resembles one of the two major classes obtained for WT channels at high CO2. These findings support the hypothesis that the K125E mutation biases channels towards the closed state, while WT channels are in an equilibrium between open and closed states even in the presence of high CO2. Consistently, a structure of K125E obtained in the absence of CO2 appeared to also represent a closed state but at lower resolution, suggesting that CO2 has other effects on the channel beyond carbamylation of K125 that also contribute to stabilizing the closed state. Structures determined for K125R channels, which are constitutively open because arginine cannot be carbamylated, and would be predicted to represent open states, yielded apparently inconclusive results.

A non-protein density was found to be trapped inside the pore in all structures obtained using both DDM and LMNG detergents, suggesting that the density represents a lipid rather than a detergent molecule. It is thought that the lipid could contribute to the process of gating, but this remains speculative. The cytoplasmic region in the tentatively closed structural class of the WT channel obtained using LMNG was better resolved. An additional portion of the cytoplasmic face could be resolved by focusing classification on a single subunit, which had a conformation that resembled the AlphaFold prediction. However, this single-subunit conformation was incompatible with a C6-symmetric arrangement. Together, the results suggest that the identified states of the channel represent open states and closed states resulting from interaction with CO2. Therefore, the observed conformational changes illuminate a possible structural mechanism for channel gating in response to CO2.

Some of the discussion involving comparisons with structures of other gap junction channels are relatively hard to follow as currently written, especially for a general readership. Also, no additional functional experiments are carried out to test any of the hypotheses arising from the data. However, structures were determined in multiple conditions, with results that were consistent with the main hypothesis of the manuscript. No discussion is provided, even if speculative, to explain the difference in behavior between hemichannels and gap junction channels. Also, no attempt was made to measure the dimensions of the pore, which is relevant because of the importance of identifying if the structures indeed represent open or closed states of the channel.

We have considerably revised the manuscript in an attempt to make it more tractable. We respond to the individual comments below.

Reviewer #2 (Public Review):

Summary:

The manuscript by Brotherton et al. describes a structural study of connexin-26 (Cx26) gap junction channel mutant K125E, which is designed to mimic the CO2-inhibited form of the channel. In the wild-type Cx26, exposure to CO2 is presumed to close the channel through carbamylation of the residue K125. The authors mutated K125 to a negatively charged residue to mimic this effect, and they observed by cryo-EM analysis of the mutated channel that the pore of the channel is constricted. The authors were able to observe conformations of the channel with resolved density for the cytoplasmic loop (in which K125 is located). Based on the observed conformations and on the position of the N-terminal helix, which is involved in channel gating and in controlling the size of the pore, the authors propose the mechanisms of Cx26 regulation.

Strengths:

This is a very interesting and timely study, and the observations provide a lot of new information on connexin channel regulation. The authors use the state of the art cryo-EM analysis and 3D classification approaches to tease out the conformations of the channel that can be interpreted as "inhibited", with important implications for our understanding of how the conformations of the connexin channels controlled.

Weaknesses:

My fundamental question to the premise of this study is: to what extent can K125 carbamylation by recapitulated by a simple K125E mutation? Lysine has a large side chain, and its carbamylation would make it even slightly larger. While the authors make a compelling case for E125-induced conformational changes focusing primarily on the negative charge, I wonder whether they considered the extent to which their observation with this mutant may translate to the carbamoylated lysine in the wild-type Cx26, considering not only the charge but also the size of the modified side-chain.

This is an important point. We agree that the difference in size will have a different effect on the structure. For kinases, aspartate or glutamate are often used as mimics of phosphorylated serine or threonine and these will have the same issues. The fact that we cannot resolve the relevant side-chains in the density may be indicative that the mutation doesn’t give the whole story. It may be able to shift the equilibrium towards the closed conformation, but not stably trap the molecule in that conformation. We include a comment to this effect in the revised manuscript.

Reviewer #3 (Public Review):

Summary:

The mechanism underlying the well-documented CO2-regulated activity of connexin 26 (Cx26) remains poorly understood. This is largely due to the labile nature of CO2-mediated carbamylation, making it challenging to visualize the effects of this reversible posttranslational modification. This paper by Brotherton et al. aims to address this gap by providing structural insights through cryo-EM structures of a carbamylation-mimetic mutant of the gap junction protein.

Strengths:

The combination of the mutation, elevated PCO2, and the use of LMNG detergent resulted in high-resolution maps that revealed, for the first time, the structure of the cytoplasmic loop between transmembrane helix (TM) 2 and 3.

Weaknesses:

The presented maps merely reinforce their previous findings, wherein wildtype Cx26 favored a closed conformation in the presence of high PCO2. While the structure of the TM2-TM3 loop may suggest a mechanism for stabilizing the closed conformation, no experimental data was provided to support this mechanism. Additionally, the cryo-EM maps were not effectively presented, making it difficult for readers to grasp the message.

We have extensively revised the manuscript so that the novelty of this study is more apparent. There are three major points

(1) The carbamylation mimetic pushes the conformation towards the closed conformation. Previously we just showed that CO2 pushes the conformation towards this conformation. Though we could show this was not due to pH, and could speculate this was due to carbamylation as suggested by previous mutagenesis studies, our data did not provide any mechanism whereby Lys125 was involved.

(2) In going from the open to closed conformations, not only is a conformational change in TM2 involved, as we saw previously, but also a conformational change in TM1, the linker to the N-terminus and the cytoplasmic loop. Thus there is a clear connection between Lys125 and the conformation of the pore-closing N-terminus.

(3) We observe for the first time in any connexin structure, density for the cytoplasmic loop. Since this loop is important in regulation, knowing how it might influence the positions of the transmembrane helices is important information if we are to understand how connexins can be regulated.

Reviewing Editor:

The reviewers have agreed on a list of suggested revisions that would improve the eLife assessment if implemented, which are as follows:

(1) For completeness, Figure 1 could be supplied with an example of how the experiment would look like in the presence of CO2 - for the wild-type and for the K125E mutant. presumably for the wild-type this has been done previously in exactly this assay format, but this control would be an important part of characterization for the mutant. Page 4, lines 105106; "unsurprisingly, Cx26K125E gap junctions remain closed at a PCO2 of 55 mmHg." The data should be presented in the manuscript.

We have now included the data with a PCO2 of 55mmH. This is now Figure 4 in our revised manuscript.

(2) Would AlphaFold predictions show any interpretable differences in the E125 mutant, compared to the K125 (the wild-type)?

We tried this in response to the reviewer’s suggestion. We did not see any interpretable differences. In general AlphaFold is not recognised as giving meaningful information around point mutations.

(3) The K125R mutant appears to be a more effective control for extracting significant features from the K125E maps. Given that the use of a buffer containing high PCO2 is essential for obtaining high-resolution maps, wildtype Cx26 is unsuitable as an appropriate control. The K125R map, obtained at a high resolution (2.1Å), supports its suitability as a robust control.

Though we are unsure what the referee is referring to here, we have rewritten this section and compare against the K125R map (figure 5a) as well as that derived from the wild-type protein. The important point is that the K125E mutant, causes a structural change that is consistent with the closure of the gap junctions that we observe in the dye-transfer assays.

(4) Likewise, the rationale for using wildtype Cx26 maps obtained in DDM is unclear. Wildtype Cx26 seems to yield much better cryo-EM maps in LMNG. We suggest focusing the manuscript on the higher-quality maps, and providing supporting information from the DDM maps to discuss consistency between observations and the likely possibility that the nonprotein density in the pore is lipid and not detergent.

The rationale for comparing the mutants against the wt Cx26 maps obtained in DDM was because the mutants were also solubilised in DDM. However, taking the lead from the referees’ comments, we have now rewritten the manuscript so that we first focus on the data we obtain from protein solubilised in LMNG. We feel this makes our message much clearer.

(5) In general, the rationale for utilizing cryo-EM maps with the entire selected particles is unclear. Although the overall resolutions may slightly improve in this approach, the regions of interest, such as the N-terminus and the cytoplasmic loop, appear to be better ordered afer further classifications. The paper would be more comprehensible if it focuses solely on the classes representing the pore-constricting N-terminus (PCN) and the pore-open flexible Nterminus (POFN) conformations. Also, the nomenclatures used in the manuscript, such as "WT90-Class1", "K125E90-1", "LMNG90-class1", "LMNG90-mon-pcn" are confusing.

LMNG90s are also wildtype; K125E-90-1 is in Class1 for this mutant and is similar to WT90Class2, which represents the PCN conformation. More consistent and intuitive nomenclatures would be helpful.

We agree with the referees’ comments. This should now be clearer with our rewritten manuscript where we have simplified this considerably. We now call the conformations NConst (N-terminus defined and constricting the pore) and NFlex (N-terminus not visible) and keep this consistent throughout.

(6) A potential salt bridge between the carbamylated K125 and R104 is proposed to account for the prevalence of Class-1 (i.e., PCN) in the majority of cryo-EM particles. However, the side chain densities are not well-defined, suggesting that such an interaction may not be strong enough to trap Cx26 in a closed conformation. Furthermore, the absence of experimental data to support this mechanism makes it unclear how likely this mechanism may be. Combining simple mutagenesis, such as R104E, with a dye transfer assay could offer support for this mechanism. Are there any published experimental results that could help address this question without the need for additional experimental work? Alternatively, as acknowledged in the discussion, this mechanism may be deemed as an "over-simplification." What is an alternative mechanism?

R104 has been mutated to alanine in gap junctions and tested in a dye transfer assay as now mentioned in the text (Nijar et al, J Physiol 2021) supporting this role. In hemichannels R104 has been mutated to both alanine and glutamate and tested through dye loading assays Meigh et al, eLife 2013). Also in hemichannels R104 and K125 have been mutated to cysteines allowing them to be cross-linked through a disulphide bond. This mutant responds to a change in redox potential in a similar way to which the wild type protein responds to CO2 (Meigh et al, Open Biol 2015). Therefore, there is no doubt that the residues are important for the mechanism and the salt-bridge interaction seems a plausible mechanism to reconcile the mutagenesis data, however we cannot be sure that there are not other interactions involved that are necessary for closure. This information has now been included in the text.

(7) The cryo-EM maps presented in the manuscript propose that gap junctions are constitutively open under normal PCO2 as the flexible N-terminus clears the solute permeation pathway in the middle of the channel. However, hemichannels appear to be closed under normal PCO2. It is puzzling how gap junctions can open when hemichannels are closed under normal PCO2 conditions. If this question has been addressed in previous studies, the underlying mechanism should be explicitly described in the introduction. If it remains an open question, differences in the opening mechanisms between hemichannels and gap junctions should be investigated.

We suspect this is due to the difference in flexibility of gap junctions relative to hemichannels. However, a discussion of this is beyond this paper and would be complete speculation based on hemichannel structures of other connexins, performed in different buffering systems. There are no high resolution structures of Cx26 hemichannels.

(8) A mystery density likely representing a lipid is abruptly introduced, but the significance of this discovery is unclear. It is hard to place the lipid on Figure S6 in the wider context of everything else that is discussed in the text. It would be helpful for readers if a figure were provided to show where the density is located in relation to all the other regions that are extensively discussed in the text.

In the revised text this section has been completely rewritten. We have now include a more informative view in a new figure (Figure 1 – figure supplement 3).

(9) Including and displaying even tentative pore-diameter measurements for the different states - this would be helpful for readers and provide a more direct visual cue as to the difference between open and closed states.

We have purposely avoided giving precise measurements to the pore-diameter, since this depends on how we model the N-terminus. The first three residues are difficult to model into the density without causing stearic clashes with the neighbouring subunits.

(10) Given that no additional experiments for channel function were carried out, it would be useful if to provide a more detailed discussion of additional mutagenesis results from the literature that are related to the experimental results presented.

We have amplified this in the discussion (see answer to point 6).

The reviewers also agreed that improvements in the presentation of the data would strengthen the manuscript. Here is a summary list of suggestions by reviewers aimed at helping improve how the data is presented:

(1) Why is the pipette bright green in the top image, but rather weakly green in the bottom image in Figure 1 - is this the case for all images?

(Now figure 4) This depends on whether the pipette was in the focal plane of view or not. The important point of these images is the difference in intensity of the donor vs the recipient cell. The graphs in figure 4c illustrate clearly the difference between the wild-type and the mutant gap junctions.

(2) In figures 2-5, labels would help a lot in understanding what is shown - while the legends do provide the information on what is presented, it would help the reader to see the models/maps with labels directly in the panel. For example, Figure 2a/b - just indicating "WT90 Cx26" in pink and "K125E90" in blue directly in the panel would reduce the work for the reader.

We have extensively modified the labels in the figures to address this issue.

(3) Figure 4 - magenta and pink are fairly close, and to avoid confusion it might be useful to use a different color selection. This is especially true when structures are overlayed, as in this figure - the presentation becomes rather complicated, so the less confusion the color code can introduce, the better.

(Now Figure 2) We have now changed pink to blue.

(4) Figure 5 - a remarkably under-labelled figure.

Now added labels.

(5) Figure 6 - it would be interesting to add a comparison to Cx32 here as well for completeness, since the structure has been published in the meantime.

Cx32 has now been included.

(6) Figure 7 - please add equivalent labels on both sides of the model, left and right. Add the connecting lines for all of the tubes TM helices - this will help trace the structural elements shown. The legend does not quite explain the colors.

We have modified the figure as suggested and explained the colours in the legend.

(8) Fig.1 legend; Unclear what mCherry fluorescence represents. State that Cx26 was expressed as a translational fusion with mCherry.

Now figure 4. We have now written “Montages each showing bright field DIC image of HeLa cells with mCherry fluorescence corresponding to the Cx26K125E-mCherry fusion superimposed (leftmost image) and the permeation of NBDG from the recorded cell to coupled cells.”

(9) Fig. 3 b); Show R104 in the figure. Also E129-R98/R99 interaction is hard to acknowledge from the figure. It seems that the side chain density of E129 is not strong enough to support the modeled orientation.

This is now Figure 1c. While the density in this region is sufficient to be confident of the main chain, we agree that the side chain density for the E129-R98/R99 interaction is not sufficiently clear to draw attention to and have removed the associated comment from the figure legend. The density is focussed on the linker between TM1 and the N-terminus and the KVRIEG motif. We prefer to omit R104, in order to keep the focus on this region. As described in the manuscript, the density for the R104 side chain is poor.

(10) Fig. 3 c); Label the N-terminus and KVRIEG motif in the figure.

Now Figure 1b. We have labelled the N-terminus. The KVRIEG motif is not visible in this map.

(11) Page 9, lines 246-248; Restate, "We note, however, density near to Lys125, between Ser19 in the TM1-N-term linker, Tyr212 of TM4 and Tyr97 on TM3 of the neighbouring subunit, which we have been unable to explain with our modelling."

We have reworded this.

(12) Page 14, line 399; Patch clamp recording is not included in the manuscript.

Patch clamp recordings were used to introduce dye into the donor cell.

(13) On the same Figure 2, clashes are mentioned but these are hard to appreciate in any of the figures shown. Perhaps would be useful to include an inset showing this.

We have modified Figure 2b slightly and added an explanation to highlight the clash. It is slightly confusing because the residues involved belong to neighbouring subunits.

(14) The discussion related to Figure 6 is very hard to follow for readers who are not familiar with the context of abbreviations included on the figure labels. This figure could be improved to allow a general readership to identify more clearly each of the features and structural differences that are discussed in the text.

We have extensively changed the text and updated the labels on the figure to make it much easier for the reader to follow.

Below, you can find the individual reviews by each of the three reviewers.

Reviewer #1 (Recommendations For The Authors):

(1) In Figure 2d-e, the text discusses differences between K125E 90-1 and WT 90-class2 (7QEW), yet the figure compares K125E with 7QEQ. I suggest including a figure panel with a comparison between the two structures discussed in the manuscript text.

This has been changed in the revised manuscript.

Other comments have been addressed above.

Author response:

The following is the authors’ response to the previous reviews.

The reviewers thoughtful comments have helped us make the manuscript both more comprehensive and clearer. Thank you for your time and effort. We know that this is a long and technical paper. In our responses we refer to three documents:

• Original: the first original submission

• Revision: the revised document (02 MillardFranklinHerzog2023 v2.pdf)

• Difference: a document that shows the changes made to text (but not figures or tables) from the original to revision (03 MillardFranklinHerzog2023 diff.pdf).

Reviewer #1 (Recommendations For The Authors):

(1) In general, the paper is well written and addresses important questions of muscle mechanics and muscle modeling. In the current version, the model limitations are briefly summarized in the abstract. However, the discussion needs a more complete description of limitations as well as a discussion of types of data (in vivo, ex vivo, single fiber, wholes muscle, MTU, etc.) that can be modeled using this approach.

Please see the response to comment 23 for more details of the limitations that have been added to the revised document.

(2) The choice of a model with several tendon parameters for simulating single muscle fiber experiments is not well justified.

A rigid-tendon model with a slack length of zero was, in fact, used for these simulations for both the VEXAT and Hill models. In case this is still not clear: a rigid-tendon model of zero length is equivalent to no tendon at all. The text that first mentions the tendon model has now been modified to make it clearer that the parameters of the model were set to be consistent with no tendon at all:

Please see the following text:

Original:

• page 17, column 1, line 28 ”... rigid tendon of zero length,”

• page 17, column 1, line 51 ”... rigid tendon of zero length.”

Revision:

• page 19, column 1, line 19 ”... we used a rigid-tendon of zero length (equivalent to ignoring the tendon)”

• page 19, column 1, line 38 ”... coupled with a rigid-tendon of zero-length.”

Difference:

• page 21, column 1, line 19 ”... we used a rigid-tendon... ”

• page 21, column 1, line 45 ”... rigid-tendon of zero length ...”

(3) A table that clarifies how all model parameters were estimated needs to be included in the main part of the manuscript.

Two tables have been added to the manuscript that detail the parameters of the elastic-tendon cat soleus model (in the main body of the text) and the rabbit psoas fibril model (in an appendix). Each table includes:

• A plain language parameter name

• The mathematical symbol for the parameter

• The value and unit of the parameter

• A coded reference to the data source that indicates both the experimental animal and how the data was used to evaluate the parameter.

Please see the following text:

Revision:

• page 11

• page 42

Difference:

• page 11

• page 46

(4) The supplemental information is not properly referenced in the main text. There are a number of smaller issues that also need to be addressed.

Thank for your attention to detail. The following problems related to Appendix referencing have been fixed:

• Appendices are now parenthetically referenced at the end of a sentence. However, a few references to figures (that are contained within anAppendix) still appear in the body of the sentence since moving these figure references makes the text difficult to understand.

• All Appendices are now referenced in the main body of the text.

(5) Abstract, line 6: While it is commonly assumed that the short range stiffness of muscle is due to cross bridges, Rack & Westbury (1974) noted that it occurs over a distance of 25-35 nm, and that many cross-bridges must be stretched even farther than this distance (their p. 348 middle). It seems unlikely that cross-bridges alone can actually account for the short-range stiffness.

There are three parts to our response to this comment:

(a) Rack & Westbury’s definition of short-range-stiffness and unrealistic cross-bridge stretches

(b) Rack & Westbury’s definition of short-range-stiffness vs. linear-timeinvariant system theory

(c) Updates to the paper

a. Rack & Westbury’s definition of short-range-stiffness and unrealistic cross-bridge stretches.

As you note, on page 348, Rack and Westbury write that ”If the short range stiffness is to be explained in terms of extension of cross-bridges, then many of them must be extended further than the 25-35 nm mentioned above.” Having re-read the paper, its not clear how these three factors are being treated in the 25−35 nm estimate:

• the elasticity of the tendon and aponeurosis,

• the elasticity of actin and myosin filaments,

• and the cycling rate of the cross-bridges.

Obviously the elasticity of the tendon, aponeurosis, actin, and myosin filaments will reduce the estimated amount of crossbridge strain during Rack and Westbury’s experiments. A potentially larger factor is the cycling rate of each cross-bridge. If each crossbridge cycles faster than 11 Hz (the maximum frequency Rack and Westbury used), then no single crossbridge would stretch by 25-35 nm. So why didn’t Rack and Westbury consider the cycling rate of crossbridges?

Rack and Westbury’s reasoned that a perfectly elastic work loop would necessarily mean that all crossbridges stayed attached: as soon as a crossbridge cycles it would release its stored elastic energy and the work loop would no longer be elastic. Since Rack and Westbury measured some nearly perfect elastic work loops (the smallest loops in Fig. 2,3, and 4), I guess they assumed crossbridges remained attached during the 25-35 nm crossbridge stretch estimate. However, even Rack and Westbury note that none of the work loops they measured were perfectly elastic and so there is room to entertain the idea that crossbridges are cycling.

Fortunately, for this discussion, crossbridge cycling rates have been measured.

In-vitro measurements by Uyeda et al. show that crossbridges are cycling at 30 Hz when moving at 0.5-1.2 length/s. At this rate, there would be enough time for a single crossbridge to cycle nearly 2.72 times for every cycle of the 11 Hz sinusoidal perturbations, reducing its expected strain from 25-35 nm down to 9.2−12.9µm. This effect becomes even more pronounced if crossbridge cycling rate is used to explain the difference in sliding velocity between Uyeda et al.’s in-vitro data (0.5-1.2 length/s) and the maximum contraction velocity of an in-situ cat soleus (4.65 lengths/s, Scott et al.).

b. Rack & Westbury’s definition of short-range-stiffness vs. linear-time-invariant system theory

Rack and Westbury defined short-range-stiffness to describe a specific kind of force response of the muscle to cyclical length changes:

• muscle force is linear with length change,

• and independent of velocity.

Rack and Westbury’s definition therefore fails when viscous forces become noticeable, because viscous forces are velocity dependent.

On line 6 of the abstract the term ‘short-range-stiffness’ is not used because Rack and Westbury’s definition is too narrow for our purposes. Instead we are using the more general approach of approximating muscle as a linear-timeinvariant (LTI) system, where it is assumed that

• the response of the system is linear

• and time invariant.

To unpack that a little, a muscle is considered in the ‘short-range’ in our work if it meets the criteria of a linear time-invariant (LTI) system:

• the force response of muscle can be accurately described as a linear function of its length and velocity (its state)

• and its response is not a function of time (which means constant stimulation, and no fatigue).

In contrast to Rack and Westbury’s definition, the ‘short-range’ in linear systems theory is general enough to accommodate both elastic and viscous forces. In physical terms, small for an LTI approximation of muscle is larger than the short-range defined by Rack and Westbury: an LTI system can include velocity dependence, while short-range-stiffness ends when velocity dependence begins.

c. Updates to the paper

To make the differences between Rack and Westbury’s ‘short-range-stiffness’ and LTI system theory clearer: - We have removed all occurrences of ‘short-range’ that were associated with Kirsch et al. and have replaced this phrase with ‘small’.

• On the first mention of Kirsch’s work we have made the wording more specific

Revision:

• page 1, column 1, lines 4,5

• page 1, column 2, lines 14-21 ”Under constant activation ...”

Difference: page 1, column 2, line 19-26

• page 1, column 1, lines 4,5

• page 1, column 2, lines 20-27 ”Under constant activation ...”

• A footnote has been added to contrast the definition of ‘small’ in the context of an linear time invariant system to ‘short-range’ in the context of Rack and Westbury’s definition of short-range-stiffness.

Revision: page 1, column 2, bottom

Difference: page 1, column 2, bottom

• In addition, we have added a brief overview of LTI system theory to make the analysis and results more easily understood:

Revision: Figure 4 paragraph beginning on page 10, column 2, line 15 ”As long as ...”

Difference: Figure 4 paragraph beginning on page 12, column 1, line 46 ”As long as ...”

(6) Page 3, lines 6-8: It also seems unlikely that 25% of cross-bridges are attached at one time (Howard, 1997) even for supramaximal isometric stimulation. The number should be less than 20%. What would the ratio of load path stiffness be for low force movements such as changing the direction of a frictionless manipulandum or slow walking? The range of relative stiffnesses is of more interest than the upper limit.

We have made the following updates to address this comment:

• A 20% duty cycle now defines the upper bound stiffness of the actinmyosin load path.

• We have also evaluated the lower bound actin-myosin stiffness when a single crossbridge is attached.

• The stiffness of titin from Kellermayer et al. has been digitized at a length of 2 µm and 4 µm to more accurately capture the length dependence of titin’s stiffness.

• We have added a new figure (Figure 14) to make it easier to compare the range of actin-myosin stiffness to titin-actin stiffness.

• The text in the main body of the paper and the Appendix has been updated.

• The script ’main ActinMyosinAndTitinStiffness.m’ used to perform the calculations and generate the figure is now a part of the code repository.

Please see the following text:

Revision

• The paragraph beginning at page 2, column 2, line 45 ”The addition of a titin element ...”

• Appendix A

• Figure 14 (in Appendix A)

Difference

• The paragraph beginning at page 3, column 1, line 6: ”The addition of a titin element ...”

• Appendix A

• Figure 14 (in Appendix A)

(7) Page 5, line 12: A word seems to be missing here, ”...together to further...”.

Thank you for your attention to detail. The sentence has been corrected.

Please see the following text:

• Revision: page 4, column 2, line 40 ”... into a single ...”

• Difference: page 5, column 1, line 18

(8) Page 5, line 24-27: These ”theories” are not mutually exclusive, and it is misleading to suggest they are. There is evidence for binding of titin to actin at multiple locations and there is no reason why evidence supporting one binding location must detract from the evidence supporting other binding locations.

The text has been modified to make it clear to readers that the different titinactin binding locations are not mutually exclusive. Please see the following text:

• Revision: page 5, column 1, lines 17-19, the sentence beginning ”As previously mentioned, ...”

• Difference: page 5, column 1, lines 41-44

(9) Page 5, lines 48-51: Should cite Kellermayer and Granzier (1996) not Kellermayer et al. (1997).

The reference to ‘Kellermayer et al.’ has been changed to ‘Kellermayer and Granzier’. The comment that the year of the reference should be changed from (1997) to (1996) is confusing: the 1996 paper is being referenced.

For further details please see:

• Revision: page 5, column 1, 39-40

• Difference: page 5, column 2, line 19-22

(10) Also, Dutta et al. (2018) should be cited as further showing that N2A titin by itself slows actin motility on myosin.

Thank you for the suggestion. The sentence has been modified to include Dutta et al.:

For further details please see:

• Revision: page 5, column 1, 40

• Difference: page 5, column 2, line 19-22

(11) Figure 2 legend and elsewhere: it is odd to say that experiments used ”a cat soleus” when more than one cat coleus was used. Change to ”cat coleus”. See also page 15, line 15.

Thank you for your attention to detail. All occurrences of ‘a cat soleus’ have been changed, with some sentence revision, to ‘cat soleus’.

(12) Page 6, line 10: It is not clear why an MTU was used to simulate single muscle fiber experiments. What is the justification for choosing this particular model? Also, the choice of model might explain why the version with stiff tendon performs better than the version with an elastic tendon, but this is never mentioned. Why not use a muscle model with no tendon (e.g., Wakeling et al., 2021 J. Biomech.)?

Please see the response to comment 2.

(13) Millard et al.’s activation dynamics model also fails to capture the lengthdependence of activation dynamics (Shue and Crago, 1998; Sandercock and Heckman, 1997), which should be noted in the discussion along with other limitations.

An additional limitations paragraph is in the revised manuscript that addresses this comment specifically. However, we have used Stephenson and Wendt as a reference for the shift in peak isometric force that comes with submaximal activation. In addition, we also reference Chow and Darling for the property that the maximum shortening velocity is reduced with submaximal activations.

• Revision: page 22, column 1, line 41 ”Finally, the VEXAT model ...”

• Difference: page 24, column 2, line 12 ”Finally, the VEXAT model ...”

In addition, please see the response to comment 23.

(14) Page 6, line 22: ”An underbar...”.

Thank you for your attention to detail, this correction has been made.

(14) Page 7, lines 27-32: This and other issues should be described in the Discussion under a heading of model limitations.

Please see the response to comment 23.

(15) Page 7, lines 43-44: Numerous papers from the last author’s laboratory contradict the claim that there is no force enhancement on the ascending limb by demonstrating that force enhancement does occur on the ascending limb (see e.g., Leonard & Herzog 2002, Peterson et al., 2004 and several papers from the Rassier laboratory).

Thank you for your attention to detail. This statement is in error and has been removed. To improve this section of the paper, a paragraph has been added to briefly mention the experimental observations of residual force enhancement before proceeding to explain how this phenomena is represented by the model.

Please see the following text:

Revision:

• the paragraph starting on page 7, column 2, line 43 ”When active muscle is lengthened, ...”

• and the following paragraph starting on page 8, column 1, line 3 “To develop RFE, ”

Difference:

• the paragraph starting on page 8, column 2, line 15

• and the following paragraph starting on page 9, column 1, line 6

(17) Figure 3 legend and elsewhere: The authors use Prado et al. (2005) to determine several titin parameters, however the simulations seem to focus on cat soleus, but Prado et al.’s paper is on rabbits. More clarity is needed about which specific results from which species and muscles were used to parameterize the model.

The new parameter table includes coded entries to indicate the literature source for experimental data, the animal it came from, and how the data was used. For example, the ‘ECM fraction’ has a source of ‘R[57]’ to show that the data came from rabbits from reference 57. For further details, please see the response to comment #3

Please see the following text:

• Revision: page 11, column 2, table section H: ‘ECM fraction’.

• Difference: page 11, column 2, table section H: ‘ECM fraction’.

To address this comment in a little more detail, we have had to use Prado et al. (2005) to give us estimates for only one parameter: P, the fraction of the passive force-length relation that is due to titin. Prado et al.’s measurements relating to P are unique to our knowledge: these are the only measurements we have to estimate P in any muscle, cat soleus or otherwise. Here we use the average of the values for P across the 5 muscles measured by Prado et al. as a plausible default value for all of our simulations.

(18) Figure 4 seems unnecessary.

Figure 4 has been removed.

(19) Page 10, lines 17-18: provide the abbreviation (VAF) here with the definition (variance accounted for).

Thank you for your attention to detail. The abbreviation has been added.

Please see these parts of the manuscripts for details:

• Revision: page 12, column 2, line 13

• Difference: page 13, column 2, line 32

(20) Page 11, lines 2-3: Here and elsewhere, it is clear that some model parameters have been optimized to fit the model. The main paper should include a table that lists all model parameters and how they were chosen or optimized, including but not limited to the information in Table 1 of the supplemental information section.

See response to comment 3.

(20) Page 17, lines 45 -49: Again, a substantial number of ad hoc adjustments to the model appear to be required. These should be described in the Discussion under limitations, and accounted for in the parameters table. See also legends to Fig. 12 and 13, page 19, lines 23-26.

Please see the response to comment #3: a coded entry now appears to indicate the data source, the animal used in the experiment, and the method used to process the data. This includes entries for parameters which were estimated

‘E’ so that the model produced acceptable results in the simulations presented. In addition, the new discussion paragraph includes a number of sentences that use the adjustment to the active-titin-damping coefficient as an opening to discuss the limitations of the VEXAT’s titin-actin bond model and the circumstances under which the model’s parameters would need to be adjusted.

Please see responses to comments 3 and 23 for additional details. In addition, please see the specific discussion text mentioning the change to βoPEVK:

• Revision: page 22, column 1, line 30 ”In Sec. 3.3 we had ...”

• Difference: page 24, column 1, line 49

(22) Page 20, lines 50-11: It should be noted here that Tahir et al.’s (2018) model has both series and parallel elastic elements, provided by superposition of rotation (series) and translation (parallel) of a pulley.

While it is true that Tahir et al.’s (2018) model has series and parallel elements, as do the other models mentioned, these models do not have the correct structure to yield a gain and phase response that mimics biological muscle. The text that I originally wrote attempted to explain this without going into the details. As you note, this explanation leaves something to be desired. The original text commenting on the models of Forcinito et al, Tahir et al, Haeufle et al., and Gunther et al. has been updated to be more specific.¨ Please see the parts of the following manuscripts for details:

• Revision: page 22, column 2, line 20, the paragraph beginning ”The models of Forcinito ...”

• Difference: page 24, column 2, line 44

(23) Discussion: This section should include a description of model limitations, including the relatively large number of ad hoc modifications and how many parameters must be found by optimization in practice. The authors should discuss what types of data are most compatible for use with the model (ex vivo, in vivo, single fiber, whole muscle, MTU), requirements for applying the model to different types of data, and impediments to using the model on different types of data.

An additional limitations paragraph has been added to the discussion.

Please see the following text:

• Revision: the paragraph beginning on page 22, column 1, line 11 ”Both the viscoelastic ...”

• Difference: the paragraph beginning on page 24, column 1, line 27.

Reviewer #2 (Recommendations For The Authors):

(1) If it is possible to compare the output of this model to other more contemporary models which incorporate titin but are also simple enough to implement in whole-body simulation (such as the winding filament model), this would seem to greatly strengthen the paper.

That’s an excellent idea, though beyond the scope of this already lengthy paper. Even though the Hill model we evaluated is a bit old it is widely used, and so, many readers will be interested in seeing the benchmark results. As benchmarking work is both difficult to fund and undertake, we do hope that others will evaluate their own models using the code and data we have provided.

(2) I’m a little unclear on the basis for the transition between short- and midrange length changes, both in reality and in the model. And also about the range of strains that qualify as ”short”. It seems like there is potential for short range stiffness, although I would have thought more in the range of 1-2% strains than >3%, to be due to currently attached crossbridges. There is clear evidence that active titin is responsible for the low stiffness at very large strains that exceed actin-myosin overlap. But I am not clear on how a transitional stiffness on the descending limb of the force-length relationship is implemented in the model, and what aspect of physiology this is replicating. It may be helpful to clarify this further and indicate where in the model this stiffness arises.

This question has several parts to it which I will paraphrase here:

A Short-range stiffness acts over smaller strains than 3.8%. How is shortrange defined?

B Where is the transition made between short-range and mid-range force response, both in reality and in the model. Also how does this change on the descending limb?

C What components in the model contribute to the stiffness of the CE?

A. Short-range stiffness acts over smaller strains than 3.8%. How is shortrange defined?

The response to Reviewer 1’s comment # 5 directly addresses this question.

B. Where is the transition made between short-range and mid-range forceresponse, both in reality and in the model. Also how does this change on the descending limb? We are going to rephrase the question because of changes in terminology that we have made in response to Reviewer 1’s comment #5.

(i) What is the basis for the transition between the muscle behaving like an LTI system? Both in reality, and in the model. (ii) What happens outside the LTI range? (iii) Also how does this change on the descending limb?

We will address this question one part at a time:

(i) What is the basis for the transition between the muscle behaving like an LTI system? Both in reality, and in the model.

A system’s response can be approximated as a linear-time-invariant (LTI) system as long as it is time-invariant, and its output can be expressed as a linear function of its input. In the context of Kirsch et al.’s experiment, the ‘system’ is the muscle, the ‘input’ is the time series of length data, and the ‘output’ is the time series of force data. Due to the requirement for timeinvariance, two experimental conditions must be met to approximate muscle as an LTI system:

• the nominal length of the muscle stays constant over long periods of time,

• and the nominal activation of the muscle stays constant.

These conditions were met by default in Kirch et al.’s experiment, and also in our simulations of this experiment. The one remaining condition to assess is whether or not the muscle’s response is linear.

To evaluate whether the muscle’s force is a linear function of the length change, Kirch et al. evaluated (Cxy)2 the coherence squared between the length and force time-series data. Even though the mathematical underpinnings of (Cxy)2 are complicated, the interpretation of (Cxy)2 is simple: muscle can be accurately approximated as a linear system if (Cxy)2 is close to 1, but the accuracy of this approximation becomes poor as (Cxy)2 approaches 0. Kirsch et al. used (Cxy)2 to identify a bandwidth in which the response of the muscle to the 1−3.8%ℓoM length changes was sufficiently linear for analysis: a lower bound of 4 Hz was identified using (Cxy)2 and the bandwidth of the input signal (15 Hz, 35 Hz, or 90 Hz) set the upper bound. In Fig. 3 of Kirsch et al. the (Cxy)2 at 4 Hz has a value of at least 0.67 for the 15 Hz and 90 Hz signals. To minimize error in our analysis and yet be consistent with Kirsch et al., we analyze the bandwidth common to both (Cxy)2 ≥ 0.67 and Kirsch et al.’s defined range. Though the bandwidth defined by the criteria (Cxy)2 ≥ 0.67 is usually larger than the one defined by Kirsch et al., there are some exceptions where the lower frequency bound of the models is higher than 4 Hz (now reported in Tables 4D and 5D).

(ii) What happens outside the LTI range?

When a muscle’s output cannot be considered a LTI it means that either that its length or activation is time-varying, or the relationship between length and force is no longer linear. In short, that the muscle is behaving as one would normally expect: time-varying and non-linearly. The wonderful part of Kirsch et al.’s work is that they found a surprisingly large region in the frequency domain where muscle behaves linearly and can be analyzed using the powerful tools of linear systems and signals.

(iii) Also how does this change on the descending limb?

Since nominal length of Kirsch et al.’s experiments is ℓoM it is not clear how the results of the perturbation experiments will change if the nominal length is moved firmly to the descending limb. However, we can see how the stiffness and damping values will change by examining Figure 9C and 9D which shows the calculated stiffness and damping of the VEXAT and Hill models as ℓM is lengthened from ℓoM down the descending limb: the stiffness and damping of the VEXAT model does not change much, while the Hill model’s stiffness changes sign and the damping coefficient changes a lot. What cannot be seen from Figure 9C and 9D is how the bandwidth over which the models are considered linear changes.

We have made a number of updates to the text to more clearly communicate these details of our response to part (i):

• Text has been edited so that it is clear that the terms ’short-range stiffness’ and ’small’ from Rack and Westbury’s work is not confused with ’stiffness’ and ’small’ from the LTI system’s analysis. Please see our response to comment # 5 for details.

• We have added text to the main body of the paper to explain how the coherence squared metric was used to select a bandwidth in which the response of the system is approximately linear:

• Revision: the paragraph that starts on page 11, column 1, line 3 ”Kirsch et al. used system identification ...”

– Difference: page 13, column 2, line 1

– Coherence is defined in Appendix D

– Coherence is now also included in the example script ‘main SystemIdentificationExample.m’

• The bandwidth over which model output can be considered linear (coherence squared > 0.67) has been added to Tables 4 and 5

– Revision: see Table 4D, and Table 5D in Appendix E

– Difference: see Table 4D, and Table 5D in Appendix E

• Figures 6 and Figures 16 are annotated now if the plotted signal does not meet the linearity requirement of Cxy > 0.67.

C. What components in the model contribute to the stiffness of the CE?

There are three components that contribute to the stiffness of the CE which are pictured in Figure 1, appear in Eqn. 15, and are listed explicitly in Eqn. 76:

(a) The XE, as represented by the afL(ℓ˜S+L˜M)k˜oX term in Eqn. 15.

(b) The elasticity of the distal segment of titin, f2(ℓ˜2). Only f2(ℓ˜2) appears in Eqn. 15 because ℓ˜1 is a model state.

(c) The extracellular matrix, as represented by the fECM(ℓ˜ECM)

There is also a compressive element fKE, but it plays no role in the simulations presented in this work because it only begins to produce force at extremely short CE lengths (ℓ˜M < 0.1ℓoM).

We have made the following changes to make these components clearer

Figure 1A has been updated:

– The symbols for a spring and a damper are now defined in Figure 1A

– The ECM now has a spring symbol. Now all springs and dampers have the correct symbol in Figure 1A.

– The caption now explicitly lists the rigid, viscoelastic, and elastic elements in the model

The equations for the VEXAT’s CE stiffness and damping are now compared and contrasted to the the Hill model’s stiffness and damping in Sec. 3.1.

– Revision: starting at page 14, column 2, line 1: Eqn. 28 and Eqn. 29 and surrounding text

– Difference: page 17, column 1, line 22

(3) This model appears to be an amalgamation of a phenomenological (forcelength and force-velocity relationships) and a mechanistic (crossbridge and titin stiffness and damping) model. While this may improve predictions, and so potentially be useful, it also seems like it limits the interpretation of physiological underpinnings of any findings. It may be helpful to explore in greater detail the implications of this approach.

We have added a limitations paragraph to the discussion which addresses this comment and can be found in:

• Revision: the paragraph beginning on page 22, column 1, line 11 ”Both the viscoelastic ...”

• Difference: the paragraph beginning on page 24, column 1, line 27

(4)As a biologist, I found the interpretation of phase and gain a little difficult and it may help the reader to show in greater detail the time series data and model predictions to highlight conditions under which the models do not accurately capture the magnitude and timing of force production.

It is important that the ideas of phase and gain are understood, especially because little information can be gleaned from the time series data directly. There is some time series data in the paper already that compares each model’s response to its spring-damper of best fit: plots of the force response of each model and its spring damper of best fit can be found in Figures 6A, 6D, 6G, 6J, 16A, 16D, 16G, and 16J in the revised manuscript. While it is clear that models with a higher VAF more closely match the spring-damper of best fit, there is not much more that can be taken from time series data: the systematic differences, particularly in phase, are just not visually apparent in the time-domain but are clear in gain and phase plots in the frequency-domain.

To make the meaning of phase and gain plots clearer, Figure 4 (Figure 5 in the first submission) has been completely re-made and includes plots that illustrate the entire process of going from two length and force timedomain signals to gain and phase plots in the frequency-domain. Included in this figure is a visual representation of transforming a signal from the time to the frequency domain (Fig. 4B and 4C), and also an illustration of the terms gain and phase (Fig. 4D). In addition, a small example file ’main SystemIdentificationExample.m’ has been added to the matlab code repository in the elife2023 branch to accompany Appendix D, which goes through the mathematics used to transform input and output time domain signals into gain and phase plots of the input-output relation. Small updates have been made to Figure 6 and 16 in the revised paper (Figures 7 and 18 in the first submission) to make the time domain signals from the spring-damper of best fit and the model output clearer. Finally, I have re-calculated the gain and phase profiles using a more advanced numerical method that trades off some resolution in frequency for more accuracy in the magnitude. This has allowed me to make Figures 6 and 16 easier to follow because the gain and phase responses are now lines rather than a scattering of points. We hope that these additions make the interpretation of gain and phase clearer.

Please see

Revision:

– Figure 4 and caption on page 12

– The opening 2 paragraphs of Sec 3.1 starting on page 10, column 2, line 4 ”In Kirsch et al.’s ...”

– Figure 6 & 16: spring damper and model annotation added, plotted the gain and phase as lines

– Appendix D: Updated to include coherence and the more advanced method used to evaluate the system transfer function, gain, and phase.

Difference:

– Figure 4 and caption on page 12

– The opening 2 paragraphs of Sec 3.1 starting on page 12, column 1, line 34 and ending on page 13, column 2, line 29

– Figure 6 & 16: spring damper and model annotation added

– Appendix D

(5) The actin-myosin and actin-titin load pathways are depicted as distinct in the model. However, given titin’s position in the center of myosin and the crossbridge connections between actin and myosin, this would seem to be an oversimplification. It seems worth considering whether the separation of these pathways is justified if it has any effect on the conclusions or interpretation.

We have reworked one of the discussion paragraphs to focus on how our simulations would be affected by two mechanisms (Nishikawa et al.’s winding filament theory and DuVall et al.’s titin entanglement hypothesis) that make it possible for crossbridges to do mechanical work on titin.

• Revision: the paragraph beginning on page 21, column 2, line 42 “The active titin model ...”

• Difference: the paragraph beginning on page 23, column 2, line 48

References

Nishikawa KC, Monroy JA, Uyeno TE, Yeo SH, Pai DK, Lindstedt SL. Is titin a ‘winding filament’? A new twist on muscle contraction. Proceedings of the royal society B: Biological sciences. 2012 Mar 7;279(1730):981-90.

DuVall M, Jinha A, Schappacher-Tilp G, Leonard T, Herzog W. I-Band Titin Interaction with Myosin in the Muscle Sarcomere during Eccentric Contraction: The Titin Entanglement Hypothesis. Biophysical Journal. 2016 Feb 16;110(3):302a.

Author response:

Reviewer #1 (Public Review):

In this manuscript, Naseri et al. present a new strategy for identifying human genetic variants with recessive effects on disease risk by the genome-wide association of phenotype with long runs-of-homozygosity (ROH). The key step of this approach is the identification of long ROH segments shared by many individuals (termed "shared ROH diplotype clusters" by the authors), which is computationally intensive for large-scale genomic data. The authors circumvented this challenge by converting the original diploid genotype data to (pseudo-)haplotype data and modifying the existing positional Burrow-Wheeler transformation (PBWT) algorithms to enable an efficient search for haplotype blocks shared by many individuals. With this method, the authors identified over 1.8 million ROH diplotype clusters (each shared by at least 100 individuals) and 61 significant associations with various non-cancer diseases in the UK Biobank dataset.

Overall, the study is well-motivated, highly innovative, and potentially impactful. Previous biobank-based studies of recessive genetic effects primarily focused on genome-wide aggregated

ROH content, but this metric is a poor proxy for homozygosity of the recessive alleles at causal loci. Therefore, searching for the association between phenotype and specific variants in the homozygous state is a key next step towards discovering and understanding disease genes/alleles with recessive effects. That said, I have some concerns regarding the power and error rate of the methods, for both identification of ROH diplotype clusters and subsequent association mapping. In addition, some of the newly identified associations need further validation and careful consideration of potential artifacts (such as cryptic relatedness and environment sharing).

1) Identification of ROH diplotype clusters.

The practice of randomly assigning heterozygous sites to a homozygous state is expected to introduce errors, leading to both false positives and false negatives. An advantage that the authors claim for this practice is to reduce false negatives due to occasional mismatch (possibly due to genotyping error, or mutation), but it's unclear how much the false positive rate is reduced compared to traditional ROH detection algorithm. The authors also justified the "random allele drawing" practice by arguing that "the rate of false positives should be low" for long ROH segments, which is likely true but is not backed up with quantitative analysis. As a result, it is unclear whether the trade-off between reducing FNs and introducing FPs makes the practice worthwhile (compared to calling ROHs in each individual with a standard approach first followed by scanning for shared diplotypes across individuals using BWT). I would like to see a combination of back-of-envelope calculation, simulation (with genotyping errors), and analysis of empirical data that characterize the performance of the proposed method.

In particular, I find the high number of ROH clusters in MHC alarming, and I am not convinced that this can be fully explained by a high density of SNPs and low recombination rate in this region. The authors may provide further support for their hypothesis by examining the genome-wide relationship between ROH cluster abundance and local recombination rate (or mutation rate).

Thanks for this insightful comment. Through additional experiments, we confirmed that the excessive number of ROH clusters in the MHC region is due to the higher density of markers per centimorgan. As discussed above at Essential Revision 2, we took this opportunity to modify our code to search for clusters with the minimum length in terms of cM instead of sites. We have also provided the genetic distance for reported clusters in the MHC region with significant association (genetic length (cM) column in Tables 1 and 2). We include the following in the main text:

“We searched for ROH clusters using a minimum target length of 0.1 cM (Figure 3–figure supplement 1). As shown in the figure, there is no excessive number of ROH clusters in chromosome 6 as was spotted using a minimum number of variant sites.”

Methods section, ROH algorithm subsection:

“We implemented ROH-DICE to allow direct use of genetic distances in addition to variant sites for L. The program can take minimum target length L directly in cM and detect all ROH clusters greater than or equal to the target length in cM. The program holds a genetic mapping table for all the available sites, and cPBWT was modified to work directly with the genetic length instead of the number of sites.”

2) Power of ROH association. Given that the authors focused on long segments only (which is a limitation of the current method), I am concerned about the power of the association mapping strategy, because only a small fraction of causal alleles are expected to be present in long, homozygous haplotypes shared by many individuals. It would be useful to perform a power analysis to estimate what fraction of true causal variants with a given effect size can be detected with the current method. To demonstrate the general utility of this method, the authors also need to characterize the condition(s) under which this method could pick up association signals missed by standard GWAS with recessive effects considered. I suspect some variants with truly additive effects can also be picked up by the ROH association, which should be discussed in the manuscript to guide the interpretation of results.

We added a new experiment in the Results section “Evaluation of ROH clusters in simulated data” under Power of ROH-DICE in association studies. We compared the power of the ROH cluster with additive, recessive, and dominant models. Our simulation shows that using ROH clusters outperforms standard GWAS when a phenotype is associated with a set of consecutive homozygous sites. We added the following text:

“...We calculated the p-values for both ROH clusters and all variant sites. We used a p-value cut-off of 0.05 divided by the number of tests for each phenotype to determine whether the calculated p-value was smaller than the threshold, indicating an association. For GWAS, only one variant site within the ROH cluster, contributing to the phenotype, was required. We tested for all additive, dominant, and recessive effects (Figure 1–figure supplement 3). The figure demonstrates that ROH-DICE outperforms GWAS when a phenotype is associated with a set of consecutive homozygous sites. The maximum effect size of 0.3 resulted in ROH clusters achieving a power of 100%, whereas the additive model only achieved 11%, and the dominant and recessive models achieved 52% and 70%, respectively. The GWAS with recessive effect yields the best results among other GWAS tests, however, its power is still lower than using ROH clusters.”

3) False positives of ROH association. GWAS is notoriously prone to confounding by population and environmental stratification. Including leading principal components in association testing alleviates this issue but is not sufficient to remove the effects of recent demographic structure and local environment (Zaidi and Mathieson 2020 eLife). Similar confounding likely applies to homozygosity mapping and should be carefully considered. For example, it is possible that individuals who share a lot of ROH diplotypes tend to be remotely related and live near each other, thus sharing similar environments. Such scenarios need to be excluded to further support the association signals.

We acknowledge that there could be confounding factors that may affect the association's results. To address this, we utilized principal component (PC) values and additional covariates while using PHESANT after our initial Chi-square tests. We also included your comments in our Discussion section:

"We used age, gender, and genetic principal components as confounding variables in the association analysis. Genetic principal components can reduce the confounding effect brought on by population structure but it may be insufficient to completely eliminate the effects of recent demographic structure and the local environment45. For example, individuals sharing excessive ROH diplotypes may share similar environments since they are closely related and reside close to one another. Since we did not rule out related individuals, some of the reported GWAS signals may not be attributable to ROH.”

4) Validation of significant associations. It is reassuring that some of the top associations are indirectly corroborated by significant GWAS associations between the same disease and individual SNPs present in the ROH region (Tables 1 and 2). However, more sanity checks should be done to confirm consistency in direction of effect size (e.g., risk alleles at individual SNPs should be commonly present in risk-increasing ROH segment, and vice versa) and the presence of dominance effect.

The beta values for effect size are now included in all reported tables. All beta values for ROH-DICE are positive indicating carriers of these ROH diplotypes may increase the risk of certain non-cancerous diseases. Moreover, we conducted the suggested sanity check to confirm the consistency of the direction of risk-inducing ROH diplotypes and risk alleles.

We also computed D’ as a measure of linkage between the reported GWAS results and ROH clusters. We found that most of the GWAS results and ROH clusters are strongly correlated. However, in a few cases, D' is small or close to zero. In such cases, the reported p-value from GWAS was also insignificant, while the ROH cluster indicated a significant association. We included these points in the Results section.

Reviewer #3 (Public Review):

A classic method to detect recessive disease variants is homozygosity mapping, where affected individuals in a pedigree are scanned for the presence of runs of homozygosity (ROH) intersecting in a given region. The method could in theory be extended to biobanks with large samples of unrelated individuals; however, no efficient method was available (to the best of my knowledge) for detecting overlapping clusters of ROH in such large samples. In this paper, the authors developed such a method based on the PBWT data structure. They applied the method to the UK biobank, finding a number of associations, some of them not discovered in single SNP associations.

Major strengths:

•           The method is innovative and algorithmically elegant and interesting. It achieves its purpose of efficiently and accurately detecting ROH clusters overlapping in a given region. It is therefore a major methodological advance.

•           The method could be very useful for many other researchers interested in detecting recessive variants associated with any phenotype.

•           The statistical analysis of the UK biobank data is solid and the results that were highlighted are interesting and supported by the data.

Major weaknesses:

•           The positions and IDs of the ROH clusters in the UK biobank are not available for other researchers. This means that other researchers will not be able to follow up on the results of the present paper.

We included the SNP IDs, positions, and consensus alleles for all reported loci in the main tables. Moreover, additional information including beta and D’ values were added. The current information should allow researchers to follow up on the results. Supplementary File 2 contains beta, D’ values for all reported clusters.

Supplementary File 3 contains the SNP IDs and consensus alleles for all reported clusters in Tables 1 and 2. The consensus allele denotes the allele with the highest occurrence in the reported clusters.

•           The vast majority of the discoveries were in regions already known to be associated with their respective phenotypes based on standard GWAS.

We agree that a majority of the ROH regions are indeed consistent with GWAS. However, some regions were missed by standard GWAS (e.g. chr6:25969631-26108168, hemochromatosis). Our message is that our method is a complementary approach to standard GWAS and will not replace standard GWAS analysis. See our response to Reviewer #2 Point Six.

•           The running time seems rather long (at least for the UK biobank), and therefore it will be difficult for other researchers to extensively experiment with the method in very large datasets. That being said, the method has a linear running time, so it is already faster than a naïve algorithm.

Thank you for your input. The algorithm used to locate matching blocks is efficient and the total CPU hours it consumed was the reported run time. Since it consumes very little memory and resources, it can be executed simultaneously for all chromosomes. We also noticed that a significant time was being spent parsing the input file and slightly modified our script to improve the parsing. We also re-ran it for all chromosomes in parallel and reported the elapsed time which was only 18 hours and 54 minutes.

“This was achieved by running the ROH-DICE program, with a wall clock time of 18 hours and 54 minutes where the program was executed for all chromosomes in parallel (total CPU hours of ~ 242.5 hours). The maximum residence size for each chromosome was approximately 180 MB.”

The reproduction of the original source giving as a result the same source code, input data, software and configurations in the system might be considered as Computational reproducibility

According to Patil, P et al. (2016) state that "everyone agrees that scientific studies should be reproducible and replicable. The problem is almost no one agrees upon what those terms mean. A major initiative in psychology used the term reproducibility'' to refer to completely re­doing experiments including data collection (1). In cancer biologyreproducibility'' has been used to refer to the re calculation of results using a fixed set of data and code." (pag. 1)

A possible approach to statistical reproducibility is to re-do experiments over and over again gathering all the data available to get the findings, but emphasising in the analysis with transparency in the information to have accurate conclusions.

References

Prasad Patil, Roger D. Peng, Jeffrey T. Leek. (2016). A statistical definition for reproducibility and replicability. see on https://www.biorxiv.org/content/10.1101/066803v1.full.pdf

Please add the legend of color code

TestRPC Related Pages

• Ganache CLI Install - <q>testRPC/ganache-cli testRPC/ganache-cli is a node.js Ethereum client for testing and development of the Smart… by technological.</q>
• Ethereum TestRPC and Tester Client - <q>This provider automatically spins up the eth-testrpc server for you so that you can test your web3.py code against an in memory EVM. This provides lightning …</q>
• Ethereum TestRPC vs. Geth - <q>Ethereum TestRPC vs. Geth explains that TestRPC is a client for testing and developing. While Geth is a full GO language client for connecting to the blockchain.</q>
• TestRPC & Ethereum Smart Contracts - <q>… testrpc. For truffle test, testrpc or geth is required to be run explicitly. Else, you get the following error: $ truffle test Could not connect to your …</q>

Comment configurer cela pour un site utilisé a l'international ? Par exemple sur Youtube, faut-il créer un code différent pour chaque langues utilisées ?

Moderate people, not code by [[Ryan Barrett]]

I need to read this book. And re-review this video with a notecard handy. (I wonder if there's a way to use hypothes.is for notes on video/audio?)

Will the code to run these analyses be available on Github? It would be helpful to follow along with the many steps used across these analyses.

Did you do anything to account for these things in your analysis?

I get a "Page not found" error when i navigate to this URL. Is the repo still private or is there a typo in the URL? I would love to look at/give feedback on the code as well!

An old code name for this product maybe?

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this manuscript, Ngo et al. report a peculiar effect where a single base mismatch (CC) can enhance the mechanical stability of a nucleosome. In previous studies, the same group used a similar state-of-the-art fluorescence-force assay to study the unwrapping dynamics of 601-DNA from the nucleosome and observed that force-induced unwrapping happens more slowly for DNA that is more bendable because of changes in sequence or chemical modification. This manuscript appears to be a sequel to this line of projects, where the effect of CC is tested. The authors confirmed that CC is the most flexible mismatch using the FRET-based cyclization assay and found that unwrapping becomes slower when CC is introduced at three different positions in the 601 sequence. The CC mismatch only affects the local unwrapping dynamics of the outer turn of nucleosomal DNA.

Strengths:

These results are in good agreement with the previously established correlation between DNA bendability and nucleosome mechanical stability by the same group. This well-executed, technically sound, and well-written experimental study contains novel nucleosome unwrapping data specific to the CC mismatch and 601 sequence, the cyclizability of DNA containing all base pair mismatches, and the unwrapping of 601-DNA from xenophus and yeast histones. Overall, this work will be received with great interest by the biophysics community and is definitely worth attention.

Weaknesses:

The scope and impact of this study are somewhat limited due to the lack of sequence variation. Whether the conclusion from this study can be generalized to other sequences and other bendability-enhancing mismatches needs further investigation.

Major questions:

(1) As pointed out by the authors, the FRET signal is not sensitive to nucleosome position; therefore, the increasing unwrapping force in the presence of CC can be interpreted as the repositioning of the nucleosome upon perturbation. It is then also possible that CC-containing DNA is not positioned exactly the same as normal DNA from the start upon nucleosome assembly, leading to different unwrapping trajectories. What is the experimental evidence that supports identical positioning of the nucleosomes before the first stretch?

We added the following and refer to our recent publication1 to address this question.

“This is consistent with a previous single nucleotide resolution mapping of dyad position from of a library of mismatches in all possible positions along the 601 sequence or a budding yeast native sequence which showed that a single mismatch (A-A or T-T) does not affect the nucleosome position27.”

(2) The authors chose a constant stretching rate in this study. Can the authors provide a more detailed explanation or rationale for why this rate was chosen? At this rate, the authors found hysteresis, which indicates that stretching is faster than quasi-static. But it must have been slow and weak enough to allow for reversible unwrapping and wrapping of a CC-containing DNA stretch longer than one helical turn. Otherwise, such a strong effect of CC at a single location would not be seen. I am also curious about the biological relevance of the magnitude of the force. Can such force arise during nucleosome assembly in vivo?

To address the comment about the magnitude of force, we added the following paragraph to Introduction. “RNA polymerase II can initiate transcription at 4 pN of hindering force2 and its elongation activity continues until it stalls at ~ 10 pN of hindering force3,4. Therefore, the transcription machinery can generate picoNewtons of force on chromatin as long as both the machinery and the chromatin segment in contact are tethered to stationary objects in the nucleus. Another class of motor protein, chromatin remodeling enzymes, was also shown to induce processive and directional sliding of single nucleosomes when the DNA is under similar amount of tension (~ 5 pN)5. Therefore, measurements of nucleosomes at a few pN of force will expand our knowledge of the physiology roles of nucleosome structure and dynamics.”

To address the comment about the stretching rate, we added the following to Results. We note that the physiological loading rate has been challenging to determine for any biomolecular interactions, and the only quantitative measurement we are aware of is that of an integrin that we are citing.

“The force increases nonlinearly and the loading rate, i.e. the rate at which the force increases, was approximately in the range of 0.2 pN/s to 6 pN/s, similar to the cellular loading rates for a mechanosensitive membrane receptor6.”

(3) In this study, the CC mismatch is the only change made to the 601 sequence. For readers to truly appreciate its unique effect on unwrapping dynamics as a base pair defect, it would be nice to include the baseline effects of other minor changes to the sequence. For example, how robust is the unwrapping force or dynamics against a single-bp change (e.g., AT to GC) at the three chosen positions?

Unfortunately, we are unable to perform the suggested unwrapping experiment in a timely manner because the instrument has been disassembled during our recent move. However, we previously performed unwrapping experiments not only as a function of sequence but also as a function of cytosine modification and showed that we can detect even more subtle effects7,8. In addition, please note that we are not claiming that simply changing basepair at the chosen sites changes the mechanical stability of a nucleosome so we do not believe the requested experiment is necessary.

(4) The last section introduces yeast histones. Based on the theme of the paper, I was expecting to see how the effect of CC is or is not preserved with a different histone source. Instead, the experiment only focuses on differences in the unwrapping dynamics. Although the data presented are important, it is not clear how they fit or support the narrative of the paper without the effect of CC.

We apologize for giving the reviewer a wrong impression. We included the data because we believe that information on how the histone core can determine the translation of DNA mechanics into nucleosome mechanical stability will be of interest to the readers of this manuscript. We now mention explicitly that the observation was made using intact DNA, i.e. no mismatch, in the abstract and elsewhere.

(5) It is stated that tRNA was excluded in experiments with yeast-expressed nucleosomes. What is the reason for excluding it for yeast nucleosomes? Did the authors rule out the possibility that tRNA causes the measured difference between the two nucleosome types?

We normally include tRNA because we found that it reduces sticking of beads to the surface over several hours of experiments. In yeast nucleosomes, we found that tRNA causes the nucleosome to disassemble. Therefore, we did not include tRNA in yeast nucleosome experiments. We now mention this in Methods as reproduced below.

“tRNA, which we normally include to reduce sticking of beads to the surface over the hours of single molecule experiments in a sealed chamber, was excluded in experiments with yeastexpressed nucleosomes because tRNA induced disassembly of nucleosomes assembled using yeast histones.”

We cannot not formally rule out the possibility that tRNA causes the measured difference between Xenopus - vs Yeast- nucleosomes. However, we have shown in our previous publication7 that the asymmetric unwrapping in Xenopus nucleosomes was modulated by the DNA sequence. When we swapped the sequence of the inner turn between the two sides, while tRNA was included in all experiments, we observed stochastic unwrapping instead. As part of our response to another reviewer’s comments, we also added the following on the relevant differences between the species in Discussion.

“The crystal structure of the yeast nucleosome suggests that yeast nucleosome architecture is subtly destabilized in comparison with nucleosomes from higher eukaryotes9. Yeast histone protein sequences are not well conserved relative to vertebrate histones (H2A, 77%; H2B, 73%; H3, 90%; H4, 92% identities), and this divergence likely contributes to differences in nucleosome stability. Substitution of three residues in yeast H3 a3-helix (Q120, K121, K125) very near the nucleosome dyad with corresponding human H3.1/H3.3 residues (QK…K replaced with MP…Q) caused severe growth defects, elevated nuclease sensitivity, reduced nucleosome positioning and nucleosome relocation to preferred locations predicted by DNA sequence alone 10. The yeast histone octamer harboring wild type H3 may be less capable of wrapping DNA over the histone core, leading to reduced resistance to the unwrapping force for the more flexible half of the 601positioning sequence.”

Reviewer #2 (Public Review):

Summary:

Mismatches occur as a result of DNA polymerase errors, chemical modification of nucleotides, during homologous recombination between near-identical partners, as well as during gene editing on chromosomal DNA. Under some circumstances, such mismatches may be incorporated into nucleosomes but their impact on nucleosome structure and stability is not known. The authors use the well-defined 601 nucleosome positioning sequence to assemble nucleosomes with histones on perfectly matched dsDNA as well as on ds DNA with defined mismatches at three nucleosomal positions. They use the R18, R39, and R56 positions situated in the middle of the outer turn, at the junction between the outer turn and inner turn, and in the middle of the inner turn, respectively. Most experiments are carried out with CC mismatches and Xenopus histones. Unwrapping of the outer DNA turn is monitored by singlemolecule FRET in which the Cy3 donor is incorporated on the 68th nucleotide from the 5'-end of the top strand and the Cy5 acceptor is attached to the 7th nucleotide from the 5' end of the bottom strand. Force is applied to the nucleosomal DNA as FRET is monitored to assess nucleosome unwrapping. The results show that a CC mismatch enhances nucleosome mechanical stability. Interestingly, yeast and Xenopus histones show different behaviors in this assay. The authors use FRET to measure the cyclization of the dsDNA substrates to test the hypothesis that mismatches enhance the flexibility of the 601 dsDNA fragment and find that CC, CA, CT, TT, and AA mismatches decrease looping time, whereas GA, GG, and GT mismatches had little to no effect. These effects correlate with the results from DNA buckling assays reported by Euler's group (NAR 41, 2013) using the same mismatches as an orthogonal way to measure DNA kinking. The authors discuss that substitution rates are higher towards the middle of the nucleosome, suggesting that mismatches/DNA damage at this position are less accessible for repair, consistent with the nucleosome stability results.

Strengths:

The single-molecule data show clear and consistent effects of mismatches on nucleosome stability and DNA persistence length.

Weaknesses:

It is unclear in the looping assay how the cyclization rate relates to the reporting looping time. The biological significance and implications such as the effect on mismatch repair or nucleosome remodelers remain untested. It is unclear whether the mutational pattern reflects the behavior of the different mismatches. Such a correlation could strengthen the argument that the observed effects are relevant for mutagenesis.

Reviewer #3 (Public Review):

Summary:

The mechanical properties of DNA wrapped in nucleosomes affect the stability of nucleosomes and may play a role in the regulation of DNA accessibility in eukaryotes. In this manuscript, Ngo and coworkers study how the stability of a nucleosome is affected by the introduction of a CC mismatched base pair, which has been reported to increase the flexibility of DNA. Previously, the group has used a sophisticated combination of single-molecule FRET and force spectroscopy with an optical trap to show that the more flexible half of a 601 DNA segment provides for more stable wrapping as compared to the other half. Here, it is confirmed with a single-molecule cyclization essay that the introduction of a CC mismatch increases the flexibility of a DNA fragment. Consistent with the previous interpretation, it also increased the unwrapping force for the half of the 601 segment in which the CC mismatch was introduced, as measured with single-molecule FRET and force spectroscopy. Enhanced stability was found up to 56 bp into the nucleosome. The intricate role of mechanical stability of nucleosomes was further investigated by comparing force-induced unwrapping profiles of yeast and Xenopus histones. Intriguingly, asymmetric unwrapping was more pronounced for yeast histones.

Strengths:

(1) High-quality single-molecule data.

(2) Novel mechanism, potentially explaining the increased prominence of mutations near the dyads of nucleosomes.

(3) A clear mechanistic explanation of how mismatches affect nucleosome stability.

Weaknesses:

(1) Disconnect between mismatches in nucleosomes and measurements comparing Xenopus and yeast nucleosome stability.

(2) Convoluted data in cyclization experiments concerning the phasing of mismatches and biotin site. ---

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Specific comments:

In Figure 1 legend, "the black diamonds on the DNA bends represent the mismatch position with R18 and R39 on minor grooves and R56 on a major groove." Minor and major grooves should be phrased as histone-facing minor and major grooves.

We fixed the problem.

In Materials and Methods, the sentence that describes the stretching rate cites reference 1, which does not seem to be relevant.

We fixed the problem.

Reviewer #2 (Recommendations For The Authors):

(1) In the introduction, the authors should also discuss the context of mismatches occurring during homologous recombination in meiosis or somatic cells in non-allelic recombination between near identical repeats.

Introduction now has the following.

“DNA base-base mismatches are generated by nucleotide misincorporation during DNA synthesis, meiotic recombination, somatic recombination between nearly identical repeats, or chemical modification such as hydrolytic deamination of cytosine.”

(2) Generally, it seems counter-intuitive in terms of biology that mismatches containing nucleosomes are more stable, as mismatches require repair and/or detection for heteroduplex rejection during recombination. Some discussion of this apparent paradox should be added.

To address this comment, we added the following to Discussion.

“The higher frequency of substitutions in the nucleosomal DNA may be attributed to the difficulty of accessing the extra-stable nucleosomes. We also note that even without an enhanced stability, a mismatch within a nucleosome would be more difficult to detect for mismatch repair machineries compared to a mismatch in a non-nucleosomal DNA. Because mismatch repair machineries accompany the replisome, most of nascent mismatches may be detected for repair before nucleosome deposition. Therefore, the decrease in accessibility predicted based on our data here may be important only in rare cases a mismatch is not detected prior to the deposition of a nucleosome on the nascent DNA or in cases where a mismatch is generated via a non-replicative mechanism.”

(3) The authors discuss that the substitution rate is higher while the indel (insertion and deletion) rate is lower nearer the center of a positioned nucleosome. Are the differences between individual mismatches reported in Figure 6 reflected in the mutagenic profile?

We cannot currently compare them because the mutagenic profile even when it is available is a complex convolution of mismatch generation, mismatch repair and selection. Mismatch generation occurs through several different processes and how they are affected by nucleosomes and their mismatch type and sequence context is unknown. Mismatch repair process itself depends on mismatch type and sequence context as recently shown by a high throughput in vivo study11. And because the population genetics does not simply reflect de novo mutation profiles due to selection, comparison between mismatch-induced DNA mechanical changes and mutagenic profiles is further complicated. We added the following to the revision.

“If and how the mismatch type-dependent DNA mechanics affects the sequence-dependent mismatch repair efficiency in vivo, as recently determined in a high through study in E. coli11, remains to be investigated. Comparison of mismatch-type dependent DNA mechanics to population genetics data is challenging because mutation profiles reflect a combined outcome of mismatch-generation, mismatch repair and selection in addition to other mutational processes.”

(4) The looping assay should be explained better, especially how the cyclization rate is related to the reported looping time.

We modified Figure 5 to include examples of looping time determination through fitting of the looped fraction vs time, and added the following to the figure caption.

“To calculate the looping time, the fraction of looped molecules (high FRET) as a function of time is fitted to an exponential function, 𝑒−𝑡⁄(𝑙𝑜𝑜𝑝𝑖𝑛𝑔 𝑡𝑖𝑚𝑒) (right panel for one run of experiments).

Furthermore, we added the following sentence to Results.

“The rate of loop formation, which is the inverse of looping time determined from an exponential fitting of loop fraction vs time, was used as a measure of apparent DNA flexibility influenced by a mismatch 12,13.”

*Reviewer #3 (Recommendations For The Authors):

I have some concerns that, when addressed upon revision, would improve the manuscript:

(1) Page 6 and Supplementary Figure S1C: Though the FRET levels are the same for all nucleosomes, the distribution between the two levels is not. The nucleosomes with CC mismatches appear to have a larger fraction in the low-FRET population. This seems to contradict the higher mechanical stability. A comment on this should clarify it, or make this conundrum explicit.

Thank you for the comment. The low FRET population also includes the nucleosomes that do not have an active acceptor the fraction of which varies between preparations. We now note this in the supplementary figure caption.

(2) It is intriguing that a more stable nucleosome forms after several pulling cycles and it is argued that this might be due to shifting of the nucleosome. This seems reasonable and has important consequences both for the interpretation of the current experimental data and for the general mechanisms involved in nucleosome maintenance and remodeling. It is puzzling though how this would work mechanistically since it only seems to happen when nucleosomes are half-wrapped and when the unwrapped half contains the mismatch. From the previous work of the group and the current manuscript, it seems that shift does not occur in DNA without mismatches (Correct?). Does shifting happen for the 601-R18 and 601-R56 nucleosomes as well?

The mismatch-containing half is the half that is mechanically less stable in an intact, mismatch-free 601 nucleosome. So indeed, that is the half that is unwrapped in an intact nucleosome. But because the introduction of mismatch makes that half more mechanically stable, it can stay wrapped until higher forces, and the resulting structural distortion may cause the shift although we acknowledge that this interpretation remains speculative. Shifting occurs for all three constructs with a mismatch but not for the intact nucleosome without a mismatch.

(3) Could the shifting be related to the differences in sub-population distribution observed in Supplementary Figure S1C?

/See our response to comment (1) above.

(4) The paper would have more impact if the mechanism of possible shifting could be clarified. This can be done experimentally with a fluorescent histone, as suggested in the manuscript. But having a FRET pair on positions in the DNA that would shift to closer proximity upon shifting, either at the ED2 or at the ED1 site will also work, is in line with the current experiments and seems feasible.

We revised the text as follows in order not to exclude labeling configurations with both fluorophores on the DNA while reporting on the shift. We are also happy to add an appropriate reference if the reviewer can help us identify an existing study that measured dyad position shifts through such a labeling configuration.

“However, since the FRET values in our DNA construct are not sensitive to the nucleosome position, further experiments with fluorophores conjugated to strategic positions that allow discrimination between different dyad positions14 will be required to test this hypothesis.”

(5) Figures 5 and 6: To appreciate the quality of the data, state the number of molecules that contributed to the cyclization essay, or better, share a figure of the number of looped molecules as a function of time as supplementary data.

We added the requested figures to Figure 5 and a new supplementary Figure 2, and added the following to Methods.

“Approximately 2500 – 3500 molecules were quantified at each timestamp during the experiment, and three independent experiments were performed for each sequence (Supplemental Figure S2).”

(6) Page 8/9: A control is added to confirm that the phasing of the biotin relative to the end affects the observed cyclization rate. However, the mismatch sites were chosen such that they included 5 bp phase shifts. This convolutes the outcomes, as the direction of flexibility due to the phasing of the mismatch relative to the biotin may also influence the rate. Was this checked?

We would like to clarify that the phasing of the biotin is not so much as with respect to the end, as it is with respect to the full molecule. Static curvature and poloidal angle associated with the DNA molecule (which is something that is ultimately determined by the full chemical composition of the molecule, including its sequence and the mismatch) could make the molecule prefer a looped configuration where the biotin points towards the “inside” of the molecule. Such a configuration would be sterically unfavoured during the single molecule looping reaction where the biotin is attached to a surface via avidin. However, if the biotin is moved by half the helical repeat (or an off multiple of half the helical repeat, essentially 16 nt as done in the manuscript), it would now point to the “outside” of the molecule. Therefore, to make sure that the difference between the looping rates of any two DNA constructs (say the 601-RH and 601-R18-RH) is a better reflection of differences in dynamic flexibility, we ensure that the difference persists even when the biotin is moved by an odd multiple of half the helical repeat. We revised the section as follows.

“For example, moving the location of the biotin tether by half the helical repeat (~ 5 bp) can lead to a large change in cyclization rate15, likely due to the preferred poloidal angle of a given DNA16 that determines whether the biotin is facing towards the inside of the circularized DNA, thereby hindering cyclization due to steric hindrance caused by surface tethering.”

(7) Page 9/10: The comparison of yeast vs Xenopus is interesting, albeit a bit disconnected. Since the single-molecule statistics are relatively small, did the nucleosomes show similar bulk FRET distributions, or did they also show a shift in FRET levels?

We included the data because we believe that information on how the histone core can determine the translation of DNA mechanics into nucleosome mechanical stability will be of interest to the readers of this manuscript. The FRET values were similarly distributed.

(8) The discussion calls for a more detailed analysis of the structural differences of the histones of the two species to rationalize the observed asymmetry in flexibility dependence: why would yeast nucleosomes be less sensitive to sequence asymmetries?

We added the following to Discussion to address this comment.

“The crystal structure of the yeast nucleosome suggests that yeast nucleosome architecture is subtly destabilized in comparison with nucleosomes from higher eukaryotes9. Yeast histone protein sequences are not well conserved relative to vertebrate histones (H2A, 77%; H2B, 73%; H3, 90%; H4, 92% identities), and this divergence likely contributes to differences in nucleosome stability. Substitution of three residues in yeast H3 3-helix (Q120, K121, K125) very near the nucleosome dyad with corresponding human H3.1/H3.3 residues (QK…K replaced with MP…Q) caused severe growth defects, elevated nuclease sensitivity, reduced nucleosome positioning and nucleosome relocation to preferred locations predicted by DNA sequence alone 10. The yeast histone octamer harboring wild type H3 may be less capable of wrapping DNA over the histone core, leading to reduced resistance to the unwrapping force for the more flexible half of the 601positioning sequence.”

(9) It would also be interesting if the increased stability due to the introduction of mismatches observed on Xenopus nucleosomes holds in yeast. Or does the reduced stability remove this effect? This is relevant to substantiate the broad claims in the context of evolution and cancer that are discussed in the manuscript.

Unfortunately, we are unable to perform the suggested unwrapping experiment in a timely manner because the instrument has been disassembled during our recent move. However, in terms of cancer relevance, our mismatch dependence experiments were performed using vertebrate nucleosomes (Xenopus) so repeating this for yeast nucleosomes would not provide relevant information.

Minor comments:

(1) Supplementary Figure S1 misses the label '(C)' in its caption.

We fixed it.

(2) The supplementary data sequences for the fleezer measurements contain entrees 'R39 construct' and miss the positions of the Cy3 and Cy labels; the color code (levels of grey) is not explained.

We fixed the labeling mistake and added detailed annotations of the highlighted features.

References

(1) Park, S., Brandani, G.B., Ha, T. & Bowman, G.D. Bi-directional nucleosome sliding by the Chd1 chromatin remodeler integrates intrinsic sequence-dependent and ATP-dependent nucleosome positioning. Nucleic Acids Res 51, 10326-10343 (2023).

(2) Fazal, F.M., Meng, C.A., Murakami, K., Kornberg, R.D. & Block, S.M. Real-time observation of the initiation of RNA polymerase II transcription. Nature 525, 274-7 (2015).

(3) Galburt, E.A., Grill, S.W., Wiedmann, A., Lubkowska, L., Choy, J., Nogales, E., Kashlev, M. & Bustamante, C. Backtracking determines the force sensitivity of RNAP II in a factor-dependent manner. Nature 446, 820-3 (2007).

(4) Schweikhard, V., Meng, C., Murakami, K., Kaplan, C.D., Kornberg, R.D. & Block, S.M. Transcription factors TFIIF and TFIIS promote transcript elongation by RNA polymerase II by synergistic and independent mechanisms. Proc Natl Acad Sci U S A 111, 6642-7 (2014).

(5) Kim, J.M., Carcamo, C.C., Jazani, S., Xie, Z., Feng, X.A., Yamadi, M., Poyton, M., Holland, K.L., Grimm, J.B., Lavis, L.D., Ha, T. & Wu, C. Dynamic 1D Search and Processive Nucleosome Translocations by RSC and ISW2 Chromatin Remodelers. bioRxiv (2024). (6) Jo, M.H., Meneses, P., Yang, O., Carcamo, C.C., Pangeni, S. & Ha, T. Determination of singlemolecule loading rate during mechanotransduction in cell adhesion. Science (in press).

(7) Ngo, T.T., Zhang, Q., Zhou, R., Yodh, J.G. & Ha, T. Asymmetric unwrapping of nucleosomes under tension directed by DNA local flexibility. Cell 160, 1135-44 (2015).

(8) Ngo, T.T., Yoo, J., Dai, Q., Zhang, Q., He, C., Aksimentiev, A. & Ha, T. Effects of cytosine modifications on DNA flexibility and nucleosome mechanical stability. Nat Commun 7, 10813 (2016).

(9) White, C.L., Suto, R.K. & Luger, K. Structure of the yeast nucleosome core particle reveals fundamental changes in internucleosome interactions. EMBO J 20, 5207-18 (2001).

(10) McBurney, K.L., Leung, A., Choi, J.K., Martin, B.J., Irwin, N.A., Bartke, T., Nelson, C.J. & Howe, L.J. Divergent Residues Within Histone H3 Dictate a Unique Chromatin Structure in Saccharomyces cerevisiae. Genetics 202, 341-9 (2016).

(11) Kayikcioglu, T., Zarb, J.S., Lin, C.-T., Mohapatra, S., London, J.A., Hansen, K.D., Rishel, R. & Ha, T. Massively parallel single molecule tracking of sequence-dependent DNA mismatch repair in vivo. bioRxiv, 2023.01.08.523062 (2023).

(12) Jeong, J., Le, T.T. & Kim, H.D. Single-molecule fluorescence studies on DNA looping. Methods 105, 34-43 (2016).

(13) Jeong, J. & Kim, H.D. Base-Pair Mismatch Can Destabilize Small DNA Loops through Cooperative Kinking. Phys Rev Lett 122, 218101 (2019).

(14) Blosser, T.R., Yang, J.G., Stone, M.D., Narlikar, G.J. & Zhuang, X. Dynamics of nucleosome remodelling by individual ACF complexes. Nature 462, 1022-7 (2009).

(15) Basu, A., Bobrovnikov, D.G., Qureshi, Z., Kayikcioglu, T., Ngo, T.T.M., Ranjan, A., Eustermann, S., Cieza, B., Morgan, M.T., Hejna, M., Rube, H.T., Hopfner, K.P., Wolberger, C., Song, J.S. & Ha, T. Measuring DNA mechanics on the genome scale. Nature 589, 462-467 (2021).

(16) Yoo, J., Park, S., Maffeo, C., Ha, T. & Aksimentiev, A. DNA sequence and methylation prescribe the inside-out conformational dynamics and bending energetics of DNA minicircles. Nucleic Acids Res 49, 11459-11475 (2021).

Imagined -> religion is about consequences and punishment

Religion to Bianca is a moral code and to the duke it is about human perceptions

Where should I put sleep in this code? Just after the print?

Author response:

The following is the authors’ response to the previous reviews.

eLife assessment

This study presents valuable data on the antigenic properties of neuraminidase proteins of human A/H3N2 influenza viruses sampled between 2009 and 2017. The antigenic properties are found to be generally concordant with genetic groups. Additional analysis have strengthened the revised manuscript, and the evidence supporting the claims is solid.

Public Reviews:

Reviewer #1 (Public Review):

Summary

The authors investigated the antigenic diversity of recent (2009-2017) A/H3N2 influenza neuraminidases (NAs), the second major antigenic protein after haemagglutinin. They used 27 viruses and 43 ferret sera and performed NA inhibition. This work was supported by a subset of mouse sera. Clustering analysis determined 4 antigenic clusters, mostly in concordance with the genetic groupings. Association analysis was used to estimate important amino acid positions, which were shown to be more likely close to the catalytic site. Antigenic distances were calculated and a random forest model used to determine potential important sites.

This revision has addressed many of my concerns of inconsistencies in the methods, results and presentation. There are still some remaining weaknesses in the computational work.

Strengths

(1) The data cover recent NA evolution and a substantial number (43) of ferret (and mouse) sera were generated and titrated against 27 viruses. This is laborious experimental work and is the largest publicly available neuraminidase inhibition dataset that I am aware of. As such, it will prove a useful resource for the influenza community.

(2) A variety of computational methods were used to analyse the data, which give a rounded picture of the antigenic and genetic relationships and link between sequence, structure and phenotype.

(3) Issues raised in the previous review have been thoroughly addressed.

Weaknesses

(1). Some inconsistencies and missing data in experimental methods Two ferret sera were boosted with H1N2, while recombinant NA protein for the others. This, and the underlying reason, are clearly explained in the manuscript. The authors note that boosting with live virus did not increase titres. Additionally, one homologous serum (A/Kansas/14/2017) was not generated, although this would not necessarily have impacted the results.

We agree with the reviewer and this point was addressed in the previous rebuttal.

(2) Inconsistency in experimental results

Clustering of the NA inhibition results identifies three viruses which do not cluster with their phylogenetic group. Again this is clearly pointed out in the paper and is consistent with the two replicate ferret sera. Additionally, A/Kansas/14/2017 is in a different cluster based on the antigenic cartography vs the clustering of the titres

We agree with the reviewer and this point was addressed in the previous rebuttal.

(3) Antigenic cartography plot would benefit from documentation of the parameters and supporting analyses

a. The number of optimisations used

We used 500 optimizations. This information is now included in the Methods section.

b. The final stress and the difference between the stress of the lowest few (e.g. 5) optimisations, or alternatively a graph of the stress of all the optimisations. Information on the stress per titre and per point, and whether any of these were outliers

The stress was obtained from 1, 5, 500, or even 5000 optimizations (resulting in stress values of respectively, 1366.47, 1366.47, 2908.60, and 3031.41). Besides limited variation or non-conversion of the stress values after optimization, the obtained maps were consistent in multiple runs. The map was obtained keeping the best optimization (stress value 1366.47, selected using the keepBestOptimization() function).

Author response image 1.

The stress per point is presented in the heat map below.

The heat map indicates stress per serum (x-axis) and strain (y-axis) in blue to red scale.

c. A measure of uncertainty in position (e.g. from bootstrapping)

Bootstrap was performed using 1000 repeats and 100 optimizations per repeat. The uncertainty is represented in the blob plot below.

Author response image 2.

(4) Random forest

The full dataset was used for the random forest model, including tuning the hyperparameters. It is more robust to have a training and test set to be able to evaluate overfitting (there are 25 features to classify 43 sera).

Explicit cross validation is not necessary for random forests as the out of bag process with multiple trees implicitly covers cross validation. In the random forest function in R this is done by setting the mtry argument (number of variables randomly sampled as candidates at each split). R samples variables with replacement (the same variable can be sampled multiple times) of the candidates from the training set. RF will then automatically take the data that is not selected as candidates as test set. Overfit may happen when all data is used for training but the RF method implicitly does use a test set and does not use all data for training.

Code:

rf <- randomForest(X,y=Y,ntree=1500,mtry=25,keep.forest=TRUE,importance=TRUE)

Reviewer #2 (Public Review):

Summary:

The authors characterized the antigenicity of N2 protein of 43 selected A(H3N2) influenza A viruses isolated from 2009-2017 using ferret and mice immune sera. Four antigenic groups were identified, which the authors claimed to be correlated with their respective phylogenic/ genetic groups. Among 102 amino acids differed by the 44 selected N2 proteins, the authors identified residues that differentiate the antigenicity of the four groups and constructed a machine-learning model that provides antigenic distance estimation. Three recent A(H3N2) vaccine strains were tested in the model but there was no experimental data to confirm the model prediction results.

Strengths:

This study used N2 protein of 44 selected A(H3N2) influenza A viruses isolated from 2009-2017 and generated corresponding panels of ferret and mouse sera to react with the selected strains. The amount of experimental data for N2 antigenicity characterization is large enough for model building.

Weaknesses:

The main weakness is that the strategy of selecting 43 A(H3N2) viruses from 2009-2017 was not explained. It is not clear if they represent the overall genetic diversity of human A(H3N2) viruses circulating during this time. In response to the reviewer's comment, the authors have provided a N2 phylogenetic tree using180 randomly selected N2 sequences from human A(H3N2) viruses from 2009-2017. While the 43 strains seems to scatter across the N2 tree, the four antigenic groups described by the author did not correlated with their respective phylogenic/ genetic groups as shown in Fig. 2. The authors should show the N2 phylogenic tree together with Fig. 2 and discuss the discrepancy observed.

The discrepancies between the provided N2 phylogenetic tree using 180 selected N2 sequences was primarily due to visualization. In the tree presented in Figure 2 the phylogeny was ordered according to branch length in a decreasing way. Further, the tree represented in the rebuttal was built with PhyML 3.0 using JTT substitution model, while the tree in figure 2 was build in CLC Workbench 21.0.5 using Bishop-Friday substitution model. The tree below was built using the same methodology as Figure 2, including branch size ordering. No discrepancies are observed.

Phylogenetic tree representing relatedness of N2 head domain. N2 NA sequences were ordered according to the branch length and phylogenetic clusters are colored as follows: G1: orange, G2: green, G3: blue, and G4: purple. NA sequences that were retained in the breadth panel are named according to the corresponding H3N2 influenza viruses. The other NA sequences are coded.

Author response image 3.

The second weakness is the use of double-immune ferret sera (post-infection plus immunization with recombinant NA protein) or mouse sera (immunized twice with recombinant NA protein) to characterize the antigenicity of the selected A(H3N2) viruses. Conventionally, NA antigenicity is characterized using ferret sera after a single infection. Repeated influenza exposure in ferrets has been shown to enhance antibody binding affinity and may affect the cross-reactivity to heterologous strains (PMID: 29672713). The increased cross-reactivity is supported by the NAI titers shown in Table S3, as many of the double immune ferret sera showed the highest reactivity not against its own homologous virus but to heterologous strains. In response to the reviewer's comment, the authors agreed the use of double-immune ferret sera may be a limitation of the study. It would be helpful if the authors can discuss the potential effect on the use of double-immune ferret sera in antigenicity characterization in the manuscript.

Our study was designed to understand the breadth of the anti-NA response after the incorporation of NA as a vaccine antigens. Our data does not allow to conclude whether increased breadth of protection is merely due to increased antibody titers or whether an NA boost immunization was able to induce antibody responses against epitopes that were not previously recognized by primary response to infection. However, we now mention this possibility in the discussion and cite Kosikova et al. CID 2018, in this context.

Another weakness is that the authors used the newly constructed a model to predict antigenic distance of three recent A(H3N2) viruses but there is no experimental data to validate their prediction (eg. if these viruses are indeed antigenically deviating from group 2 strains as concluded by the authors). In response to the comment, the authors have taken two strains out of the dataset and use them for validation. The results is shown as Fig. R7. However, it may be useful to include this in the main manuscript to support the validity of the model.

The removal of 2 strains was performed to illustrate the predictive performance of the RF modeling. However, Random Forest does not require cross-validation. The reason is that RF modeling already uses an out-of-bag evaluation which, in short, consists of using only a fraction of the data for the creation of the decision trees (2/3 of the data), obviating the need for a set aside the test set:

“…In each bootstrap training set, about one-third of the instances are left out. Therefore, the out-of-bag estimates are based on combining only about one- third as many classifiers as in the ongoing main combination. Since the error rate decreases as the number of combinations increases, the out-of-bag estimates will tend to overestimate the current error rate. To get unbiased out-of-bag estimates, it is necessary to run past the point where the test set error converges. But unlike cross-validation, where bias is present but its extent unknown, the out-of-bag estimates are unbiased…” from https://www.stat.berkeley.edu/%7Ebreiman/randomforest2001.pdf

Reviewer #3 (Public Review):

Summary:

This paper by Portela Catani et al examines the antigenic relationships (measured using monotypic ferret and mouse sera) across a panel of N2 genes from the past 14 years, along with the underlying sequence differences and phylogenetic relationships. This is a highly significant topic given the recent increased appreciation of the importance of NA as a vaccine target, and the relative lack of information about NA antigenic evolution compared with what is known about HA. Thus, these data will be of interest to those studying the antigenic evolution of influenza viruses. The methods used are generally quite sound, though there are a few addressable concerns that limit the confidence with which conclusions can be drawn from the data/analyses.

Strengths:

• The significance of the work, and the (general) soundness of the methods. -Explicit comparison of results obtained with mouse and ferret sera

Weaknesses:

• Approach for assessing influence of individual polymorphisms on antigenicity does not account for potential effects of epistasis (this point is acknowledged by the authors).

We agree with the reviewer and this point was addressed in the previous rebuttal.

• Machine learning analyses neither experimentally validated nor shown to be better than simple, phylogenetic-based inference.

We respectfully disagree with the reviewer. This point was addressed in the previous rebuttal as follows.

This is a valid remark and indeed we have found a clear correlation between NAI cross reactivity and phylogenetic relatedness. However, besides achieving good prediction of the experimental data (as shown in Figure 5 and in FigureR7), machine Learning analysis has the potential to rank or indicate major antigenic divergences based on available sequences before it has consolidated as new clade. ML can also support the selection and design of broader reactive antigens. “

Recommendations for the authors:

Reviewer #2 (Recommendations For The Authors):

(1) Discuss the discrepancy between Fig. 2 and the newly constructed N2 phylogenetic tree with 180 randomly selected N2 sequences of A(H3N2) viruses from 2009-2017. Specifically please explain the antigenic vs. phylogenetic relationship observed in Fig. 2 was not observed in the large N2 phylogenetic tree.

Discrepancies were due to different method and visualization. A new tree was provided.

(2) Include a sentence to discuss the potential effect on the use of double-immune ferret sera in antigenic characterization.

We prefer not to speculate on this.

(3) Include the results of the exercise run (with the use of Swe17 and HK17) in the manuscript as a way to validate the model.

The exercise was performed to illustrate predictive potential of the RF modeling to the reviewer. However, cross-validation is not a usual requirement for random forest, since it uses out-of-bag calculations. We prefer to not include the exercise runs within the main manuscript.

I added this because linewar_fit was not declared causing an error in the code further down

Author response:

The following is the authors’ response to the original reviews.

eLife assessment

This study presents a new and valuable theoretical account of spatial representational drift in the hippocampus. The evidence supporting the claims is convincing, with a clear and accessible explanation of the phenomenon. Overall, this study will likely attract researchers exploring learning and representation in both biological and artificial neural networks.

We would like to ask the reviewers to consider elevating the assessment due to the following arguments. As noted in the original review, the study bridges two different fields (machine learning and neuroscience), and does not only touch a single subfield (representational drift in neuroscience). In the revision, we also analysed data from four different labs, strengthening the evidence and the generality of the conclusions.

Public Reviews:

Reviewer #1 (Public Review):

The authors start from the premise that neural circuits exhibit "representational drift" -- i.e., slow and spontaneous changes in neural tuning despite constant network performance. While the extent to which biological systems exhibit drift is an active area of study and debate (as the authors acknowledge), there is enough interest in this topic to justify the development of theoretical models of drift.

The contribution of this paper is to claim that drift can reflect a mixture of "directed random motion" as well as "steady state null drift." Thus far, most work within the computational neuroscience literature has focused on the latter. That is, drift is often viewed to be a harmless byproduct of continual learning under noise. In this view, drift does not affect the performance of the circuit nor does it change the nature of the network's solution or representation of the environment. The authors aim to challenge the latter viewpoint by showing that the statistics of neural representations can change (e.g. increase in sparsity) during early stages of drift. Further, they interpret this directed form of drift as "implicit regularization" on the network.

The evidence presented in favor of these claims is concise. Nevertheless, on balance, I find their evidence persuasive on a theoretical level -- i.e., I am convinced that implicit regularization of noisy learning rules is a feature of most artificial network models. This paper does not seem to make strong claims about real biological systems. The authors do cite circumstantial experimental evidence in line with the expectations of their model (Khatib et al. 2022), but those experimental data are not carefully and quantitatively related to the authors' model.

We thank the reviewer for pushing us to present stronger experimental evidence. We now analysed data from four different labs. Two of those are novel analyses of existing data (Karlsson et al, Jercog et al). All datasets show the same trend - increasing sparsity and increasing information per cell. We think that the results, presented in the new figure 3, allow us to make a stronger claim on real biological systems.

To establish the possibility of implicit regularization in artificial networks, the authors cite convincing work from the machine-learning community (Blanc et al. 2020, Li et al., 2021). Here the authors make an important contribution by translating these findings into more biologically plausible models and showing that their core assumptions remain plausible. The authors also develop helpful intuition in Figure 4 by showing a minimal model that captures the essence of their result.

We are glad that these translation efforts are appreciated.

In Figure 2, the authors show a convincing example of the gradual sparsification of tuning curves during the early stages of drift in a model of 1D navigation. However, the evidence presented in Figure 3 could be improved. In particular, 3A shows a histogram displaying the fraction of active units over 1117 simulations. Although there is a spike near zero, a sizeable portion of simulations have greater than 60% active units at the end of the training, and critically the authors do not characterize the time course of the active fraction for every network, so it is difficult to evaluate their claim that "all [networks] demonstrated... [a] phase of directed random motion with the low-loss space." It would be useful to revise the manuscript to unpack these results more carefully. For example, a histogram of log(tau) computed in panel B on a subset of simulations may be more informative than the current histogram in panel A.

The previous figure 3A was indeed confusing. In particular, it lumped together many simulations without proper curation. We redid this figure (now Figure 4), and added supplementary figures (Figures S1, S2) to better explain our results. It is now clear that the simulations with a large number of active units were either due to non-convergence, slow timescale of sparsification or simulations featuring label noise in which the fraction of active units is less affected. Regarding the log(tau) calculation, while it could indeed be an informative plot, it could not be calculated in a simple manner for all simulations. This is because learning curves are not always exponential, but sometimes feature initial plateaus (see also Saxe et al 2013, Schuessler et al 2020). We added a more detailed explanation of this limitation in the methods section, and we believe the current figure exemplifies the effect in a satisfactory manner.

Reviewer #2 (Public Review):

Summary:

In the manuscript "Representational drift as a result of implicit regularization" the authors study the phenomenon of representational drift (RD) in the context of an artificial network that is trained in a predictive coding framework. When trained on a task for spatial navigation on a linear track, they found that a stochastic gradient descent algorithm led to a fast initial convergence to spatially tuned units, but then to a second very slow, yet directed drift which sparsified the representation while increasing the spatial information. They finally show that this separation of timescales is a robust phenomenon and occurs for a number of distinct learning rules.

Strengths:

This is a very clearly written and insightful paper, and I think people in the community will benefit from understanding how RD can emerge in such artificial networks. The mechanism underlying RD in these models is clearly laid out and the explanation given is convincing.

We thank the reviewer for the support.

Weaknesses:

It is unclear how this mechanism may account for the learning of multiple environments.

There are two facets to the topic of multiple environments. First, are the results of the current paper relevant when there are multiple environments? Second, what is the interaction between brain mechanisms of dealing with multiple environments and the results of the current paper?

We believe the answer to the first question is positive. The near-orthogonality of representations between environments implies that changes in one can happen without changes in the other. This is evident, for instance, in Khatib et al and Geva et al - in both cases, drift seems to happen independently in two environments, even though they are visited intermittently and are visually similar.

The second question is a fascinating one, and we are planning to pursue it in future work. While the exact way in which the brain achieves this near-independence is an open question, remapping is one possible window into this process.

We extended the discussion to make these points clear.

The process of RD through this mechanism also appears highly non-stationary, in contrast to what is seen in familiar environments in the hippocampus, for example.

The non-stationarity noted by the reviewer is indeed a major feature of our observations, and is indeed linked to familiarity. We divide learning into three phases (now more clearly stated in Table 1 and Figure 4C). The first, rapid phase, consists of improvement of performance - corresponding to initial familiarity with the environment. The third phase, often reported in the literature of representational drift, is indeed stationary and obtained after prolonged familiarity. Our work focuses on the second phase, which is not as immediate as the first one, and can take several days. We note in the discussion that experiments which include a long familiarization process can miss this phase (see also Table 3). Furthermore, we speculate that real life is less stationary than a lab environment, and this second phase might actually be more relevant there.

Reviewer #3 (Public Review):

Summary:

Single-unit neural activity tuned to environmental or behavioral variables gradually changes over time. This phenomenon, called representational drift, occurs even when all external variables remain constant, and challenges the idea that stable neural activity supports the performance of well-learned behaviors. While a number of studies have described representational drift across multiple brain regions, our understanding of the underlying mechanism driving drift is limited. Ratzon et al. propose that implicit regularization - which occurs when machine learning networks continue to reconfigure after reaching an optimal solution - could provide insights into why and how drift occurs in neurons. To test this theory, Ratzon et al. trained a Feedforward Network trained to perform the oft-utilized linear track behavioral paradigm and compare the changes in hidden layer units to those observed in hippocampal place cells recorded in awake, behaving animals.

Ratzon et al. clearly demonstrate that hidden layer units in their model undergo consistent changes even after the task is well-learned, mirroring representational drift observed in real hippocampal neurons. They show that the drift occurs across three separate measures: the active proportion of units (referred to as sparsification), spatial information of units, and correlation of spatial activity. They continue to address the conditions and parameters under which drift occurs in their model to assess the generalizability of their findings.

However, the generalizability results are presented primarily in written form: additional figures are warranted to aid in reproducibility.

We added figures, and a Github with all the code to allow full reproducibility.

Last, they investigate the mechanism through which sparsification occurs, showing that the flatness of the manifold near the solution can influence how the network reconfigures. The authors suggest that their findings indicate a three-stage learning process: 1) fast initial learning followed by 2) directed motion along a manifold which transitions to 3) undirected motion along a manifold.

Overall, the authors' results support the main conclusion that implicit regularization in machine learning networks mirrors representational drift observed in hippocampal place cells.

We thank the reviewer for this summary.

However, additional figures/analyses are needed to clearly demonstrate how different parameters used in their model qualitatively and quantitatively influence drift.

We now provide additional figures regarding parameters (Figures S1, S2).

Finally, the authors need to clearly identify how their data supports the three-stage learning model they suggest.

Their findings promise to open new fields of inquiry into the connection between machine learning and representational drift and generate testable predictions for neural data.

Strengths:

(1) Ratzon et al. make an insightful connection between well-known phenomena in two separate fields: implicit regularization in machine learning and representational drift in the brain. They demonstrate that changes in a recurrent neural network mirror those observed in the brain, which opens a number of interesting questions for future investigation.

(2) The authors do an admirable job of writing to a large audience and make efforts to provide examples to make machine learning ideas accessible to a neuroscience audience and vice versa. This is no small feat and aids in broadening the impact of their work.

(3) This paper promises to generate testable hypotheses to examine in real neural data, e.g., that drift rate should plateau over long timescales (now testable with the ability to track single-unit neural activity across long time scales with calcium imaging and flexible silicon probes). Additionally, it provides another set of tools for the neuroscience community at large to use when analyzing the increasingly high-dimensional data sets collected today.

We thank the reviewer for these comments. Regarding the hypotheses, these are partially confirmed in the new analyses we provide of data from multiple labs (new Figure 3 and Table 3) - indicating that prolonged exposure to the environment leads to more stationarity.

Weaknesses:

(1) Neural representational drift and directed/undirected random walks along a manifold in ML are well described. However, outside of the first section of the main text, the analysis focuses primarily on the connection between manifold exploration and sparsification without addressing the other two drift metrics: spatial information and place field correlations. It is therefore unclear if the results from Figures 3 and 4 are specific to sparseness or extend to the other two metrics. For example, are these other metrics of drift also insensitive to most of the Feedforward Network parameters as shown in Figure 3 and the related text? These concerns could be addressed with panels analogous to Figures 3a-c and 4b for the other metrics and will increase the reproducibility of this work.

We note that the results from figures 3 and 4 (original manuscript) are based on abstract tasks, while in figure 2 there is a contextual notion of spatial position. Spatial position metrics are not applicable to the abstract tasks as they are simple random mapping of inputs, and there isn’t necessarily an underlying latent variable such as position. This transition between task types is better explained in the text now. In essence the spatial information and place field correlation changes are simply signatures of the movements in parameter space. In the abstract tasks their change becomes trivial, as the spatial information becomes strongly correlated with sparsity and place fields are simply the activity vectors of units. These are guaranteed to change as long as there are changes in the activity statistics. We present here the calculation of these metrics averaged over simulations for completeness.

Author response image 1.

PV correlation between training time points averaged over 362 simulations. (B) Mean SI of units normalized to first time step, averaged over 362 simulations. Red line shows the average time point of loss convergence, the shaded area represents one standard deviation.

(2) Many caveats/exceptions to the generality of findings are mentioned only in the main text without any supporting figures, e.g., "For label noise, the dynamics were qualitatively different, the fraction of active units did not reduce, but the activity of the units did sparsify" (lines 116-117). Supporting figures are warranted to illustrate which findings are "qualitatively different" from the main model, which are not different from the main model, and which of the many parameters mentioned are important for reproducing the findings.

We now added figures (S1, S2) that show this exactly. We also added a github to allow full reproduction.

(3) Key details of the model used by the authors are not listed in the methods. While they are mentioned in reference 30 (Recanatesi et al., 2021), they need to be explicitly defined in the methods section to ensure future reproducibility.

The details of the simulation are detailed in the methods sections. We also added a github to allow full reproducibility.

(4) How different states of drift correspond to the three learning stages outlined by the authors is unclear. Specifically, it is not clear where the second stage ends, and the third stage begins, either in real neural data or in the figures. This is compounded by the fact that the third stage - of undirected, random manifold exploration - is only discussed in relation to the introductory Figure 1 and is never connected to the neural network data or actual brain data presented by the authors. Are both stages meant to represent drift? Or is only the second stage meant to mirror drift, while undirected random motion along a manifold is a prediction that could be tested in real neural data? Identifying where each stage occurs in Figures 2C and E, for example, would clearly illustrate which attributes of drift in hidden layer neurons and real hippocampal neurons correspond to each stage.

Thanks for this comment, which urged us to better explain these concepts.

The different processes (reduction in loss, reduction in Hessian) happen in parallel with different timescales. Thus, there are no sharp transitions between the phases. This is now explained in the text in relation to figure 4C, where the approximate boundaries are depicted.

The term drift is often used to denote a change in representation without a change in behavior. In this sense, both the second and third phases correspond to drift. Only the third stage is stationary. This is now emphasized in the text and in the new Table 1. Regarding experimental data, apart from the new figure 3 with four datasets, we also summarize in Table 3 the relation between duration of familiarity and stationarity of the data.

Recommendations for the authors:

The reviewers have raised several concerns. They concur that the authors should address the specific points below to enhance the manuscript.

(1) The three different phases of learning should be clearly delineated, along with how they are determined. It remains unclear in which exact phase the drift is observed.

This is now clearly explained in the new Table 1 and Figure 4C. Note that the different processes (reduction in loss, reduction in Hessian) happen in parallel with different timescales. Thus, there are no sharp transitions between the phases. This is now explained in the text in relation to figure 4C, where the approximate boundaries are depicted.

The term drift is often used to denote a change in representation without a change in behavior. In this sense, both the second and third phases correspond to drift. Only the third stage is stationary. This is now emphasized in the text and in the new Table 1. Regarding experimental data, apart from the new figure 3 with four datasets, we also summarize in Table 3 the relation between duration of familiarity and stationarity of the data.

(2) The term "sparsification" of unit activity is not fully clear. Its meaning should be more explicitly explained, especially since, in the simulations, a significant number of units appear to remain active (Fig. 3A).

We now define precisely the two measures we use - Active Fraction, and Fraction Active Units. There is a new section with an accompanying figure in the Methods section. As Figure S2 shows, the noise statistics (label noise vs. update noise) differentially affects these two measures.

(3) While the study primarily focuses on one aspect of representational drift-the proportion of active units-it should also explore other features traditionally associated with representational drift, such as spatial information and the correlation between place fields.

This absence of features is related to the abstract nature of some of the tasks simulated in our paper. In our original submission the transition between a predictive coding task to more abstract tasks was not clearly explained, creating some confusion regarding the measured metrics. We now clarified the motivation for this transition.

Both the initial simulation and the new experimental data analysis include spatial information (Figures 2,3). The following simulations (Figure 4) with many parameter choices use more abstract tasks, for which the notion of correlation between place cells and spatial information loses its meaning as there is no spatial ordering of the inputs, and every input is encountered only once. Spatial information becomes strongly correlated with the inverse of the active fraction metric. The correlation between place cells is also directly linked to increase in sparseness for these tasks.

(4) There should be a clearer illustration of how labeling noise influences learning dynamics and sparsification.

This was indeed confusing in the original submission. We removed the simulations with label noise from Figure 4, and added a supplementary figure (S2) illustrating the different effects of label noise.

(5) The representational drift observed in this study's simulations appears to be nonstationary, which differs from in vivo reports. The reasons for this discrepancy should be clarified.

We added experimental results from three additional labs demonstrating a change in activity statistics (i.e. increase in spatial information and increase in sparseness) over a long period of time. We suggest that such a change long after the environment is already familiar is an indication for the second phase, and stress that this change seems to saturate at some point, and that most drift papers start collecting data after this saturation, hence this effect was missed in previous in vivo reports. Furthermore, these effects are become more abundant with the advent on new calcium imaging methods, as the older electrophysiological regording methods did not usually allow recording of large amounts of cells for long periods of time. The new Table 3 surveys several experimental papers, emphasizing the degree of familiarity with the environment.

(6) A distinctive feature of the hippocampus is its ability to learn different spatial representations for various environments. The study does not test representational drift in this context, a topic of significant interest to the community. Whether the authors choose to delve into this is up to them, but it should at least be discussed more comprehensively, as it's only briefly touched upon in the current manuscript version.

There are two facets to the topic of multiple environments. First, are the results of the current paper relevant when there are multiple environments? Second, what is the interaction between brain mechanisms of dealing with multiple environments and the results of the current paper?

We believe the answer to the first question is positive. The near-orthogonality of representations between environments implies that changes in one can happen without changes in the other. This is evident, for instance, in Khatib et al and Geva et al - in both cases, drift seems to happen independently in two environments, even though they are visited intermittently and are visually similar.

The second question is a fascinating one, and we are planning to pursue it in future work. While the exact way in which the brain achieves this near-independence is an open question, remapping is one possible window into this process.

We extended the discussion to make these points clear.

(7) The methods section should offer more details about the neural nets employed in the study. The manuscript should be explicit about the terms "hidden layer", "units", and "neurons", ensuring they are defined clearly and not used interchangeably..

We changed the usage of these terms to be more coherent and made our code publicly available. Specifically, “units” refer to artificial networks and “neurons” to biological ones.

In addition, each reviewer has raised both major and minor concerns. These are listed below and should be addressed where possible.

Reviewer #1 (Recommendations For The Authors):

I recommend that the authors edit the text to soften their claims. For example:

In the abstract "To uncover the underlying mechanism, we..." could be changed to "To investigate, we..."

Agree. Done

On line 21, "Specifically, recent studies showed that..." could be changed to "Specifically, recent studies suggest that..."

Agree. Done

On line 100, "All cases" should probably be softened to "Most cases" or more details should be added to Figure 3 to support the claim that every simulation truly had a phase of directed random motion.

The text was changed in accordance with the reviewer’s suggestion. In addition, the figure was changed and only includes simulations in which we expected unit sparsity to arise (without label noise). We also added explanations and supplementary figures for label noise.

Unless I missed something obvious, there is no new experimental data analysis reported in the paper. Thus, line 159 of the discussion, "a phenomenon we also observed in experimental data" should be changed to "a phenomenon that recently reported in experimental data."

We thank the reviewer for drawing our attention to this. We now analyzed data from three other labs, two of which are novel analyses on existing data. All four datasets show the same trends of sparseness with increasing spatial information. The new Figure 3 and text now describe this.

On line 179 of the Discussion, "a family of network configurations that have identical performance..." could be softened to "nearly identical performance." It would be possible for networks to have minuscule differences in performance that are not detected due to stochastic batch effects or limits on machine precision.

The text was changed in accordance with the reviewer’s suggestion.

Other minor comments:

Citation 44 is missing the conference venue, please check all citations are formatted properly.

Corrected.

In the discussion on line 184, the connection to remapping was confusing to me, particularly because the cited reference (Sanders et al. 2020) is more of a conceptual model than an artificial network model that could be adapted to the setting of noisy learning considered in this paper. How would an RNN model of remapping (e.g. Low et al. 2023; Remapping in a recurrent neural network model of navigation and context inference) be expected to behave during the sparsifying portion of drift?

We now clarified this section. The conceptual model of Sanders et al includes a specific prediction (Figure 7 there) which is very similar to ours - a systematic change in robustness depending on duration of training. Regarding the Low et al model, using such mechanistic models is an exciting avenue for future research.

Reviewer #2 (Recommendations For The Authors):

I only have two major questions.

(1) Learning multiple representations: Memory systems in the brain typically must store many distinct memories. Certainly, the hippocampus, where RD is prominent, is involved in the ongoing storage of episodic memories. But even in the idealized case of just two spatial memories, for example, two distinct linear tracks, how would this learning process look? Would there be any interference between the two learning processes or would they be largely independent? Is the separation of time scales robust to the number of representations stored? I understand that to answer this question fully probably requires a research effort that goes well beyond the current study, but perhaps an example could be shown with two environments. At the very least the authors could express their thoughts on the matter.

There are two facets to the topic of multiple environments. First, are the results of the current paper relevant when there are multiple environments? Second, what is the interaction between brain mechanisms of dealing with multiple environments and the results of the current paper?

We believe the answer to the first question is positive. The near-orthogonality of representations between environments implies that changes in one can happen without changes in the other. This is evident, for instance, in Khatib et al and Geva et al - in both cases, drift seems to happen independently in two environments, even though they are visited intermittently and are visually similar.

The second question is a fascinating one, and we are planning to pursue it in future work. While the exact way in which the brain achieves this near-independence is an open question, remapping is one possible window into this process.

We extended the discussion to make these points clear.

(2) Directed drift versus stationarity: I could not help but notice that the RD illustrated in Fig.2D is not stationary in nature, i.e. the upper right and lower left panels are quite different. This appears to contrast with findings in the hippocampus, for example, Fig.3e-g in (Ziv et al, 2013). Perhaps it is obvious that a directed process will not be stationary, but the authors note that there is a third phase of steady-state null drift. Is the RD seen there stationary? Basically, I wonder if the process the authors are studying is relevant only as a novel environment becomes familiar, or if it is also applicable to RD in an already familiar environment. Please discuss the issue of stationarity in this context.

The non-stationarity noted by the reviewer is indeed a major feature of our observations, and is indeed linked to familiarity. We divide learning into three phases (now more clearly stated in Table 1 and Figure 4C). The first, rapid, phase consists of improvement of performance - corresponding to initial familiarity with the environment. The third phase, often reported in the literature of representational drift, is indeed stationary and obtained after prolonged familiarity. Our work focuses on the second phase, which is not as immediate as the first one, and can take several days. We note in the discussion that experiments which include a long familiarization process can miss this phase (see also Table 3). Furthermore, we speculate that real life is less stationary than a lab environment, and this second phase might actually be more relevant there.

Reviewer #3 (Recommendations For The Authors):

Most of my general recommendations are outlined in the public review. A large portion of my comments regards increasing clarity and explicitly defining many of the terms used which may require generating more figures (to better illustrate the generality of findings) or modifying existing figures (e.g., to show how/where the three stages of learning map onto the authors' data).

Sparsification is not clearly defined in the main text. As I read it, sparsification is meant to refer to the activity of neurons, but this needs to be clearly defined. For example, lines 262-263 in the methods define "sparseness" by the number of active units, but lines 116-117 state: "For label noise, the dynamics were qualitatively different, the fraction of active units did not reduce, but the activity of the units did sparsify." If the fraction of active units (defined as "sparseness") did not change, what does it mean that the activity of the units "sparsified"? If the authors mean that the spatial activity patterns of hidden units became more sharply tuned, this should be clearly stated.

We now defined precisely the two measures we use - Active Fraction, and Fraction Active Units. There is a new section with an accompanying figure in the Methods section. As Figure S2 shows, the noise statistics (label noise vs. update noise) differentially affects these two measures.

Likewise, it is unclear which of the features the authors outlined - spatial information, active proportion of units, and spatial correlation - are meant to represent drift. The authors should clearly delineate which of these three metrics they mean to delineate drift in the main text rather than leave it to the reader to infer. While all three are mentioned early on in the text (Figure 2), the authors focus more on sparseness in the last half of the text, making it unclear if it is just sparseness that the authors mean to represent drift or the other metrics as well.

The main focus of our paper is on the non-stationarity of drift. Namely that features (such as these three) systematically change in a directed manner as part of the drift process. This is in The new analyses of experimental data show sparseness and spatial information.

The focus on sparseness in the second half of the paper is because we move to more abstract These are also easy to study in the more abstract tasks in the second part of the paper. In our original submission the transition between a predictive coding task to more abstract tasks was not clearly explained, creating some confusion regarding the measured metrics. We now clarified the motivation for this transition.

It is not clear if a change in the number of active units alone constitutes "drift", especially since Geva et al. (2023) recently showed that both changes in firing rate AND place field location drive drift, and that the passage of time drives changes in activity rate (or # cells active).

Our work did not deal with purely time-dependent drift, but rather focused on experience-dependence. Furthermore, Geva et al study the stationary phase of drift, where we do not expect a systematic change in the total number of cells active. They report changes in the average firing rate of active cells in this phase, as a function of time - which does not contradict our findings.

"hidden layer", "units", and "neurons" seem to be used interchangeably in the text (e.g., line 81-85). However, this is confusing in several places, in particular in lines 83-85 where "neurons" is used twice. The first usage appears to refer to the rate maps of the hidden layer units simulated by the authors, while the second "neurons" appears to refer to real data from Ziv 2013 (ref 5). The authors should make it explicit whether they are referring to hidden layer units or actual neurons to avoid reader confusion.

We changed the usage of these terms to be more coherent. Specifically, “units” refer to artificial networks and “neurons” to biological ones.

The authors should clearly illustrate which parts of their findings support their three-phase learning theory. For example, does 2E illustrate these phases, with the first tenth of training time points illustrating the early phase, time 0.1-0.4 illustrating the intermediate phase, and 0.4-1 illustrating the last phase? Additionally, they should clarify whether the second and third stages are meant to represent drift, or is it only the second stage of directed manifold exploration that is considered to represent drift? This is unclear from the main text.

The different processes (reduction in loss, reduction in Hessian) happen in parallel with different timescales. Thus, there are no sharp transitions between the phases. This is now explained in the text in relation to figure 4C, where the approximate boundaries are depicted.

The term drift is often used to denote a change in representation without a change in behavior. In this sense, both the second and third phases correspond to drift. Only the third stage is stationary. This is now emphasized in the text and in the new Table 1. Regarding experimental data, apart from the new figure 3 with four datasets, we also summarize in Table 3 the relation between duration of familiarity and stationarity of the data.

Line 45 - It appears that the acronym ML is not defined above here anywhere.

Added.

Line 71: the ReLU function should be defined in the text, e.g., sigma(x) = x if x > 0 else 0.

Added.

106-107: Figures (or supplemental figures) to demonstrate how most parameters do not influence sparsification dynamics are warranted. As written, it is unclear what "most parameters" mean - all but noise scale. What about the learning rule? Are there any interactions between parameters?

We now removed the label noise from Figure 4, and added two supplementary figures to clearly explain the effect of parameters. Figure 4 itself was also redone to clarify this issue.

2F middle: should "change" be omitted for SI?

The panel was replaced by a new one in Figure 3.

116-119: A figure showing how results differ for label noise is warranted.

This is now done in Figure S1, S2.

124: typo, The -> the

Corrected.

127-129: This conclusion statement is the first place in the text where the three stages are explicitly outlined. There does not appear to be any support or further explanation of these stages in the text above.

We now explain this earlier at the end of the Introduction section, along with the new Table 1 and marking on Figure 4C.

132-133 seems to be more of a statement and less of a prediction or conclusion - do the authors mean "the flatness of the loss landscape in the vicinity of the solution predicts the rate of sparsification?"

We thank the reviewer for this observation. The sentence was rephrased:

Old: As illustrated in Fig. 1, different solutions in the zero-loss manifold might vary in some of their properties. The specific property suggested from theory is the flatness of the loss landscape in the vicinity of the solution.

New: As illustrated in Fig. 1, solutions in the zero-loss manifold have identical loss, but might vary in some of their properties. The authors of [26] suggest that noisy learning will slowly increase the flatness of the loss landscape in the vicinity of the solution.

135: typo, it's -> its

Corrected.

Line 135-136 "Crucially, the loss on the 136 entire manifold is exactly zero..." This appears to contradict the Figure 4A legend - the loss appears to be very high near the top and bottom edges of the manifold in 4A. Do the authors mean that the loss along the horizontal axis of the manifold is zero?

The reviewer is correct. The manifold mentioned in the sentence is indeed the horizontal axis. We changed the text and the figure to make it clearer.

Equation 6: This does not appear to agree with equation 2 - should there be an E_t term for an expectation function?

Corrected.

Line 262-263: "Sparseness means that a unit has become inactive for all inputs." This should also be stated explicitly as the definition of sparseness/sparsification in the main text.

We now define precisely the two measures we use - Active Fraction, and Fraction Active Units. There is a new section with an accompanying figure in the Methods section. As Figure S2 shows, the noise statistics (label noise vs. update noise) differentially affects these two measures.

Author response:

The following is the authors’ response to the original reviews.

eLife assessment

This is a valuable computational study that applies the machine learning method of bilinear modeling to the problem of relating gene expression to connectivity. Specifically, the author attempts to use transcriptomic data from mouse retinal neurons to predict their known connectivity. The results are promising, although the reviewers felt that demonstration of the general applicability of the approach required testing it against a second data set. Hence the present results were felt to provide borderline incomplete support for a key premise of the paper.

We thank the reviewers for their insightful and constructive feedback. In response to the reviews, we have undertaken a comprehensive revision of our manuscript, incorporating changes and improvements as outlined below:

(1) New results have been included showcasing the application of our bilinear model to a seconddataset focusing on C. elegans gap junction connectivity. This extension validates our model with a biological context other than mouse retina and facilitates a direct comparison with the spatial connectome model (SCM).

(2) A new section titled "Previous Approaches" has been added to background, situating our studywithin the broader landscape of existing modeling methodologies.

(3) The discussion sections have been expanded to fully incorporate the suggestions and insightsoffered by the reviewers. This includes a deeper exploration of the implications of our findings, potential applications of our model, and a more thorough consideration of its limitations and future directions.

(4) To streamline the main text and ensure that the core narrative remains focused and accessible, select figures and tables have been relocated to the "Supplementary Materials" section.

Reviewer 1 (Public Review):

Summary of what the author was trying to achieve: In this study, the author aimed to develop a method for estimating neuronal-type connectivity from transcriptomic gene expression data, specifically from mouse retinal neurons. They sought to develop an interpretable model that could be used to characterize the underlying genetic mechanisms of circuit assembly and connectivity.

Strengths:

The proposed bilinear model draws inspiration from commonly implemented recommendation systems in the field of machine learning. The author presents the model clearly and addresses critical statistical limitations that may weaken the validity of the model such as multicollinearity and outliers. The author presents two formulations of the model for separate scenarios in which varying levels of data resolution are available. The author effectively references key work in the field when establishing assumptions that affect the underlying model and subsequent results. For example, correspondence between gene expression cell types and connectivity cell types from different references are clearly outlined in Tables 1-3. The model training and validation are sufficient and yield a relatively high correlation with the ground truth connectivity matrix. Seemingly valid biological assumptions are made throughout, however, some assumptions may reduce resolution (such as averaging over cell types), thus missing potentially important single-cell gene expression interactions.

Thank you for recognizing the strengths of our work, particularly the clarity of the model presentation and its foundation in recommendation systems. In the revised manuscript we have also extended the model’s capabilities to analyze gene interactions for neural connectivity at single-cell resolution, when gene expression and connectivity of each cell are known simultaneously.

Weaknesses:

The main results of the study could benefit from replication in another dataset beyond mouse retinal neurons, to validate the proposed method. Dimensionality reduction significantly reduces the resolution of the model and the PCA methodology employed is largely non-deterministic. This may reduce the resolution and reproducibility of the model. It may be worth exploring how the PCA methodology of the model may affect results when replicating. Figure 5, ’Gene signatures associated with the two latent dimensions’, lacks some readability and related results could be outlined more clearly in the results section. There should be more discussion on weaknesses of the results e.g. quantification of what connectivity motifs were not captured and what gene signatures might have been missed.

We acknowledge the significance of validating our method across different datasets. In line with this, our revised manuscript now includes an expanded analysis utilizing a C. elegans gap junction connectivity dataset, which not only broadens the method’s demonstrated applicability but also underscores its versatility across varied neuronal systems.

To address the concern of resolution and reproducibility associated with PCA preprocessing, we have conducted a comparative analysis from five replicates of the bilinear model, presenting the results in the revised manuscript (Figure S3). This analysis confirms the consistency of the solutions, as evidenced by the similarity metrics. Furthermore, we discussed alternative methodologies, such as L1 or L2 regularization, to tackle multicollinearity, offering flexibility in preprocessing choices.

In response to feedback on the original Figure 5’s clarity, we have replaced the original Figure 5e-h with Table S4, which summarizes the gene ontology (GO) enrichment results and quantifies the number of genes associated with aspects of neural development and synaptic organization. This revision aims to improve the interpretability and accessibility of the results, ensuring a clearer presentation of the model’s insights.

Finally, we have expanded our discussion to address the study’s limitations more comprehensively. This includes exploration of potentially missed connections and gene signatures, such as transcription factors, which might not be captured by a linear model due to its inherent preference for predictors with strong correlations to the target variable.

The main weakness is the lack of comparison against other similar methods, e.g. methods presented in Barabási, Dániel L., and Albert-László Barabási. "A genetic model of the connectome." Neuron 105.3 (2020): 435-445. Kovács, István A., Dániel L. Barabási, and Albert-László Barabási. "Uncovering the genetic blueprint of the C. elegans nervous system." Proceedings of the National Academy of Sciences 117.52 (2020): 33570-33577. Taylor, Seth R., et al. "Molecular topography of an entire nervous system." Cell 184.16 (2021): 4329-4347.

We value your suggestion to compare our model with established methods. The revised manuscript now includes a comparative analysis with the spatial connectome model (SCM) using the same C. elegans dataset. In addition, a section reviewing previous approaches has been included in the background part, and the discussion part has been extended for the comparison.

Appraisal of whether the author achieved their aims, and whether results support their conclusions: The author achieved their aims by recapitulating key connectivity motifs from single-cell gene expression data in the mouse retina. Furthermore, the model setup allowed for insight into gene signatures and interactions, however could have benefited from a deeper evaluation of the accuracy of these signatures. The author claims the method sets a new benchmark for single-cell transcriptomic analysis of synaptic connections. This should be more rigorously proven. (I’m not sure I can speak on the novelty of the method)

In the revised manuscript. we emphasized the bilinear model’s innovative application in the context of neuronal connectivity analysis, inspired by collaborative filtering in recommendation systems. We present quantitative performance metrics, such as the ROC-AUC score and Pearson correlation coefficient, as well as its comparison with the SCM, to benchmark our model’s efficacy in reconstructing connectivity matrices. We also quantified the overlap of the genetic interactions revealed by the bilinear model and the SCM (using the C. elegans dataset), and reported the percentage of the top genes associated with neural development and synaptic organization (using the mouse retina dataset). These numbers set a precedent for future methodological comparisons.

Discussion of the likely impact of the work on the field, and the utility of methods and data to the community : This study provides an understandable bilinear model for decoding the genetic programming of neuronal type connectivity. The proposed model leaves the door open for further testing and comparison with alternative linear and/or non-linear models, such as neural networkbased models. In addition to more complex models, this model can be built on to include higher resolution data such as more gene expression dimensions, different types of connectivity measures, and additional omics data.

We are grateful for your recognition of the study’s potential impact. The bilinear model indeed offers a foundation for future explorations, allowing for integration with more complex models, higher-resolution data, and diverse connectivity measures.

Reviewer 1 (Recommendations For The Authors):

The inclusion of predicted connectivity (Figure 6) of unknown BC neurons is useful as it shows that this is a strong hypothesis generation tool. This utility should potentially be showcased more as it is also brought up in the abstract, "genetic manipulation of circuit wiring", with an explanation of how the model could be leveraged as such. The discussion may benefit from a summarizing sentence regarding which key gene signatures were identified and are in line with the literature, which key gene signatures/connectivity motifs may have been missed, and which gene signatures are novel.

Thank you for the insightful recommendation on emphasizing the model’s utility in generating hypotheses, particularly regarding predicting connectivity. In the revised manuscript, we have expanded the discussion on how our model can be leveraged to guide genetic manipulations at altering circuit wiring and highlighted its potential impact in the field.

We have discussed key gene signatures identified from our model that are in line with existing literature, such as plexins and cadherins, which have been previously recognized for their involvement in synaptic connection formation and maintenance. We have also introduced potential new candidates, such as delta-protocadherins. In the revised manuscript, we summarized potentially missed gene signatures or synaptic connections, to provide a comprehensive view of our findings.

Reviewer 2 (Public Review):

Summary:

In this study, Mu Qiao employs a bilinear modeling approach, commonly utilized in recommendation systems, to explore the intricate neural connections between different pre- and post-synaptic neuronal types. This approach involves projecting single-cell transcriptomic datasets of pre- and post-synaptic neuronal types into a latent space through transformation matrices. Subsequently, the cross-correlation between these projected latent spaces is employed to estimate neuronal connectivity. To facilitate the model training, connectomic data is used to estimate the ground-truth connectivity map. This work introduces a promising model for the exploration of neuronal connectivity and its associated molecular determinants. However, it is important to note that the current model has only been tested with Bipolar Cell and Retinal Ganglion Cell data, and its applicability in more general neuronal connectivity scenarios remains to be demonstrated.

Strengths:

This study introduces a succinct yet promising computational model for investigating connections between neuronal types. The model, while straightforward, effectively integrates singlecell transcriptomic and connectomic data to produce a reasonably accurate connectivity map, particularly within the context of retinal connectivity. Furthermore, it successfully recapitulates connectivity patterns and helps uncover the genetic factors that underlie these connections.

Thank you for your positive assessment of the paper.

Weaknesses:

(1) The study lacks experimental validation of the model’s prediction results.

We recognize the importance of experimental validation in substantiating the predictions made by computational models. While the primary focus of this study remains computational, we have dedicated a section in the revised manuscript, titled "Experimental Validation of Candidate Genes", to outline proposed methodologies for the empirical verification of our model’s predictions. This section specifically discusses the experimental exploration of novel candidate genes, such as deltaprotocadherins, within the mouse retina using AAV-mediated CRISPR/Cas9 genetic manipulation. We plan to collaborate with experimental laboratories to facilitate the validation. Given the extensive nature of experimental work, both in terms of time and resources, it is more pragmatic to present a comprehensive experimental investigation in a follow-up study.

(2) The model’s applicability in other neuronal connectivity settings has not been thoroughly explored.

The question of the model’s broader applicability is well-taken. In response, we have expanded our analysis to include additional neuronal data and connectivity settings. Specifically, the revised manuscript includes results where we apply the model to a dataset of C. elegans gap junction connectivity, demonstrating its potential in different neuronal systems. This extension serves to illustrate the model’s adaptability and potential applicability to a broader range of neuronal connectivity studies.

(3) The proposed method relies on the availability of neuronal connectomic data for model training,which may be limited or absent in certain brain connectivity settings.

We acknowledge the limitations posed by the model’s dependency on comprehensive connectomic data, which may not be readily available across all research contexts. To address this, we have discussed in the revised manuscript several alternative strategies to adapt our model to the available data. This includes exploring the potential of applying the model to available data such as projectome, and integrating other data modalities such as electrophysiological measurements. These initiatives aim to enhance the model’s applicability and ensure its utility in a broader spectrum of brain connectivity studies, especially in scenarios where detailed connectomic data are not available.

Reviewer 2 (Recommendations For The Authors):

Q1. In this work, the author has mainly been studying the retina neuronal type connectivity, it will be interesting to see whether the model works for other brain regions or other neuronal type connectivity as well.

We value your interest in the model’s applicability to other brain regions and neuronal types. To address this, we have extended our analysis in the revised manuscript to include a study on gap junction connectivity between C. elegans neurons. This extension demonstrates the model’s versatility and its potential applicability across various nervous systems and connectivity types.

Q2. Whether the authors can use the same transformation matrices trained from the retina data to predict neuronal connectivity in other brain regions? Or an easier case, the connectivity between RGC types to the neuronal types in SC, dLGN, or other post-RGC-synaptic brain regions. As the neuronal connection mechanisms are conserved and widely shared between different neuronal types, one would expect the same transformation matrices may work in predicting other neuronal type connectivity as well (at least to some extent).

The idea to use the same transformation matrices for predicting connectivity in other brain regions is intriguing. While direct application of these matrices to different regions remains challenging, we discussed the potential scalability of our model to other brain areas. By applying the model to combined datasets from various regions, we could uncover conserved neuronal connection mechanisms. This approach is theoretically feasible and is supported by the demonstrated scalability of the bilinear model and its deep learning variants in industrial applications.

Q3. Section 5.2 Connectivity metric generation: in this work, the author uses the stratification profiles of the neurons to estimate the connectivity metric, how reliable this method is? There will be a scenario where though two neuronal types project to a similar inner plexiform layer, they may not have any connection. Have the authors considered combining other experimental data (like electrophysiology data or neuron tracing data)?

We discussed the reliability of using stratification profiles for estimating connectivity metrics, acknowledging potential limitations. In the revised manuscript, we added discussion on how the integration of additional experimental data, such as electrophysiological and neuron tracing data, could enhance the accuracy of the connectivity metrics.

Q4. Section 6 Model training and validation: does the author have a potential hypothesis as to why 2 dimensions are the best latent feature spaces dimensionality? One would imagine with more dimensionality, the model will give better results. Could it be that the connectivity data that is used to train the model is only considering the two-dimensional space of the neuronal stratification?

The selection of two dimensions for the latent feature space was informed by 5-fold cross-validation, aimed at optimizing model generalization to unseen data. Here while increasing dimensionality improves performance on the training set, it does not necessarily enhance generalization to the validation set. Thus, the choice of two dimensions ensures good performance without overfitting to the training data.

Q5. Could the author provide the source code for the analysis? Or could the author make it a python/R package so that non-computational biologists can easily apply the method to their own data?

We have included a "Data and Code Availability" section in the revised manuscript. This section provides a link to the source code with pointers to datasets used in our study, facilitating the application of our methods by researchers from various backgrounds.

Q6. I know it may be difficult for the author to do, but is it possible to design and perform some experiments to validate the model prediction results, either connectivity partners of transcriptomicallydefined RGC types or the function of the key genetic molecules (which hasn’t been discovered before)? The author may consider collaborating with some experimental labs. The author may even consider predicting the connectivity between RGC with some of its post-synaptic neurons in the brain regions, like SC or dLGN, as recently there are a lot of single-cell sequencing data as well as connectivity data.

We appreciate your suggestion regarding experimental validation. As a future direction, we have discussed potential experimental approaches to validate the model’s predictions in the "Experimental Validation of Candidate Genes" section. Specifically, we propose an experimental design involving the manipulation of delta-protocadherins using AAV-mediated CRISPR/Cas9 and subsequent examination of connectivity phenotypes. We are also open to collaborating with experimental labs to further explore the model’s predictions, particularly in predicting connectivity between RGCs and their post-synaptic neurons in other brain regions.

Reviewer #3 (Public Review):

Summary:

The manuscript addresses a question inspired by the Baroceptor Hypothesis and its links to visual awareness and interoception. Specifically, the reported study aimed to determine if the effects of cardiac contraction (systole) on binocular rivalry (BR) are facilitatory or suppressive. The main experiment - relying on a technically challenging procedure of presenting stimuli synchronised with the heartbeats of participants - has been conducted with great care, and numerous manipulation checks the authors report convincingly show that the methods they used work as intended. Moreover, the control experiment allows for excluding alternative explanations related to participants being aware of their heartbeats. Therefore, the study convincingly shows the effect of cardiac activity on BR - and this is an important finding. The results, however, do not allow for unambiguously determining if this effect is facilitatory or suppressive (see details below), which renders the study not as informative as it could be.

While the authors strongly focus on interoception and awareness, this study will be of interest to researchers studying BR as such. Moreover, the code and the data the authors share can facilitate the adoption of their methods in other labs.

Strengths:

(1) The study required a complex technical setup and the manuscript both describes it well and demonstrates that it was free from potential technical issues (e.g. in section 3.3. Manipulation check).

(2) The sophisticated statistical methods the authors used, at least for a non-statistician like me, appear to be well-suited for their purpose. For example, they take into account the characteristics of BR (gamma distributions of dominance durations). Moreover, the authors demonstrate that at least in one case their approach is more conservative than a more basic one (Binomial test) would be.

(3) Finally, the control experiment, and the analysis it enabled, allow for excluding a multitude of alternative explanations of the main results.

(4) The authors share all their data and materials, even the code for the experiment.

(5) The manuscript is well-written. In particular, it introduces the problem and methods in a way that should be easy to understand for readers coming from different research fields.

Weaknesses:

(1) The interpretation of the main result in the context of the Baroceptor hypothesis is not clear. The manuscript states: The Baroreceptor Hypothesis would predict that the stimulus entrained to systole would spend more time suppressed and, conversely, less time dominant, as cortical activity would be suppressed each time that stimulus pulses. The manuscript does not specify why this should be the case, and the term 'entrained' is not too helpful here (does it refer to neural entrainment? or to 'being in phase with'?). The answer to this question is provided by the manuscript only implicitly, and, to explain my concern, I try to spell it out here in a slightly simplified form.

During systole (cardiac contraction), the visual system is less sensitive to external information, so it 'ignores' periods when the systole-synchronised stimulus is at the peak of its pulse. Conversely, the system is more sensitive during diastole, so the stimulus that is at the peak of its pulse then should dominate for longer, because its peaks are synchronised with the periods of the highest sensitivity of the visual system when the information used to resolve the rivalry is sampled from the environment. This idea, while indeed being a clever test of the hypothesis in question, rests on one critical assumption: that the peak of the stimulus pulse (as defined in the manuscript) is the time when the stimulus is the strongest for the visual system. The notion of 'stimulus strength' is widely used in the BR literature (see Brascamp et al., 2015 for a review). It refers to the stimulus property that, simply speaking, determines its tendency to dominate in the BR. The strength of a stimulus is underpinned by its low-level visual properties, such as contrast and spatial frequency content. Coming back to the manuscript, the pulsing of the stimuli affected at least spatial frequency (and likely other low-level properties), and it is unknown if it was in phase with the pulsing of the stimulus strength, or not. If my understanding of the premise of the study is correct, the conclusions drawn by the authors stand only if it was.

In other words, most likely the strength of one of the stimuli was pulsating in sync with the systole, but is it not clear which stimulus it was. It is possible that, for the visual system, the stimulus meant to pulse in sync with the systole was pulsing strength-wise in phase with the diastole (and the one intended to pulse with in sync with the diastole strength-wise pulsed with the systole). If this is the case, the predictions of the Baroceptor Hypothesis hold, which would change the conclusion of the manuscript.

(2) Using anaglyph goggles necessitates presenting stimuli of a different colour to each eye. The way in which different colours are presented can impact stimulus strength (e.g. consider that different anaglyph foils can attenuate the light they let through to different degrees). To deal with such effects, at least some studies on BR employed procedures of adjusting the colours for each participant individually (see Papathomas et al., 2004; Patel et al., 2015 and works cited there). While I think that counterbalancing applied in the study excludes the possibility that colour-related effects influenced the results, the effects of interest still could be stronger for one of the coloured foils.

(3) Several aspects of the methods (e.g. the stimuli), are not described at the level of detail some readers might be accustomed to. The most important issue here is the task the participants performed. The manuscript says that they pressed a button whenever they experienced a switch in perception, but it is only implied that there were different buttons for each stimulus.

Brascamp, J. W., Klink, P. C., & Levelt, W. J. M. (2015). The 'laws' of binocular rivalry: 50 years of Levelt's propositions. Vision Research, 109, 20-37. https://doi.org/10.1016/j.visres.2015.02.019<br /> Papathomas, T. V., Kovács, I., & Conway, T. (2004). Interocular grouping in binocular rivalry: Basic attributes and combinations. In D. Alais & R. Blake (Eds.), Binocular Rivalry (pp. 155-168). MIT Press<br /> Patel, V., Stuit, S., & Blake, R. (2015). Individual differences in the temporal dynamics of binocular rivalry and stimulus rivalry. Psychonomic Bulletin and Review, 22(2), 476-482. https://doi.org/10.3758/s13423-014-0695-1