Spring AOP 的一个例子就是 @Transactional 注解，它会在方法的执行中开启一个事务，在方法结束后提交事务，这就是通过一个 proxy 来在原始类前后添加功能，让它成为 transactional 的。

Although we now are able to write titles that are not confined by the physicality of the book form due to electronics, we still seem to avoid being "guided" by titles, or rather we prefer brevity which I believe speaks a lot about the current culture and the digital age.

I never thought of titles specifically as a coded message in a market situation, in other words, the meeting point of the linguistic realm and the economic realm but I feel that it is very true: we all buy a book with a look at its cover/title. In addition, I didn't consider the titles of books as of a cultural product themselves.

DOI: 10.1016/j.isci.2025.112207

Resource: RRID:IMSR_CRL:027

Curator: @areedewitt04

SciCrunch record: RRID:IMSR_CRL:027

What is this?

This goes back to what I said earlier that Putin and the political leaders of the country are trying to lay this new ideology into the groundworks of society rather than making an institutionalized system due to it technically not being allowed. Putin wants the society of Russia to have one congruent goal together.

Reviewer #2 (Public review):

Summary:

In "Founder effects arising from gathering dynamics systematically bias emerging pathogen surveillance" Bradford and Hang present an extension to the SIR model to account for the role of larger than pairwise interactions in infectious disease dynamics. They explore the impact of accounting for group interactions on the progression of infection through the various sub-populations that make up the population as a whole. Further, they explore the extent to which interaction heterogeneity can bias epidemiological inference from surveillance data in the form of IFR and variant growth rate dynamics. This work advances the theoretical formulation of the SIR model and may allow for more realistic modeling of infectious disease outbreaks in the future.

Strengths:

(1) This work addresses an important limitation of standard SIR models. While this limitation has been addressed previously in the form of network-based models, those are, as the authors argue, difficult to parameterize to real-world scenarios. Further, this work highlights critical biases that may appear in real-world epidemiological surveillance data. Particularly, over-estimation of variant growth rates shortly after emergence has led to a number of "false alarms" about new variants over the past five years (although also to some true alarms).

(2) While the results presented here generally confirm my intuitions on this topic, I think it is really useful for the field to have it presented in such a clear manner with a corresponding mathematical framework. This will be a helpful piece of work to point to to temper concerns about rapid increases in the frequency of rare variants.

(3) The authors provide a succinct derivation of their model that helps the reader understand how they arrived at their formulation starting from the standard SIR model.

(4) The visualizations throughout are generally easy to interpret and communicate the key points of the authors' work.

(5) I thank the authors for providing detailed code to reproduce manuscript figures in the associated GitHub repo.

Weaknesses:

(1) The authors argue that network-based SIR models are difficult to parameterize (line 66), however, the model presented here also has a key parameter, mainly P_n, or the distribution of risk groups in the population. I think it is important to explore the extent to which this parameter can be inferred from real-world data to assess whether this model is, in practice, any easier to parameterize.

(2) The authors explore only up to four different risk groups, accounting for only four-wise interactions. But, clearly, in real-world settings, there can be much larger gatherings that promote transmission. What was the justification for setting such a low limit on the maximum group size? I presume it's due to computational efficiency, which is understandable, but it should be discussed as a limitation.

(3) Another key limitation that isn't addressed by the authors is that there may be population structure beyond just risk heterogeneity. For example, there may be two separate (or, weakly connected) high-risk sub-groups. This will introduce temporal correlation in interactions that are not (and can not easily be) captured in this model. My instinct is that this would dampen the difference between risk groups shown in Figure 2A. While I appreciate the authors's desire to keep their model relatively simple, I think this limitation should be explicitly discussed as it is, in my opinion, relatively significant.

Author response:

Reviewer #1 (Public review):

Summary:

This work considers the biases introduced into pathogen surveillance due to congregation effects, and also models homophily and variants/clades. The results are primarily quantitative assessments of this bias but some qualitative insights are gained e.g. that initial variant transmission tends to be biased upwards due to this effect, which is closely related to classical founder effects.

Strengths:

The model considered involves a simplification of the process of congregation using multinomial sampling that allows for a simpler and more easily interpretable analysis.

Weaknesses:

This simplification removes some realism, for example, detailed temporal transmission dynamics of congregations.

We appreciate Reviewer #1's comments. We hope our framework, like the classic SIR model, can be adapted in the future to build more complex and realistic models.

Reviewer #2 (Public review):

Summary:

In "Founder effects arising from gathering dynamics systematically bias emerging pathogen surveillance" Bradford and Hang present an extension to the SIR model to account for the role of larger than pairwise interactions in infectious disease dynamics. They explore the impact of accounting for group interactions on the progression of infection through the various sub-populations that make up the population as a whole. Further, they explore the extent to which interaction heterogeneity can bias epidemiological inference from surveillance data in the form of IFR and variant growth rate dynamics. This work advances the theoretical formulation of the SIR model and may allow for more realistic modeling of infectious disease outbreaks in the future.

Strengths:

(1) This work addresses an important limitation of standard SIR models. While this limitation has been addressed previously in the form of network-based models, those are, as the authors argue, difficult to parameterize to real-world scenarios. Further, this work highlights critical biases that may appear in real-world epidemiological surveillance data. Particularly, over-estimation of variant growth rates shortly after emergence has led to a number of "false alarms" about new variants over the past five years (although also to some true alarms).

(2) While the results presented here generally confirm my intuitions on this topic, I think it is really useful for the field to have it presented in such a clear manner with a corresponding mathematical framework. This will be a helpful piece of work to point to to temper concerns about rapid increases in the frequency of rare variants.

(3) The authors provide a succinct derivation of their model that helps the reader understand how they arrived at their formulation starting from the standard SIR model.

(4) The visualizations throughout are generally easy to interpret and communicate the key points of the authors' work.

(5) I thank the authors for providing detailed code to reproduce manuscript figures in the associated GitHub repo.

Weaknesses:

(1) The authors argue that network-based SIR models are difficult to parameterize (line 66), however, the model presented here also has a key parameter, mainly P_n, or the distribution of risk groups in the population. I think it is important to explore the extent to which this parameter can be inferred from real-world data to assess whether this model is, in practice, any easier to parameterize.

(2) The authors explore only up to four different risk groups, accounting for only four-wise interactions. But, clearly, in real-world settings, there can be much larger gatherings that promote transmission. What was the justification for setting such a low limit on the maximum group size? I presume it's due to computational efficiency, which is understandable, but it should be discussed as a limitation.

(3) Another key limitation that isn't addressed by the authors is that there may be population structure beyond just risk heterogeneity. For example, there may be two separate (or, weakly connected) high-risk sub-groups. This will introduce temporal correlation in interactions that are not (and can not easily be) captured in this model. My instinct is that this would dampen the difference between risk groups shown in Figure 2A. While I appreciate the authors's desire to keep their model relatively simple, I think this limitation should be explicitly discussed as it is, in my opinion, relatively significant.

We appreciate Reviewer 2's thoughtful comments and wish to address some of the weaknesses:

We agree that inferring P_n from real data will be challenging, but think this is an important direction for future research. Further, we’d like to reframe our claim that our approach is "easier to parameterize" than network models. Rather, P_n has fewer degrees of freedom than analogous network models, just as many different networks can share the same degree distribution. Fewer degrees of freedom mean that we expect our model to suffer from fewer identifiability issues when fitting to data, though non-identifiability is often inescapable in models of this nature (e.g., \beta and \gamma in the SIR model are not uniquely identifiable during exponential growth). Whether this is more or less accurate is another question. Classic bias-variance tradeoffs argue that a model with a moderate complexity trained on one data set can better fit future data than overly simple or overly complex models.

We chose four risk groups for purposes of illustration, but this can be increased arbitrarily. It should be noted that the simulation bottleneck when increasing the numbers of risk groups is numerical due the stiffness of the ODEs. This arises because the nonlinearity of infection terms scales with the number of risk groups (e.g., ~ \beta * S * I^3 for 4 risk groups). As such, a careful choice of numerical solvers may be required when integrating the ODEs. Meanwhile, this is not an issue for stochastic, individual based implementation (e.g., Gillespie). As for how well this captures super-spreading, we believe choosing smaller risk groups does not hinder modeling disease spread at large gatherings. Consider a statistical interpretation, where individuals at a large gathering engage in a series of smaller interactions over time (e.g., 2/3/4/etc person conversations). The key determinants of the resulting gathering size distribution at any one large gathering are the number of individuals within some shared proximity over time and the infectiousness/dispersal of the pathogen. Of course, whether this interpretation is a sufficient approximation for classic super-spreading events (e.g., funerals during 2014-2015 West Africa Ebola outbreak) is a matter of debate. Our framework is best interpreted at a population level where the effects of any single gathering are washed out by the overall gathering distribution, P_n. As the prior weakness highlighted, establishing P_n is challenging, but we believe empirically measuring proxies of it may provide future insight in how behavior impacts disease spread. For example, prior work has combined contact tracing and co-location data from connection to WiFi networks to estimate the distribution of contacts per individual, and its degree of overdispersion (Petros et al. Med 2022).

We chose to introduce our framework in a simple SIR context familiar to many readers. This decision does not in any way limit applying it to settings with more population structure. Rather, we believe our framework is easily adaptable and that our presentation (hopefully) makes it clear how to do this. For example, two weakly connected groups could be easily achieved by (for each gathering) first sampling the preferred group and then sampling from the population in a biased manner. The biased sampling could even be a function of gathering sizes, time, etc. The resulting infection terms are still (sums of) multinomials. More generally, the sampling probabilities for an individual of some type need not be its frequency (e.g., S/N, I/N). Indeed, we believe generating models with complex social interactions is both simplified and made more robust by focusing on modeling the generative process of attending gatherings.

The bare bones of the reality without cultural interpretation

Briefing Document : Le Cyberharcèlement au Périscope Source : Excerpts de l'émission "Le cyberharcèlement au périscope" diffusée sur l'IH2EF.

Date d'émission : 2024.

Participants :

• Fabrice Poli : Inspecteur général de l'éducation, du sport et de la recherche, membre de la mission enseignement primaire et du groupe des lettres.
• Séraphin Alava : Professeur émérite en sciences de l'éducation à l'Université de Toulouse 2 Jean Jaurès.
• Frédéric Vedren : Chef d'établissement du collège André Chénier au Bouscat (Académie de Bordeaux).
• Anne Philipson : Directrice territoriale de Canopé Occitanie (en visioconférence).

Thèmes Principaux :

Définition et formes du cyberharcèlement :

• Le cyberharcèlement est défini comme un "acte malveillant répétitif qui est commis par une personne ou par un groupe de personnes de manière électronique dans le but de nuire à cette personne et à l'égard justement d'une personne qui a du mal à se défendre toute seule" (Fabrice Poli).
• Les formes incluent : intimidations, insultes, moqueries, menaces, propagation de rumeurs, usurpation d'identité, focalisation groupée sur une personne, et transmission de photos ou vidéos dévalorisantes ou intimes.

Prise en charge du cyberharcèlement par l'Éducation Nationale :

• Le programme PHARE (programme de lutte contre le harcèlement à l'école) est le dispositif principal. Ses objectifs sont : éduquer pour prévenir, développer les compétences psycho-sociales (empathie), former une communauté protectrice, intervenir pour résoudre les problèmes, et associer parents et communauté éducative.
• Une prise en charge "nourrie" et "extrêmement attentive" est mise en œuvre (Fabrice Poli).

Contexte du monde adolescent et numérique :

• Les adolescents sont immergés dans la culture numérique depuis leur plus jeune âge, avec ses bons usages mais aussi ses pratiques "plus intimes, plus dangereuses, plus violentes" (Séraphin Alava).
• Le rapport à l'image et à la perception par les autres ("peau numérique") est fondamental à l'adolescence. Les actions en ligne, même impulsives, peuvent avoir des répercussions importantes.
• Le monde numérique adolescent est parfois perçu comme un "nom numérique de la chambre", un espace d'intimité où les parents se sentent désarmés. Paradoxalement, ces vies numériques sont souvent publiques, et les jeunes manquent de compétences pour en mesurer les dangers.

Distinction entre agressivité, violence et harcèlement :

• Agressivité : naturelle, liée à la différenciation des groupes.
• Violence : dépasse la simple altercation, devient ciblée et violente.
• Harcèlement : présence évidente d'une cible, de malveillance et d'une volonté de nuire.
• Environ 20% des jeunes se disent témoins de propos dangereux ou violents, mais les formes de harcèlement violent sont estimées autour de 5 pour 1000 (Séraphin Alava). La perception de la violence est plus forte en ligne.

Formes spécifiques de violence dans le cyberharcèlement :

• Violences discriminatoires (racisme, grossophobie) : près de 30% des actes violents (Séraphin Alava).
• Violences de genre : centrées sur la relation sexuelle ou homme-femme.
• Violences de refus du vivre ensemble : liées à des oppositions identitaires (origine, religion).
• Le quotidien : même des disputes entre amis sur les réseaux peuvent devenir très violentes.

Articulation entre harcèlement scolaire et cyberharcèlement :

• Il n'y a pas de cloisonnement étanche. Un acte de harcèlement (physique, verbal) peut être révélateur d'un cyberharcèlement et inversement (Fabrice Poli). Les deux vont souvent de pair.
• Le cyberharcèlement peut venir de l'intérieur ou de l'extérieur de l'établissement et impacter ce qui s'y passe.
• Cadre juridique du cyberharcèlement (Minute Juris) :
• Le cyberharcèlement relève de la catégorie juridique générale du harcèlement, défini dans le Code pénal comme le fait de "faire subir à autrui des propos ou des comportements négatifs voir violent" (Raphaël Mata du Vigot).
• Le harcèlement peut être moral, physique ou sexuel et entraîne une dégradation des conditions de vie de la victime. La répétition n'est pas obligatoire.
• Le cyberharcèlement est le harcèlement exercé via les nouvelles technologies (téléphones, messageries, forums, réseaux sociaux, etc.). Il permet une diffusion massive et répétée de messages humiliants. Exemples : envois répétés de messages insultants, publication sans consentement de contenus intimes, diffusion de rumeurs.
• La loi du 7 octobre 2016 a introduit un article sur le "revenge porn". La loi du 3 août 2018 a renforcé la répression du harcèlement en ligne (article 222-33-2-2 du Code pénal).
• Le harcèlement scolaire est une catégorie juridique spécifique (Loi du 26 juillet 2019 et Loi du 2 mars 2022). L'article L 111-6 du code de l'éducation stipule qu'"aucun élève ou étudiant ne doit subir de fait de harcèlement résultant de propos ou comportement commis au sein de l'établissement d'enseignement ou en marge de la vie scolaire universitaire".
• Le cyberharcèlement scolaire est une faute pénale spécifique (article 222-33-2-3 du Code pénal), s'ajoutant aux sanctions disciplinaires. Les peines peuvent aller jusqu'à 10 ans d'emprisonnement et 150 000 € d'amende en cas de suicide ou tentative de suicide de la victime.
• L'arsenal répressif inclut des peines complémentaires (bannissement des réseaux sociaux), la saisie de matériel, des stages de sensibilisation. Les chefs d'établissement ont l'obligation d'engager des actions disciplinaires. Les personnels ont l'obligation de signaler les faits de harcèlement (article 40 du Code de procédure pénale). Les personnels sont également protégés contre le harcèlement et le cyberharcèlement.

Témoignages et leviers pour lutter contre le cyberharcèlement :

• Anne Philipson (Canopé) : La formation est essentielle pour accompagner les personnels éducatifs et ouvrir le dialogue avec les jeunes. Canopé propose des ressources, webinaires et formations. Une veille active est nécessaire face à l'évolution rapide des technologies et des problématiques (IA, fake news). Impliquer les parents est crucial.
• Frédéric Vedren (Chef d'établissement) : Réagir vite aux signalements (sans précipitation) est primordial pour rassurer la victime et sa famille. Mettre en place des processus clairs (comme le programme PHARE) et professionnaliser la gestion du harcèlement. Former et informer tous les personnels (enseignants, vie scolaire, agents) et les partenaires extérieurs. Agir en classe et individuellement avec les élèves. Les ambassadeurs et la journalisation autour du programme PHARE portent leurs fruits. La question des parents est centrale, souvent axés sur la sanction immédiate alors que l'objectif est la compréhension de l'erreur et la non-répétition.

Acteurs de la lutte contre le cyberharcèlement :

Formateurs (Canopé, Clémi), collectivités, personnels périscolaires, conseillers pédagogiques, inspecteurs, réseau régalien (IPREV), cellules départementales/académiques de prévention du harcèlement, gendarmerie/police (BPDJ), associations, centres sociaux. * Fabrice Poli : Importance de la prévention et de l'éducation dès le plus jeune âge (séances d'empathie dès la maternelle). 10 heures annuelles dédiées à la lutte contre le harcèlement (programme PHARE). Intégration dans l'enseignement moral et civique (respect d'autrui) et l'éducation aux médias et à l'information (Clémi). Le collège est un lieu central de détection. La dépendance des jeunes aux boucles de messagerie est forte, car constitutive de leur appartenance au groupe. * Séraphin Alava : L'alliance parents-enseignants-chefs d'établissement est obligatoire. Agir rapidement mais sans empressement (protéger, instruire, éduquer). Le temps de résolution peut être une angoisse pour les parents, le dialogue est essentiel. Une grande partie du cyberharcèlement se passe en dehors de l'établissement. Le cyberharcèlement progresse avec l'âge d'accès aux portables. Les formes spécifiques (vidéos/images fausses, rumeurs) sont très violentes car persistantes et largement diffusées. Les enseignants en sont aussi victimes.

Stratégies d'approche dans les établissements :

Éducation des élèves et des parents. Travail au sein du CESC (comité d'éducation à la santé, la citoyenneté et l'environnement) pour un programme d'action.

Pédagogie pour apprendre à reconnaître les situations, le rôle du témoin, les sanctions.

Progression et cohérence des actions selon l'âge des élèves. Mobilisation des enseignants en lien avec leur discipline (conseil pédagogique, comité numérique, égalité filles-garçons, EMI).

Acculturation des personnels aux outils numériques. Rôle des cadres dans l'accompagnement et la formation.

Cercles restauratifs comme approche complémentaire.

• Résistances rencontrées : Crainte des enseignants de s'éparpiller, surcharge de travail perçue, sentiment d'incompétence face aux outils numériques et à la gestion émotionnelle des situations et des parents, attente de solutions immédiates parfois irréalistes.

Évolution et impact du cyberharcèlement :

• Le cyberharcèlement pourrait potentiellement descendre en âge avec l'usage plus précoce des réseaux sociaux.
• L'essor de l'IA et la création massive de contenu pourraient complexifier la perception et l'impact du cyberharcèlement (désensibilisation ?), mais l'éducation reste cruciale pour développer un esprit critique. Les jeunes victimes sont souvent dans une phase sensible de leur développement.

Développement de la citoyenneté numérique :

• Nécessité de développer une citoyenneté juvénile et numérique, en s'appuyant sur les valeurs de l'école et de la République. Aider les élèves à maîtriser leur image, leur discours et leur empreinte numérique. Ne pas moraliser mais accompagner. Des outils comme les passeports numériques existent. L'IA représente un nouveau défi à intégrer dans la réflexion.

Temps consacré à l'éducation numérique :

• C'est une préoccupation transdisciplinaire. Des temps spécifiques sont dédiés (ex : 10 heures annuelles dans le programme PHARE), mais chaque heure de cours peut être l'occasion d'aborder le sujet. Des ressources pédagogiques existent pour accompagner les enseignants (programme PHARE, Clémi, kit empathie).

Élèves harcelés devenant harceleurs :

Un élève victime peut devenir agresseur (perception de la loi de la jungle, vengeance, mimétisme). Il est important de prendre en compte cette dynamique pour éviter le surharcèlement et identifier les causes profondes (problèmes familiaux). L'objectif est que l'auteur comprenne son erreur et retrouve une voie pacifique.

• Vulnérabilité des élèves en situation de handicap :

• Ils constituent une proie plus facile pour le harcèlement et nécessitent une attention particulière.

Citation Clé :

• "Le cyberharcèlement c'est un acte malveillant répétitif qui est commis par une personne ou par un groupe de personnes de manière électronique dans le but de de nuire à cette personne et à l'égard justement d'une personne qui a du mal à se défendre toute seule" (Fabrice Poli).

• "Aujourd'hui les jeunes sont dans la culture numérique ils sont totalement immergés dans la culture numérique et dans cette culture numérique ils en on les bons usages et l'éducation nationale fait beaucoup pour les aider à être des citoyens numériques mais ils en on aussi les pratiques plus intimes plus dangereuses plus violentes" (Séraphin Alava).

• "Il faut former les personnels il faut fermer former et informer les parents les élèves" (Frédéric Vedren).

• "Liker c'est harceler et il faut éduquer les les jeunes à tous les gestes qu'ils peuvent avoir parfois parce qu'ils pensent être soucouvert d'anonymat" (Anne Philipson).

Conclusion :

L'émission "Le cyberharcèlement au périscope" met en lumière la complexité et l'omniprésence du cyberharcèlement dans le monde adolescent.

Elle souligne l'engagement de l'Éducation Nationale à travers des programmes comme PHARE, la nécessité d'une approche globale impliquant l'ensemble de la communauté éducative et les parents, et l'importance cruciale de la formation et de la sensibilisation.

Le cadre juridique se renforce pour mieux appréhender et punir ces actes.

Face à l'évolution constante des technologies et des usages, une veille active et une adaptation des stratégies de prévention et d'intervention sont indispensables pour protéger les jeunes et promouvoir une citoyenneté numérique responsable.

Voici un sommaire de la vidéo "Le cyberharcèlement au périscope" avec des indications temporelles approximatives basées sur le déroulement de l'émission :

• [0:00 - 4:20] Introduction et définition du cyberharcèlement et de ses enjeux : Présentation des intervenants. Fabrice Poli donne une définition simple du cyberharcèlement comme un acte malveillant répétitif commis électroniquement dans le but de nuire à une personne ayant du mal à se défendre. Il cite différentes formes de cyberharcèlement : intimidations, insultes, moqueries, menaces, propagation de rumeurs, usurpation d'identité, focalisation groupée, transmission de photos ou vidéos dévalorisantes ou intimes. Il mentionne la prise en charge du cyberharcèlement par l'Éducation nationale, notamment via le programme PHARE axé sur l'éducation, la prévention, la formation d'une communauté protectrice, l'intervention et l'association des parents.

• [4:20 - 9:30] Contexte du monde adolescent et distinction entre agressivité, violence et harcèlement : Séraphin Alava évoque le contexte du monde adolescent, immergé dans la culture numérique avec ses bons et mauvais usages. Il souligne l'importance de la perception et de la présentation de soi pour les adolescents dans leur "peau numérique". Il distingue l'agressivité naturelle, la violence ciblée et l'harcèlement caractérisé par une cible et une malveillance. Il mentionne que 20% des jeunes disent avoir été témoins de propos dangereux, tandis que les formes de harcèlement violent concernent environ 5 pour 1000. Il aborde les formes de violence dans le cyberharcèlement : discriminations (racisme, grossophobie), violences de genre et violences liées au refus du vivre ensemble. Fabrice Poli souligne qu'on ne peut pas cloisonner harcèlement et cyberharcèlement, car ils sont souvent liés.

• [9:30 - 15:30] Cadre juridique du harcèlement et du cyberharcèlement : Présentation de la "minute juris" par Raphaël Mata du Vigot. Il rappelle que le harcèlement n'est pas nouveau mais a pris une dimension numérique. Le cyberharcèlement relève de la catégorie juridique générale du harcèlement, défini dans le code pénal comme le fait de subir des propos ou comportements négatifs ou violents, fondé sur le rejet de la différence. Le harcèlement peut être moral, physique ou sexuel et entraîne une dégradation des conditions de vie de la victime. Contrairement aux idées reçues, la répétition n'est pas obligatoire pour constituer du harcèlement. Le cyberharcèlement s'exerce via les technologies numériques et porte atteinte à la dignité de la victime, créant une situation intimidante ou hostile. Il cite différentes formes de cyberharcèlement (messages humiliants, diffusion de contenus intimes sans consentement, publication d'insultes, divulgation d'informations personnelles). Il mentionne les lois de 2016 et 2018 renforçant la lutte contre le "revenge porn" et le harcèlement en ligne. Il aborde ensuite le harcèlement scolaire, introduit dans le code de l'éducation, et la loi de 2022 visant spécifiquement à combattre le harcèlement scolaire. Le cyberharcèlement est constitutif d'une faute pénale dans le cadre scolaire, avec des sanctions pouvant aller jusqu'à 10 ans d'emprisonnement en cas de suicide de la victime. Des mesures répressives complémentaires existent (bannissement des réseaux sociaux, confiscation de matériel, stages de sensibilisation). Les chefs d'établissement ont l'obligation d'engager une action disciplinaire en cas de harcèlement, et les personnels ont l'obligation de signaler ces faits. Le cyberharcèlement est une forme du harcèlement scolaire et est puni par la loi.

• [15:30 - 22:50] Témoignages et leviers pour lutter contre le cyberharcèlement : Anne Philipson (Canopé) partage la vision des intervenants et souligne le rôle de la formation des personnels éducatifs. Elle insiste sur la nécessité de rester en veille face à l'évolution du cyberharcèlement et à l'impact de l'IA et des "fake news". Elle rappelle le slogan "liker c'est harceler" et l'importance d'éduquer les jeunes à leurs gestes en ligne. Frédéric Vedren (chef d'établissement) insiste sur la nécessité de réagir vite face aux signalements des parents et de mettre en place des process (programme PHARE). Il souligne l'importance de la formation et de l'information de tous les personnels, y compris la vie scolaire. Il évoque les acteurs pouvant intervenir dans la lutte contre le cyberharcèlement : formateurs, Clémi, collectivités, personnels du périscolaire. Il détaille les acteurs au sein de l'établissement (enseignants, direction, vie scolaire, assistants sociaux, infirmières) et les partenaires externes (collectivités, centres sociaux, forces de l'ordre, associations, cellules de prévention du harcèlement). Fabrice Poli met en avant le travail de prévention et d'éducation de l'Éducation nationale, notamment les séances d'empathie dès le premier degré, les 10 heures annuelles dédiées à la lutte contre le harcèlement dans le programme PHARE et l'enseignement moral et civique intégrant l'éducation aux médias et à l'information (ÉMI). Il souligne l'importance du collège comme lieu de détection et le besoin d'appartenance des adolescents aux groupes en ligne. Séraphin Alava confirme la progression du cyberharcèlement dès l'école primaire avec l'accès plus précoce aux téléphones portables. Il insiste sur l'alliance nécessaire entre parents et enseignants, soulignant les principes d'action de l'Éducation nationale (protéger, instruire, éduquer) et le temps nécessaire à la résolution des situations, souvent en dehors de l'établissement. Il évoque un dispositif étranger intégrant des adultes dans les groupes de messagerie des élèves à des fins de prévention. Frédéric Vedren revient sur les stratégies d'abord du cyberharcèlement dans les établissements, insistant sur l'éducation des élèves et des parents plutôt que la simple sanction. Il met en avant le rôle du CESCE et la pédagogie pour apprendre aux élèves à reconnaître les situations de harcèlement et le rôle des témoins. Il détaille les types de sanctions et l'importance de l'acceptabilité par les élèves. Il souligne l'adaptation des actions de prévention aux différents niveaux (6e, 5e, 4e, 3e) et aux problématiques spécifiques (égalité filles-garçons, consentement, réputation en ligne). Il cite des exemples d'intervenants externes (BPDJ, associations) et de partenaires locaux (centres sociaux) pour la sensibilisation et la responsabilisation. Il insiste sur la formation de tous les personnels pour repérer les signaux faibles de mal-être pouvant être liés au harcèlement. Séraphin Alava met en lumière la spécificité des actes de cyberharcèlement avec la production et la diffusion de fausses vidéos et images, amplifiées par l'IA, et le fait que les enseignants en sont aussi victimes. Il nuance l'idée d'une réduction de l'impact du cyberharcèlement avec l'augmentation des contenus, car les victimes sont souvent dans une phase sensible de leur vie. Anne Philipson plaide pour le développement d'une citoyenneté juvénile et numérique, basée sur les valeurs de l'école et de la République, et l'importance d'aider les jeunes à maîtriser leur image et leur empreinte sociale. Elle évoque les passeports numériques et la nécessité de se pencher sur les enjeux de l'IA. Fabrice Poli précise le temps consacré à l'éducation numérique et à la lutte contre le cyberharcèlement : 10 heures annuelles dans le cadre du programme PHARE, de la maternelle à la terminale, intégrées à l'enseignement moral et civique et à l'éducation aux médias et à l'information. Il souligne que ces interventions sont accompagnées de ressources pédagogiques. Il met en garde contre le fait qu'un élève harcelé peut devenir harceleur et inversement, et que l'harcèlement peut être le symptôme de problèmes familiaux. Il insiste sur l'importance de la compréhension et de la responsabilisation de l'agresseur. Il attire l'attention sur la vulnérabilité des élèves en situation de handicap face au harcèlement. Frédéric Vedren rappelle que chaque heure de présence dans l'établissement peut être un moment pour travailler sur le harcèlement. Il évoque les résistances des enseignants (crainte de s'éparpiller, manque de compétences perçu) et l'importance de les accompagner et de les rassurer. Séraphin Alava identifie trois inquiétudes subjectives des enseignants : la gestion de la relation aux parents, le sentiment d'incompétence à la décision et l'inquiétude face au retour de l'agresseur dans l'établissement. Anne Philipson conclut sur l'importance de la communauté éducative, du dialogue entre adultes et de la reconnaissance des besoins en formation. Elle rappelle que le cyberharcèlement est transversal aux disciplines (ÉMI, EMC, égalité filles-garçons, éducation affective et sexuelle). Fabrice Poli synthétise en soulignant la nature polymorphe du cyberharcèlement, touchant à différents aspects de la personne, et l'importance de l'éducation pour distinguer la moquerie de l'harcèlement et faire prendre conscience aux élèves de leurs actes.

• [29:30 - 32:30] Minute bibliographique : Présentation de ressources sur le cyberharcèlement : site Eduscol et kit d'accompagnement pédagogique sur l'empathie, site education.gouv.fr et guide sur la prévention de la cyberviolence, site Réseau Canopé et ressources bibliographiques ainsi qu'un parcours de formation "Cyberharcèlement" sur Magister, site du Clémi et les aventures de la famille Tout Écran, site Internet sans crainte et les fiches conseils de Vin et Lou, ouvrage "Cyber harcèlement sortir de la violence à l'école et sur les écrans" de Béranger Stassin et son blog, jeu sérieux "Le quartier des légendes" de Séraphin Alava et l'association militant des savoirs.

• [32:30 - Fin] Conclusion de l'émission.

Maven 运行某一个 phase 会把之前的 phase 全部运行。

和测试相关的两个 phase 是 test 和 verify，一个是单元测试，一个是集成测试。

Briefing Doc : Prévention et gestion des conflits au périscope

Source : Excerpts de l'émission "Prévenir et gérer les conflits au périscope" (Transcription textuelle)

Date de diffusion (implicite) : Avant février 2025

Thématique principale : La prévention et la gestion des conflits dans divers environnements, notamment l'éducation nationale et le secteur de la santé.

Participants :

• Animation : (Nom non mentionné explicitement, mais présente le studio Joseph et l'émission "Au Périscope")
• Leticia Chardavoine : Chef d'établissement au lycée général et technologique Joseph des Fontaines à Melle (79).
• Michel Keré : Inspecteur général de l'éducation, du sport et de la recherche.
• Stéphane Maré : Directeur de la communication et du mécénat du CHU de Poitiers et médiateur diplômé.
• Jean Pralon : Enseignant-chercheur en gestion des ressources humaines à l'École de Management de Normandie (en distanciel).
• Raphaël Mata du Vigot : (Présentateur de la "minute juris").
• Sylvain Paul : (Propose le "point ressources" ou "minute biblie").

Structure de l'émission :

• Introduction et enjeux : Présentation des invités et introduction de la thématique par Michel Keré et Jean Pralon.
• Minute Juris : Présentation des aspects juridiques de la gestion des conflits, notamment les modes alternatifs de règlement des différends (MARD) dans l'administration.
• Débat : Discussion entre les quatre invités sur la prévention et la gestion des conflits, avec des exemples concrets et des perspectives variées.
• Point Ressources (Minute Biblie) : Présentation de ressources documentaires et d'ouvrages pertinents sur la thématique.

I. Enjeux de la prévention et de la gestion des conflits (Introduction)

• Universalité du conflit : Michel Keré souligne que le conflit est une caractéristique universelle de l'activité humaine, présente même dans les sociétés animales (violence ou évitement). Cependant, dans les sociétés humaines civilisées, le conflit est organisé par des règles (lois, acceptation des comportements).
• "Cette question, cette thématique est effectivement universelle, elle existe depuis toujours, euh c'est le propre de l'activité humaine, le conflit n'ayons pas peur de le dire parce que si on fait un peu j'allais même dire si on fait de l'anthropologie de bazar on pourrait dire que le conflit c'est y compris le propre des sociétés animales où le conflit se réglait de deux manières hein soit par la violence le combat ou par l'évitement on allait dans une autre tribu quand on était euh animal le cas échéant ou entre espèces animalière d'ailleurs aussi."
• Le conflit comme interaction humaine : La parole et l'échange distinguent les conflits humains, offrant des outils d'argumentation et de résolution.
• "La différence avec les sociétés animales d'ailleurs c'est nous avons la parole et nous avons du coup l'argumentation et l'échange et la parole et l'échanges sont le propre de l'interaction de l'activité humaine et donc de caractère universel du conflit par définition."
• Organisation et canalisation du conflit dans la société : Les règles collectives (lois) lissent et diminuent le conflit. Des institutions comme l'entreprise, l'association, l'éducation nationale et l'hôpital sont des "creusets de sociétés" où les conflits se manifestent. L'éducation nationale a un rôle particulier dans la formation aux valeurs qui font société et canalisent les conflits.
• "Dans l'état de société que nous formons le conflit est organisé c'est-à-dire que il y a des règles comme point de repère, il y a une acceptation des comportements de l'autre dans une certaine mesure et qui l'interaction individuelle est régie finalement par des règles collectives qui lissent qui diminuent nécessairement le conflit qu'il s'agisse de la loi..."
• Définition du conflit : Nécessite un émetteur, un récepteur, une interaction (verbale, écrite) et une argumentation non partagée, menant à des frictions et potentiellement à une gradation (échanges d'opinion, tensions interpersonnelles/professionnelles, conflits explicites, crise).
• Déterminants du conflit : Manque d'écoute, argumentations excessives, enjeux de négociation.
• Dualité du conflit : Peut être salutaire et positif (source d'innovation et de convergence de vues) mais aussi négatif et destructeur (rupture de communication, situations frictionnelles structurelles).
• Perspective du management (Jean Pralon) :Les conflits basés sur les goûts personnels sont moins importants. Dans les organisations, les conflits prennent des formes institutionnelles liées à la distribution du pouvoir.
• "Si le conflit porte sur des goûts des préférences des intérêts personnels c'est pas très grave au fond... en revanche ce qui serait important c'est de considérer que dans les les organisations dans une institution le conflit prend des formes un peu différentes et que pour ça il faut essayer de de décoder les choses d'une façon un peu plus institutionnelle."
• Les organisations distribuent du pouvoir, et les conflits sont souvent l'expression d'un conflit de pouvoir entre ceux qui tentent de l'exercer et ceux qui n'ont pas envie de le subir.
• Il faut se méfier de l'interprétation qui voit toute résistance ou opposition comme une résistance au changement basée sur des intérêts personnels, car elle peut ignorer les questions de légitimité des contraintes imposées.
• "Très souvent la question du conflit il faut la penser dans un contexte qui va être avant tout un contexte de distribution du pouvoir... c'est l'expression d'un conflit de pouvoir entre des gens sur lesquels on essaie d'exercer du pouvoir et qui n'ont pas envie qu'on les qu'on les manipule et qu'on les maltraite..."
• Le conflit peut également être lié à des conflits de statut (ex: chef d'établissement exprimant son rôle).

II. Gestion des conflits : Aspects juridiques (Minute Juris)

• Évolution de la gestion des conflits : Du duel judiciaire à un règlement plus procédural et respectueux du droit.
• Définition juridique du conflit : Une décision administrative contestée par son destinataire.
• Modes Alternatifs de Règlement des Différends (MARD) : Existent en dehors du recours au juge, permettant un règlement plus rapide, souple, accessible et personnalisé.
• Types de MARD :Recours administratif (gracieux ou hiérarchique) : Mode classique régi par le Code de justice administrative.
• Arbitrage : Un tiers tranche la contestation.
• Conciliation : Un tiers rapproche les parties et fait des propositions.
• Médiation : Un tiers restaure la communication pour que les parties trouvent elles-mêmes une solution. L'accord peut prendre la forme d'une transaction (concessions réciproques) et parfois nécessiter une homologation judiciaire.
• Encouragement des MARD par les pouvoirs publics : Raisons sociales (règlement pacifique, redonner le pouvoir aux citoyens) et économiques (désengorger les juridictions).
• Focus sur la médiation administrative :Médiateur de l'éducation nationale et de l'enseignement supérieur (depuis 1999) : Nommé par arrêté ministériel.
• Médiateurs académiques : Institués dans chaque ressort académique, coordonnés par le médiateur national.
• Compétences : Le médiateur national traite les réclamations relatives aux services centraux et aux établissements non rattachés à un recteur ; les médiateurs académiques sont compétents pour les services et établissements de leur circonscription.
• Statut : Organe intégré au ministère, mais agissant de manière impartiale et neutre (adhère à la charte du club des médiateurs du service public).
• Mission : Médiation et traitement extrajudiciaire des réclamations (questions financières, affectation, carrière, statut, recrutement, retraite, protection sociale pour les personnels ; inscription, orientation, vie scolaire/universitaire, examens, questions sociales, inclusion pour les usagers).
• Procédure de saisine : La réclamation doit être précédée de démarches auprès de l'établissement concerné.
• Pouvoirs d'investigation du médiateur : Peut faire appel aux services du ministère et aux inspections générales.
• Issues de l'intervention du médiateur : Refus d'appuyer la réclamation ou appui avec recommandations (non contraignantes) à l'établissement ou au service concerné.
• Articulation avec le Défenseur des droits : La saisine du Défenseur des droits met fin à la procédure devant le médiateur de l'Éducation si la réclamation entre dans sa compétence.
• Médiation préalable obligatoire (MPO) : Expérimentée puis pérennisée par la loi pour la confiance dans l'institution judiciaire.
• Principes fixés par le Code de justice administrative : Articles R213-10 et suivants.
• Déclenchement : Engagée auprès du médiateur compétent dans les délais du recours contentieux (2 mois suivant la notification de la décision contestée, qui doit mentionner l'obligation de saisir le médiateur et ses coordonnées).
• Effets de la saisine : Interruption du délai de recours contentieux et suspension des délais de prescription.
• Champ d'application : Recours formés par les agents publics à l'encontre de certaines décisions administratives individuelles défavorables (éléments de rémunération, refus de détachement/disponibilité/congé parental, réemploi d'un contractuel, refus de promotion interne/accès à la formation).
• Développement de l'éducation aux MARD : Nécessaire pour leur progression, inscrit dans l'objectif de l'article L11-1 du Code de l'éducation.
• Dispositifs existants : Enseignement moral et civique, médiation par les pairs dans certains établissements.
• Limites : La généralisation des MARD est encore un chemin long.

III. Débat : Prévention et gestion des conflits (Points clés)

• Prévention des conflits (Leticia Chardavoine) :Poser un cadre de confiance et de respect : Cadre de communication entre tous les personnels, respect de la diversité, reconnaissance de l'humain avant le professionnel, prise en compte des parcours et expériences différents.
• "Pour moi le ça passe par euh le fait de poser un cadre euh de confiance et de respect c'est-à-dire un cadre de communication entre les personnels entre la direction et les personnels les personnels la direction et entre l'ensemble des personnels qui soit vraiment euh respectueux de la diversité des personnes qui constituent ses personnels."
• Modélisation par le chef d'établissement : Être soi-même respectueux et garder son calme pour être légitime à exiger cela des autres.
• "Ce cadre je vais moi-même m'astreindre à le respecter c'est-à-dire qu'à partir du moment où j'explique à mes personnels que on peut tous dire mais qu'il y a la manière de le dire ça sous-entend que que moi à aucun moment je ne peux me permettre de perdre mon calme envers un personnel..."
• Développement des "soft skills" : Écoute active, empathie, assertivité, communication claire et précise (orale et écrite), formulations respectueuses, donner du sens aux demandes.
• "Le deuxième point il est très en lien avec ce qu'on appelle les soft skill c'est-à-dire les les compétences douces avec la notion d'écoute active d'empathie d'assertivité tout ce qui tourne autour de de la communication..."
• Cohérence entre le dire et le faire : Expliciter les actions et les décisions, respecter les engagements pour éviter l'incertitude et le terrain propice aux conflits.
• "L'expression dire ce qu'on fait et faire ce qu'on dit c'est-à-dire que les personnels aujourd'hui ne fonctionnent plus à l'aveugle... on veut savoir pourquoi on fonctionne dans quel but et quand une décision est prise euh savoir qu'elle est prise et savoir pourquoi elle est prise et dans quel objectif à quelle durée..."
• Attention au poids du non-verbal : Être conscient de sa communication non verbale et de celle des autres, veiller à la cohérence avec le verbal dans une attitude de respect.
• "Je leur rappelle que dans une communication euh orale il n'y a que 30 % de la communication qui va être verbal et 70 % qui va être non verbal..."
• Valorisation des instances : Les instances (conseil d'administration, etc.) sont des lieux de prévention en permettant l'expression des divergences d'opinion et en régulant les conflits potentiels. Le conflit d'idées est vu comme fondamental pour l'évolution.
• "Les instances pour moi c'est le lieu euh de prévention des conflits dans le sens où très souvent les conflits ils apparaissent parce qu'ils ne sont pas exprimés..."
• Être attentif aux signaux faibles : Écouter les remarques, les tensions, les incompréhensions pour réagir rapidement avant l'escalade.
• "Le dernier point je dirais sur la prévention du conflit c'est d'être très attentif aux signaux faibles..."
• Analogie entre le secteur de la santé et l'éducation (Stéphane Maré) : Les deux environnements impliquent la prise en charge d'humains par des humains, créant des contextes où les conflits peuvent survenir. La médiation est un outil développé plus récemment dans la santé pour la prévention et la gestion des conflits entre agents.
• Importance du contexte (Michel Keré) : Les solutions standardisées pour gérer les conflits sont limitées. Il faut tenir compte du contexte spatial et temporel. L'écoute empathique et le respect partagé sont fondamentaux. L'exemplarité et la conviction du responsable sont essentielles.
• "Je suis un peu frappé par la multiplicité des solutions standardisées qu'on vous propose les 60 recettes pour gérer les conflits les 30 conseils pour et cetera et et moi je je suis pas là c'est-à-dire que je trouve que la manière dont Madame Chardavoine l'a amené est tout à fait juste c'est-à-dire que le le conflit qu'on essaie de le prévenir ou de le gérer il est contextualisé à la fois dans l'espace et dans le temps..."
• La capacité à convaincre est une garantie de la prévention et de la gestion des conflits. L'argument d'autorité est un dernier recours moins efficace que le pouvoir de conviction.
• Gestion des conflits par la médiation (Stéphane Maré) :La médiation est un outil structuré proposé au personnel (parmi d'autres MARD comme la conciliation).
• Vertus de la médiation : Permettre aux personnes en difficulté de communication de se retrouver, de s'exprimer, de s'écouter et de se comprendre ; permettre aux intéressés de trouver eux-mêmes des solutions durables à leurs conflits.
• Rôle du médiateur : Neutre, impartial, indépendant, accompagnateur (pas un arbitre).
• Processus de médiation : Entretiens individuels préalables (expliquer le processus, s'assurer de l'engagement volontaire), séances plénières (réunion des acteurs, durée de 2-3 heures, plusieurs séances possibles). Durée moyenne d'une médiation : 3 mois.
• Confidentialité : Garantie de sécurité pour l'expression des participants.
• Médiation de projet : Application du processus de médiation pour accompagner des équipes dans la conduite de projets, anticiper les réticences au changement et faciliter la communication.
• Nombre de participants : Peut concerner deux personnes ou des groupes importants (jusqu'à 40 avec plusieurs médiateurs).
• Taux de réussite : Environ 70% (évalué par la concrétisation d'un accord écrit). Le simple fait de reprendre le dialogue peut être considéré comme un succès.
• Gestion des conflits par le chef d'établissement (Leticia Chardavoine) :Objectif : Revenir à une situation sans conflit et sans perdant (éviter la rancœur et la reprise du conflit).
• Différer la gestion : Laisser le temps pour que les émotions s'apaisent.
• Recueillir le ressenti individuel : Permettre à chaque partie d'exprimer son point de vue et ses émotions.
• Entretiens individuels et réunions spécifiques : Organiser des moments pour la verbalisation des ressentis et des opinions divergentes.
• Accompagnement à la formulation : Aider à trouver des expressions respectueuses, distinguer les faits des ressentis, identifier les problèmes de communication et les mauvaises interprétations.
• Laisser la solution émerger des parties : Le chef d'établissement n'est pas un juge, mais un facilitateur pour trouver un compromis ou des excuses.
• Retour en différé : Vérifier si la situation est complètement résolue et prendre la température pour éviter une reprise du conflit.
• Approche systémique du conflit (Stéphane Maré) : Le conflit est un système où chacun se positionne différemment. Il est important de considérer les besoins et d'aider les personnes à sortir de ce système. L'aspect émotionnel est crucial.
• Rôle des enquêtes administratives (Michel Keré) : L'Inspection Générale peut mener des enquêtes en cas de dysfonctionnement organisationnel, offrant un regard externe et une analyse de la gestion des conflits. Cela permet de libérer la parole et d'apaiser l'organisation.
• Principes de la gestion des conflits (Jean Pralon) :Distinguer l'accord sur les idées du conflit de position : On peut être d'accord avec quelqu'un mais le conflit peut venir d'une question de position (statut, intérêts personnels).
• Ne pas hésiter à ne pas négocier : Lorsque des directives institutionnelles s'imposent, la marge de négociation peut être limitée.
• Faire sens aujourd'hui : Ne plus seulement rappeler les grands principes, mais aussi faire appel à des justifications alternatives et variées qui trouvent un écho dans les intérêts personnels des individus, tout en respectant les grands principes.
• Décoder les intérêts personnels et les positions : La lecture du conflit doit se faire au niveau où il se pose (rarement des conflits de personne, souvent des conflits de statut et de position).
• L'assertivité : Règle de comportement impliquant de ne ni agresser, ni se soumettre, ni manipuler dans sa communication. Assumer son rôle et son statut, expliquer les choses en tenant compte des légitimités institutionnelles et des besoins des personnels.
• Formation à l'assertivité : Possible pour améliorer sa communication.
• Importance de la culture professionnelle (Michel Keré) : L'adhésion, la volonté d'avancer sur des projets partagés, la conviction de l'utilité de son action dans le service public (éducation et santé) facilitent la gestion et la prévention des conflits. Le partage de valeurs et de convictions est une force.

IV. Point Ressources (Minute Biblie)

• Ressources institutionnelles :Site education.gouv.fr : Informations sur le médiateur de l'éducation nationale et de l'enseignement supérieur, carte des médiateurs académiques.
• Actes du colloque des 20 ans de la médiation de l'éducation nationale (2018) : "La médiation pour une société de la confiance".
• Rapports annuels de la médiation de l'éducation nationale :
• 2019 : "Prendre soin, une autre voie pour prévenir les conflits".
• 2023 : "Faire alliance, redonner confiance".
• Ouvrages suggérés :"Les modes alternatifs de règlement des conflits" (3e édition, 2019) de Loïc Cadiet et Thomas Clay (Dalloz).
• "La médiation" (Nouvelle édition, 2015) de Jacques Faget (Érès).
• "La médiation pour tous en France : Comment gérer relations et conflits" (2022) (L'Harmattan).
• Définition du mot "conflit" dans "Les 100 mots de la sociologie" (2022) de Sandrine Rui.
• "Les mots sont des fenêtres ou bien ce sont des murs : Initiation à la communication non violente" (2016) de Marshall B. Rosenberg (La Découverte).
• "Chevaucher son tigre ou comment résoudre des problèmes compliqués avec des solutions simples" (2008) de Giorgio Nardone (Seuil).

Conclusion (implicite) :

L'émission met en lumière la complexité et la diversité des approches pour prévenir et gérer les conflits.

Que ce soit par la mise en place d'un cadre relationnel basé sur le respect et la communication (dans les établissements scolaires), par l'utilisation d'outils structurés comme la médiation (dans les secteurs de la santé et de l'éducation), par une analyse managériale des enjeux de pouvoir et de statut, ou par le recours à des dispositifs juridiques alternatifs, la gestion des conflits nécessite une compréhension des dynamiques humaines et organisationnelles, ainsi qu'une adaptation des méthodes au contexte spécifique.

La valorisation du dialogue, de l'écoute, et de la recherche de solutions mutuellement acceptables apparaît comme essentielle pour construire des environnements de travail plus sereins et productifs.

chronologie détaillée, la liste des personnages et leurs bios, basés sur les informations contenues dans la source fournie :

Chronologie des principaux événements liés à l'école inclusive en France

• Début des années 1900 : Premières recherches pour la création de classes spécialisées pour les enfants handicapés, perçues rétrospectivement comme une forme d'exclusion et une école de la séparation.
• 1975 : Étape importante avec la construction de l'intégration des enfants en situation de handicap dans le milieu ordinaire. Une loi de 1975 fait évoluer (légèrement) l'approche.
• 2005 : Loi marquant un tournant véritable dans la mise en place d'une école inclusive.
• Changement de la notion de handicap : il n'est plus uniquement lié à la personne mais aussi au contexte.
• Transfert de la gestion de l'éducation des enfants handicapés de l'Éducation Nationale aux Maisons Départementales des Personnes Handicapées (MDPH).
• Émergence de trois droits fondamentaux :
• Le droit de vivre parmi les autres.
• Le droit de participer sans exclusion à la vie collective.
• Le droit de décider de sa vie et de choisir son projet de vie.
• 1994 (International) : Déclaration de Salamanque, où la France et 90 autres pays réfléchissent à la question des besoins éducatifs spéciaux, prônant la construction d'une école pour tous, au-delà de la seule question du handicap.
• Début des années 2000 (International) : L'UNESCO publie des principes directeurs en faveur de l'inclusion, insistant sur le fait que l'inclusion concerne toutes et tous et définit l'inclusion comme un processus en construction.
• 2010 (International) : La France ratifie la Convention relative aux droits des personnes handicapées, qui pose le principe d'accessibilité (physique, services, communication) et engage la France à faire évoluer ses lois.
• 2013 : Loi de refondation pour l'école qui inscrit la notion d'école inclusive dans le code de l'éducation et établit la nécessité de la réaliser (obligation de résultat).
• 8 juillet 2013 : La loi intègre à l'article L 111-1 du code de l'éducation la disposition selon laquelle le service public de l'éducation veille à l'inclusion scolaire de tous les enfants sans aucune distinction.
• 2017 : Rapport de l'ONU sur la mise en œuvre de la Convention relative aux droits des personnes handicapées pointe le fait que la France doit passer d'une approche individuelle compensatrice à une approche collective de mise en accessibilité et enjoint la France de fermer ses établissements spécialisés.
• 26 juillet 2019 : Loi pour une école de la confiance qui confirme le principe de l'école inclusive et précise de nombreux points :
• Réaffirmation du droit à l'école inclusive (une scolarité pour tous).
• Institution des Pôles Inclusifs d'Accompagnement Localisés (PIAL) pour un accompagnement au plus près des enfants en situation de handicap.
• Favorisation de la coopération avec le médico-social.
• Création des Services Départementaux de l'École Inclusive (SDUI, mentionnés comme ayant succédé aux CDSEI).
• 29 septembre 2021 : Circulaire Blanquer consacrant des développements fondamentaux au droit à l'éducation des élèves transgenres, notamment concernant l'utilisation du prénom d'usage.
• 2022 : Loi visant à combattre le harcèlement scolaire, inscrivant le droit à une scolarité sans harcèlement à l'article L111-6 du code de l'éducation et prévoyant un délit spécifique de harcèlement scolaire.
• 2023 : Conférence Nationale du Handicap (CNH 2023) avec 10 engagements, dont le premier est celui de l'école pour tous, marquant l'engagement dans l'acte 2 de l'école inclusive.

Cast of Characters et Bios

• Yannick Ten : Inspecteur général de l'éducation, du sport et de la recherche. Il intervient dans l'émission pour donner un propos introductif et historique sur la thématique de l'école inclusive en France.
• Virginie Legal : Principale adjointe au collège Trémolière à Cholet. Elle apporte le témoignage du terrain, expliquant comment l'école inclusive est mise en œuvre au sein de son établissement, notamment en termes de pilotage, de collaboration, d'adaptation et de formation.
• Stéphane Bertrou : Conseiller technique école inclusive pour la rectrice de l'académie de Nantes. Il présente les orientations nationales et leur mise en œuvre au niveau local (l'académie de Nantes), en abordant les aspects de pilotage, de collaboration avec le médico-social, de maillage territorial et de formation des personnels.
• Frédéric Dupré : Maître de conférences à l'INSEI (Institut national supérieur de recherche et de formation pour l'éducation inclusive). Il apporte une perspective internationale sur le mouvement inclusif, évoquant la Déclaration de Salamanque et la Convention relative aux droits des personnes handicapées, et analyse les enjeux et les points de blocage en France, notamment le passage d'une approche compensatrice à une approche privilégiant l'accessibilité.
• Raphaël Mata Duvgot : Non présent physiquement mais sa contribution est la "minute juris". Il est l'auteur de cette séquence qui décrypte les aspects juridiques de l'inclusion scolaire, abordant notamment le droit à l'éducation des élèves en situation de handicap et les questions de genre (élèves transgenres).

Briefing Document : Analyse des enjeux et perspectives de l'école inclusive en France

Ce document de briefing synthétise les principaux thèmes, idées et faits saillants issus de l'émission "L'école inclusive au périscope".

L'émission a exploré l'évolution, les enjeux, les défis et les perspectives de l'école inclusive en France, en s'appuyant sur l'expertise d'acteurs clés du secteur.

Thèmes Principaux et Idées Clés :

1. Évolution Historique de l'École Inclusive en France : Un Long Cheminement

• L'idée d'une école inclusive est le fruit d'une longue évolution historique en France.
• Début du 20ème siècle : Création de classes spécialisées, perçue rétrospectivement comme une forme d'exclusion des enfants handicapés ("une école de la séparation").
• 1975 : Étape de l'intégration des enfants en situation de handicap dans le milieu ordinaire avec une loi importante. Cependant, cette loi n'a qu'un impact limité initialement.
• 2005 : Loi marquant un tournant fondamental. Deux changements majeurs sont introduits :
• La notion de handicap évolue : elle n'est plus uniquement liée à la personne mais aussi au contexte.
• La gestion de l'accompagnement est confiée à la Maison Départementale des Personnes Handicapées (MDPH), auparavant relevant uniquement de l'Éducation Nationale.
• Ces évolutions ont fait émerger trois droits fondamentaux :
• Le droit de vivre parmi les autres.
• Le droit de participer sans exclusion à la vie collective.
• Le droit de décider de sa vie et de choisir son projet de vie.
• Confirmations législatives :2013 (Loi de refondation pour l'école) : Inscrit la notion d'école inclusive dans le code de l'éducation et établit la nécessité de la réaliser (obligation de résultat).
• 2019 (Loi pour une école de la confiance) : Réaffirme le principe de l'école inclusive, précise de nombreux points, notamment :
• Le droit à une scolarité pour tous les enfants.
• L'institution des Pôles Inclusifs d'Accompagnement Localisés (PIAL) pour un accompagnement au plus près des enfants.
• La favorisation de la coopération avec le médico-social.
• La création d'un service départemental de l'école inclusive.

2. L'École Inclusive dans un Cadre International : Un Mouvement Global

• Le mouvement inclusif n'est pas spécifique à la France et s'inscrit dans un cadre international depuis une trentaine d'années.
• 1994 : Déclaration de Salamanque. La France, avec de nombreux autres pays, réfléchit à la question des besoins éducatifs spéciaux et émerge l'idée de construire une "école pour tous", allant au-delà de la seule question du handicap pour inclure tous les enfants à risque d'exclusion.
• Début des années 2000 : UNESCO. Publication de principes directeurs en faveur de l'inclusion, insistant sur l'idée de processus et s'adressant à tous les élèves, pas seulement ceux en situation de handicap.
• 2010 : Convention relative aux droits des personnes handicapées (ratifiée par la France). Pose le principe d'accessibilité (pas seulement physique mais aussi aux services et systèmes de communication), engageant la France à rendre son école accessible. Cette convention a une valeur juridique supérieure aux lois nationales.

3. Bilan et Points de Blocage Depuis 2005 : Entre Avancées et Défis Persistants

• Évolutions Positives :Augmentation significative du nombre d'enfants en situation de handicap scolarisés en milieu ordinaire (environ 500 000, dont près de 90% en classes ordinaires).
• Amélioration de l'accueil avec des structures médico-sociales et des classes spécialisées plus ouvertes (passage de l'intégration à l'inclusion).
• Développement de l'accompagnement humain, qui a permis l'accueil de nombreux enfants.
• Points de Blocage et Défis :Accompagnement Humain : Développement exponentiel parfois peu maîtrisé (statuts, salaires, conditions de travail).
• Déséquilibre Accessibilité/Compensation : La France a peut-être trop mis l'accent sur la compensation (accompagnement humain) au détriment de l'accessibilité de l'environnement scolaire.
• Relation Médico-Social/École : Questionnement et retard de la France selon l'ONU quant au rapprochement entre ces deux secteurs.
• Relation MDPH/Éducation Nationale : Nécessité d'amélioration et fortes disparités territoriales en termes de délais et d'accueil.

4. Engagements Internationaux et Réalités Nationales : Un Décalage Persistant

• Bien qu'il y ait des évolutions quantitatives (nombre d'élèves scolarisés, modalités), la France peine à répondre pleinement à ses engagements internationaux, notamment en matière d'accessibilité et de réduction des exclusions.
• Rapport de l'ONU (2017) : Pointe la nécessité pour la France de passer d'une approche individuelle compensatrice à une approche collective de mise en accessibilité du système éducatif et enjoint la fermeture des établissements spécialisés.
• D'autres rapports d'inspections générales vont dans le même sens.
• La construction de l'école inclusive est un "processus transformatif" visant à faire évoluer le système pour accueillir a priori tous les élèves et leur permettre de prendre leur place.

5. La Minute Juris : Cadre Législatif et Évolutions du Droit

• Principe d'inclusion scolaire (devenu scolarisation inclusive) : L'école doit s'adapter à l'élève, et non l'inverse. Ce droit à l'éducation a une valeur constitutionnelle et est exigible de la collectivité.
• Obligation de l'État : Assurer une formation scolaire aux enfants présentant un handicap, avec les moyens financiers et humains nécessaires à la scolarisation en milieu ordinaire. Jusqu'en 1975, l'obligation scolaire n'était pas effective pour ces enfants.
• Responsabilité de l'État : La justice administrative reconnaît l'obligation légale de l'État d'offrir une prise en charge éducative au moins équivalente à celle du milieu ordinaire, adaptée aux besoins spécifiques. Le défaut de scolarisation d'un enfant handicapé engage la responsabilité de l'État.
• Scolarisation en établissement de référence : Principe de proximité, l'orientation en établissement spécialisé ne se fait qu'en cas de besoin et avec l'accord des parents.
• Collaboration Médico-Sociale : Institutionnalisée par décret, avec des conventions précisant les modalités d'intervention.
• Droit Créance des AESH : Reconnu par le Conseil d'État, la privation de scolarisation adaptée constituant une atteinte grave à une liberté fondamentale.
• Questions de Genre (Élèves Trans) : Formalisation progressive du droit à l'éducation, axée sur :
• Protection contre le harcèlement : Inclus dans les dispositifs généraux de lutte contre le harcèlement scolaire (lois de 2013 et 2022). Le harcèlement moral est reconnu comme atteinte à une liberté fondamentale.
• Reconnaissance du droit : Respect du prénom d'usage choisi par l'élève (avec accord parental pour les mineurs) dans les documents internes. Aménagements pratiques possibles concernant les espaces d'intimité (simple possibilité, pas obligation de résultat).
• Si la consécration juridique de l'inclusion ne fait aucun doute, sa pleine effectivité demeure un idéal.

6. Témoignages de Terrain : Mise en Œuvre et Défis Concrets (Académie de Nantes)

• Pilotage : Intégration de l'école inclusive dans le projet d'établissement, coordination des moyens, sensibilisation dès la rentrée, utilisation d'outils de communication (Pronote), adaptation aux dispositifs (PIAL), évaluations régulières. Rôle central du chef d'établissement.
• Collaboration/Coopération/Coordination : Travail avec le médico-social (CESSAD), les familles (rendez-vous, entretiens), intégration des AESH à l'équipe éducative (participation aux ESS, conseils de classe, réunions). Rapprochement Rectorat/ARS au niveau académique (stratégies communes, unités d'enseignement externalisées, équipes mobiles d'appui à la scolarisation). Nécessité d'apprendre à travailler en intermétiers.
• Maillage Territorial et Dispositifs d'Accompagnement : Complexité des niveaux d'intervention et diversité des territoires. Enjeu de rendre l'environnement accessible pour limiter le recours à la compensation. Expérimentation des pôles d'appui à la scolarité. Importance de la coopération avec les familles et de la lisibilité des dispositifs. Intégration des PIAL dans les établissements. Travail en réseau avec différents services.
• Formation et Sensibilisation : Nombreuses formations déployées (AESH, enseignants, cadres). Nécessité de mieux former pour répondre à la diversité des besoins. Accent sur l'accessibilité plutôt que sur la connaissance exhaustive des troubles. Importance de la formation initiale et continue, ainsi que de la sensibilisation des personnels de service.

7. Conclusion et Perspectives

• Points Positifs : Engagement des personnels, augmentation du nombre d'enfants accueillis, engagement localisé.
• Défis Majeurs : Augmentation continue du nombre d'enfants en situation de handicap à accueillir, nécessité d'améliorer la communication avec les familles (parcours perçu comme un "parcours du combattant").
• Enjeux d'Accessibilité : Passer d'une logique de compensation individuelle à une logique d'accessibilité collective de l'environnement scolaire et pédagogique. Les mesures compensatoires peuvent parfois être sources d'obstacles supplémentaires.
• Transformation Nécessaire : L'école inclusive nécessite une transformation réelle de l'école, impliquant une réflexion à tous les niveaux du système éducatif (politiques publiques, pilotage, établissement, pratiques pédagogiques) et une action conjointe des différents acteurs. Reconnaissance des "petits pas" quotidiens.
• Ce briefing met en lumière la complexité et la richesse des débats autour de l'école inclusive en France, soulignant à la fois les progrès accomplis et les défis importants qui restent à relever pour garantir une éducation de qualité et véritablement inclusive pour tous les élèves.

Voici un sommaire de la vidéo "L'école inclusive au périscope" avec des estimations de timestamps basées sur le déroulement de l'émission :

• [0:00 - 1:15] Introduction de l'émission et des intervenants

  • Présentation de l'émission "Au Périscope" consacrée à l'école inclusive.
  • Introduction des quatre intervenants : Yannick Ten (Inspecteur général de l'éducation, du sport et de la recherche), Virginie Legal (Principale adjointe au collège Trémolière à Cholet), Stéphane Bertrou (Conseiller technique école inclusive pour la rectrice de l'académie de Nantes), et Frédéric Dupré (Maître de conférence à l'INSEI).

• [1:15 - 2:00] Introduction à la thématique de l'école inclusive

  • Annonce d'une première partie consacrée aux enjeux et d'une seconde partie aux témoignages de terrain.

• [2:00 - 4:25] Retour historique sur la thématique de l'école inclusive (Yannick Ten)

  • Début des années 1900 : Création de classes spécialisées perçue comme de l'exclusion.
  • 1975 : Étape de l'intégration des enfants en situation de handicap en milieu ordinaire (loi de 75).
  • 2005 : Loi posant véritablement les principes d'une école inclusive, avec deux changements majeurs : la notion de handicap liée au contexte et la gestion confiée à la Maison départementale des personnes handicapées (MDPH).
  • Émergence de trois droits : vivre parmi les autres, participer sans exclusion, et décider de sa vie.
  • 2013 (Loi de refondation pour l'école) et 2019 (Loi pour une école de la confiance) : Confirmation et précision du principe de l'école inclusive, obligation de résultat, institution des pôles inclusifs d'accompagnement localisés, coopération avec le médico-social, et création du service départemental de l'école inclusive.

• [4:25 - 6:50] Vision internationale de l'école inclusive (Frédéric Dupré)

  • Mouvement inclusif global depuis une trentaine d'années.
  • 1994 : Déclaration de Salamanque sur les besoins éducatifs spéciaux et l'idée d'une école pour tous.
  • Début des années 2000 : Principes directeurs de l'UNESCO en faveur de l'inclusion.
  • 2010 : Ratification par la France de la Convention relative aux droits des personnes handicapées, principe d'accessibilité (non uniquement physique) et valeur juridique supérieure aux lois nationales.

• [6:50 - 10:00] Évolutions depuis 2005 et points de blocage (Yannick Ten & Frédéric Dupré)

  • Évolutions positives : augmentation du nombre d'enfants en situation de handicap scolarisés en milieu ordinaire (environ 500 000, majoritairement à 90% en classes ordinaires), accueil amélioré avec des structures médico-sociales et des classes spécialisées plus ouvertes.
  • Point de blocage : l'accompagnement humain (développement exponentiel mais parfois non maîtrisé, questions de statuts, salaires, conditions de travail).
  • Déséquilibre entre accessibilité et compensation (la France n'a pas assez développé l'accessibilité).
  • Relation avec le médico-social qui pose question (retard de la France selon l'ONU).
  • Relation MDPH et Éducation Nationale à améliorer, avec de fortes disparités territoriales.
  • Engagement international : évolutions quantitatives mais nécessité de progresser sur l'accessibilité.
  • Rapports (ONU, inspections générales) pointant la nécessité de passer d'une approche compensatrice à une approche privilégiant l'accessibilité et enjoignant à fermer les établissements spécialisés.
  • La construction de l'école inclusive est un processus transformatif visant à accueillir a priori tous les élèves.

• [10:00 - 11:05] La Minute Juris (Raphaël Mata duvgot)

  • Loi du 8 juillet 2013 intégrant l'inclusion scolaire dans le code de l'éducation.
  • Loi du 26 juillet 2019 (école de la confiance) transformant "inclusion scolaire" en "scolarisation inclusive" : ce n'est plus l'élève qui s'adapte à l'école, mais l'école qui s'adapte à l'élève.
  • Droit à l'éducation à valeur constitutionnelle.
  • Obligation pour l'État d'assurer l'égalité de traitement et des dispositions appropriées pour l'accès de chacun en fonction de ses besoins particuliers.
  • Focus sur les élèves souffrant d'un handicap : obligation de formation scolaire, moyens financiers et humains nécessaires à la scolarisation en milieu ordinaire (loi de 2005).
  • Responsabilité de l'État en cas de défaut de scolarisation adaptée.
  • Collaboration avec le secteur médico-social institutionnalisée (décret de 2009).
  • Droit créance à la charge de l'éducation nationale pour les AESH (ordonnance de 2010).
  • Focus sur les questions de genre et le droit à l'éducation des élèves trans : protection contre le harcèlement (lois de 2013 et 2022), reconnaissance du droit (circulaire Blanquer sur le prénom d'usage), aménagements pratiques (usage des espaces d'intimité).
  • La pleine effectivité de l'inclusivité demeure un idéal.

• [11:05 - 12:10] Transition vers les témoignages de terrain

  • Introduction de la seconde partie avec Stéphane Bertrou et Virginie Legal.

• [12:10 - 13:10] Orientations nationales et mise en œuvre locale (Stéphane Bertrou)

  • L'école inclusive comme question socialement vive, nécessitant un cadrage du supranational au national.
  • Conférence nationale du handicap de 2023 (CNH 23) et ses 10 engagements, dont l'école pour tous.
  • Engagement dans l'acte 2 de l'école inclusive.
  • Déploiement sur le territoire de l'Académie de Nantes (cinq départements, environ 25 000 jeunes reconnus handicapés scolarisés en établissements ordinaires et 3 500 dans le médico-social).
  • Quatre thématiques pour le transfert du national au territoire : pilotage, travailler avec le médico-social, maillage territorial, et formation.

• [13:10 - 14:00] Pilotage de l'école inclusive (Stéphane Bertrou & Virginie Legal)

  • Au niveau académique : projet académique intégrant l'école inclusive, création des services départementaux de l'école inclusive (SDUI) et des comités départementaux de suivi de l'école inclusive (CDSI).
  • Au niveau de l'établissement (Collège Trémolière) : intégration dans le projet d'établissement, rôle central du chef d'établissement, sessions de sensibilisation, utilisation de Pronote pour le partage d'informations, intégration des PIAL, évaluations régulières.

• [14:00 - 16:00] Collaboration, coopération, coordination (Stéphane Bertrou & Virginie Legal)

  • Nécessité de "travailler avec" différents acteurs (médico-social, familles, PIAL).
  • Rapprochement rectorat-ARS, stratégies communes (troubles neurodéveloppementaux, unités d'enseignement autisme, dispositifs d'autorégulation).
  • Équipes mobiles d'appui à la scolarisation (EMAS).
  • Apprendre à travailler ensemble entre enseignement et médico-social, exemples de plateaux du médico-social implantés dans des établissements scolaires.
  • Dans l'établissement : collaboration avec les CESSAD et les familles, intégration des AESH à l'équipe éducative, rôle crucial du coordinateur ULIS.

• [16:00 - 17:30] Réaction de Frédéric Dupré sur les pratiques effectives et la complexité

  • Constatation de la présence du vocabulaire inclusif.
  • Importance de regarder les pratiques effectives et leur correspondance aux enjeux d'accessibilité.
  • Évolution de la forme scolaire classique avec plusieurs professionnels travaillant ensemble.
  • Complexité des dispositifs et du travail conjoint (enseignant-AESH, etc.).

• [17:30 - 18:00] Complexité des niveaux d'intervention et des dispositifs (Yannick Ten)

  • Complexité des niveaux d'intervention (académie, département, établissement, écoles) et nécessité d'une compréhension partagée.
  • Complexité des dispositifs (acronymes, etc.) pour les enseignants, les acteurs et les familles.
  • Importance de visualiser le parcours de l'élève de la maternelle au lycée.
  • Question de la mesure et de l'évaluation de l'autonomie progressive des enfants en situation de handicap.

• [18:00 - 19:45] Maillage territorial et dispositifs d'accompagnement (Stéphane Bertrou & Virginie Legal)

  • Le maillage territorial en écho à la complexité et aux différents niveaux d'intervention, permettant de visualiser le parcours.
  • Enjeu premier : rendre l'environnement (spatial, classe, enseignement) accessible.
  • Pôle d'appui à la scolarité (expérimentation nationale) : concentration des regards pour rendre les établissements plus accessibles.
  • Nécessité pour l'Éducation Nationale d'apporter les réponses de premier niveau.
  • Coopération avec les familles, rendre lisible le parcours.
  • Dans l'établissement : intégration des PIAL, travail en réseau avec les établissements environnants et les services éducatifs et de soins, anticipation de la mise en place des pôles d'appui.

• [19:45 - 22:40] Formation et sensibilisation des personnels (Stéphane Bertrou, Yannick Ten & Virginie Legal)

  • Importance de la formation face à la diversité des élèves.
  • Nombreuses formations déployées (AESH, enseignants, cadres) au niveau académique et départemental.
  • Accent mis sur l'accessibilité plutôt que sur la connaissance approfondie de chaque trouble.
  • Bémol de Yannick Ten : évolution positive des processus de formation mais insuffisance et écart entre intentions et réalité.
  • Nécessité de renforcer la formation initiale et continue, et de former les personnels de service et d'intervention des collectivités territoriales.
  • Dans l'établissement : importance accordée à la formation et à la sensibilisation de tous les personnels, sessions de sensibilisation dès la rentrée, réunions pluridisciplinaires, encouragement à la participation aux formations académiques et départementales, rôle central du chef d'établissement dans l'impulsion de cette dynamique.

• [22:40 - 23:30] Autres aspects essentiels de l'école inclusive (Stéphane Bertrou)

  • Importance de la communication pour rendre l'existant lisible et mieux accueillir les familles.
  • Nécessité d'anticiper pour que l'école soit suffisamment accessible et éviter les demandes de compensation à la MDPH.
  • Enjeu des adaptations des pratiques pédagogiques, appui sur les ressources existantes (enseignants spécialisés), transformation réelle de l'école nécessaire.

• [23:30 - 25:45] Conclusion (Yannick Ten & Frédéric Dupré)

  • Yannick Ten : engagement des personnels, augmentation du nombre d'enfants en situation de handicap accueillis (tendance à la hausse), importance de comprendre le "parcours du combattant" des familles.
  • Frédéric Dupré : la loi de 2005 évalue les besoins pour proposer des mesures compensatoires et non une mise en accessibilité de l'environnement, choix du système éducatif français s'appuyant sur l'accompagnement humain et les dispositifs "inclusifs", vigilance sur le fait que les mesures compensatoires peuvent être sources d'obstacles supplémentaires, importance de réfléchir à l'accessibilité (à l'école, pédagogique, didactique) à tous les échelons du système éducatif, reconnaissance des petits pas réalisés au quotidien, nécessité d'une action conjointe pour une école pour tous.

• [25:45 - 28:30] La Minute Biblique (Sylvien Paul)

  • Présentation de ressources sur l'école inclusive : site education.gouv.fr, ouvrages (Agir pour les réussites scolaires, Inclusion scolaire dispositif et pratique pédagogique, L'autisme à l'école, Éducation inclusive privilège ou droit, Gouvernance et inclusion scolaire), revue (La Nouvelle Revue éducation et société inclusive), portail Canopé (Cap école inclusive), et les anciens numéros du Périscope.

• [28:30 - Fin] Mot de la fin et annonce du prochain numéro

  • Remerciements aux invités et annonce du prochain numéro du Périscope le 11 mars sur le parcours de l'élève .

Author response:

In view of the suggestions of the referees, we wish to underline that a user can interact with celldetective at two levels: a non-coder can analyse data and train models without coding, but is necessarily offered pre-determined choices and flexibility. An advanced user however has practically limitless flexibility to extend the fully-open source celldetective, aided by its modularity and detailed manual.

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this manuscript, Torro et al. presented CellDetective, an open-source software designed for a user-friendly execution of single-cell segmentation, tracking, and analysis of time-lapse microscopy data. The authors demonstrated the applications of the software by measuring NK cell spreading events acquired with reflection interference contrast microscopy (RICM), as well as detecting target cell death events and their interaction with neighboring NK cells in a multichannel widefield microscopy dataset.

Strengths:

The segmentation (StarDist, Cellpose) and tracking (bTrack) modules implemented were based on existing and published software packages. The authors added the event detection, classification, and analysis modules to enable an end-to-end time-lapse microscopy data processing and analysis pipeline, complete with a graphical user interface (GUI). This minimizes the coding experience required from the user. The documentation that accompanies CellDetective is also adequate.

Weaknesses:

Given that the software was designed to improve user experience, such an approach also limits its scope and functionality and is currently capable of handling very specific types of experiments. Additionally, this reviewer has also encountered many technical difficulties (see documented bugs/crashes below) that have prevented an extensive exploration of all the functionality of CellDetective.

We apologize for the technical difficulties and bugs; the ones mentioned have been already corrected. New users have also tested the installation and reported it to be bug-free.

We fully agree on the compromise that has to be found between user experience and versatility. We have already tested celldetective in other biological contexts, such as microbiology, but made a choice to showcase it in the article for immunological applications. We invite the reader to consult the software documentation and online examples to learn about more options.

Specifics:

(1) The software can only handle 2D 'widefield' time-lapse imaging datasets. It should be noted that many studies that examine cell-cell interactions in vitro also used confocal microscopy and acquired the time-lapse images in 3D z-stacks to enable the reconstruction of entire cell volumes from multiple optical sections along the z-axis.

Given that almost all of the implemented segmentation (StarDist, Cellpose) and tracking (bTrack) packages already support the handling of 3D datasets, it is unclear why CellDetective was designed to only work with 2D datasets.

As noted above, extending the support for 3D images would allow the scope and utility of this software to be further extended for imaging studies acquired in z-stacks. As an example, the dense clustering of effector cells in Figure 4 had prevented accurate segmentation due to the 2D nature of the experimental dataset. More importantly, support for a 3D dataset could also allow for the tracking of fluorescent protein-based sub-cellular as well as membrane protein localization during cell-cell interactions.

Furthermore, it also widens the potential applicability for analyzing datasets from 3D organoid imaging and perhaps even intravital two-photon microscopy.

We thank the reviewer for this suggestion. Indeed, extension to 3-dimensions is a natural development, since we have chosen segmentation and tracking methods which are compatible with 3D. However, two important strengths of celldetective are: harnessing statistical power of cell populations together with multiplexing biological conditions, and dynamic analysis of fast events.

For both, 2D is advantageous. Our own focus is on analyzing cellular events with minute time resolution, relevant in immunology. By our estimate (experience and literature), 3D timelapse acquisition would reduce the time resolution, as well as throughput (in terms of events and conditions) to below acceptable level. While we don’t envisage this upgrade in the immediate future, we encourage advanced users to contribute to further develop the open-source code in this direction. As a mitigation solution, a 2.5D approach on a flat sample by combining two z planes (in order to address issues of cell superposition for example), could be readily implemented with minimal change.

(2) The software in its current form only allows the broad demarcation of the cells examined into two populations: targets and effectors. This limits the number of cell populations that can be examined for their interactions. It might be more useful to just allow multiple user-defined populations instead of restricting the populations to target and effector cells only.

We thank the reviewer for this suggestion. There is little architectural limitation to its implementation; this will be proposed in the future version. This updated version will allow more than two user-defined populations, labelled directly by the user, which will also facilitate the natural extension to more varied biological applications. Three-way interactions are much more complex, and, to our knowledge, not currently addressed by biologists. The interactions will for the moment be limited to 2 populations interactions, as multipartite ones involve a higher level of code modifications, not immediately envisaged.

(3) Similarly, subsetting of each of the populations could be made more intuitive. Although it is possible to define subsets of cells using the "Custom classification" function under the "Measure" module with user-defined parameters, visualization of multiple groups remains unintuitive and it appears that only one custom classified group can be selected and visualized at any given time in the Signal Annotator under Measurement instead of allowing visualization of multiple (custom defined) groups of cells in different colors. It is also unclear how, if possible at all, to visualize a custom group of cells in the Signal Annotator under the Detect Events module.

The simultaneous visualization of several classes poses problems in the choice of colors and symbols, and may render the tool difficult to use. The time propagation option in the classification tool allows to define event classes as opposed to groups, that are compatible with the Signal Annotator. For more complex classifications, a simple solution is to work with composite classifications, which are already supported by using logical AND/OR operators on the condition defining the class. We believe that this feature is sufficient to address this issue.

Software issues:

(4) When initially tested on v1.3.9, the Segment module could not be initiated (with the error message AttributeError: 'WindowsPath' object has no attribute 'endswith' when attempting to run segmentation).

Update: this has been fixed in v1.3.9.post4 dated February 7th, 2025.

(5) Further testing was then performed by downgrading the software to v1.3.1. While testing the ADCC demo experiment (https://celldetective.readthedocs.io/en/latest/adcc-example.html), the workflow was stuck at attempts to initiate the Detect Events step:

AssertionError: No signal matches with the requirements of the model ['dead_nuclei_channel_mean', 'area']. Please pass the signals manually with the argument selected_signals or add measurements. Abort.

(Update: fixed in the latest v1.3.9.post4 version dated February 7th, 2025)

(6) Random bugs causing the software to crash. Example: switching characteristic to 'status_color' in the Signal Annotator under Measurement caused the software to crash (v1.3.9.post4):

TypeError: ufunc 'isnan' is not supported for the input types, and the inputs could not be safely coerced to any supported types according to the casting rule 'safe'

(7) Overall, when exploring the functionality of the software, there have been multiple instances of software crashes when clicking/switching around to show different parameters, etc.

This reviewer understands the difficulties and time involved in bug fixing and hopes that the experience could have been much smoother and that the software behaves much more stably in order to maximize its useability.

We apologize again for the various technical issues encountered during the review process, and thank the reviewer for mentioning that several bugs were already fixed in the last software release. The open source and software maintenance protocol enabled by github should help to resolve any further emerging issue.

Reviewer #2 (Public review):

Summary:

Immune assays enable the analysis of immune responses in vitro. These assays generate time series image data across several experimental conditions. The imaging parameters such as the imaging modality and the number of channels can vary across experiments. A challenge in the field is the lack of (open source) tools to process and analyze these data. R. Torro, et. al. developed an open source end-to-end pipeline for the analysis of image data from these immune assays. The pipeline is designed with a GUI and is suited for experimental biologists with no coding experience. The authors have incorporated several existing methods and tools for individual tasks such as for segmentation and cell tracking, and incorporated them with custom methods where necessary such as for tracking cell state transitions.

Strengths:

(1) The tool is extremely well-documented and easy to install.

(2) Applicable to a wide variety of imaging modalities and analysis.

(3) There are several different options for each step, such as segmentation using traditional methods or deep learning methods, and all the analysis steps are integrated in one place with a GUI. The no-coding requirement makes this a very powerful tool for biologists and has the potential to enable a wide variety of analyses.

Weakness:

(1) It would be good to provide documentation on how to make the tool applicable for applications and analysis other than for immune profiling since most methods integrated here are applicable well beyond immune profiling. For example, a user might want to use the tool just for the segmentation of their IF microscopy-images.

This is an important suggestion that we will implement as short demonstrations using data from the public domain. These will be proposed as examples in the online documentation.

(2) They applied Celldetective to two immune assays. The authors present the results from these assays and use the results to validate their assay. However, they have not included data that demonstrates results obtained via this pipeline are comparable to results obtained with other pipelines and/or if these results are consistent with what is expected in the literature.

In the final version of the article, we shall compare celldetective with existing literature, including our previous work, when possible. However, we emphasize that most of the presented data are original and don’t have any published equivalent in the literature. Concerning the immunotherapy assays, data presented already show expected trends (see for example Fig. 2 and Fig. 5). We reserve for future publications the systematic comparison with traditional (non microscopy-based) methods, as we consider it out-of-scope here. Additionally, there is, to our knowledge no existing open pipeline performing the full end-to-end analysis.

Error code is missing " Bad request error"

other child params are missing - name, bank code etc

The I/O control level transfers data between memory and storage using device drivers, which convert commands into hardware instructions. The basic file system handles read/write operations, I/O scheduling, and caching. The file-organization module manages file storage and free space. The logical file system tracks metadata, directories, and file protection. Layering reduces code duplication but may slow performance due to added overhead.

The file-organization module structures files into logical blocks, assigning each block a number and managing free-space allocation. It interacts with the logical file system, which handles metadata such as file ownership, permissions, and directory structures. The logical file system maintains file-control blocks (FCBs), storing essential information about files. It also enforces security measures and ensures protection policies are upheld. Layered file-system structures minimize code duplication, allowing multiple file systems to share basic components. However, layering can introduce overhead, affecting performance. Designing an efficient layered architecture requires balancing modularity, reusability, and operational efficiency in file-system implementations.

The I/O control level is responsible for handling the transfer of data between memory and storage devices. It includes device drivers, which translate high-level commands into low-level instructions for hardware controllers. These drivers communicate with the I/O device by writing specific bit patterns to controller memory. The block I/O subsystem issues generic read and write commands to the appropriate driver while managing request scheduling, buffers, and caches. Buffers store temporary data blocks during transfers, and caches hold frequently accessed metadata to improve performance. Efficient buffer and cache management is crucial to maintaining system responsiveness and optimizing I/O operations.

File types define their structure, which some operating systems enforce for proper execution. While supporting multiple file structures adds flexibility, it can make the system complex, so many OSs, like UNIX, treat files as simple byte sequences, leaving interpretation to applications.

A file’s content is decided by its creator and can store different types of data, like programs, text, images, music, or videos. Each file has a specific structure based on its type. A text file contains characters arranged in lines, a source file has functions with declarations and instructions, and an executable file has code that can be loaded into memory and run by the computer.

This section describes various storage devices such as HDDs, SSDs, magnetic tapes, and optical disks. These devices differ in speed, capacity, and durability, which impact the efficiency of file systems. For example, SSDs provide faster data access than HDDs due to their lack of moving parts, making them ideal for performance-driven applications. Magnetic tapes, on the other hand, are used for archival storage because of their low cost and high capacity, despite their slow sequential access. Understanding the relationship between storage devices and file systems is essential when choosing an appropriate file management solution. The operating system abstracts the differences between these devices, offering a uniform interface for storing and retrieving files seamlessly across different hardware platforms.

File structures help the operating system and applications organize and manage data effectively. Some files, like source and object files, follow a defined structure expected by the programs that use them. Operating systems often impose structures on certain files, such as executable files, to facilitate their loading and execution. However, supporting multiple file structures increases system complexity and can restrict flexibility. Some systems, like UNIX, adopt a minimal approach, treating all files as byte sequences without enforcing structure. This approach enhances flexibility but places the burden on applications to interpret data. A rigid file structure system can create issues when new types of files, such as encrypted files, do not conform to predefined formats, requiring workarounds or alternative storage methods.

Shared Memory in the Windows API Windows provides built-in support for memory-mapped files to enable shared memory between processes. The process begins with the creation of a file mapping using CreateFile() and CreateFileMapping(), followed by mapping a view of the file into a process’s address space with MapViewOfFile(). A second process can then access the same memory-mapped file, allowing seamless data sharing. This mechanism is particularly useful for producer-consumer scenarios, where one process writes to shared memory while another reads from it. The example code in this section demonstrates how a producer writes a message to a shared-memory object, which a consumer then reads. This implementation avoids the overhead of traditional IPC methods like pipes or message queues, leveraging efficient memory management instead.

[(https://cscircles.cemc.uwaterloo.ca/java_visualize/#code=public+class+TwoDArr%7B%0A+++public+static+void+main(String%5B%5D+args)+%7B%0A++++++int%5B%5D%5B%5D+points+%3D+%7B+%7B11,+12,+13,+14,+15%7D,%0A++++++++++++++++++++%7B21,+22,+23,+24,+25%7D,%0A++++++++++++++++++++%7B31,+32,+33,+34,+35%7D,%0A++++++++++++++++++++%7B41,+42,+43,+44,+45%7D%7D%3B%0A+for+(int+row+%3D+0%3B+row+%3C+points.length%3B+row%2B%2B)%0A+%7B%0A+++++for+(int+col+%3D+points%5B0%5D.length+-+1%3B+col+%3E%3D+row%3B+col--)%0A+++++%7B%0A++++++++++System.out.print(points%5Brow%5D%5Bcol%5D+%2B+%22+%22)%3B%0A+++++%7D%0A+++++System.out.println()%3B%0A%7D%0A+++%7D%0A%7D&mode=display&curInstr=0)]

Reviewer #2 (Public review):

Summary:

The manuscript from Hammond et al., investigates the modularity of the segmentation clock and morphogenesis in early vertebrate development, focusing on how these processes might independently evolve to influence the diversity of segment numbers across vertebrates.

Methodology: The study uses a previously published computational model, parameterized for zebrafish, to simulate and analyse the interactions between the segmentation clock and the morphogenesis of the pre-somitic mesoderm (PSM). Their model integrates cell advection, motility, compaction, cell division, and the synchronization of the embryo clock. Three alternative scenarios of PSM morphogenesis were modeled to examine how these changes affect the segmentation clock.

Model System: The computational model system combines a representation of cell movements and the phase oscillator dynamics of the segmentation clock within a three-dimensional horseshoe-shaped domain mimicking the geometry of the vertebrate embryo PSM. The parameters used for the mathematical model are mostly estimated from previously published experimental findings.

Key Findings and Conclusions: (1) The segmentation clock was found to be broadly robust against variations in morphogenetic processes such as cell ingression and motility; (2) Changes in the length of the PSM and the strength of phase coupling within the clock significantly influenced the system's robustness; (3) The authors conclude that the segmentation clock and PSM morphogenesis exhibited developmental modularity (i.e. relative independence), allowing these two phenomena to evolve independently, and therefore possibly contributing to the diverse segment numbers observed in vertebrates.

Major comments from the original round of review:

(1) The key conclusion drawn by the authors (that there is robustness, and therefore modularity, between the morphogenetic cellular processes modeled and the embryo clock synchronization) stems directly from the modeling results appropriately presented and discussed in the manuscript.

The model comprises some strong assumptions, however all have been clearly explained and the parameterization choices are supported by experimental findings, providing biological meaning to the model. Estimated parameters are well explained, and seem reasonable assumptions (from the embryology perspective).

(2) This study, as is, achieves its proposed goal of evaluating the potential robustness of the embryo clock to changes in (some) morphogenetic processes. The authors do not claim that the model used is complete, and they properly identify some limitations, including the lack of cell-cell interactions. Given the recognized importance of cellular physical interactions for successful embryo development, including them in the model would be a significant addition in future studies.

(3) The authors have deposited all the code used for analysis in a public GitHub repository that is updated and available for the research community.

(4) In page 6, the authors justify their choice of clock parameters for cells ingressing the PSM: "As ingressing cells do not appear to express segmentation clock genes (Mara et al. (2007)), the position at which cells ingress into the PSM can create challenges for clock patterning, as only in the 'off' phase of the clock will ingressing cells be in-phase with their neighbors."

However, there are several lines of evidence (in chick and mouse), that some oscillatory clock genes are already being expressed as early as in the gastrulation phase (so prior to PSM ingression) (Feitas et al, 2001 [10.1242/dev.128.24.5139]; Jouve et al, 2002 [10.1242/dev.129.5.1107]; Maia-Fernandes at al, 2024 [10.1371/journal.pone.0297853]).

Question: Is this also true in zebrafish? (I.e. is there any recent experimental evidence that the clock genes are not expressed at ingression, since the paper cited to support this assumption is from 2007).

If they are expressed in zebrafish (as they are in mouse and chick), then the cell addition should have random clock gene periods when they enter the PSM and not start all with a constant initial phase of zero. Probably this will not impact the results since the cells will also be out of phase with their neighbors when they "ingress", however, it will model more closely the biological scenario (and avoid such criticism).

Significance:

GENERAL ASSESSMENT

This study uses a previously published model to simulate alternative scenarios of morphogenetic parameters to infer the potential independence (termed here modularity) between the segmentation clock and a set of morphogenetic processes, arguing that such modularity could allow the evolution of more flexible body plans, therefore partially explaining the variability in the number of segments observed in the vertebrates. This question is fundamental and relevant, yet still poorly researched. This work provides a comprehensive simulation with a model that tries to simplify the many morphogenetic processes described in the literature, reducing it to a few core fundamental processes that allow drawing the conclusions sought. It provides theoretical insight to support a conceptual advance in the field of evolutionary vertebrate embryology.

ADVANCE

This study builds on a model recently published by Uriu et al. (eLife, 2021) that incorporates quantitative experimental data within a modeling framework including cell and tissue-level parameters, allowing the study of multiscale phenomena active during zebrafish embryo segmentation. Uriu's publication reports many relevant and often non-intuitive insights uncovered by the model, most notably the description of phase vortices formed by the synchronizing genetic oscillators interfering with the traveling-wave front pattern.

However, this model can be further explored to ask additional questions beyond those described in the original paper. A good example is the present study, which uses this mathematical framework to investigate the potential independence between two of the modeled processes, thereby extracting extra knowledge from it. Accordingly, the present study represents a step forward in the direction of using relevant theoretical frameworks to quantitatively explore the landscape of complex molecular hypotheses in silico, and with it shed some light on fundamental open questions or inform the design of future experiments in the lab.

The study incorporates a wide range of existing literature on the developmental biology of vertebrates. It comprehensively cites prior work, such as the foundational studies by Cooke and Zeeman on the segmentation clock and the role of FGF signaling in PSM development as discussed by Gomez et al. The literature properly covers the breadth of knowledge in this field.

AUDIENCE

Target audience: This study is relevant for fundamental research in developmental biology, specifically targeting researchers who focus on early embryo development and morphogenesis from both experimental and theoretical perspectives. It is also relevant for evolutionary biologists investigating the genetic factors that influence vertebrate evolution, as well as to computational biologists and bioinformatics researchers studying developmental processes and embryology.

Developmental researchers studying the segmentation clock in other vertebrate model organisms (namely mouse and chick), will find this publication especially valuable since it provides insights that can help them formulate new hypotheses to elucidate the molecular mechanisms of the clock (for example finding a set of evolutionarily divergent genes that might interfere with PSM length).<br /> Additionally, this study provides a set of cellular parameters that have yet to be measured in mouse and chick, therefore guiding the design of future experiments to measure them, allowing the simulation of the same model with sets of parameters from different vertebrate model organisms, therefore testing the robustness of the findings reported for zebrafish.

Author response:

Reviewer #1 (Evidence, reproducibility and clarity):

Summary:

In this manuscript, Hammond et al. study robustness of the vertebrate segmentation clock against morphogenetic processes such as cell ingression, cell movement and cell division to ask whether the segmentation clock and morphogenesis are modular or not. The modularity of these two would be important for evolvability of the segmenting system. The authors adopt a previously proposed 3D model of the presomitic mesoderm (Uriu et al. 2021 eLife) and include new elements; different types of cell ingression, tissue compaction and cell cycles. Based on the results of numerical simulations that synchrony of the segmentation clock is robust, the authors conclude that there is a modularity in the segmentation clock and morphogenetic processes. The presented results support the conclusion. The manuscript is clearly written. I have several comments that could help the authors further strengthen their arguments.

Major comment: 

[Optional] In both the current model and Uriu et al. 2021, coupling delay in phase oscillator model is not considered. Given that several previous studies (e.g. Lewis 2003, Herrgen et al. 2010, Yoshioka-Kobayashi et al. 2020) suggested the presence of coupling delays in DeltaNotch signaling, could the authors analyze the effect of coupling delay on robustness of the segmentation clock against morphogenetic processes?

We thank the reviewer for the suggestion. Owing to the computational demands of including such a delay in the model, we cannot feasibly repeat every simulation analysed here in the presence of delay, and would like to note that the increased computational demand that delays put on the simulations is also the reason why Uriu et al 2021 did not include it, as stated in their published exchange with reviewers. However, analogous to our analysis in figure 7, we can analyse how varying the position of progenitor cell ingression affects synchrony in the presence of the coupling delay measured in zebrafish by Herrgen et al. (2010). We show this analysis in a new figure 8 (8B, specifically), on page 21, and discuss its implications in the text on pages 2022. Our analysis reveals that the model cannot recover synchrony using the default parameters used by Uriu et al. (2021) and reveal a much stronger dependence on the rate of cell mixing (vs) than shown in the instantaneous coupling case (cf. figure 7). However, by systematically varying the value of the delay we find that a relatively minor increase in the delay is sufficient to recover synchrony using the parameter set of Uriu et al. (see figure 8C). Repeating this across the three scenarios of cell ingression we see that the combination of coupling strength and delay determine the robustness of synchrony to varying position of cell ingression. This suggests that the combination of these two parameters constrain the evolution of morphogenesis.

Minor comments: 

-  PSM radius and oscillation synchrony are both denoted by the same alphabet r. The authors should use different alphabets for these two to avoid confusion.

We thank the reviewer for spotting this. This has now been changed throughout to rT, as shorthand for ‘radius of tissue’.

-  page 5 Figure 1 caption: (x-x_a/L) should be (x-x_a)/L.

We thank the reviewer for spotting this. This has now been corrected.

-  Figure 3C: Description of black crosses in the panels is required in the figure legend.

Thank you for spotting this. The legend has now been corrected.

-  Figure 3C another comment: In this panel, synchrony r at the anterior PSM is shown. It is true that synchrony at anterior PSM is most relevant for normal segment formation. However, in this case, the mobility profile is changed, so it may be appropriate to show how synchrony at mid and posterior PSM would depend on changes in mobility profile. Is synchrony improved by cell mobility at the region where cell ingression happens?

We thank the reviewer for the suggestion. We have now plotted the synchrony along the AP axis for varying motility profiles, and this can be seen in figure 3 supplement 1, and is briefly discussed in the text on page 11. We show that while the synchrony varies with x-position (as already expected, see figure 2), there is no trend associated with the shape of the motility profile.

-  In page 12, the authors state that "the results for the DP and DP+LV cases are exactly equal for L = 185 um, as .... and the two ingression methods are numerically equivalent in the model". I understood that in this case two ingression methods are equivalent, but I do not understand why the results are "exactly" equal, given the presence of stochasticity in the model.

These results can be exactly equal despite the simulations being stochastic because they were both initialised using the same ‘seed’ in the source code. However, we now see that this might be confusing to the reader, and we have re-generated this figure but this time initialising the simulations for each ingression scenario using a different seed value. This is now reflected in the text on page 12 and in figure 4.

-  The authors analyze the effect of cell density on oscillation synchrony in Fig. 4 and they mention that higher density increases robustness of the clock by increasing the average number of interacting neighbours. I think it would be helpful to plot the average number of neighbouring cells in simulations as a function of density to quantitatively support the claim.

We thank the reviewer for their suggestion. Distributions of neighbour numbers for exemplar simulations with varying density can now be found in  figure 4 supplementary figure 1 and are referred to in the text on page 11.

-  The authors analyze the effect of PSM length on synchrony in Fig. 4. I think kymographs of synchrony r as shown in Fig. 2D would also be helpful to show that indeed cells get synchronized while advecting through a longer PSM.

We thank the reviewer for their suggestion and agree that visualising the data in this way is an excellent idea. We have generated the suggested kymographs and added them to figure 4 as supplements 2 and 4, and discussed these results in the text on page 12.

-  I understand that cells in M phase can interact with neighboring cells with the same coupling strength kappa in the model, although their clocks are arrested. If so, this aspect should be also mentioned in the main text in page 16, as this coupling can be another noise source for synchrony.

We agree this is an important clarification. We explicitly state this, and briefly justify our choice, in the text on page 16.

-  Figure 5-figure supplement 2: panel labels A, B, C are missing. 

Thank you for bringing this to our attention. These have now been added.

– Figure 5-figure supplement 3: panel labels A, B, C are missing.

Thank you for bringing this to our attention. These have now been added.

Reviewer #1 (Significance):

Synchronization of the segmentation clock has been studied by mathematical modeling, but most previous studies considered cells in a static tissue without morphogenesis. In the previous study by Uriu et al. 2021, morphogenetic processes such as cell advection due to tissue elongation, tissue shortening, and cell mobility were considered in synchronization. The current manuscript provides methodological advances in this aspect by newly including cell ingression, tissue compaction and cell cycle. In addition, the authors bring a concept of modularity and evolvability to the field of the vertebrate segmentation clock, which is new. On the other hand, the manuscript confirms that the synchronization of the segmentation clock is robust by careful simulations, but it does not propose or reveal new mechanisms for making it robust or modular. The main targets of the manuscript will be researchers working on somitogenesis and evolutionary biologists who are interested in evolution of developmental systems. The manuscript will also be interested by broader audiences, like developmental biologists, biophysicists, and physicists and computer scientists who are working on dynamical systems.

We thank the reviewer for their interest in our manuscript and for acknowledging us as one of the first to address the modularity and evolvability of somitogenesis. We hope that this work will encourage others to think about these concepts in this system too.  

In the original submission, we identified a high enough coupling strength as the main mechanism underlying the identified modularity in somitogenesis. Since, we have included an analysis of the coupling delay and find that it is the interplay between coupling strength and coupling delay that mediate the identified modularity, allowing PSM morphogenesis and the segmentation clock to evolve independently in regions of parameter space that are constrained and determined by the interplay between these two parameters. We have now added an extra figure (figure 8) where we explore this interplay and have discussed it at length in the last section of the results and in the discussion. We again thank the reviewer for encouraging us to include delays in our analysis.

Reviewer #2 (Evidence, reproducibility and clarity):

SUMMARY 

The manuscript from Hammond et al., investigates the modularity of the segmentation clock and morphogenesis in early vertebrate development, focusing on how these processes might independently evolve to influence the diversity of segment numbers across vertebrates.

Methodology: The study uses a previously published computational model, parameterized for zebrafish, to simulate and analyse the interactions between the segmentation clock and the morphogenesis of the pre-somitic mesoderm (PSM). Their model integrates cell advection, motility, compaction, cell division, and the synchronization of the embryo clock. Three alternative scenarios of PSM morphogenesis were modeled to examine how these changes affect the segmentation clock.

Model System: The computational model system combines a representation of cell movements and the phase oscillator dynamics of the segmentation clock within a three-dimensional horseshoe-shaped domain mimicking the geometry of the vertebrate embryo PSM. The parameters used for the mathematical model are mostly estimated from previously published experimental findings.

Key Findings and Conclusions: (1) The segmentation clock was found to be broadly robust against variations in morphogenetic processes such as cell ingression and motility; (2) Changes in the length of the PSM and the strength of phase coupling within the clock significantly influenced the system's robustness; (3) The authors conclude that the segmentation clock and PSM morphogenesis exhibited developmental modularity (i.e. relative independence), allowing these two phenomena to evolve independently, and therefore possibly contributing to the diverse segment numbers observed in vertebrates.

MAJOR COMMENTS

(1) The key conclusion drawn by the authors (that there is robustness, and therefore modularity, between the morphogenetic cellular processes modeled and the embryo clock synchronization) stems directly from the modeling results appropriately presented and discussed in the manuscript. The model comprises some strong assumptions, however all have been clearly explained and the parameterization choices are supported by experimental findings, providing biological meaning to the model. Estimated parameters are well explained and seem reasonable assumptions (from the embryology perspective).

We thank the reviewer for their positive comments about our work

(2) This study, as is, achieves its proposed goal of evaluating the potential robustness of the embryo clock to changes in (some) morphogenetic processes. The authors do not claim that the model used is complete, and they properly identify some limitations, including the lack of cellcell interactions. Given the recognized importance of cellular physical interactions for successful embryo development, including them in the model would be a significant addition in future studies.

We would like to clarify that the model does include cell-cell interactions as cells interact with their neighbours’ clock phase to synchronise and to avoid occupying the same physical space. 

(3) The authors have deposited all the code used for analysis in a public GitHub repository that is updated and available for the research community.

We support open source coding practices.

(4) In page 6, the authors justify their choice of clock parameters for cells ingressing the PSM: "As ingressing cells do not appear to express segmentation clock genes (Mara et al. (2007)), the position at which cells ingress into the PSM can create challenges for clock patterning, as only in the 'off' phase of the clock will ingressing cells be in-phase with their neighbours."  However, there are several lines of evidence (in chick and mouse), that some oscillatory clock genes are already being expressed as early as in the gastrulation phase (so prior to PSM ingression) (Feitas et al, 2001 [10.1242/dev.128.24.5139]; Jouve et al, 2002 [10.1242/dev.129.5.1107]; Maia-Fernandes at al, 2024 [10.1371/journal.pone.0297853]) Question: Is this also true in zebrafish? (I.e. is there any recent experimental evidence that the clock genes are not expressed at ingression, since the paper cited to support this assumption is from 2007). If they are expressed in zebrafish (as they are in mouse and chick), then the cell addition should have random clock gene periods when they enter the PSM and not start all with a constant initial phase of zero. Probably this will not impact the results since the cells will also be out of phase with their neighbours when they "ingress", however, it will model more closely the biological scenario (and avoid such criticism).

We thank the reviewer for their comments. While it is known that in zebrafish the clock begins oscillating during epiboly and before the onset of segmentation (Riedel-Kruse et al., 2007), to our knowledge no-one has examined whether posteriorly or laterally ingressing progenitor cells express clock genes prior to their ingression into the PSM, which occurs later in development than the first oscillations which give rise to the first somites. We have not found any published evidence of her/hes gene expression in the dorsal donor tissues or lateral tissues surrounding the PSM, however we acknowledge that this has not been actively studied before and our assumption relies on an absence of evidence, rather than evidence of absence. 

However, we agree with the reviewer that one should include such an analysis for completeness, and we have now generated additional simulations where progenitor cells ingress with a random clock phase. This data is presented in figure 2 supplement 1 and mentioned in the main text on page 9.

MINOR COMMENTS 

(1) The citations are appropriate and cover the major labs that have published work related to this study (although with some overrepresentation of the lab that published the model used).

We have cited the vast literature on somitogenesis to the best of our ability and do recognise that the work of the Oates lab appears prominently, but this is probably because their experimental data were originally used to parametrise the model in Uriu et al. 2021.

(2) The text is clear, carefully written, and both the methods and the reasoning behind them are clearly explained and supported by proper citations.

We are very glad to see that the reviewer found that the manuscript was clearly presented.

(3) The figures are comprehensive, properly annotated, with explanatory self-contained legends. I have no comments regarding the presentation of the results.

Thank you

(4) Minor suggestions: 

a. Page 26: In the Cell addition sub-section of the Methods section, correct all instances where the word domain is used, but subdomain should be used (for clarity and coherence with the description of the model, stated as having a single domain comprising 3 subdomains).

We thank the reviewer for raising this, this is a good point. We have now corrected to ‘subdomain’ where appropriate.

b. Page 32: Table 1. Parameter values used in our work, unless otherwise stated -> Suggestion: Add a column with the individual citations used for each parameter (to facilitate the confirmation of each corresponding reference).

Thank you for the suggstion, we have now done this (see table 1 page 36).

Reviewer #2 (Significance):

GENERAL ASSESSMENT 

This study uses a previously published model to simulate alternative scenarios of morphogenetic parameters to infer the potential independence (termed here modularity) between the segmentation clock and a set of morphogenetic processes, arguing that such modularity could allow the evolution of more flexible body plans, therefore partially explaining the variability in the number of segments observed in the vertebrates. This question is fundamental and relevant, yet still poorly researched. This work provides a comprehensive simulation with a model that tries to simplify the many morphogenetic processes described in the literature, reducing it to a few core fundamental processes that allow drawing the conclusions seeked. It provides theoretical insight to support a conceptual advance in the field of evolutionary vertebrate embryology.

ADVANCE

This study builds on a model recently published by Uriu et al. (eLife, 2021) that incorporates quantitative experimental data within a modeling framework including cell and tissue-level parameters, allowing the study of multiscale phenomena active during zebrafish embryo segmentation. Uriu's publication reports many relevant and often non-intuitive insights uncovered by the model, most notably the description of phase vortices formed by the synchronizing genetic oscillators interfering with the traveling-wave front pattern.  However, this model can be further explored to ask additional questions beyond those described in the original paper. A good example is the present study, which uses this mathematical framework to investigate the potential independence between two of the modeled processes, thereby extracting extra knowledge from it. Accordingly, the present study represents a step forward in the direction of using relevant theoretical frameworks to quantitatively explore the landscape of complex molecular hypotheses in silico, and with it shed some light on fundamental open questions or inform the design of future experiments in the lab.

The study incorporates a wide range of existing literature on the developmental biology of vertebrates. It comprehensively cites prior work, such as the foundational studies by Cooke and Zeeman on the segmentation clock and the role of FGF signaling in PSM development as discussed by Gomez et al. The literature properly covers the breadth of knowledge in this field.

AUDIENCE

Target audience | This study is relevant for fundamental research in developmental biology, specifically targeting researchers who focus on early embryo development and morphogenesis from both experimental and theoretical perspectives. It is also relevant for evolutionary biologists investigating the genetic factors that influence vertebrate evolution, as well as to computational biologists and bioinformatics researchers studying developmental processes and embryology.

Developmental researchers studying the segmentation clock in other vertebrate model organisms (namely mouse and chick), will find this publication especially valuable since it provides insights that can help them formulate new hypotheses to elucidate the molecular mechanisms of the clock (for example finding a set of evolutionarily divergent genes that might interfere with PSM length). Additionally, this study provides a set of cellular parameters that have yet to be measured in mouse and chick, therefore guiding the design of future experiments to measure them, allowing the simulation of the same model with sets of parameters from different vertebrate model organisms, therefore testing the robustness of the findings reported for zebrafish.

Reviewer #3 (Evidence, reproducibility and clarity): 

In this manuscript, Verd and colleagues explored how various biologically relevant factors influence the robustness of clock dynamics synchronization among neighboring cells within the context of somatogenesis, adapting a mathematical model presented by Urio et. al in 2021 in a similar context. Specifically they show that clock dynamics is robust to different biological mechanisms such as cell infusion, cellular motility, compaction-extension and cell-division. On the other hand , the length of Presomitic Mesoderm (PSM) and density of cells in it has a significant role in the robustness of clock dynamics. While the manuscript is well-written and provides clear descriptions of methods and technical details, it tends to be somewhat lengthy.

Below are the comments I would like the authors to address:

(1) The authors mention that "...the model is three dimensional and so can quantitatively recapture the rates of cell mixing that we observe in the PSM". I am not convinced with this justification of using a 3D model. None of the effects the authors explore in this manuscript requires a three dimensional model or full physical description of the cellular mechanics such as excluded volume interaction etc. A one-dimensional model characterized by cell position along the arclength of PSM and somatic region and segmentation clock phase θ can incorporate all the physics authors described in this manuscript as well as significantly computationally cheap allowing the authors to explore the effect of different parameters in greater detail.

One of the main objectives of the work we present in this manuscript is to assess how the evolution of PSM morphogenesis affects, or does not affect, segment patterning. The PSM is a three-dimensional tissue with differing cell rearrangement dynamics along its anterior-posterior axis. In addition, PSM dimension, density, the rearrangement rate, and patterns of cell ingression all vary across vertebrate species, and they are functional, especially cell mixing as it promotes synchronisation and drives elongation. In order to answer questions on the modularity of somitogenesis we therefore consider it absolutely necessary to include a three-dimensional representation of the PSM that captures single cells and their movements. In addition, this will allow us, as Reviewer #2 also pointed out, to reparametrize our model using species-specific data as it becomes available. 

While the reviewer is right in that lower dimensional representations would be computationally more efficient, and are generally more tractable, it would not be possible to represent cell mixing in one dimension, as this happens in three dimensions. One could perhaps encode the synchrony-promoting effect of cell mixing via some coupling function κ(x) that increases towards the posterior, however it is unclear what existing biological data one could use to parameterise this function or determine its form. Cell mixing can be modelled in a two-dimensional framework, however this cannot quantitatively recapture the rate of cell mixing observed in vivo, which is an advantage of this model. 

Furthermore, it is unclear how one would simulate processes such as compactionextension using a one-dimensional model. The two different scenarios of cell ingression which we consider can also not be replicated in a one-dimensional model, as having a population of cells re-acquiring synchrony on the dorsal surface of the tissue while new material is added to the ventral side, creating asynchrony, is qualitatively different than a one-dimensional scenario where cells are introduced continuously along the spatial axis.

(2) I am not sure about the justification for limiting the quantification of phase synchrony in a very limited (one cell diameter wide) region at one end of the somatic part (Page 33 below Fig. 9). From my understanding of the manuscript, the segments appear in significant length anterior to this region. Wouldn't an ensemble average of multiple such one cell diameter wide regions in the somatic region be a more accurate metric for quantifying synchrony?

Indeed, such a metric (e.g. as that used by Uriu et al. to quantify synchrony along the xaxis) would be more accurate for determining synchrony within the PSM. However, as per the clock and wavefront model of somitogenesis, only synchrony at the very anterior of the PSM (or at the wavefront, equivalently) is functional for somitogenesis and thus evolution. Therefore, we restrict our analysis to the anterior-most region of the PSM. We now further justify this in the main text on page 9.

(3) While studying the effect of cellular ingression, the authors study three discrete modes- random, DP and DP+LV and show that in the DP+LV mode the clock synchrony becomes affected. I would like the authors to explore this in a continuous fashion from a pure DP ingression to Pure LV ingression and intermediates.

We thank the reviewer for this suggestion; this is a very interesting question. We are currently working on a related computational and experimental project to address the question of how PSM morphogenesis can change over evolutionary time to evolve the different modes that we see across species. As part of this work, we are running precisely the simulations suggested by the reviewer to find regions of parameter space in which all the relevant morphogenetic processes can freely evolve.  While interesting, this work is however outside the scope of the current manuscript.

(4) While studying the effect of length and density of cells in PSM on cellular synchrony, the authors restrict to 3 values of density and 6 values of PSM length keeping the other parameter constant. I would be interested to see a phase diagram similar to Fig. 7 in the two-dimensional parameter space of L and ρ0. I am curious if a scaling relation exists for the parameter values that partition the parameter space with and without synchrony.

We thank the reviewer for their suggestion and agree that this would constitute an interesting addition to the manuscript. We have now generated these data, which are shown in figure 4 supplement 5 and mentioned on page 13. We see no clear relationship between these two variables when co-varying in the presence of random ingression. 

(5) Both in the abstract and introduction, the authors discuss at a great length about the variability in the number of segments. I am curious how the number and width of the segments observed depend on different parameters related to cellular mechanics and the segmentation clock ?

We thank the reviewer for this question. It was not clear to us if this was something the reviewer wants us to address in the study’s background and introduction, or an analysis we should include in the results. Therefore, we have responded to both comprehensively below:

The prevailing conceptual framework for understanding this is the clock and wavefront model (Cooke and Zeeman, 1976), which posits that the somite length is inversely proportional to the frequency of the clock relative to the speed of the wavefront, and that the total number of segments is the relative frequency multiplied by the total duration of somitogenesis.

Experimentally we know that the frequency is determined in part by the coupling strength (Liao, Jorg, and Oates, 2016), and from comparative embryological studies (Gomez et al., 2008; Steventon et al., 2016) we know that changes in the elongation dynamics of the PSM correlate with changes in somite number, presumably by altering the total duration of somitogenesis (Gomez et al., 2009). These changes in elongation are thought to be driven by the changes in cell and tissue mechanics we test in our manuscript. 

Within our model, we cannot in general predict how the number of segments responds to changes in either clock parameters or cell mechanical parameters, as we lack understanding of what causes somitogenesis to cease; this is thus not encoded in our model and segmentation can in principle proceed indefinitely. Therefore, we have not performed this analysis.

Similarly, we have not included an analysis of somite length. This is for two reasons: 1) as per the clock and wavefront model, the frequency at the PSM anterior (which we analyse) is equivalent to this measurement, as we assume (in general) the wavefront ($x = x_{a}$) is inertial. 2) the length of the nascent somite is not thought to be of much relevance to the adult phenotype, and by extension evolution. Somites undergo cell division and growth soon after their patterning by the segmentation clock, therefore their final size does not majorly depend on the dynamics of the segmentation clock. Rather, the main function of the clock is to control their number (and polarity).

(6) The authors assume that the phase dynamics of the chemical network may be described by an oscillator with constant frequency. For the completeness of the manuscript, the author should discuss in detail, for which chemical networks this is a good assumption.

We thank the reviewer for their suggestion and now justify this assumption in the methods on page 31. 

Such an assumption is appropriate for the segmentation clock, as the clock in the posterior of the PSM is thought to oscillate with a constant frequency, at least for the majority of somitogenesis although the frequency of somite formation slows towards the end of this process in zebrafish (Giudicelli et al., 2007, PLoS Biol.). In addition, PSM cells isolated and cultured in the presence of FGF (thus replicating the signalling environment of the posterior PSM) will continue to exhibit her1 oscillations with an apparently constant frequency (Webb et al., 2016). 

We note that such formulations are widely used within the segmentation clock literature (e.g. Riedel-Kruse et al., 2007, Morelli et al., 2009).

(7) Figure 3 and the associated text shows no effect of the cellular motility profile in the synchrony of the segmentation clock. This may be moved to the supplementary considering the length of this manuscript.

Thank you for the suggestion. However, we would argue that the lack of effect is a crucial result when discussing modularity. Reviewer #2 agrees with this assessment.

Reviewer #3 (Significance): 

The manuscript answers some important questions in the synchrony of segmentation clock in the vertebrates utilizing a model published earlier. However, the presented result is incomplete in some aspects (points 2 to 5 of section A) and that could be overcome by a more detailed analysis using a simpler one dimensional (point 1 of section A). I believe this manuscript could be of interest to an intersecting audience of developmental biologists, systems biologists, and physicists/engineers interested in dynamical systems.

I'm skeptical of this paragraph being defined as an act of colonization for two main reasons: the first reason is that programming, as a tool that is needed all over the world, a unified language would be easier to communicate on a large scale, and English, as a more widely used language, is a suitable choice to become a programming language. The second reason is that the original code was compiled in English, so I don't think there's anything wrong with keeping English as a programming language.

docker e2b sandbox

using multi-agents in an E2B andbox. mono agent use might look like:

from smolagents import HfApiModel, CodeAgent

agent = CodeAgent(model=HfApiModel(), tools=[], executor_type="e2b")

agent.run("Can you give me the 100th Fibonacci number?")

When an AI (like an LLM) writes code and runs it, you don’t want it to have direct access to your personal files, computer, or sensitive data. Instead, you should run the code in a safe, separate space where it can’t harm your system if something goes wrong.

Why is this important? 🛑 LLM-generated code might have bugs 🐞

The AI isn’t perfect and can write code with mistakes that could crash your system. Security risks 🔓

If the AI accidentally generates malicious code (like deleting files or accessing sensitive data), you don’t want it to touch your real files. Protection from infinite loops or crashes ⏳

Bad code can get stuck running forever or use up all your computer’s power. How do you isolate execution? 🏰 To keep things safe, you can run AI-generated code in a controlled environment, such as:

A Virtual Machine (VM) – A fake computer inside your real computer, which you can reset anytime. A Docker container – A lightweight, temporary environment for running code separately. A Cloud sandbox – A secure online space where code runs without touching your real computer. Restricted Execution Tools (like Pyodide or Firejail) – These limit what the code can do.

no additional imports, consider re

Author response:

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

Recommendations for the authors:

Reviewer #1:

The authors have thoroughly changed the manuscript and addressed most of my concerns. I appreciate adding the activity assays of the C115/120S mutants, however, I suggest that the authors embed and also discuss these data more clearly. It also escaped my attention earlier that the positioning of the disulfide bond is 117-122 in the deposited PDBs instead of 115-120. The authors should carefully check which positioning is correct here.

We thank reviewer #1 for his or her careful assessment of our revised manuscript. As suggested, we detailed the results section “CrSBPase enzymatic activity” with additional numerical values, and discussed more clearly the comparisons of results for activity assays of mutants C115S and C120S in the section “Oligomeric states of CrSBPase”. Residues numbering was carefully proof-checked throughout the manuscript for correctness and homogeneity. C115 and C120 are numbered according to best databases consensus, ie. GenBank and Uniprot, and may differ from one database to another (including PDB) due to varying numbering rules. We clarified the chosen nomenclature in methods section “Cloning and mutagenesis of CrSBPase expression plasmids”.

Line 246-250: I think it is evident that the two SBPase structures superpose well given the sequence identity of more than 70%. However, it would be great to include a superposition of the two structures in Figure 1, especially with regard to the region harboring C115 and C120.

We added a panel showing superimposition of CrSBPase 7b2o and PpSBPase 5iz3 and made a close-up view around the region C115-C120 in supplementary figure 5. Given the density in information of figure 1 we prefer not to add additional images on it. Supplementary figure 5 was initially intended to illustrate sequence conservation/variation among homologs, thus fitting with the objective to compare past and present XRC results.

Line 255-266: I am again missing a panel in Figure 1 here, e.g. a side-by-side view of Xray vs AF2/3 structure.

We added another panel in supplementary figure 5 to visually compare side-by-side SBPase crystallographic structure 7b2o and our AF3 model. Again, for the sake of clarity we prefer not to overload figure 1 with additional panels. This will also enable thorough comparison of past XRC of PpSBPase, present XRC of CrSBPase, and various AF models (see below, oligomer comparisons).

Line 261-266: Did the authors predict dimers and tetramers using AF3? What are the confidence metrics in this case? Do the authors see differences to the monomer prediction in case a multimer is confidently predicted?

We modeled dimers and tetramers using AF3 and added them on supplementary figure 5 side by side with protomer of XRC model 7b2o and with monomer predicted by AF3. Color code for supplementary figure 5 panels F-H is according to AF standard representation of plDDT. Confidence metrics per residue correspond to very high reliability (navy blue) or, locally, confident prediction (cyan) and overall prediction scores range from pTM=0.85-0.91, a high-quality prediction. Interface prediction score is high for both dimer (ipTM=0.9) and tetramer (ipTM=0.82). We reported these data in supplementary figure 5 and corresponding updated legend. XRC and AF models all align with RMSD<0.5 Å, indicating a globally unchanged structure of the protomer in the various methods and oligomeric states.

Line 441: How does the oligomeric equilibrium change in C115/120S mutants? This information should be added for the mutants. Besides, the mAU units in Fig. 6 could be normalized to allow an easier comparison between the chromatograms of wt and mutants.

Change in oligomeric equilibrium is assessed by size-exclusion chromatography of WT and mutants C115S, C120S as reported in figure 6A. We made quantitative estimation of WT, and C115S and C120S mutants equilibrium by comparing maximal peak intensity and added this information in the text. Briefly, the oligomer ratio on a scale of 100 is 9:48:43 for WT, 42:25:33 for mutant C115S, and 29:17:54 for mutant C120S (ratio expressed as tetramer:dimer:monomer). We prefer not to normalize values of absorbance, but rather keep the actual measurement of absorbance at 280 nm on the chromatogram of figure 6, for the sake of consistency with the added text and for a more transparent report of the experiment.

Line 447: WT activity is 12.15+-2.15 and both mutants have a higher activity. The authors should check if their values (96% and 107%) are correct. Besides, did the authors check if the increase in C120S is statistically significant? My impression is that both mutants have a higher activity than the wildtype, in both correlating with increased fractions of the tetramer. This would also make sense, as the corresponding region is part of the tetramer interface in the crystal packing.

The reported activity values were checked for correctness. Wild-type SBPase specific activity at 12.5 ±2.15 µmol(NADPH) min^-1 mg(SBPase)^-1 was obtained by pre-incubating the enzyme with 1 µM CrTRXf2 supplemented with 1 mM DTT and 10 mM Mg²⁺, while the results of supplementary figure 14 reporting the comparison of activation of WT and mutants, with a variation of 107 or 96 %, were obtained with a slightly different protocol for pre-incubation of the enzyme with 10 mM DTT and 10 mM Mg²⁺. Please note that whether WT enzyme was assayed in 10 mM DTT 10 mM Mg or in 1 µM TRX 1 mM DTT 10 mM Mg, its specific activity appears equal within experimental error. Both mutants have nearly the same activity than the WT in the assay reported in supplementary figure 14: we fully agree that 107% (and 96%) variation is indeed not significant considering the uncertainty of the measurement (see error bars representing standard deviations of the mean in supplementary figure 14). We added this important information in the text. Even though both mutations stabilize the most active tetramer in untreated recombinant protein, we think that after reducting treatment both WT and mutants all reach the same maximal activity because they all form an equivalent proportion of the active tetramer versus alternative oligomeric states. We furhter interprete this piece of data as a decoupling of reduction and catalysis: in physiological conditions we assume that SBPase would initiate activation upon the reduction of disulfide bridges, including but not limited to C115-C120 that restricts the entry into fully active tetramer, at which point SBPase in reduced form reaches maximal activity until another post-translational signal eventually changes its conformation and oligomerisation.

We thank again reviewer 1 for his or her assessment and valuable suggestions.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

The authors used a subset of a very large, previously generated 16S dataset to:<br /> (1) Assess age-associated features; and (2) develop a fecal microbiome clock, based on an extensive longitudinal sampling of wild baboons for which near-exact chronological age is known. They further seek to understand deviation from age-expected patterns and uncover if and why some individuals have an older or younger microbiome than expected, and the health and longevity implications of such variation. Overall, the authors compellingly achieved their goals of discovering age-associated microbiome features and developing a fecal microbiome clock. They also showed clear and exciting evidence for sex and rank-associated variation in the pace of gut microbiome aging and impacts of seasonality on microbiome age in females. These data add to a growing understanding of modifiers of the pace of age in primates, and links among different biological indicators of age, with implications for understanding and contextualizing human variation. However, in the current version, there are gaps in the analyses with respect to the social environment, and in comparisons with other biological indicators of age. Despite this, I anticipate this work will be impactful, generate new areas of inquiry, and fuel additional comparative studies.

Thank you for the supportive comments and constructive reviews.

Strengths:

The major strengths of the paper are the size and sampling depth of the study population, including the ability to characterize the social and physical environments, and the application of recent and exciting methods to characterize the microbiome clock. An additional strength was the ability of the authors to compare and contrast the relative age-predictive power of the fecal microbiome clock to other biological methods of age estimation available for the study population (dental wear, blood cell parameters, methylation data). Furthermore, the writing and support materials are clear, informative and visually appealing.

Weaknesses:

It seems clear that more could be done in the area of drawing comparisons among the microbiome clock and other metrics of biological age, given the extensive data available for the study population. It was confusing to see this goal (i.e. "(i) to test whether microbiome age is correlated with other hallmarks of biological age in this population"), listed as a future direction, when the authors began this process here and have the data to do more; it would add to the impact of the paper to see this more extensively developed.

Comparing the microbiome clock to other metrics of biological age in our population is a high priority (these other metrics of biological age are in Table S5 and include epigenetic age measured in blood, the non-invasive physiology and behavior clock (NPB clock), dentine exposure, body mass index, and blood cell counts (Galbany et al. 2011; Altmann et al. 2010; Jayashankar et al. 2003; Weibel et al. 2024; Anderson et al. 2021)). However, we have opted to test these relationships in a separate manuscript. We made this decision because of the complexity of the analytical task: these metrics were not necessarily collected on the same subjects, and when they were, each metric was often measured at a different age for a given animal. Further, two of the metrics (microbiome clock and NPB clock) are measured longitudinally within subjects but on different time scales (the NPB clock is measured annually while microbiome age is measured in individual samples). The other metrics are cross-sectional. Testing the correlations between them will require exploration of how subject inclusion and time scale affect the relationships between metrics.

We now explain the complexity of this analysis in the discussion in lines 447-450. In addition, we have added the NPB clock (Weibel et al. 2024) to the text in lines 260-262 and to Table S5.

An additional weakness of the current set of analyses is that the authors did not explore the impact of current social network connectedness on microbiome parameters, despite the landmark finding from members of this authorship studying the same population that "Social networks predict gut microbiome composition in wild baboons" published here in eLife some years ago. While a mother's social connectedness is included as a parameter of early life adversity, overall the authors focus strongly on social dominance rank, without discussion of that parameter's impact on social network size or directly assessing it.

Thank you for raising this important point, which was not well explained in our manuscript. We find that the signatures of social group membership and social network proximity are only detectable our population for samples collected close in time. All of the samples analyzed in  Tung et al. 2015 (“Social networks predict gut microbiome composition in wild baboons”) were collected within six weeks of each other. By contrast, the data set analyzed here spans 14 years, with very few samples from close social partners collected close in time. Hence, the effects of social group membership and social proximity are weak or undetectable. We described these findings in Grieneisen et al. 2021 and Bjork et al. 2022, and we now explain this logic on line 530, which states, “We did not model individual social network position because prior analyses of this data set find no evidence that close social partners have more similar gut microbiomes, probably because we lack samples from close social partners sampled close in time (Grieneisen et al. 2021; Björk et al. 2022).”

We do find small effects of social group membership, which is included as a random effect in our models of how each microbiome feature is associated with host age (line 529) and our models predicting microbiome Dage (line 606; Table S6).

Reviewer #2 (Public review):

Summary:

Dasari et al present an interesting study investigating the use of 'microbiota age' as an alternative to other measures of 'biological age'. The study provides several curious insights into biological aging. Although 'microbiota age' holds potential as a proxy of biological age, it comes with limitations considering the gut microbial community can be influenced by various non-age related factors, and various age-related stressors may not manifest in changes in the gut microbiota. The work would benefit from a more comprehensive discussion, that includes the limitations of the study and what these mean to the interpretation of the results.

We agree and have text to the discussion that expands on the limitations of this study and what those limitations mean for the interpretation of the results. For instance, lines 395-400 read, “Despite the relative accuracy of the baboon microbiome clock compared to similar clocks in humans, our clock has several limitations. First, the clock’s ability to predict  individual age is lower than for age clocks based on patterns of DNA methylation—both for humans and baboons (Horvath 2013; Marioni et al. 2015; Chen et al. 2016; Binder et al. 2018; Anderson et al. 2021). One reason for this difference may be that gut microbiomes can be influenced by several non-age-related factors, including social group membership, seasonal changes in resource use, and fluctuations in microbial communities in the environment”

In addition, lines 405-411 now reads, “Third, the relationships between potential socio-environmental drivers of biological aging and the resulting biological age predictions were inconsistent. For instance, some sources of early life adversity were linked to old-for-age gut microbiomes (e.g., males born into large social groups), while others were linked to young-for-age microbiomes (e.g., males who experienced maternal social isolation or early life drought), or were unrelated to gut microbiome age (e.g., males who experienced maternal loss; any source of early life adversity in females).”

Strengths:

The dataset this study is based on is impressive, and can reveal various insights into biological ageing and beyond. The analysis implemented is extensive and high-level.

Weaknesses:

The key weakness is the use of microbiota age instead of e.g., DNA-methylation-based epigenetic age as a proxy of biological ageing, for reasons stated in the summary. DNA methylation levels can be measured from faecal samples, and as such epigenetic clocks too can be non-invasive. I will provide authors a list of minor edits to improve the read, to provide more details on Methods, and to make sure study limitations are discussed comprehensively.

Thank you for this point. In response, we have deleted the text from the discussion that stated that non-invasive sampling is an advantage of microbiome clocks. In addition, we now propose a non-invasive epigenetic clock from fecal samples as an important future direction for our population (see line 450).

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Abstract - The opening 2 sentences are not especially original or reflective of the potential value/ premise of the study. Members of this team have themselves measured variation in biological age in many different ways, and the implication that measuring a microbiome clock is easy or straightforward is not compelling. This paper is very interesting and provides unique insight, but I think overall there is a missed opportunity in the abstract to emphasize this, given the innovative science presented here. Furthermore, the last 2 sentences of the abstract are especially interesting - but missing a final statement on the broader significance of research outside of baboons.

We appreciate these comments and have revised the Abstract accordingly. The introductory sentences now read, “Mammalian gut microbiomes are highly dynamic communities that shape and are shaped by host aging, including age-related changes to host immunity, metabolism, and behavior. As such, gut microbial composition may provide valuable information on host biological age.” (lines 31-34). The last two sentences of the abstract now read, “Hence, in our host population, gut microbiome age largely reflects current, as opposed to past, social and environmental conditions, and does not predict the pace of host development or host mortality risk. We add to a growing understanding of how age is reflected in different host phenotypes and what forces modify biological age in primates.” (lines 40-43).

If possible, it would be highly useful to present some comments on concordance in patterns at different levels. Are all ASVs assessed at both the family and genus levels? Do they follow similar patterns when assessed at different levels? What can we learn about the system by looking at different levels of taxonomic assignment?

The section on relationships between host age and individual microbiome features is already lengthy, so we have not added an analysis of concordance between different taxonomic levels. However, we added a justification for why we tested for age signatures in different levels of taxa to line 171, which reads, “We tested these different taxonomic levels in order to learn whether the degree to which coarse and fine-grained designations categories were associated with host age.”

To calculate the delta age - please clarify if this was done at the level of years, as suggested in Figure 3C, or at the level of months or portion months, etc?

Delta age is measured in years. This is now clarified in lines 294, 295, and 578.

Spelling mistake in table S12, cell B4 (Octovber)

Thank you. This typo has been corrected.

Given the start intro with vertebrates, the second paragraph needs some tweaking to be appropriate. Perhaps, "At least among mammals, one valuable marker of biological aging may lie in the composition and dynamics of the mammalian gut microbiome (7-10)." Or simply remove "mammalian".

We have updated this sentence based on your suggestions in line 54. It reads, “In mammals, one valuable marker of biological aging may lie in the composition and dynamics of the gut microbiome (Claesson et al. 2012; Heintz and Mair 2014; O’Toole and Jeffery 2015; Sadoughi et al. 2022).”

A rewrite at the end of the introduction is needed to avoid the almost direct repetition in lines 115-118 and 129-131 (including lit cited). One potentially effective way to approach this is to keep the predictions in the earlier paragraph and then more clearly center the approach and the overarching results statement in the latter paragraph. (I.e., "we find that season and social rank have stronger effects on microbiome age than early life events. Further, microbiome age does not predict host development or mortality.").

Thank you for pointing this out. We have re-organized the predictions in the introduction based on your suggestion. The alternative “recency effects” model now appears in the paragraph that starts in line 110. The final paragraph then centers on the overall approach and the results statement (lines 128-140)

Be clear in each case where taxon-level trends are discussed if it's at Family, Genus, or other level. It's there most, but not all, of the time.

We have gone through the text and clarified what taxa or microbiome feature was the subject of our analyses in any places where this was not clear.

In the legend for Figure 2, add clarification for how values to right versus left of the centered value should be interpreted with respect to age (e.g. "values to x of the center are more abundant in older individuals").

We now clarify in Figure 2C and 2D that “Positive values are more abundant in older hosts”.

Figure 3 - Are Panels A, B, and C all needed - can the value for all individuals not also be overlaid in the panel showing sex differences and the same point showing individuals with "old" and "young" microbiomes be added in the same plot if it was slightly larger?

We agree and have simplified Figure 3. We reduced the number of panels from three to two, and we added the information about how to calculate delta age to Panel A. We also moved the equation from the top of Panel C to the bottom right of Panel A.

Reviewer #2 (Recommendations for the authors):

Dasari et al present an interesting study investigating the use of 'microbiota age' as an alternative to other measures of 'biological age'. The study provides several curious insights which in principle warrant publication. However, I do think the manuscript should be carefully revised. Below I list some minor revisions that should be implemented. Importantly, the authors should discuss in the Discussion the pros and cons of using 'microbiota age' as a proxy of 'biological age'. Further, the authors should provide more information on Methods, to make sure the study can be replicated.

Thank you for these important points. Based on your comments and those of the first reviewer, we have expanded our discussion of the limitations of using microbiota age as a proxy for biological age (see edits to the paragraph starting in line 395).

We have also expanded our methods around sample collection, DNA extraction, and sequencing to describe our sampling methods, strategies to mitigate and address possible contamination, and batch effects. See lines 483-490 and our citations to the original papers where these methods are described in detail.

(1) Lines 85-99: I think this paragraph could be revisited to make the assumptions clearer. For instance, the last sentence is currently a little confusing: are authors expecting males to exhibit old-for-age microbiomes already during the juvenile period?

This prediction has been clarified. Line 96 now reads, “Hence, we predicted that adult male baboons would exhibit gut microbiomes that are old-for-age, compared to adult females (by contrast, we expected no sex effects on microbiome age in juvenile baboons).”

(2) Lines 118-121: Could the authors discuss this assumption in relation to what has been observed e.g., in humans in terms of delays in gut microbiome development? Delayed/accelerated gut microbiome development has been studied before, so this assumption would be stronger if related to what we know from previous studies.

This comment refers to the sentence which originally stated, “However, we also expected that some sources of early life adversity might be linked to young-for-age gut microbiota. For instance, maternal social isolation might delay gut microbiome development due to less frequent microbial exposures from conspecifics.” We have slightly expanded the text here (line 117) to explain our logic. We now include citations for our predictions. We did not include a detailed discussion of prior literature on microbiome development in the interest of keeping the same level of detail across all sections on our predictions.

(3) As the authors discuss, various adversities can lead to old-for-age but also young-for-age microbiome composition. This should be discussed in the limitations.

We agree. This is now discussed in the sentence starting at line 371, which reads, “…deviations from microbiome age predictions are explained by socio-environmental conditions experienced by individual hosts, especially recent conditions, although the effect sizes are small and are not always directionally consistent.” In addition, the text starting at line 405 now reads, “Third, the relationships between potential socio-environmental drivers of biological aging and the resulting biological age predictions were inconsistent. For instance, some sources of early life adversity were linked to old-for-age gut microbiomes (e.g., males born into large social groups), while others were linked to young-for-age microbiomes (e.g., males who experienced maternal social isolation or early life drought), or were unrelated to gut microbiome age (e.g., males who experienced maternal loss; any source of early life adversity in females).”

(4) In various places, e.g., lines 129-131, it is a little unclear at what chronological age authors are expecting microbiota to appear young/old-for-age.

This sentence was removed while responding to the comments from the first reviewer.

(5) Lines 132-133: this statement could be backed by stating that this is because the gut microbiota can change rapidly e.g., when diet changes (or whatever the authors think could be behind this).

We have added an expository sentence at line 123, including new citations. This sentence reads, “Indeed, gut microbiomes are highly dynamic and can change rapidly in response to host diet or other aspects of host physiology, behavior, or environments”.

We now cite:

· Hicks, A.L., et al. (2018). Gut microbiomes of wild great apes fluctuate seasonally in response to diet. Nature Communications 9, 1786.

· Kolodny, O., et al. (2019). Coordinated change at the colony level in fruit bat fur microbiomes through time. Nature Ecology & Evolution 3, 116-124.

· Risely, A., et al. (2021) Diurnal oscillations in gut bacterial load and composition eclipse seasonal and lifetime dynamics in wild meerkats. Nat Commun 12, 6017.

(6) Lines 135-137: current or past season and social rank? This paragraph introduces the idea that it could be past rather than current socio-environmental factors that might predict microbiota age, so the authors should clarify this sentence.

We have clarified the information in this sentence. line 135 now reads, “In general, our results support the idea that a baboon’s current socio-environmental conditions, especially their current social rank and the season of sampling, have stronger effects on microbiome age than early life events—many of which occurred many years prior to sampling.”

(7) Lines 136-137: this sentence could include some kind of a conclusion of this finding. What might this mean?

We have added a sentence at line 138, which speculates that, “…the dynamism of the gut microbiome may often overwhelm and erase early life effects on gut microbiome age.”

(8) Use 'microbiota' or 'microbiome' across the manuscript; currently, the terms are used interchangeably. I don't have a strong opinion on this, although typically 'microbiota' is used when data comes from 16S rRNA.

We have updated the text to replace any instance of “microbiota” with “microbiome”. We use the term microbiome in the sense of this definition from the National Human Genome Research Institute, which defines a microbiome as “the community of microorganisms (such as fungi, bacteria and viruses) that exists in a particular environment”.

(9) Figure 1 legend: make sure to unify formatting; e.g., present sample sizes as N= or n=, rather than both, and either include or do not include commas in 4-digit values (sample sizes).

We have checked the formatting related to sample sizes and the use of commas in 4-digits in the main text and supplement. The formats are now consistent.

(10) Line 166: relative abundances surely?

Following Gloor et al. (2017), our analyses use centered log-ratio (CLR) transformations of read counts, which is the recommended approach for compositional data such as 16S rRNA amplicon read counts. CLR transformations are scale-invariant, so the same ratio is obtained in a sample with few read versus many reads. We now cite Gloor et al. (2017) at line 169 and in the methods in line 517, which reads “centered log ratio (CLR) transformed abundances (i.e., read counts) of each microbial phyla (n=30), family (n=290), genus (n=747), and amplicon sequence variance (ASV) detected in >25% of samples (n=358). CLR transformations are a recommended approach for addressing the compositional nature of 16S rRNA amplicon read count data (Gloor et al. 2017).”  

(11) Lines 167-172: were technical factors, e.g., read depth or sequencing batch, included as random effects?

Thank you for catching this oversight in the text. We did model sequencing depth and batch effects. The sentence starting at line 173 now reads, “For each of these 1,440 features, we tested its association with host age by running linear mixed effects models that included linear and quadratic effects of host age and four other fixed effects: sequencing depth, the season of sample collection (wet or dry), the average maximum temperature for the month prior to sample collection, and the total rainfall in the month prior to sample collection (Grieneisen et al. 2021; Björk et al. 2022; Tung et al. 2015). Baboon identity, social group membership, hydrological year of sampling, and sequencing plate (as a batch effect) were modeled as random effects.”

(12) Lines 175-180: When discussing how these alpha diversity results relate to previous findings, the authors should be clear about whether they talk about weighted or non-weighted measures of alpha diversity. - also maybe this should be included in the discussion rather than the results? Please consider this when revisiting the manuscript (see how it reads after edits).

Richness is the only unweighted metric, which we now clarify in line 181. We opted to retain the interpretation in the text in its original location to maintain the emphasis in the discussion on the microbiome clock results.

(13) Table S1 is very hard to interpret in the provided PDF format as columns are not presented side-by-side. It is currently hard to check model output for e.g., specific families. This needs to be revisited.

We agree. We believe that eLife’s submission portal automatically generates a PDF for any supplementary item. However, we also include the supplementary tables as an Excel workbook which has the columns presented side-by-side.

(14) Line 184: taxa meaning what? Unclear what authors refer to with this sentence, taxa across taxonomic levels, or ASVs, or what does the 51.6% refer to?

We have edited line 191 to clarify that this sentence refers to taxa at all taxonomic levels (phyla to ASVs).

(15) Line 191: a punctuation mark missing after ref (81).

We have added the missing period at the end of this sentence.

(16) Lines 189-197: this should go into the discussion in my opinion.

We have opted to retain this interpretation, now at line 183.

(17) Lines 215-219: Not sure what this means; do the authors mean features were not restricted to age-associated taxa, ie also e.g., diversity and other taxa-independent patterns were included? If so, the rest of the highlighted lines should be revisited to make this clear, currently to me it is very unclear what 'These could include features that are not strongly age-correlated in isolation' means. Currently, that sounds like some features included were only age-associated in combination with other features, but unclear how this relates to taxa-dependency/taxa-independency.

We agree this was not clear. We have revised line 224 to read, “We included all 9,575 microbiome features in our age predictions, as opposed to just those that were statistically significantly associated with age because removing these non-significant features could exclude features that contribute to age prediction via interactions with other taxa.”

(18) Line 403-407: There is now a paper showing epigenetic clocks can be built with faecal samples, so this argument is not valid. Please revisit in light of this publication: https://onlinelibrary.wiley.com/doi/epdf/10.1111/mec.17330

Thank you for bringing this paper to our attention. We deleted the text that describes epigenetic clocks as invasive, and we now cite this paper in line 450, which reads, “We also hope to measure epigenetic age in fecal samples, leveraging methods developed in Hanski et al. 2024.”

(19) Line 427: a punctuation mark/semicolon missing before However.

We have corrected this typo.

(20) Lines 419-428: I don't quite understand this speculation. Why would the priority of access to food lead to an old-looking gut microbiome? This paragraph needs stronger arguments, currently unclear and also not super convincing.

We agree this was confusing. We have revised this text to clarify the explanation. The text starting at line 424 now reads, “This outcome points towards a shared driver of high social status in shaping gut microbiome age in both males and females. While it is difficult to identify a plausible shared driver, one benefit shared by both high-ranking males and females is priority of access to food. This access may result in fewer foraging disruptions and a higher quality, more stable diet. At the same time, prior research in Amboseli suggests that as animals age, their diets become more canalized and less variable (Grieneisen et al. 2021). Hence aging and priority of access to food might both be associated with dietary stability and old-for-age microbiomes. However, this explanation is speculative and more work is needed to understand the relationship between rank and microbiome age.”

(21) Line 434: remove 'be'.

We have corrected this typo.

(22) Line 478: add information on how samples were collected; e.g., were samples collected from the ground? How was cross-contamination with soil microbiota minimised? Were samples taken from the inner part of depositions? These factors can influence microbiota samples quite drastically so detailed info is needed. Also what does homogenisation mean in this context? How soon were samples freeze-dried after sample collection?

We have expanded our methods with respect to sample collection. This text starts in line 483 and reads, “Samples were collected from the ground within 15 minutes of defecation. For each sample, approximately 20 g of feces was collected into a paper cup, homogenized by stirring with a wooden tongue depressor, and a 5 g aliquot of the homogenized sample was transferred to a tube containing 95% ethanol. While a small amount of soil was typically present on the outside of the fecal sample, mammalian feces contains 1000 times the number of microbial cells in a typical soil sample (Sender, Fuchs, and Milo 2016; Raynaud and Nunan 2014), which overwhelms the signal of soil bacteria in our analyses (Grieneisen et al. 2021). Samples were transported from the field in Amboseli to a lab in Nairobi, freeze-dried, and then sifted to remove plant matter prior to long term storage at -80°C.”

(23) Line 480 onwards: were negative controls included in extraction batches? Were samples randomised into extraction batches?

Yes, we included extraction blanks. These are now described in lines 495-500. This text reads, “We included one extraction blank per batch, which had significantly lower DNA concentrations than sample wells (t-test; t=-50, p < 2.2x10-16; Grieneisen et al. 2021). We also included technical replicates, which were the same fecal sample sequenced across multiple extraction and library preparation batches. Technical replicates from different batches clustered with each other rather than with their batch, indicating that true biological differences between samples are larger than batch effects.”

(24) Were extraction, library prep, and sequencing negative controls included? Is data available?

We included extraction blanks (described above) and technical replicates, which were the same sample sequenced across multiple extraction and library preparation batches. Technical replicates from different batches clustered with each other rather than with their batch, indicating that true biological differences between samples are larger than batch effects.

We have updated the data availability statement to read, “All data for these analyses are available on Dryad at https://doi.org/10.5061/dryad.b2rbnzspv. The 16S rRNA gene sequencing data are deposited on EBI-ENA (project ERP119849) and Qiita (study 12949). Code is available at the following GitHub repository: https://github.com/maunadasari/Dasari_etal-GutMicrobiomeAge”.

(25) Line 562: how were corrected microbiome delta ages calculated? Currently, the authors state x, y and z factors were corrected for, but it is unclear how this was done.

The paragraph starting at line 577 describes how microbiome delta age was calculated. We have made only a few changes to this text because we were not sure which aspects of these methods confused the reviewer. However, briefly, we calculated sample-specific microbiome Dage in years as the difference between a sample’s microbial age estimate, age_m from the microbiome clock, and the host’s chronological age in years at the time of sample collection, age_c. Higher microbiome Dages indicate old-for-age microbiomes, as age_m > age_c, and lower values (which are often negative) indicate a young-for-age microbiome, where age_c > age_m (see Figure 3).

(26) Line 579: typo 'as'.

We have corrected this typo.

Works Cited

Altmann, Jeanne, Laurence Gesquiere, Jordi Galbany, Patrick O Onyango, and Susan C Alberts. 2010. “Life History Context of Reproductive Aging in a Wild Primate Model.” Annals of the New York Academy of Sciences 1204:127–38. https://doi.org/10.1111/j.1749-6632.2010.05531.x.

Anderson, Jordan A, Rachel A Johnston, Amanda J Lea, Fernando A Campos, Tawni N Voyles, Mercy Y Akinyi, Susan C Alberts, Elizabeth A Archie, and Jenny Tung. 2021. “High Social Status Males Experience Accelerated Epigenetic Aging in Wild Baboons.” Edited by George H Perry. eLife 10 (April):e66128. https://doi.org/10.7554/eLife.66128.

Binder, Alexandra M., Camila Corvalan, Verónica Mericq, Ana Pereira, José Luis Santos, Steve Horvath, John Shepherd, and Karin B. Michels. 2018. “Faster Ticking Rate of the Epigenetic Clock Is Associated with Faster Pubertal Development in Girls.” Epigenetics 13 (1): 85–94. https://doi.org/10.1080/15592294.2017.1414127.

Björk, Johannes R., Mauna R. Dasari, Kim Roche, Laura Grieneisen, Trevor J. Gould, Jean-Christophe Grenier, Vania Yotova, et al. 2022. “Synchrony and Idiosyncrasy in the Gut Microbiome of Wild Baboons.” Nature Ecology & Evolution, June, 1–10. https://doi.org/10.1038/s41559-022-01773-4.

Chen, Brian H., Riccardo E. Marioni, Elena Colicino, Marjolein J. Peters, Cavin K. Ward-Caviness, Pei-Chien Tsai, Nicholas S. Roetker, et al. 2016. “DNA Methylation-Based Measures of Biological Age: Meta-Analysis Predicting Time to Death.” Aging (Albany NY) 8 (9): 1844–59. https://doi.org/10.18632/aging.101020.

Claesson, Marcus J., Ian B. Jeffery, Susana Conde, Susan E. Power, Eibhlís M. O’Connor, Siobhán Cusack, Hugh M. B. Harris, et al. 2012. “Gut Microbiota Composition Correlates with Diet and Health in the Elderly.” Nature 488 (7410): 178–84. https://doi.org/10.1038/nature11319.

Galbany, Jordi, Jeanne Altmann, Alejandro Pérez-Pérez, and Susan C. Alberts. 2011. “Age and Individual Foraging Behavior Predict Tooth Wear in Amboseli Baboons.” American Journal of Physical Anthropology 144 (1): 51–59. https://doi.org/10.1002/ajpa.21368.

Gloor, Gregory B., Jean M. Macklaim, Vera Pawlowsky-Glahn, and Juan J. Egozcue. 2017. “Microbiome Datasets Are Compositional: And This Is Not Optional.” Frontiers in Microbiology 8. https://doi.org/10.3389/fmicb.2017.02224.

Grieneisen, Laura E., Mauna Dasari, Trevor J. Gould, Johannes R. Björk, Jean-Christophe Grenier, Vania Yotova, David Jansen, et al. 2021. “Gut Microbiome Heritability Is Nearly Universal but Environmentally Contingent.” Science 373 (6551): 181–86. https://doi.org/10.1126/science.aba5483.

Hanski, Eveliina, Susan Joseph, Aura Raulo, Klara M. Wanelik, Áine O’Toole, Sarah C. L. Knowles, and Tom J. Little. 2024. “Epigenetic Age Estimation of Wild Mice Using Faecal Samples.” Molecular Ecology 33 (8): e17330. https://doi.org/10.1111/mec.17330.

Heintz, Caroline, and William Mair. 2014. “You Are What You Host: Microbiome Modulation of the Aging Process.” Cell 156 (3): 408–11. http://dx.doi.org/10.1016/j.cell.2014.01.025.

Horvath, Steve. 2013. “DNA Methylation Age of Human Tissues and Cell Types.” Genome Biology 14 (10): R115. https://doi.org/10.1186/gb-2013-14-10-r115.

Jayashankar, Lakshmi, Kathleen M. Brasky, John A. Ward, and Roberta Attanasio. 2003. “Lymphocyte Modulation in a Baboon Model of Immunosenescence.” Clinical and Vaccine Immunology 10 (5): 870–75. https://doi.org/10.1128/CDLI.10.5.870-875.2003.

Marioni, Riccardo E., Sonia Shah, Allan F. McRae, Brian H. Chen, Elena Colicino, Sarah E. Harris, Jude Gibson, et al. 2015. “DNA Methylation Age of Blood Predicts All-Cause Mortality in Later Life.” Genome Biology 16 (1): 25. https://doi.org/10.1186/s13059-015-0584-6.

O’Toole, Paul W., and Ian B. Jeffery. 2015. “Gut Microbiota and Aging.” Science 350 (6265): 1214–15. https://doi.org/10.1126/science.aac8469.

Raynaud, Xavier, and Naoise Nunan. 2014. “Spatial Ecology of Bacteria at the Microscale in Soil.” PLOS ONE 9 (1): e87217. https://doi.org/10.1371/journal.pone.0087217.

Sadoughi, Baptiste, Dominik Schneider, Rolf Daniel, Oliver Schülke, and Julia Ostner. 2022. “Aging Gut Microbiota of Wild Macaques Are Equally Diverse, Less Stable, but Progressively Personalized.” Microbiome 10 (1): 95. https://doi.org/10.1186/s40168-022-01283-2.

Sender, Ron, Shai Fuchs, and Ron Milo. 2016. “Revised Estimates for the Number of Human and Bacteria Cells in the Body.” PLoS Biology 14 (8): e1002533. https://doi.org/10.1371/journal.pbio.1002533.

Tung, J, L B Barreiro, M B Burns, J C Grenier, J Lynch, L E Grieneisen, J Altmann, S C Alberts, R Blekhman, and E A Archie. 2015. “Social Networks Predict Gut Microbiome Composition in Wild Baboons.” Elife 4. https://doi.org/10.7554/eLife.05224.

Weibel, Chelsea J., Mauna R. Dasari, David A. Jansen, Laurence R. Gesquiere, Raphael S. Mututua, J. Kinyua Warutere, Long’ida I. Siodi, Susan C. Alberts, Jenny Tung, and Elizabeth A. Archie. 2024. “Using Non-Invasive Behavioral and Physiological Data to Measure Biological Age in Wild Baboons.” GeroScience 46 (5): 4059–74. https://doi.org/10.1007/s11357-024-01157-5.

the colours used to colour code this section are all very similar - I had a hard time differentiating between the shades of yellow and green - to make this more accessible, it would be better to use more contrasting and differentiated colours

An evangelical saying that what the church is telling guys is "just wrong"?!? This requires some self-reflection on the part of the speaker...

I'm curious what is the right thing in his framing?

Humans without community and thus without a sense of moral responsibility are evil creatures. Also justice decides what is just is an interesting statement. Using the prior sentence, we can say that justice is the upholding of the community's moral code. Justice is what is necessary to maintain the community.

també hi ha la help aquí (ho dic perquè és algo que parlem en totes les sessions)

Afegiria que és on surt l'output del que writes en el source

Do we want to expand this section with what we planned out in our brainstorming RE how we want to handle code review based on the plan for individual SMEs for each project/repo but also having shared repos?

The benefit to ssh-agent is that you only need to enter your passphrase once. If your private RSA key is not encrypted with a passphrase, then ssh-agent is not necessary

The irony

Reviewer #1 (Public review):

This work derives a general theory of optimal gain modulation in neural populations. It demonstrates that population homeostasis is a consequence of optimal modulation for information maximization with noisy neurons. The developed theory is then applied to the distributed distributional code (DDC) model of the primary visual cortex to demonstrate that homeostatic DDCs can account for stimulus-specific adaptation.

What I consider to be the most important contribution of this work is the unification of efficient information transmission in neural populations with population homeostasis. The former is an established theoretical framework, and the latter is a well-known empirical phenomenon - the relationship between them has never been fully clarified. I consider this work to be an interesting and relevant step in that direction.

The theory proposed in the paper is rigorous and the analysis is thorough. The manuscript begins with a general mathematical setting to identify normative solutions to the problem of information maximization. It then gradually builds towards questions about approximate solutions, neural implementation and plausibility of these solutions, applications of the theory to specific models of neural computation (DDC), and finally comparisons to experimental data in V1. Such a connection of different levels of abstraction is an obvious strength of this work.

Overall I find this contribution interesting and assess it positively. At the same time, I have three major points of criticism, which I believe the authors should address. I list them below, followed by a number of more specific comments and feedback.

Major comments:

(1) Interpretation of key results and relationship between different parts of the manuscript. The manuscript begins with an information-transmission ansatz which is described as "independent of the computational goal" (e.g. p. 17). While information theory indeed is not concerned with what quantity is being encoded (e.g. whether it is sensory periphery or hippocampus), the goal of the studied system is to *transmit* the largest amount of bits about the input in the presence of noise. In my view, this does not make the proposed framework "independent of the computational goal". Furthermore, the derived theory is then applied to a DDC model which proposes a very specific solution to inference problems. The relationship between information transmission and inference is deep and nuanced. Because the writing is very dense, it is quite hard to understand how the information transmission framework developed in the first part applies to the inference problem. How does the neural coding diagram in Figure 3 map onto the inference diagram in Figure 10? How does the problem of information transmission under constraints from the first part of the manuscript become an inference problem with DDCs? I am certain that authors have good answers to these questions - but they should be explained much better.

(2) Clarity of writing for an interdisciplinary audience. I do not believe that in its current form, the manuscript is accessible to a broader, interdisciplinary audience such as eLife readers. The writing is very dense and technical, which I believe unnecessarily obscures the key results of this study.

(3) Positioning within the context of the field and relationship to prior work. While the proposed theory is interesting and timely, the manuscript omits multiple closely related results which in my view should be discussed in relationship to the current work. In particular:

A number of recent studies propose normative criteria for gain modulation in populations:

- Duong, L., Simoncelli, E., Chklovskii, D. and Lipshutz, D., 2024. Adaptive whitening with fast gain modulation and slow synaptic plasticity. Advances in Neural Information Processing Systems<br /> - Tring, E., Dipoppa, M. and Ringach, D.L., 2023. A power law describes the magnitude of adaptation in neural populations of primary visual cortex. Nature Communications, 14(1), p.8366.<br /> - Młynarski, W. and Tkačik, G., 2022. Efficient coding theory of dynamic attentional modulation. PLoS Biology<br /> - Haimerl, C., Ruff, D.A., Cohen, M.R., Savin, C. and Simoncelli, E.P., 2023. Targeted V1 co-modulation supports task-adaptive sensory decisions. Nature Communications<br /> - The Ganguli and Simoncelli framework has been extended to a multivariate case and analyzed for a generalized class of error measures:<br /> - Yerxa, T.E., Kee, E., DeWeese, M.R. and Cooper, E.A., 2020. Efficient sensory coding of multidimensional stimuli. PLoS Computational Biology<br /> - Wang, Z., Stocker, A.A. and Lee, D.D., 2016. Efficient neural codes that minimize LP reconstruction error. Neural Computation, 28(12),

More detailed comments and feedback:

(1) I believe that this work offers the possibility to address an important question about novelty responses in the cortex (e.g. Homann et al, 2021 PNAS). Are they encoding novelty per-se, or are they inefficient responses of a not-yet-adapted population? Perhaps it's worth speculating about.

(2) Clustering in populations - typically in efficient coding studies, tuning curve distributions are a consequence of input statistics, constraints, and optimality criteria. Here the authors introduce randomly perturbed curves for each cluster - how to interpret that in light of the efficient coding theory? This links to a more general aspect of this work - it does not specify how to find optimal tuning curves, just how to modulate them (already addressed in the discussion).

(3) Figure 8 - where do Hz come from as physical units? As I understand there are no physical units in simulations.

(4) Inference with DDCs in changing environments. To perform efficient inference in a dynamically changing environment (as considered here), an ideal observer needs some form of posterior-prior updating. Where does that enter here?

(5) Page 6 - "We did this in such a way that, for all ν, the correlation matrices, ρ(ν), were derived from covariance matrices with a 1/n power-law eigenspectrum (i.e., the ranked eigenvalues of the covariance matrix fall off inversely with their rank), in line with the findings of Stringer et al. (2019) in the primary visual cortex." This is a very specific assumption, taken from a study of a specific brain region - how does it relate to the generality of the approach?

Reviewer #2 (Public review):

Summary:

Using the theory of efficient coding, the authors study how neural gains may be adjusted to optimize coding by noisy neural populations while minimizing metabolic costs. The manuscript first presents mathematical results for the general case where the computational goals of the neural population are not specified (the computation is implicit in the assumed tuning curves) and then develops the theory for a specific probabilistic coding scheme. The general theory provides an explanation for firing rate homeostasis at the level of neural clusters with firing rate heterogeneity within clusters, and the specific application further captures stimulus-specific and neuron-specific adaptation in the visual cortex.

The mathematical derivations, simulations, and application to visual cortex data are solid as far as I can tell.

In the current format, the significance is difficult to assess fully: the manuscript is a bit sprawling, in the first half the general theory is lengthy and technical, and then in the second half a few phenomena are addressed without a clear relation between them (rate homeostasis, rate heterogeneity, synaptic homeostasis, V1 adaptation, divisive normalization), requiring several ad-hoc choices and assumptions.

Strengths:

The problem of efficient coding is a long-standing and important one. This manuscript contributes to that field by proposing a theory of efficient coding through gain adjustments, independent of the computational goals of the system. The main result is a normative explanation for firing rate homeostasis at the level of neural clusters (groups of neurons that perform a similar computation) with firing rate heterogeneity within each cluster. Both phenomena are widely observed, and reconciling them under one theory is important.

The mathematical derivations are thorough as far as I can tell. Although the model of neural activity is artificial, the authors make sure to include many aspects of cortical physiology, while also keeping the models quite general.

Section 2.5 derives the conditions in which homeostasis would be near-optimal in the cortex, which appear to be consistent with many empirical observations in V1. This indicates that homeostasis in V1 might be indeed close to the optimal solution to code efficiently in the face of noise.

The application to the data of Benucci et al 2013 is the first to offer a normative explanation of stimulus-specific and neuron-specific adaptation in V1.

Weaknesses:

The novelty and significance of the work are not presented clearly. The relation to other theoretical work, particularly Ganguli and Simoncelli and other efficient coding theories, is explained in the Discussion but perhaps would be better placed in the Introduction, to motivate some of the many choices of the mathematical models used here.

The manuscript is very hard to read as is, it almost feels like this could be two different papers. The first half seems like a standalone document, detailing the general theory with interesting results on homeostasis and optimal coding. The second half, from Section 2.7 on, presents a series of specific applications that appear somewhat disconnected, are not very clearly motivated nor pursued in-depth, and require ad-hoc assumptions.

For instance, it is unclear if the main significant finding is the role of homeostasis in the general theory or the demonstration that homeostatic DDC with Bayes Ratio coding captures V1 adaptation phenomena. It would be helpful to clarify if this is being proposed as a new/better computational model of V1 compared to other existing models.

Early on in the manuscript (Section 2.1), the theory is presented as general in terms of the stimulus dimensionality and brain area, but then it is only demonstrated for orientation coding in V1.

The manuscript relies on a specific response noise model, with arbitrary tuning curves. Using a population model with arbitrary tuning curves and noise covariance matrix, as the basis for a study of coding optimality, is problematic because not all combinations of tuning curves and covariances are achievable by neural circuits (e.g. https://pubmed.ncbi.nlm.nih.gov/27145916/ )

The paper Benucci et al 2013 shows that homeostasis holds for some stimulus distributions, but not others i.e. when the 'adapter' is present too often. This manuscript, like the Benucci paper, discards those datasets. But from a theoretical standpoint, it seems important to consider why that would be the case, and if it can be predicted by the theory proposed here.

This work has been peer reviewed in GigaScience (https://doi.org/10.1093/gigascience/giaf007), which carries out open, named peer-review. These reviews are published under a CC-BY 4.0 license and were as follows:

Reviewer #2: Anuradha Wickramarachchi

Overall comments.

Authors of the manuscript have developed an iterative overlap graph construction algorithm to support genome assembly. This is both an interesting and a demanding area of research due to very recent advancements in sequencing technologies.

Although the text in the manuscript is interesting, grammar must be rechecked and revised. At some point it is difficult to keep track of the content and references to supplementary to make sense out of the content.

Specific comments

Page 1 Line 13: I believe the authors are talking about assembly sizes and not genome sizes. The sentences here could be a bit short to make them easy to understand.

Page 2 Line 19: Theoretical time complexity O(m2n2) is bit of an overstatement due to the heuristics employed by most assemblers. For example, mash distance, minimisers and k-mer bins are there to prevent this explosion of complexity. Either acknowledge such methods or provide a range for the time complexity. I would be interesting to know the time complexities of the methods expressed in sentence starting Line 15.

Page 5 Line 11: Was this performed with overlapping windows of 1gb? Otherwise, simulations may not have reads spanning across such regions.

Page 5 Line 14: It seems you are simulating 9 + 4 + 4 datasets. This is unclear, please make this into bullet points or separate paragraphs and explain clearly. Include simulator information in the table itself by may be making it landscape (in supplementary).

Fig 2: I believe authors should expand their analysis to more recent and popular assemblers. For example, wtdbg2 is designed for noisy reads and not specifically for more accurate R10/ HiFi reads. So please include, HiFi-asm, Flye where appropriate. Flye supports ONT out of the box and in my experience does produce good assemblies.

Although, you are evaluating read overlaps, it is hard to ignore assemblers themselves just because they do not produce intermediate overlaps graphs.

Page 5-9: In the benchmarks section, please include how True Positives and False Positives were labelled. Was this from simulation data?

Page 11: Use of xRead has been evaluated on genome assemblies. This is a very important and it is a bit unfortunate that existing assemblers are not very flexible in terms of plugging in new intermediate steps. It might be worth exploring into creating a new assembler using the wtpoa2 cli command of wtdbg2.

Page 16: What will happen if you only capture reads from a single chromosome due to longer length? I believe the objective is to gather longest reads capturing as much as possible covering the whole genome. Please comment on this.

Page 19: In the Github Readme the download URL was wrong. Please correct it to the latest release

Correct: https://github.com/tcKong47/xRead/releases/download/xRead-v1.0.0.1/xRead-v1.0.0.tar.gz Existing: https://github.com/tcKong47/xRead/releases/download/v1.0.0/xRead-v1.0.0.tar.gz

Make command failed with make: *** No rule to make target main.h', needed bymain.o'. Stop.

It seems the release does not have source code, but rather the compiled version. Please update github instructing how to compile code properly with a git clone.

This work has been peer reviewed in GigaScience (https://doi.org/10.1093/gigascience/giaf008), which carries out open, named peer-review. These reviews are published under a CC-BY 4.0 license and were as follows:

Reviewer 2: Markus Müller

The authors present TransHLA, a deep learning tool to predict whether a peptide is an HLA binder or not. They use the ESM2 language model to create peptide embeddings for structural and sequence features and then use transformers and CNNs for the binding prediction. The article is well-written and clear. However, the authors must better justify the choice of their model and its potential application.

Major comments:

1) In personalized medicine, the HLA alleles of a patient can be obtained via WES and there is no need for such a HLA agnostic binding predictor. Could you briefly outline the most important medical applications where your TransHLA predictor could be most useful?

2) Could you give more information about your IEDB training set? What are the frequencies of the HLA alleles, and the number of peptides per allele? How did you perform the splits into training, validation, and test sets? Were peptides from the same allele all present in all 3 sets? How does TransHLA perform for peptides binding to alleles not present in the training set compared to peptides binding to alleles present in the training set? How does the performance depend on the number of peptides of the allele in the training set? Is the model biased to these frequent alleles?

3) Peptides are processed by many steps before being presented on HLA molecules. These include cleavage in the proteasome, transport via TAP to the ER, cleavage by ERADs and finally loading on the HLA complex. Why don't you perform your study on extended peptide sequences, where you take into account several amino acids before and after the peptide termini? Like this, you could also include the other processing steps. It would be interesting to see whether this sequence extension would improve prediction.

4) Could you compare your approach with a 'simpler' approach, where you calculate all biopython features (such as flexibility), ev. choose the n most informative ones by feature selection, and use a standard classifier such as logistic regression or XGBoost to predict the HLA binding. This method has the advantage that it tells you directly which features are most relevant.

5) Please provide the results of the ablation study in a table in the main text, where you compare the ablated models to the base model.

6) Could you briefly explain what the different terms in the TIM loss are and why they are important?

7) Does the flexibility depend on the length of the peptides? Peptides longer than 10 often bulge out of the binding groove, and naively one would expect them to be less stiff than peptides of length 8 or 9.

Minor:

1) In Equation 10, please define ^p_k. In the text, you use T for the number of classes, in the formulae K.

analytical method: example, breaking down code napoleon into sort of philosophical concepts and conceptaul frameworks

This can be applied back to a couple different ethical implications as the trust that is needed in this case to be truthful to physical therapists is important. Along with that the rule of law is something to pay attention to as well. Minding everything in your report to be accurate to physical therapy is crucial.

This one is very interesting to me because reading the code of ethics is easy. understanding and following them is different and not everyone honors or understand it. What is expected of you in your field should be small part of professional field. what makes you professional is understanding youre job is the education and learning time that goes into it before getting hired.

another example of data shown in a photograph

Deep software engineering

YOOO US AND CANADA TWINNING LEGAL FAMS

This is so different compared to the other societies that we have studied throughout this course. It is interesting that during the code of Hammurabi times, it required contract and receipts for every purchase, and then this was based more on tradition than concrete contracts.

Teaching some or all of these skills is undoubtedly a time-consuming task. However, many of them are potentially crucial future skills that we cannot simply assume students will fully acquire through mere exposure.

The code creates an ArrayList<integer> called myList and adds the integers 50, 30, and 20 to it.

It then computes the sum of all the elements in myList using an enhanced for loop and prints it out with the message "Sum of all elements: ".

The output will be the sum of 50, 30, and 20, which is 100.</integer>

To make the code work, we need to correctly set the found variable to true when the name is found and removed, and false when the name is not found after the loop completes.

Fix: Set found = true; when a match is found and the name is removed. Initially set found = false; before starting the loop, because we assume the name is not found unless we find it.

You need to change the loop condition to i < myList.size() to ensure that the loop stops before accessing an index that is out of bounds.

Need to explain any code you put in

I don't understand why this isn't just because of the state of the program rather than the interactions.

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 comments

• *

Major comments:

• *

- Neither data nor code was made available for review. There's only a mention of them being in Figshare with no link. As a consequence and a matter of principle, this study is not publishable without both public data and code. I would recommend using adequate repositories for data and code. Image data can be deposited in a public image data repository such as the BioImage Archive which would ensure that minimal metadata are provided and code could go to a public code repository (e.g. GitLab...) so that it is discoverable and eventual changes can be tracked and visible (for example should any bug be fixed after publication). Also consider depositing the models into the BioImage Model Zoo (https://bioimage.io).

• *

We will upload all the code used in the article in GitHub while image data will be deposited in BioImage Archive as suggested by the referee. Method section will be also rewritten.

• *

- The use of the term morphology is misleading. Like I expect most readers would, I understand morphology in this context as being related to shape. However, there is no indication that any specific type of information (like shape, texture, size/scale...) is used or learned by the described method. To understand what information the classifiers rely on, it would be interesting to compare with human engineered features extracted from the same ROIs.

All references to morphology in the text must be removed unless indication can be provided as to what type of information is used by the models.

• *

We understand the concern regarding the use of "morphology" and will revise the manuscript to be more precise. Instead of referring broadly to "morphology," we will specify "image-derived features" or "texture and structural features" where applicable.

Additionally, to address this concern directly, we have performed an analysis comparing our learned features to classical human-engineered features (such as texture and shape descriptors) to better understand what type of information is utilized by the model. These results will be incorporated into the revised manuscript.

• *

- The method should be described with more details:

- How are the window sizes to use determined? Are the two sizes listed in the methods section used simultaneously? What is the effect of this parameter on the performance?

- How are the ROIs determined? In a grid pattern? Do they overlap? i.e. how does the windowing function work?

- Predictions seem to be made at the ROI level but it isn't clear if this is always the case. Can inference be made at the level of individual cells?

• *

Window Sizes: We will clarify that the two window sizes were chosen based on empirical performance assessments. We will include a specific figure evaluating the impact of window size on classification performance, by expanding the analysis to multiple window sizes and number of training regions.

ROI Determination: We will describe thoroughly the ROI selection in the Method Section. We will include a comparison between overlapped and non-overlapping grid selection.

Inference at the Cell Level: While predictions are made at the ROI level (we will clarify the text), we will discuss an additional approach that aggregates ROI-level predictions into a final cell-level classification, which we will add as an optional post-processing step.

• *

- What would be the advantages of the proposed subcellular approach compared to learning to classify whole images?

• *

We will detail a comparison between subcellular and cellular or whole image classification; the main advantage of this subcellular technique (that will be remarked in the text) is the reduction in the number of images that are required to learn to classify cell types. Nevertheless, other advantages are the robustness to confluency variations (whole-image classification can be biased by confluency differences, while subcellular regions focus on individual cell features) and a fine-grained feature learning.

• *

- When fluorescent markers are used, the text isn't clear on what measures have been taken to prevent these markers from bleeding through into the brightfield image. To rule out the possibility that the models learn from bleed-through of the marker into the brightfield image, the staining should be performed after the brightfield image acquisition. Without this, conclusions of the related experiments are fatally flawed.

• *

__We appreciate this important point and confirm that all fluorescent staining was performed after brightfield image acquisition, ensuring that no fluorescence contamination influenced model training. We will have explicitly stated this in the Methods section. __

• *

- How robust are the models e.g. with respect to culture age and batch effects? Use of a different microscope is mentioned in the methods section. This should be shown, i.e. can a model trained on one microscope accurately predict on data acquired from a different microscope? Does mixing images from different sources for training improve robustness?

• *

We have used different cellular batches without any effect on accuracy. We will also include the experiment using another microscope, and we will add new data with/without combination of mixed images from different figures. In summary, we include a new supplementary figure that address the use of distinct and mixed cellular batches and microscopies in terms of accuracy and trained models.

• *

- Why not use the Mahalanobis distance in feature space? This would be the natural choice given that PCA has been selected for visualization and would allow to show uncertainty regions in the PCA plots. Could other dimensionality reduction methods show better separation of the groups? Why not train the network for further dimensionality reduction if the goal is to learn a useful feature space?

We appreciate this suggestion and will include a comparison of Mahalanobis distance-based classification with our existing approach. Regarding dimensionality reduction, we will test additional methods including t-SNE and UMAP as supplementary figures. Finally, while training a network specifically for dimensionality reduction is an interesting alternative, our current pipeline was focused on simplicity and the ample range of techniques that allow to address. However, we include include a discussion on potential future directions where such an approach could be explored.

Minor comments:

• *

- Make sure the language used is clear, e.g. The text describes the method as involving a transformation to black and white followed by thresholding. This doesn't make sense. What is meant by "the set of 300 genes was subjected to Gene Ontology"? Use percent instead of permille in the text for easier reading.

These minor changes will be addressed in the text, including the percent instead of permille as it was a common point suggested by the referees.

• *

- To provide more context, cite previous work that indicates that brightfield images contain exploitable information, e.g.

- Cross-Zamirski, J.O., Mouchet, E., Williams, G. et al. Label-free prediction of cell painting from brightfield images. Sci Rep 12, 10001 (2022). https://doi.org/10.1038/s41598-022-12914-x

- Harrison PJ, Gupta A, Rietdijk J, Wieslander H, Carreras-Puigvert J, et al. (2023) Evaluating the utility of brightfield image data for mechanism of action prediction. PLOS Computational Biology 19(7): e1011323. https://doi.org/10.1371/journal.pcbi.1011323

• *

We will cite these references in the introduction of the paper.

Reviewer #2

• *

Major comments:

• *

• • Place this study in context of previous studies that classify cell types. Here are two relevant recent papers, which could provide a good start for properly crediting previous work and placing your contribution in context: PMID: 39819559 (note the "Nucleocentric" approach) and PMID: 38837346. Please seek for papers that use label free for similar applications (which is the main contribution of the current manuscript).*
• *

We appreciate this suggestion (shared by reviewer #1) and we will include references to these and other relevant studies on label-free cell classification. We specifically discuss how our approach differs from the "nucleocentric" method in PMID: 39819559 and how our method complements existing work in label-free imaging. We will update both the Introduction and Discussion sections to reflect this improved contextualization.

• *

• Many experiments were performed, but we found it hard to follow them and the logic behind each experiment. Please include a Table summarizing the experiments and their full statistics (see below) and also please provide more comprehensive justifications for designing these specific experiments and regarding the experimental details. This will make the reading more fluent.*

• *

We will include a summary table in the Methods section that provides an overview of all experiments, detailing:

-The purpose of each experiment

-The dataset used

-The number of images/cells

-Objective used

-Cellular confluence

-Reference to BioImage Archive

-Model used (reference to Github)

-Technical / Biological replicates

-The main conclusions drawn

-Figure that presents the data

Additionally, we will revise the Results section to provide clearer justifications for each experiment, improving the logical flow of the manuscript.

• *

• The experiments, data acquisition and data reporting details are lacking. 10x objective is reported in the Results and 20x in the Methods. Please explain how the co-culturing (mixed) experiments were performed including co-culturing experiments with varying fractions of each cell type and on what data were the models trained on (Fig. 2F). Differential confluency experiments are not described in the Methods (and not on what confluency levels were the models trained on), this is also true for the detachment experiment. How many cells were acquired in each experiment (it says "20 and 40 images per cell line" but this is a wide range + it is not clear how many cells appear in each image)? How many biological/technical replicates were performed for each experiment? Please report these for each experiment in the corresponding figure legend and show the results on replicates (can be included as Supplementary). "Using a different microscope with the same objective produced similar results (data not shown)" (lines #370-371), please report these results (including what is the "different microscope") in the SI.*

• *

We will carefully review and expand the Methods section to provide complete details, as with the Table that we will prepare to address the previous comment and this one. In addition, the co-culturing experiments will explicitly describe how cell fractions were varied and how training data were generated for Fig. 2F. The differential confluency and detachment experiments will be fully described, including confluency levels used during model training. The secondary microscopy data will be added as part of a new figure that was commented for reviewer #1.

• *

• The machine learning details are lacking. The train-validation-test strategy is not described, which could be critical in excluding concerns for data leakage (e.g., batch effects) which could be a major concern in this study. It is not always clear what network architecture was used. What were the parameters used for training? Accuracy is reported in % (and sometimes in an awkward representation, 990‰). Proper evaluation will use measurements that are not sensitive to unbalanced data (e.g., ROC-AUC). What are the controls (i.e., could the accuracy reported be by chance?). Reporting accuracy at the pixel/patch level and not at the cell level is a weakness. Estimation of cell numbers (in methods) is helpful but I did not see when it was used in the Results - a better alternative is using fluorescent nuclear markers to move to a cell level (not necessary to implement if it was not imaged).*

• *

We will significantly expand the Machine Learning method and result sections, providing:

-A detailed description of the train-validation-test split strategy, (that explicitly rules out batch effects as a confounding factor). A clarification of the network architecture used for different tasks and their parameters (always the same one).

-We will expand the evaluation metrics, including ROC-AUC scores to account for class imbalances, and baseline models as controls, ensuring that model performance is not due to chance as a new supplementary figure.

- Accuracy will be reported to use percentage instead of permille as suggested by other referees.

- We will clarify the use of cell number estimation in the specific figures in which we use it, including new data in the first figure for the generalization of patch-to-cell estimation.

• *

• Downstream analyses lacking sufficient information to enable us to follow and interpret the results, please provide more information.*

• The PCA ellipses visualizations reference to previous papers. Please explain what was done, how the ellipses were calculated and from how much data? If they are computed from a small number of data points - please show the actual data. It would also be useful to briefly include the information regarding the representation and dimensionality reduction in the Results and not only in the Methods. No biologically-meaningful interpretation is provided - perhaps providing cell images along the PCs projections can help interpret what are the features that distinguish between different experimental conditions.*

• *

We will include a clearer explanation in methods as well as results for PCA and dimensionality reduction, as well as the use of Mahalanobis distance as another metric, another visualization for improved interpretation, and a supplementary figure related to tSNE reduction. We will update the figure for inclusion of real subcelullar images that help the biological interpretation of the results.

• *

• • How were the pairwise accuracies calculated? How did the authors avoid potential batch effects driving classification.*
• *

We have used different cellular batches without any effect on accuracy. In the new revised manuscript, we will clarify batch normalization techniques used in training and include additional control analyses ensuring that batch effects are not driving classification results (new figure as suggested by reviewer #1 with mixed and separate cellular batches).

• *

• • "suggesting that the current workflow can handle four cell lines simultaneously" (lines #126-127) - how were the cell lines determined for each analysis? We assume that the performance will depend on the cell types (e.g., two similar morphology cell types will be hard to distinguish). Fig. 2F is not clear: the legend should report a mixture of four cell types, and this should be translated to clear visualization in the figure panel itself: what do the data points mean? Where are the different cell types?*
• *

We will include additional experiments with other cell lines, and we will explicitly describe the rationale for cell line pairings, considering morphological similarities. Fig. 2F will be redesigned for clarity, ensuring data points are clearly labeled by cell type.

• *

• • Lines 232 and onwards use #pixels as a subcellular size measurement when referring to cell nucleus, cytoplasm and membrane, please report the actual physical size and show specific examples of these patches. This visualization and analysis of patch sizes should appear much earlier in the manuscript because it relates to the method's robustness and interpretability.*
• *

We will explicitly report patch sizes in microns and include a supplementary figure illustrating different subcellular regions to enhance interpretability.

• *

• • Analysis of co-cultured (mixed) experiments is not clear. Was the fluorescent marker used to define ground truth? Was the model trained and evaluated on co-cultures or trained on cultures of a single cell type and evaluated on mixed cultures? We assume that the models were still evaluated on the label-free data? "...obtain subcellular ROIs only from regions positive in the red channel. Using these labeled ROIs,.." (138-139) - shouldn't both positive and negative ROIs be used to have both cell types? What are the two quantifications in the bottom of Fig. 1E? Did the "labeled cells" trained another classifier for the fluorescent labels?*
• *

We will clarify both the method and results section regarding the co-culture experiment from the first figure. In that specific case, the model learned from positive ROIs in order to demonstrate that this approach can also be used from a mixed culture. In order to become clearer, we will transfer this experiment to a supplementary figure.

• *

• • Please interpret the results from Fig. 3C-D - should we expect to see passage-related changes in cells (that lead to deterioration in classification) or is it a limitation of the current study?*
• *

We will explicitly discuss whether passage-related changes affect cell morphology. In addition, we will include novel RNA-seq data comparing passage and batch effects, in order to correlate them to the image-based deterioration as part of the figure.

• *

• • In general, as we mentioned a couple of times. It would be useful to visualize different predictions (or use explainability methods such as GradCam) to try to interpret what the model has learned.*
• *

We will perform a GradCAM analysis, highlighting which subcellular regions contribute most to classification, improving interpretability.

• *

• • The correlation analysis between transcriptional profiles and morphological profiles is not clear. There are not sufficient details to follow the genetic algorithm (and its justification). What was the control for this analysis? Would shuffling the cells' labels (identities) and repeating the analysis will not yield a correlation?*
• *

We agree with the concern of the reviewer. We will expand the Methods section to clarify how the correlation was calculated, as well as the genetic algorithm. We will perform a control analysis using shuffled cell identities, trying to demonstrate that correlations do not arise by chance.

• *

• Please use proper scientific terms. For example, "white-light microscopy" and "live cell red marker".*

• *

We will change the text accordingly, making a global review of the manuscript.

• *

• This is a "Methods" manuscript and thus should open the source code and data, along with some examples on how to use it in order to enable others to replicate the results and to enable others to use it.*

• *

We acknowledge that our manuscript is more a ‘Methods’ manuscript instead of a general article (that it was conceptualized by us). Probably most of the critical points arose by the referees at the end are explained by this reason. We will deposit image data in the BioImage Archive with proper metadata, and we will published our code in GitHub as well as the models.

• *

• Please improve the figures. Fonts are tiny and in some places even clipped (e.g., Fig. 1D,E, Fig.2 E, E', and many more), some labels are missing (e.g., units of the color bar in Fig. 1B).*

• *

Figures will be redesigned accordingly.

• *

• Discussion. Please place this work in context of other studies that tackled a similar challenge of classifying cell types and discuss cons and pros of the different measurements. For example, there are clear benefits of using label-free data to reduce the number of fluorescent labels and enable long-term live cell imaging following a process without photobleaching and phototoxicity (Fig. 2G) but it is more difficult to interpret these differences in label-free image patches rather than fluorescently labeled single cells. One solution to bridge this gap that could be discussed is using silico labeling (PMID: 38838549).*

• *

The Discussion will be significantly expanded to compare our work with other methods, including in silico labeling (PMID: 38838549).

• *

• The idea of using the pairwise correlation distance of different cell types to model unseen cell types is interesting and promising. Why did these specific pairwise networks were used? How robust is this representation to inclusion of other/additional models?*

• *

As the referees are very interested in pairwise correlation distance, we will include a sensitivity analysis, testing alternative model selections to assess robustness.

Reviewer #3

• *

## General

• *

- It is often unclear if a sample in the particular experiment is a patch or a pixel. Please be more specific on this in the text.

• *

Manuscript will be rewritten for clarification of pixel/patch.

• *

- It is unclear which patch size was used and if it was consistent throughout the experiments. Please add this information.

• *

We will include a new figure with comparison between different patch sizes, as suggested by reviewer #1.

• *

- It is often unclear which data was used for training/validation and final readout. Did you do train/val splits? Did you predict on the same data or new samples? This should be stated more specifically.

• *

We will clarify in the Methods section the strategy of training/testing (90% - 10%, same data) with new samples used for final readout. All reported classification results come from that set, ensuring that the model was evaluated on unseen data.

• *

- Also, it is a little bit unclear what you mean by patch or by ROI or by region, please be more consistent and explain what you mean by adding definitions.

• *

We will standardize the use of these terms, leaving only ROI.

• *

- Please compare your method to other approaches and to baselines (see also our comment above).

• *

We will compare our approach with whole-image classification, showing that our subcellular approach provides better generalization. A new supplementary analysis will explore the feasibility of alternative feature extraction techniques and their relative performance. Several baselines will be incorporated in order to assess random accuracies (following the suggestions of other reviewers).

• *

- In general, if possible, please add more concrete examples of how you envision your method to be used in practice. There are general ideas presented in the discussion section, but we feel those could be substantiated by more concrete implementation suggestions.

• *

We will provide three specific case studies in the Discussion section, demonstrating how our approach can be applied in real-world scenarios:

-Drug Screening: Identifying cellular responses to drug treatments in high-throughput screening pipelines.

-Stem Cell Differentiation Monitoring: Tracking changes in subcellular morphology during differentiation to assess developmental stages.

-Cancer Cell Classification: Distinguishing between different subtypes of cancer cells in heterogeneous populations.

• *

Minor comments (grouped and summarized for clarification):

• *

General Clarifications & Wording Improvements

• *

Line 18: Clarify if the study is based on morphological features and specify the novelty (e.g., subcellular features).

Lines 25 & 29: The wording suggests that the workflow was extended before being validated. Improve clarity.

Line 92: Add a brief explanation of "subcellular region."

• *

We will clarify in the Introduction that our study is based on morphological features but specifically focuses on subcellular features, which distinguishes it from whole-cell analysis. We will rephrase the relevant sentences to make it clear that the workflow was first validated and then extended. We will provide a brief definition of "subcellular region" and ensure consistency throughout the manuscript.

• *

Experimental Setup & Methodological Details

• *

Lines 100-141: Clarify the use of validation and test sets, and discuss potential batch effects.

Line 113: Missing training details (loss function, data volume, epochs).

Line 117: Clarify if "pairwise classification" is meant.

Line 119: Accuracy should be reported in percent instead of permille.

Lines 136-141: Justify why two cell lines were mixed but only one was analyzed.

• *

We will add a clear explanation of the train-validation-test split, ensuring reproducibility and ruling out batch effects. Additional batch effect control experiments will be performed and included in Supplementary Figures as suggested by other reviewers.

We will include training details (e.g., loss function, number of epochs, data volume) in the Methods section and referenced it in the Results section for clarity. The terminology will be updated to "pairwise classification" where appropriate. We will report accuracy in percent (%) as suggested by other reviewers. The rationale for mixing two cell lines but analyzing only one is now explicitly stated: we used a mixed culture to simulate realistic conditions but focused on one cell type to test classification specificity. Nevertheless, following other reviewer suggestion this experiment will be placed in a supplementary figure in order to become clearer.

• *

Technical & Experimental Design Clarifications

• *

Line 105: Replace "white light microscopy" with "brightfield microscopy."

Line 107: Be specific about "transformation to black and white" and "contrast thresholding algorithm."

Line 125: Explain why performance dropped—did you try a larger network?

Line 133: Clarify how confluency was estimated.

• *

"White light microscopy" will be replaced with "brightfield microscopy." The thresholding method will be explicitly described, with a reference to the Methods section where details are provided. We will discuss the possible reasons for performance drop. Confluency estimation will be described, explaining that it was calculated using automated image segmentation and validated manually.

• *

Data Representation & Interpretation

• *

Line 143-158: Clarify the ground truth—was it based on dye labeling, thresholding, or human annotation?

Line 156: What is meant by "magnification"? Higher resolution? Different microscope? Crops?

Lines 163-166: Sudden switch to pixels instead of ROIs—explain why.

Line 191 & 192: If a strong correlation is claimed, include a statistical test.

Lines 211-214: If differences are claimed, add a quantitative analysis.

Lines 396-404: Clarify how the test set was chosen and what "in situ prediction" means.

Lines 407-409: What do you mean by "binarizing the image"? What threshold was used?

• *

We will clearly explain terms like “ground truth”, "magnification", “in situ prediction” and ‘binarization”. Consistent terminology will be ensured, regarding ROIs throughout the text. Statistical analyses will be added to correlation results and morphological feature comparisons to support claims.

• *

Biological Interpretation & Feature Space Analysis

• *

Line 226-228: You show classification in feature space but not whether distances in feature space correlate with real-world differences between cell types.

Line 234-236: What do you mean by "detect potentially more informative subcellular regions"?

Line 302-303: The claimed application (estimating cell types in an unseen culture) was not shown—please add an experiment.

• *

We now include an experiment comparing three cell types, where two are closely related and one is more distinct, to test if feature space distance corresponds to real-world differences. The concept of "informative subcellular regions" will be rephrased. We will add an experiment demonstrating the ability of our model to estimate the number of cell types in an unseen culture, as suggested.

• *

Figure & Visualization Improvements

• *

• Improve figure readability (tiny fonts, clipped text).*

Line 653-655: Show actual data points in PCA ellipses, not just ellipses.

Line 672-677: Add a quantification of performance differences between different categories.

• *

All figures will be revised for better readability, ensuring that text is legible, axes are labeled, and color bars are clear. We will overlay data points onto PCA ellipses for better visualization of feature distribution, as suggested by other reviewers. Performance differences between different experimental conditions will be quantified, with statistical comparisons provided.

• *

Model Training & Data Reproducibility

• *

Lines 386-392: Add exact details on model architecture, loss function, number of images used per experiment.

• *

A complete breakdown of model architecture, loss function, training set size, and validation details will be included in the Methods section, ensuring full reproducibility.

Dimensionality Reduction & Feature Space Interpretation

• *

Line 438-439: Consider using UMAP or t-SNE in addition to PCA. Report variance explained by PCA components.

Line 439-440: Provide more details on how eigenvectors were used to calculate ellipses.

Line 442-443: Clarify which correlation method was used.

• *

We will include t-SNE visualizations in Supplementary Figures and report the variance explained by PCA components, as well as Mahalanobis distance, as suggested by other reviewers. The eigenvector-based ellipse calculation will be described in more detail in the Methods section, and the specific correlation metric used will be explicitly stated.

• *

Code & Data Accessibility

• *

Line 491: Provide a direct URL to the code and data. Consider using GitHub for code and BioImage Archive for data.

• *

We will include the code to GitHub and image data to the BioImage Archive, following the reviewers recommendation, with direct URLs.

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 #3

Evidence, reproducibility and clarity

Summary

The authors present a computational workflow that automatically classifies patches of transmission microscopy images of cultured cells into different cell types.

Comments to the Manuscript

General

• It is often unclear if a sample in the particular experiment is a patch or a pixel. Please be more specific on this in the text.
• It is unclear which patch size was used and if it was consistent throughout the experiments. Please add this information.
• It is often unclear which data was used for training/validation and final readout. Did you do train/val splits? Did you predict on the same data or new samples? This should be stated more specifically.
• Also, it is a little bit unclear what you mean by patch or by ROI or by region, please be more consistent and explain what you mean by adding definitions.
• Please compare your method to other approaches and to baselines (see also our comment above).
• In general, if possible, please add more concrete examples of how you envision your method to be used in practice. There are general ideas presented in the discussion section, but we feel those could be substantiated by more concrete implementation suggestions.

Specific

Line 18

• Isn't this study also based on morphological features? Eventually, you could be more specific what the novelty is, it might be the fact that your features are subcellular?

Lines 25 & 29

• In general one would expect a workflow to be validated first and extended afterwards. You could improve the wording here to make this clear for the reader.

Line 92

• Please add a short explanation of what is meant by "subcellular region".

Lines 100-141

• Did you validate the classification results with a validation and test set? Maybe with cross validation? Please add more details on how this was done.
• It could be that the model exploits batch effects of different imaging runs (e.g. different overall intensity in patches). It would be nice if this could be checked by an additional experiment.

Line 105

• "white light microscopy" is an unusual term, can you be more specific, e.g. bright-field?

Line 107

• It is unclear what a "transformation to black and white" and a "contrast thresholding algorithm" are, please be more specific (and potentially point the reader to a corresponding Methods section).

Line 113

• How does training work? Which loss is used? How much data? How many epochs? ... All of this information is missing which makes the study non-reproducible. Please add this here or point to an appropriate method section.

Line 117

• Do you mean pairwise classification?

Line 119

• It is unusual to use permille as a unit to report, percent is more common
• Also, it is unclear if accuracy is the correct read-out here, are all the data sets balanced?
• more information about the data sets could be added in a methods section and the decision to use accuracy as a measure could be explained

Line 121

• this hypothesis was never stated before, please explain this to the reader first and then check your hypothesis by experiments

Line 125

• do you have a hypothesis why the performance dropped? Did you for example try a larger network?

Line 133

• how is the confluence estimated?

Lines 136-141

• It is unclear why two cell lines were mixed when only data of one of them is used for analysis afterwards. Could you explain this in more detail or specify why this approach is used?

Lines 143-158

• We think you are trying to establish a ground truth here. Unfortunately, there are two things mixed here, the labeling with an additional dye combined with thresholding and human annotation. It is unclear which is considered the ground truth or if both are considered true. Could you explain this in more detail or be more specific?

Line 156

• What do you mean by magnification? Images with a higher resolution (different microscope with higher magnification)? Crops of the same data? Something else? Could you explain this in more detail or be more specific?

Lines 163-166

• Suddenly you talk about pixels instead of ROIs, where are they coming from? Maybe point the reader to a method section and explain the switch here.
• Also, why is the pixel size cell line dependent, didn't you use the same microscope for all of them? Could you define what you mean by pixel size?
• You say you compared different cell lines, how is this summarized in one plot? Please explain in more detail.

Lines 177-214

• Again, it is unclear which data was used for training, validation and analysis in the end. Please add this.

Lines 191 & 192

• If you claim a strong correlation please add a statistical analysis that shows this.

Lines 211-214

• If you claim these differences you should add a quantitative read-out with a statistical analysis. You could use distances in your representation space as a basis for this.

Lines 226-228

• What is shown here is that the morphological features can be used to classify cell types. You show that these classes are distant in feature space. But you don't show any correlation between the distance in feature space and the distance in real space (a.k.a how different the cell types are). It would be nice to have an experiment with at least three classes where 2 are closer to each other than to the third one. This would be a stronger claim that your features actually capture meaningful distances/differences.

Lines 234-236

• What do you mean by "detect potentially more informative subcellular regions within the cell"? Please describe in more detail what the training task was for the model and how you interpret the results.

Lines 296-298

• It is a little bit confusing what you mean here since you do train a network for each pair of cell lines. What you are describing is a foundation model. Please explain in more detail what you mean.

Lines 302-303

• The application you are claiming here was never shown in the experiments. Could you please add this experiment where a model estimates the number of cell types in an unseen culture.

Line 323

• Could you please elaborate how you would identify "specific cellular compartments"?

Lines 323-326

• Are there other studies that suggest that such malignant cells show features that are recognizable by your approach?

Lines 342-365

• Did you use biological replicates? This would be interesting and also a nice way to validate your models.

Lines 369-373

• Why do you claim that a similar microscope produces similar images? Can you give more details why this is relevant. And if that is the case it would be nice to show them. Maybe in some supplementary material.
• How big is one image? How many cells can you see in one image? What is the resolution? What is the pixel size? ... Also, for which experiment did you use how many images? Please add all these details.
• Also, please show some example images to make it clear for the reader what the data looks like. Could be done in supplementary material.

Lines 386-392

• Again, please add details. As it is right now the study is not reproducible. How many images were used for each experiment? How many for training, validation, analysis? Give the exact architecture of the model used. Which loss was used for training?

Lines 396-404

• Please add more details and clarify. How was the test set chosen? What do you mean by "in situ prediction"? What do you mean by "running ROIs"? What do you mean by "if the cell type was predicted to be more than 50 % of the times"? Was the human annotation or the life cell marker used for the final accuracy? Humans are never unbiased.

Lines 404-407

• This sounds like the ground truth for a segmentation task - is this what you mean? Since you are solving a classification task this is confusing. Please clarify.

Lines 407-409

• This sentence is confusing and it is unclear what was done. Please clarify. Do you mean the image was binarized? If yes, which threshold was used? What do you mean by "accuracy was estimated as with the prediction"? The accuracy should be estimated by comparing the prediction to the ground truth.

Lines 413-422

• Please give more details. What are these specific numbers? What do you mean by "pixel size of each cell type"? The pixel size is metadata given by the microscope/image and should not be cell type specific. We also did not understand what is meant by "fitting the percentages" and what the aim of this is. Please consider rewriting this to make it more clear.

Lines 426-430

• Please provide the oligo sequence.

Line 435

• Please consider rephrasing to: "the output of the last max pooling layer"

Lines 438-439

• It would be interesting to visualize the data based on a different dimensionality reduction algorithm that is non-linear like UMAP or t-SNE. If you use PCA, could you give a measure on how much of the variance is captured in the first two PCs.

Lines 439-440

• Please give some more details on how you use eigenvectors to calculate ellipses.

Lines 442-443

• Please give more details on which correlation you calculated.

Lines 447-457

• It would be nice if you could rephrase this a little bit to make clear that the preprocessing itself stays the same but you basically establish different data sets by separating ROIs based on their distance to the closest nucleus.

Lines 455-457

• Please be more precise here. The networks still learn to classify patches and are not aware of the fact that these ROIs fall in a certain category. You exploit this fact afterwards for your analysis.

Lines 464-474

• Please add more details why this experiment is done. Why is a genetic algorithm needed? Could not the same analysis be done on the original transcriptomics data?

Line 486

• Do you mean technical or biological replicates? If that is the case, could you please clearly state that you report mean values and also give the standard deviation.
• "test" should be experiment

Line 491

• Could you please provide a URL to the code and the data.
• Also, it is common practice to upload code to GitHub and image data to the Bioimage Archive. Please consider doing this.

Lines 627-633

• Panel A could be improved by making the ROIs larger since it is hard to see them.
• Also, please make sure that it is clear that one ROI at a time is given to the model.

Line 638

• What does "magnification" mean here - see above.
• Why do you not show the same region?

Line 640-642

• This basically shows that your approach is as good as simple thresholding. What do you want to show with this?

Lines 643-644

• Please clarify. It is unclear what percentage you present here.

Lines 652-653 (Fig. 3C)

• Please clarify. It is unclear what statistical analysis was performed here and to what end.

Lines 653-655

• It would be interesting to see not only the ellipses but also the actual data points plotted.

LInes 658-661

• Please add a statistical analysis of what you want to show here.
• It is clear that the correlation is not as clear for higher values on the x-axis, why is this?

Lines 661-662

• Please clarify. It is unclear what statistical analysis was performed here.

Lines 662-664

• Please add a statistical analysis of what you want to show here.

Lines 672-677

• Please also plot the actual data points
• Also, if possible it would be nice to quantify the differences in performance between the different categories.

Code and data availability

We could not see how to access example image data. To our best knowledge it is current best practice to upload image data to the Bioimage Archive: https://www.ebi.ac.uk/bioimage-archive/

Specifically for this kind of study the reader should have access to the training and test data that was used to train the classifier.

We also could not see how to reproduce the analysis. To our best knowledge it is current best practice to make all code publicly accessible, e.g. in a GitHub repository.

Please see https://www.nature.com/articles/s41592-023-01987-9 for general guidelines of publishing bioimage data and analysis.

Significance

The ability to use label free microscopy for extracting biologically meaningful information is very valuable and it is very interesting to learn that simple transmission microscopy contains enough information to reveal cell types. In this study the authors trained a neural network for this task and demonstrated that it works with rather high accuracy.

In its current form, we could not access the data nor the code. We could thus not fully judge the quality of the presented work. For a future revision, access to data and code will be essential.

We also found it difficult to judge how difficult the classification task is, because the size of the cells in the current figures does not allow one to see texture detail in the images. Since we did not manage to access the image data, we could not assess whether the classification task is very hard (and indeed requires an AI approach) or whether the differences are rather obvious and could be quantified with classical image analysis. To enable the interested reader to better assess this important information we would like to recommend to (a) add figures that allow one to better see the cells and their texture, at least for some of the cell types, and (b) provide easy download access to the raw image data.

Along those lines, we think it would be very interesting to actually test whether training a neural network is required or whether other methods would yield similar results. For instance, we would recommend to simply compute the mean and variance of the intensities in each patch and check whether this information also can perform some of the classification tasks. Depending on the outcome of this analysis this could be either added to some of the main figures of the article or to the supplemental material.

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:

Automatic classification of single cell types and cell states in heterogeneous mixed cell populations has many applications in cell biology and screening. The authors present a machine learning workflow to distinguish between different cell types or cell states from label-free microscopy image patches of subcellular size. The authors evaluate their ability to identify different cell types and molecular profiles on many applications.

Major comments:

The application of classifying cell type and states from label-free data is promising and useful, but this manuscript requires major rewriting to enable us comprehensive assessment. Specifically, provide all technical details necessary for its evaluation, improve clarity and justification for the methodology used and the results obtained, and to better place this study in context of other studies in the field. Two crucial points are excluding the concern of the possibility that batch effects are contributing to the classification results and providing stronger evidence for a link between transcriptional and morphological profiles. Some efforts to interpret the classification decision making could help understand what morphological information was used for classification and reduce the concerns for the model using non-biologically meaningful information for the classification (e.g., illumination changes due to batch effects). Finally, making the source code and data publicly available would be important to enable others to apply the method (code) and to benchmark other methods (data).

1. Place this study in context of previous studies that classify cell types. Here are two relevant recent papers, which could provide a good start for properly crediting previous work and placing your contribution in context: PMID: 39819559 (note the "Nucleocentric" approach) and PMID: 38837346. Please seek for papers that use label free for similar applications (which is the main contribution of the current manuscript).
2. Many experiments were performed, but we found it hard to follow them and the logic behind each experiment. Please include a Table summarizing the experiments and their full statistics (see below) and also please provide more comprehensive justifications for designing these specific experiments and regarding the experimental details. This will make the reading more fluent.
3. The experiments, data acquisition and data reporting details are lacking. 10x objective is reported in the Results and 20x in the Methods. Please explain how the co-culturing (mixed) experiments were performed including co-culturing experiments with varying fractions of each cell type and on what data were the models trained on (Fig. 2F). Differential confluency experiments are not described in the Methods (and not on what confluency levels were the models trained on), this is also true for the detachment experiment. How many cells were acquired in each experiment (it says "20 and 40 images per cell line" but this is a wide range + it is not clear how many cells appear in each image)? How many biological/technical replicates were performed for each experiment? Please report these for each experiment in the corresponding figure legend and show the results on replicates (can be included as Supplementary). "Using a different microscope with the same objective produced similar results (data not shown)" (lines #370-371), please report these results (including what is the "different microscope") in the SI.
4. The machine learning details are lacking. The train-validation-test strategy is not described, which could be critical in excluding concerns for data leakage (e.g., batch effects) which could be a major concern in this study. It is not always clear what network architecture was used. What were the parameters used for training? Accuracy is reported in % (and sometimes in an awkward representation, 990‰). Proper evaluation will use measurements that are not sensitive to unbalanced data (e.g., ROC-AUC). What are the controls (i.e., could the accuracy reported be by chance?). Reporting accuracy at the pixel/patch level and not at the cell level is a weakness. Estimation of cell numbers (in methods) is helpful but I did not see when it was used in the Results - a better alternative is using fluorescent nuclear markers to move to a cell level (not necessary to implement if it was not imaged).
5. Downstream analyses lacking sufficient information to enable us to follow and interpret the results, please provide more information.

a. The PCA ellipses visualizations reference to previous papers. Please explain what was done, how the ellipses were calculated and from how much data? If they are computed from a small number of data points - please show the actual data. It would also be useful to briefly include the information regarding the representation and dimensionality reduction in the Results and not only in the Methods. No biologically-meaningful interpretation is provided - perhaps providing cell images along the PCs projections can help interpret what are the features that distinguish between different experimental conditions.

b. How were the pairwise accuracies calculated? How did the authors avoid potential batch effects driving classification.

c. "suggesting that the current workflow can handle four cell lines simultaneously" (lines #126-127) - how were the cell lines determined for each analysis? We assume that the performance will depend on the cell types (e.g., two similar morphology cell types will be hard to distinguish). Fig. 2F is not clear: the legend should report a mixture of four cell types, and this should be translated to clear visualization in the figure panel itself: what do the data points mean? Where are the different cell types?

d. Lines 232 and onwards use #pixels as a subcellular size measurement when referring to cell nucleus, cytoplasm and membrane, please report the actual physical size and show specific examples of these patches. This visualization and analysis of patch sizes should appear much earlier in the manuscript because it relates to the method's robustness and interpretability.

e. Analysis of co-cultured (mixed) experiments is not clear. Was the fluorescent marker used to define ground truth? Was the model trained and evaluated on co-cultures or trained on cultures of a single cell type and evaluated on mixed cultures? We assume that the models were still evaluated on the label-free data? "...obtain subcellular ROIs only from regions positive in the red channel. Using these labeled ROIs,.." (138-139) - shouldn't both positive and negative ROIs be used to have both cell types? What are the two quantifications in the bottom of Fig. 1E? Did the "labeled cells" trained another classifier for the fluorescent labels?

f. Please interpret the results from Fig. 3C-D - should we expect to see passage-related changes in cells (that lead to deterioration in classification) or is it a limitation of the current study?

g. In general, as we mentioned a couple of times. It would be useful to visualize different predictions (or use explainability methods such as GradCam) to try to interpret what the model has learned.

h. The correlation analysis between transcriptional profiles and morphological profiles is not clear. There are not sufficient details to follow the genetic algorithm (and its justification). What was the control for this analysis? Would shuffling the cells' labels (identities) and repeating the analysis will not yield a correlation? 6. Please use proper scientific terms. For example, "white-light microscopy" and "live cell red marker". 7. This is a "Methods" manuscript and thus should open the source code and data, along with some examples on how to use it in order to enable others to replicate the results and to enable others to use it. 8. Please improve the figures. Fonts are tiny and in some places even clipped (e.g., Fig. 1D,E, Fig.2 E, E', and many more), some labels are missing (e.g., units of the color bar in Fig. 1B). 9. Discussion. Please place this work in context of other studies that tackled a similar challenge of classifying cell types and discuss cons and pros of the different measurements. For example, there are clear benefits of using label-free data to reduce the number of fluorescent labels and enable long-term live cell imaging following a process without photobleaching and phototoxicity (Fig. 2G) but it is more difficult to interpret these differences in label-free image patches rather than fluorescently labeled single cells. One solution to bridge this gap that could be discussed is using silico labeling (PMID: 38838549).<br /> 10. The idea of using the pairwise correlation distance of different cell types to model unseen cell types is interesting and promising. Why did these specific pairwise networks were used? How robust is this representation to inclusion of other/additional models?

Significance

Automated classification of cell types and cell states in mixed cell populations using label-free images has important applications in academic research and in industry (e.g., cell profiling). This paper applies standard machine learning toward this technical goal, and demonstrates it on many different experimental systems, exceeding the common standard in terms of quantity and variability, and with the potential of being a nice contribution to the field. However, we were not able to properly evaluate these results due to lacking experimental and methodological details as detailed above and thus can not make a strong point regarding validity and significance before a major revision. Our expertise is in computational biology, and specifically applications of machine learning in microscopy. We are not familiar with the specific cell types, states and perturbations used in this manuscript.

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

Summary:

This paper presents a method to classify cells in brightfield images using information from subcellular regions. The approach consists in first thresholding a brightfield image then splitting the resulting binary image into small ROIs which are then fed to a CNN-based classifier. The authors demonstrate application to the identification of cell types in pure cultures and in cultures with mixed types. They then show that features learned by the classifier correlate with expression of cell type-specific genes and explore what information can be learned from networks trained on subcellular regions selected based on distance from the nucleus. The authors conclude that subcellular ROIs extracted from brightfield images contain useful information about the identity and state of the cells in the image.

Major comments:

• Neither data nor code was made available for review. There's only a mention of them being in Figshare with no link. As a consequence and a matter of principle, this study is not publishable without both public data and code.
• I would recommend using adequate repositories for data and code. Image data can be deposited in a public image data repository such as the BioImage Archive which would ensure that minimal metadata are provided and code could go to a public code repository (e.g. GitLab...) so that it is discoverable and eventual changes can be tracked and visible (for example should any bug be fixed after publication). Also consider depositing the models into the BioImage Model Zoo (https://bioimage.io).
• The use of the term morphology is misleading. Like I expect most readers would, I understand morphology in this context as being related to shape. However, there is no indication that any specific type of information (like shape, texture, size/scale...) is used or learned by the described method. To understand what information the classifiers rely on, it would be interesting to compare with human engineered features extracted from the same ROIs. All references to morphology in the text must be removed unless indication can be provided as to what type of information is used by the models.
• The method should be described with more details:
• How are the window sizes to use determined? Are the two sizes listed in the methods section used simultaneously? What is the effect of this parameter on the performance?
• How are the ROIs determined? In a grid pattern? Do they overlap? i.e. how does the windowing function work?
• Predictions seem to be made at the ROI level but it isn't clear if this is always the case. Can inference be made at the level of individual cells?
• What would be the advantages of the proposed subcellular approach compared to learning to classify whole images?
• When fluorescent markers are used, the text isn't clear on what measures have been taken to prevent these markers from bleeding through into the brightfield image. To rule out the possibility that the models learn from bleed-through of the marker into the brightfield image, the staining should be performed after the brightfield image acquisition. Without this, conclusions of the related experiments are fatally flawed.
• How robust are the models e.g. with respect to culture age and batch effects? Use of a different microscope is mentioned in the methods section. This should be shown, i.e. can a model trained on one microscope accurately predict on data acquired from a different microscope? Does mixing images from different sources for training improve robustness?
• Why not use the Mahalanobis distance in feature space? This would be the natural choice given that PCA has been selected for visualization and would allow to show uncertainty regions in the PCA plots. Could other dimensionality reduction methods show better separation of the groups? Why not train the network for further dimensionality reduction if the goal is to learn a useful feature space?

Minor comments:

• Make sure the language used is clear, e.g.
• The text describes the method as involving a transformation to black and white followed by thresholding. This doesn't make sense.
• What is meant by "the set of 300 genes was subjected to Gene Ontology"?
• Use percent instead of permille in the text for easier reading.
• To provide more context, cite previous work that indicates that brightfield images contain exploitable information, e.g.
• Cross-Zamirski, J.O., Mouchet, E., Williams, G. et al. Label-free prediction of cell painting from brightfield images. Sci Rep 12, 10001 (2022). https://doi.org/10.1038/s41598-022-12914-x
• Harrison PJ, Gupta A, Rietdijk J, Wieslander H, Carreras-Puigvert J, et al. (2023) Evaluating the utility of brightfield image data for mechanism of action prediction. PLOS Computational Biology 19(7): e1011323. https://doi.org/10.1371/journal.pcbi.1011323

Referees cross-commenting

I support comments from reviewers 2 and 3 around the lack of sufficient details fro interpretability and reproducibility. Some of the necessary information could be communicated through well documented re-usable code and computational workflows as well as properly documented data sets.

Jean-Karim Hériché (heriche@embl.de)

Significance

This is an interesting study that adds to a growing body of evidence showing that information contained in brightfield images can be usefully exploited, potentially replacing the expensive and time-consuming use of fluorescent markers and is therefore of interest to a broad audience of cell biologists.

Even though my comments elsewhere mention NPM, etc., I myself don't think they quite fit to what McIlroy (and Cox) were talking about. If I recall correctly, both mention configuring components in ways that are not generally seen with modern packages. McIlroy in particular talks about parameterization that suggests to me compile-time (or pre-compile-time) configuration, whereas the junk on NPM traditionally blobs all the behavior together and selects behavior at program runtime.

See also: the trend of an overabundance of triple equals operators (===) in NPM programmers' code.

It's important when working for any organization or company to review their code of ethics because they will vary from company to company.

Try to be objective and not subjective

i recommend you keep the game jam project since hackathon projects check off a lot of benchmarks. collaboratively working on code, working in fast-paced environment, playing different roles in a team , etc...

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

This study reports that spatial frequency representation can predict category coding in the inferior temporal cortex.

Thank you for taking the time to review our manuscript. We greatly appreciate your valuable feedback and constructive comments, which have been instrumental in improving the quality and clarity of our work.

The original conclusion was based on likely problematic stimulus timing (33 ms which was too brief). Now the authors claim that they also have a different set of data on the basis of longer stimulus duration (200 ms).

One big issue in the original report was that the experiments used a stimulus duration that was too brief and could have weakened the effects of high spatial frequencies and confounded the conclusions. Now the authors provided a new set of data on the basis of a longer stimulus duration and made the claim that the conclusions are unchanged. These new data and the data in the original report were collected at the same time as the authors report.

The authors may provide an explanation why they performed the same experiments using two stimulus durations and only reported one data set with the brief duration. They may also explain why they opted not to mention in the original report the existence of another data set with a different stimulus duration, which would otherwise have certainly strengthened their main conclusions.

Thank you for your comments regarding the stimulus duration used in our experiments. We appreciate the opportunity to clarify and provide further details on our methodology and decisions.

In our original report, we focused on the early phase of the neuronal response, which is less affected by the duration of the stimulus. Observations from our data showed that certain neurons exhibited high firing rates even with the brief 33 ms stimulus duration, and the results we obtained were consistent across different durations. To avoid redundancy, we initially chose not to include the results from the 200 ms stimulus duration, as they reiterated the findings of the 33 ms duration.

However, we acknowledge that the brief stimulus duration could raise concerns regarding the robustness of our conclusions, particularly concerning the effects of high spatial frequencies. Upon reflecting on the reviewer’s comments during the first revision, we recognized the importance of addressing these potential concerns directly. Therefore, we have included the data from the 200 ms stimulus duration in our revised manuscript.

Furthermore, Our team is actively investigating the differences between fast (33 ms) and slow (200 ms) presentations in terms of SF processing. Our preliminary observations suggest similar processing of HSF in the early phase of the response for both fast and slow presentations, but different processing of HSF in the late phase. This was another reason we initially opted to publish the results from the brief stimulus duration separately, as we intended to explore the different aspects of SF processing in fast and slow presentations in subsequent studies.

I suggest the authors upload both data sets and analyzing codes, so that the claim could be easily examined by interested readers.

Thank you for your suggestion to make both data sets and the analyzing codes available for examination by interested readers.

We have created a repository that includes a sample of the dataset along with the necessary codes to output the main results. While we cannot provide the entire dataset at this time due to ongoing investigations by our team, we are committed to ensuring transparency and reproducibility. The data and code samples we have provided should enable interested readers to verify our claims and understand our analysis process.

Repository: https://github.com/ramintoosi/spatial-frequency-selectivity

Reviewer #2 (Public Review):

Summary:

This paper aimed to examine the spatial frequency selectivity of macaque inferotemporal (IT) neurons and its relation to category selectivity. The authors suggest in the present study that some IT neurons show a sensitivity for the spatial frequency of scrambled images. Their report suggests a shift in preferred spatial frequency during the response, from low to high spatial frequencies. This agrees with a coarse-to-fine processing strategy, which is in line with multiple studies in the early visual cortex. In addition, they report that the selectivity for faces and objects, relative to scrambled stimuli, depends on the spatial frequency tuning of the neurons.

Strengths:

Previous studies using human fMRI and psychophysics studied the contribution of different spatial frequency bands to object recognition, but as pointed out by the authors little is known about the spatial frequency selectivity of single IT neurons. This study addresses this gap and shows spatial frequency selectivity in IT for scrambled stimuli that drive the neurons poorly. They related this weak spatial frequency selectivity to category selectivity, but these findings are premature given the low number of stimuli they employed to assess category selectivity.

Thank you for your thorough review and insightful feedback on our manuscript. We greatly appreciate your time and effort in providing valuable comments and suggestions, which have significantly contributed to enhancing the quality of our work.

The authors revised their manuscript and provided some clarifications regarding their experimental design and data analysis. They responded to most of my comments but I find that some issues were not fully or poorly addressed. The new data they provided confirmed my concern about low responses to their scrambled stimuli. Thus, this paper shows spatial frequency selectivity in IT for scrambled stimuli that drive the neurons poorly (see main comments below). They related this (weak) spatial frequency selectivity to category selectivity, but these findings are premature given the low number of stimuli to assess category selectivity.

While we acknowledge that the number of instances per condition is relatively low, the overall dataset is substantial. Specifically, our study includes a total of 180 stimuli (6 spatial frequencies × 2 scrambled/non-scrambled conditions × 15 instances, including 9 fixed and 6 non-fixed) and 5400 trials (180 stimuli × 2 durations × 15 repetitions). Conducting these trials requires approximately one hour of experimental time per session.

Extending the number of stimuli, while potentially addressing this limitation, would significantly compromise the quality of the experiment by increasing the duration and introducing potential fatigue effects in the subjects. Despite this limitation, our findings lay important groundwork by offering novel insights into object recognition through the lens of spatial frequency. We believe this work can serve as a foundation for future experiments designed to further explore and validate these theories with expanded stimulus sets.

Main points.

(1) They have provided now the responses of their neurons in spikes/s and present a distribution of the raw responses in a new Figure. These data suggest that their scrambled stimuli were driving the neurons rather poorly and thus it is unclear how well their findings will generalize to more effective stimuli. Indeed, the mean net firing rate to their scrambled stimuli was very low: about 3 spikes/s. How much can one conclude when the stimuli are driving the recorded neurons that poorly? Also, the new Figure 2- Appendix 1 shows that the mean modulation by spatial frequency is about 2 spikes/s, which is a rather small modulation. Thus, the spatial frequency selectivity the authors describe in this paper is rather small compared to the stimulus selectivity one typically observes in IT (stimulus-driven modulations can be at least 20 spikes/s).

To address the concerns regarding the firing rates and the modulation of neuronal responses by spatial frequency (SF), we emphasize several key points:

(1) Significance of Firing Rate Differences: While it is true that the mean net firing rate to our scrambled stimuli was relatively low, the firing rate differences observed were statistically significant, with p-values approximately at 1e-5. This indicates that despite the low firing rates, the observed differences are reliable and unlikely to have occurred by chance.

(2) Classification Rate and Modulation by SF: Our analysis showed that the difference between various SF responses led to a classification rate of 44.68%, which is 24.68% higher than the chance level. This substantial increase above the chance level demonstrates that SF significantly modulates IT responses, even if the overall firing rates are modest.

(3) Effect Size and SF Modulation: While the effect size in terms of firing rate differences may be small, it is significant. The significant modulation of IT responses by SF, as evidenced by our statistical analyses and classification rate, supports our conclusions regarding the role of SF in driving IT responses.

(4) Expectations for Noise-like Pure SF Stimuli: We acknowledge that IT responses are typically higher for various object stimuli. Given the nature of our pure SF stimuli, which resemble noise-like patterns, we did not anticipate high responses in terms of spikes per second. The low firing rates are consistent with the expectation for such stimuli and do not undermine the significance of the observed modulation by SF.

We believe that these points collectively support the validity of our findings and the significance of SF modulation in IT responses, despite the low firing rates. We appreciate your insights and hope this clarifies our stance on the data and its implications.

We added the following description to the Appendix 1 - “Strength of SF selectivity” section:

“While the firing rates and net responses to scrambled stimuli were modest (e.g., 2.9 Hz in T1), the differences across spatial frequency (SF) bands were statistically significant (p ≈ 1e-5) and led to a classification accuracy 24.68\% above chance. This demonstrates the robustness of SF modulation in IT neurons despite low firing rates. The modest responses align with expectations for noise-like stimuli, which are less effective in driving IT neurons, yet the observed SF selectivity highlights a fundamental property of IT encoding.”

(2) Their new Figure 2-Appendix 1 does not show net firing rates (baseline-subtracted; as I requested) and thus is not very informative. Please provide distributions of net responses so that the readers can evaluate the responses to the stimuli of the recorded neurons.

We understand the reviewer’s concern about the presentation of net firing rates. In T2 (the late time interval), the average response rate falls below the baseline, resulting in negative net firing rates, which might confuse readers. To address this, we have added the net responses to the text for clarity. Additionally, we have included the average baseline response in the figure to provide a more comprehensive view of the data.

“To check the SF response strength, the histogram of IT neuron responses to scrambled, face, and non-face stimuli is illustrated in this figure. A Gamma distribution is also fitted to each histogram. To calculate the histogram, the neuron response to each unique stimulus is calculated for each neuron in spike/seconds (Hz). In the early phase, T1, the average firing rate to scrambled stimuli is 26.3 Hz which is significantly higher than the response in -50 to 50ms which is 23.4 Hz. In comparison, the mean response to intact face stimuli is 30.5 Hz, while non-face stimuli elicit an average response of 28.8 Hz. The average net responses to the scrambled, face, and non-face stimuli are 2.9 Hz, 7.1 Hz, and 5.4 Hz, respectively. Moving to the late phase, T2, the responses to scrambled, face, and object stimuli are 19.5 Hz, 19.4 Hz, and 22.4 Hz, respectively. The corresponding average net responses are 3.9 Hz, 4.0 Hz, and 1.0 Hz below the baseline response.”

(3) The poor responses might be due to the short stimulus duration. The authors report now new data using a 200 ms duration which supported their classification and latency data obtained with their brief duration. It would be very informative if the authors could also provide the mean net responses for the 200 ms durations to their stimuli. Were these responses as low as those for the brief duration? If so, the concern of generalization to effective stimuli that drive IT neurons well remains.

The firing rates for the 200 ms stimulus duration are as follows: 27.7 Hz, 30.7 Hz, and 30.4 Hz for scrambled, face, and object stimuli in T1), respectively; and 26.2 Hz, 29.1 Hz, and 33.9 Hz in T2. The average baseline firing rate (−50 to 50 ms) is 23.4 Hz. Therefore, the net responses are 4.3 Hz, 7.3 Hz, and 7.0 Hz for T1; and 2.8 Hz, 5.7 Hz, and 10.5 Hz for T2 for scrambled, face, and object stimuli, respectively.

Notably, the impact of stimulus duration is more pronounced in T2, which is consistent with the time interval of the T2 compared to T1. However, the firing rates in T1 do not show substantial changes with the longer duration. As we discussed in our response to the first comment, it is important to note that high net responses are not typically expected for scrambled or noise-like stimuli in IT neurons. Instead, the key findings of this study lie in the statistical significance of these responses and their meaningful relationship to category selectivity. These results highlight the broader implications for understanding the role of spatial frequency in object recognition.

We added the firing rates to the, Appendix 1, “Extended stimulus duration supports LSF-preferred tuning” part as follows.

“For the 200 ms stimulus duration, the firing rates were 27.7 Hz, 30.7 Hz, and 30.4 Hz for scrambled, face, and object stimuli in T1, respectively, and 26.2 Hz, 29.1 Hz, and 33.9 Hz in T2. The corresponding net responses were 4.3 Hz, 7.3 Hz, and 7.0 Hz in T1, and 2.8 Hz, 5.7 Hz, and 10.5 Hz in T2. While the longer stimulus duration did not substantially increase firing rates in T1, its impact was more pronounced in T2.”

(4) I still do not understand why the analyses of Figures 3 and 4 provide different outcomes on the relationship between spatial frequency and category selectivity. I believe they refer to this finding in the Discussion: "Our results show a direct relationship between the population's category coding capability and the SF coding capability of individual neurons. While we observed a relation between SF and category coding, we have found uncorrelated representations. Unlike category coding, SF relies more on sparse, individual neuron representations.". I believe more clarification is necessary regarding the analyses of Figures 3 and 4, and why they can show different outcomes.

Figure 3 explores the relationship between SF coding and category coding at both the single-neuron and population levels.

● Figures 3(a) and 3(b) examine the relationship between a single neuron’s response pattern and object decoding in the population.

● Figure 3(c) investigates the relationship between a single neuron’s SF decoding capabilities and object decoding in the population.

● Figure 3(d) assesses the relationship between a single neuron’s object decoding capabilities and SF decoding in the population.

In summary, Figure 3 demonstrates a relation between SF coding/response pattern at the single level and category coding at the population level.

Figure 4, on the other hand, addresses the uncorrelated nature of SF and category coding.

● Figure 4(a) shows the uncorrelated relation between a single neuron’s SF decoding capability and its object decoding capability. This suggests that a neuron's ability to decode SF does not predict its ability to decode object categories.

● Figure 4(b) illustrates that the contribution of a neuron to the population decoding of SF is uncorrelated with its contribution to the population decoding of object categories. This further supports the idea that the mechanisms behind SF coding and object coding are uncorrelated.

In summary, Figure 4 suggests that while there is a relation between SF coding and category coding as illustrated in Figure 3, the mechanisms underlying SF coding and object coding operate independently (in terms of correlation), highlighting the distinct nature of these processes.

We hope this explanation clarifies why the analyses in Figures 3 and 4 present different outcomes. Figure 3 provides insight into the relationship between SF and category coding, while Figure 4 emphasizes the uncorrelated nature of these processes. We also added the following explanation in the “Uncorrelated mechanisms for SF and category coding” section.

Based on your command, to clarify the presentation of the work, we added the following description to the “Uncorrelated mechanisms for SF and category coding” section:

“Figures 3 and 4 examine different aspects of the relationship between SF and category coding. Figure 3 highlights a relationship between SF coding at the single-neuron level and category coding at the population level. Conversely, Figure 4 demonstrates the uncorrelated mechanisms underlying SF and category coding, showing that a neuron’s ability to decode SF is not predictive of its ability to decode object categories. This distinction underscores that while SF and category coding are related at broader levels, their underlying mechanisms are independent, emphasizing the distinct processes driving each form of coding.”

(5) The authors found a higher separability for faces (versus scrambled patterns) for neurons preferring high spatial frequencies. This is consistent for the two monkeys but we are dealing here with a small amount of neurons. Only 6% of their neurons (16 neurons) belonged to this high spatial frequency group when pooling the two monkeys. Thus, although both monkeys show this effect I wonder how robust it is given the small number of neurons per monkey that belong to this spatial frequency profile. Furthermore, the higher separability for faces for the low-frequency profiles is not consistent across monkeys which should be pointed out.

We appreciate the reviewer’s concern regarding the relatively small number of neurons in the high spatial frequency group (16 neurons, 6% of the total sample across the two monkeys) and the consistency of the results. While we acknowledge this limitation, it is important to note that findings involving sparse subsets of neurons can still be meaningful. For example, Dalgleish et al. (2020) demonstrated that perception can arise from the activity of as few as ~14 neurons in the mouse cortex, supporting the sparse coding hypothesis. This underscores the potential robustness of results derived from small neuronal populations when the activity is statistically significant and functionally relevant.

Regarding the higher separability for faces among neurons preferring high spatial frequencies, the consistency of this finding across both monkeys suggests that this effect is robust within this subgroup. For neurons preferring low spatial frequencies, we agree that the lack of consistency across monkeys should be explicitly noted. These differences may reflect individual variability or differences in sampling across subjects and merit further investigation in future studies.

To address this concern, we have updated the text to explicitly discuss the small size of the high spatial frequency group, its implications, and the observed inconsistency in the low spatial frequency profiles between monkeys. We have added the following description to the discussion.

“Next, according to Figure 3(a), 6% of the neurons are HSF-preferred and their firing rate in HSF is comparable to the LSF firing rate in the LSF-preferred group. This analysis is carried out in the early phase of the response (70-170ms). While most of the neurons prefer LSF, this observation shows that there is an HSF input that excites a small group of neurons. Importantly, findings involving small neuronal populations can still be meaningful, as studies like Dalgleish et al. (2020) have demonstrated that perception can arise from the activity of as few as ~14 neurons in the mouse cortex, emphasizing the robustness of sparse coding.”

Regarding the separability of faces for the low-frequency profiles, we added the following to the appendix section,

“For neurons preferring LSF, LP profile, it is important to note the lack of consistency in responses across monkeys. This variability may reflect individual differences in neural processing or variations in sampling between subjects.”

And in the discussion:

“Our results are based on grouping the neurons of the two monkeys; however, the results remain consistent when looking at the data from individual monkeys as illustrated in Appendix 2. However, for neurons preferring LSF, we observed inconsistency across monkeys, which may reflect individual differences or sampling variability. These findings highlight the complexity of SF processing in the IT cortex and suggest the need for further research to explore these variations.”

* Henry WP Dalgleish, Lloyd E Russel, lAdam M Packer, Arnd Roth, Oliver M Gauld, Francesca Greenstreet, Emmett J Thompson, Michael Häusser (2020) How many neurons are sufficient for perception of cortical activity? eLife 9:e58889.

(6) I agree that CNNs are useful models for ventral stream processing but that is not relevant to the point I was making before regarding the comparison of the classification scores between neurons and the model. Because the number of features and trial-to-trial variability differs between neural nets and neurons, the classification scores are difficult to compare. One can compare the trends but not the raw classification scores between CNN and neurons without equating these variables.

We appreciate the reviewer’s follow-up comment and agree that differences in the number of features and trial-to-trial variability between IT neurons and CNN units make direct comparisons of raw classification scores challenging. As the reviewer suggests, it is more appropriate to focus on comparing trends rather than absolute scores when analyzing the similarities and differences between these systems. In light of this, we have revised the text to clarify that our intention was not to equate raw classification scores but to highlight the qualitative patterns and trends observed in spatial frequency encoding between IT and CNN units.

“SF representation in the artificial neural networks

We conducted a thorough analysis to compare our findings with CNNs. To assess the SF coding capabilities and trends of CNNs, we utilized popular architectures, including ResNet18, ResNet34, VGG11, VGG16, InceptionV3, EfficientNetb0, CORNet-S, CORTNet-RT, and CORNet-z, with both pre-trained on ImageNet and randomly initialized weights. Employing feature maps from the four last layers of each CNN, we trained an LDA model to classify the SF content of input images. Figure 5(a) shows the SF decoding accuracy of the CNNs on our dataset (SF decoding accuracy with random (R) and pre-trained (P) weights, ResNet18: P=0.96±0.01 / R=0.94±0.01, ResNet34 P=0.95±0.01 / R=0.86±0.01, VGG11: P=0.94±0.01 / R=0.93±0.01, VGG16: P=0.92±0.02 / R=0.90±0.02, InceptionV3: P=0.89±0.01 / R=0.67±0.03, EfficientNetb0: P=0.94±0.01 / R=0.30±0.01, CORNet-S: P=0.77±0.02 / R=0.36±0.02, CORTNet-RT: P=0.31±0.02 / R=0.33±0.02, and CORNet-z: P=0.94±0.01 / R=0.97±0.01). Except for CORNet-z, object recognition training increases the network's capacity for SF coding, with an improvement as significant as 64\% in EfficientNetb0. Furthermore, except for the CORNet family, LSF content exhibits higher recall values than HSF content, as observed in the IT cortex (p-value with random (R) and pre-trained (P) weights, ResNet18: P=0.39 / R=0.06, ResNet34 P=0.01 / R=0.01, VGG11: P=0.13 / R=0.07, VGG16: P=0.03 / R=0.05, InceptionV3: P=<0.001 / R=0.05, EfficientNetb0: P=0.07 / R=0.01). The recall values of CORNet-Z and ResNet18 are illustrated in Figure 5(b). However, while the CNNs exhibited some similarities in SF representation with the IT cortex, they did not replicate the SF-based profiles that predict neuron category selectivity. As depicted in Figure 5(c) although neurons formed similar profiles, these profiles were not associated with the category decoding performances of the neurons sharing the same profile.”

Discussion:

“Finally, we compared SF's representation trends and findings within the IT cortex and the current state-of-the-art networks in deep neural networks.”

Recommendations for the authors:

Reviewer #2 (Recommendations For The Authors):

The mean baseline firing rate of their neurons (23.4 Hz) was rather high for single IT neurons (typically around 10 spikes/s or lower). Were these well-isolated units or mainly multiunit activity?

We confirm that the recordings in our study were from both well-isolated single units and multi-unit activities (remaining after isolation neurons) sorted based on our spike sorting toolbox. The higher baseline firing rate is likely due to the experimental design, particularly the inclusion of the responsive neurons from the selectivity phase. We added the following statement to the methods section.

“In our analysis, we utilized both well-isolated single units and multi-unit activities (which represent neural activities that could not be further sorted into single units), ensuring a comprehensive representation of neural responses across the recorded population.”

Reviewer #1 (Public review):

Summary:

Epiney et al. use single-nuclei RNA sequencing (snRNA-seq) to characterize the lineage of Type-2 (T2) neuroblasts (NBs) in the adult Drosophila brain. To isolate cells born from T2 NBs, the authors used a genetic tool that specifically allows the permanent labeling of T2-derived cell types, which are then FAC-sorted for snRNA-seq. This effective labeling approach also allows them to compare the isolated T2 lineage cells with T1-derived cell types by a simple exclusion method. The authors begin by describing a transcriptomic atlas for all T1 and T2-derived neuronal and glia clusters, reporting that the T2-derived lineage comprises 161 neuronal clusters, in contrast to the T1 lineage which comprises 114 of them. The authors then use the expression of VAChT, VGlut, Gad1, Tbh, Ple, SerT, and Tdc2 to show that T2 neuroblasts generate all major neuron classes of fast-acting neurotransmitters. Strikingly, they show that a subset of glia and neuronal clusters have disproportionate enrichment in males or females, suggesting that T2 neuroblasts generate sex-biased cell types. The authors then proceed to characterize neuropeptide expression across T2-derived neuronal clusters and argue that the same neuropeptide can be expressed across different cell types, while similar cell types can express distinct neuropeptides. The functional implication of both observations, however, remains to be tested. Furthermore, the authors describe combinatorial transcription factor (TF) codes that are correlated with neuropeptide expression for T2-derived neurons along with an overall TF code for all T2-derived cell types, both of which will serve as an important starting point for future investigations. Finally, the authors map well-studied neuronal types of the central complex to the clusters of their T2-derived snRNA-seq dataset. They use known marker combinations, bulk RNA-seq data and highly specific split-GAL4 driver lines to annotate their T2-derived atlas, establishing a comprehensive transcriptomic atlas that would guide future studies in this field.

Strengths:

This study provides an in-depth transcriptomic characterization of neurons and glia derived from Type-2 neuroblast lineages. The results of this manuscript offer several future directions to investigate the mechanisms of diversifying neuronal identity. The datasets of T1-derived and T2-derived cells will pave the way for studies focused on the functional analysis of combinatorial TF codes specifying cell identity, sex-based differences in neurogenesis and gliogenesis, the relationship between neuropeptide (co)expression and cell identity, and the differential contributions of distinct progenitor populations to the same cell type.

Weaknesses:

The study presents several important observations based on the characterization of Type II neuroblast-derived lineages. However, a mechanistic insight is missing for most observations. The idea that there is a sex-specific bias to certain T2-derived neurons and glial clusters is quite interesting, however, the functional significance of this observation is not tested or discussed extensively. Finally, the authors do not show whether the combinatorial TF code is indeed necessary for neuropeptide expression or if this is just a correlation due to cell identity being defined by TFs. Functional knockdown of some candidate TFs for a subset of neuropeptide-expressing cells would have been helpful in this case.

Reviewer #2 (Public review):

Summary:

The authors apply the recently developed VARX model, which explicitly models intrinsic dynamics and the effect of extrinsic inputs, to simulated data and intracranial EEG recordings. This method provides a directed method of 'intrinsic connectivity'. They argue this model is better suited to the analysis of task neuroimaging data because it separates the intrinsic and extrinsic activity. They show: that intrinsic connectivity is largely unaltered during a movie-watching task compared to eyes open rest; intrinsic noise is reduced in the task; and there is intrinsic directed connectivity from sensory to higher-order brain areas.

Strengths:

(1) The paper tackles an important issue with an appropriate method.

(2) The authors validated their method on data simulated with a neural mass model.

(3) They use intracranial EEG, which provides a direct measure of neuronal activity.

(4) Code is made publicly available and the paper is written well.

Weaknesses:

It is unclear whether a linear model is adequate to describe brain data. To the author's credit, they discuss this in the manuscript. Also, the model presented still provides a useful and computationally efficient method for studying brain data - no model is 'the truth'.

Appraisal of whether the authors achieve their aims:

As a methodological advancement highlighting a limitation of existing approaches and presenting a new model to overcome it, the authors achieve their aim. Generally, the claims/conclusions are supported by the results.

The wider neuroscience claims regarding the role of intrinsic dynamics and external inputs in affecting brain data could benefit from further replication with another independent dataset and in a variety of tasks - but I understand if the authors wanted to focus on the method rather than the neuroscientific claims in this manuscript.

Impact:

The authors propose a useful new approach that solves an important problem in the analysis of task neuroimaging data. I believe the work can have a significant impact on the field.

Reviewer #2 (Public review):

Summary:

In this paper, the authors use numerical simulations to try to understand better a major experimental discovery in songbird neuroscience from 2002 by Richard Hahnloser and collaborators. The 2002 paper found that a certain class of projection neurons in the premotor nucleus HVC of adult male zebra finch songbirds, the neurons that project to another premotor nucleus RA, fired sparsely (once per song motif) and precisely (to about 1 ms accuracy) during singing.

The experimental discovery is important to understand since it initially suggested that the sparsely firing RA-projecting neurons acted as a simple clock that was localized to HVC and that controlled all details of the temporal hierarchy of singing: notes, syllables, gaps, and motifs. Later experiments suggested that the initial interpretation might be incomplete: that the temporal structure of adult male zebra finch songs instead emerged in a more complicated and distributed way, still not well understood, from the interaction of HVC with multiple other nuclei, including auditory and brainstem areas. So at least two major questions remain unanswered more than two decades after the 2002 experiment: What is the neurobiological mechanism that produces the sparse precise bursting: is it a local circuit in HVC or is it some combination of external input to HVC and local circuitry? And how is the sparse precise bursting in HVC related to a songbird's vocalizations?

The authors only investigate part of the first question, whether the mechanism for sparse precise bursts is local to HVC. They do so indirectly, by using conductance-based Hodgkin-Huxley-like equations to simulate the spiking dynamics of a simplified network that includes three known major classes of HVC neurons and such that all neurons within a class are assumed to be identical. A strength of the calculations is that the authors include known biophysically deduced details of the different conductances of the three major classes of HVC neurons, and they take into account what is known, based on sparse paired recordings in slices, about how the three classes connect to one another. One weakness of the paper is that the authors make arbitrary and not well-motivated assumptions about the network geometry, and they do not use the flexibility of their simulations to study how their results depend on their network assumptions. A second weakness is that they ignore many known experimental details such as projections into HVC from other nuclei, dendritic computations (the somas and dendrites are treated by the authors as point-like isopotential objects), the role of neuromodulators, and known heterogeneity of the interneurons. These weaknesses make it difficult for readers to know the relevance of the simulations for experiments and for advancing theoretical understanding.

Strengths:

The authors use conductance-based Hodgkin-Huxley-like equations to simulate spiking activity in a network of neurons intended to model more accurately songbird nucleus HVC of adult male zebra finches. Spiking models are much closer to experiments than models based on firing rates or on 2-state neurons.

The authors include information deduced from modeling experimental current-clamp data such as the types and properties of conductances. They also take into account how neurons in one class connect to neurons in other classes via excitatory or inhibitory synapses, based on sparse paired recordings in slices by other researchers.

The authors obtain some new results of modest interest such as how changes in the maximum conductances of four key channels (e.g., A-type K+ currents or Ca-dependent K+ currents) influence the structure and propagation of bursts, while simultaneously being able to mimic accurately current-clamp voltage measurements.

Weaknesses:

One weakness of this paper is the lack of a clearly stated, interesting, and relevant scientific question to try to answer. In the introduction, the authors do not discuss adequately which questions recent experimental and theoretical work have failed to explain adequately, concerning HVC neural dynamics and its role in producing vocalizations. The authors do not discuss adequately why they chose the approach of their paper and how their results address some of these questions.

For example, the authors need to explain in more detail how their calculations relate to the works of Daou et al, J. Neurophys. 2013 (which already fitted spiking models to neuronal data and identified certain conductances), to Jin et al J. Comput. Neurosci. 2007 (which already discussed how to get bursts using some experimental details), and to the rather similar paper by E. Armstrong and H. Abarbanel, J. Neurophys 2016, which already postulated and studied sequences of microcircuits in HVC. This last paper is not even cited by the authors.

The authors' main achievement is to show that simulations of a certain simplified and idealized network of spiking neurons, which includes some experimental details but ignores many others, match some experimental results like current-clamp-derived voltage time series for the three classes of HVC neurons (although this was already reported in earlier work by Daou and collaborators in 2013), and simultaneously the robust propagation of bursts with properties similar to those observed in experiments. The authors also present results about how certain neuronal details and burst propagation change when certain key maximum conductances are varied.

However, these are weak conclusions for two reasons. First, the authors did not do enough calculations to allow the reader to understand how many parameters were needed to obtain these fits and whether simpler circuits, say with fewer parameters and simpler network topology, could do just as well. Second, many previous researchers have demonstrated robust burst propagation in a variety of feed-forward models. So what is new and important about the authors' results compared to the previous computational papers?

Also missing is a discussion, or at least an acknowledgment, of the fact that not all of the fine experimental details of undershoots, latencies, spike structure, spike accommodation, etc may be relevant for understanding vocalization. While it is nice to know that some models can match these experimental details and produce realistic bursts, that does not mean that all of these details are relevant for the function of producing precise vocalizations. Scientific insights in biology often require exploring which of the many observed details can be ignored and especially identifying the few that are essential for answering some questions. As one example, if HVC-X neurons are completely removed from the authors' model, does one still get robust and reasonable burst propagation of HVC-RA neurons? While part of the nucleus HVC acts as a premotor circuit that drives the nucleus RA, part of HVC is also related to learning. It is not clear that HVC-X neurons, which carry out some unknown calculation and transmit information to area X in a learning pathway, are relevant for burst production and propagation of HVC-RA neurons, and so relevant for vocalization. Simulations provide a convenient and direct way to explore questions of this kind.

One key question to answer is whether the bursting of HVC-RA projection neurons is based on a mechanism local to HVC or is some combination of external driving (say from auditory nuclei) and local circuitry. The authors do not contribute to answering this question because they ignore external driving and assume that the mechanism is some kind of intrinsic feed-forward circuit, which they put in by hand in a rather arbitrary and poorly justified way, by assuming the existence of small microcircuits consisting of a few HVC-RA, HVC-X, and HVC-I neurons that somehow correspond to "sub-syllabic segments". To my knowledge, experiments do not suggest the existence of such microcircuits nor does theory suggest the need for such microcircuits.

Another weakness of this paper is an unsatisfactory discussion of how the model was obtained, validated, and simulated. The authors should state as clearly as possible, in one location such as an appendix, what is the total number of independent parameters for the entire network and how parameter values were deduced from data or assigned by hand. With enough parameters and variables, many details can be fit arbitrarily accurately so researchers have to be careful to avoid overfitting. If parameter values were obtained by fitting to data, the authors should state clearly what the fitting algorithm was (some iterative nonlinear method, whose results can depend on the initial choice of parameters), what the error function used for fitting (sum of least squares?) was, and what data were used for the fitting.

The authors should also state clearly the dynamical state of the network, the vector of quantities that evolve over time. (What is the dimension of that vector, which is also the number of ordinary differential equations that have to be integrated?) The authors do not mention what initial state was used to start the numerical integrations, whether transient dynamics were observed and what were their properties, or how the results depended on the choice of the initial state. The authors do not discuss how they determined that their model was programmed correctly (it is difficult to avoid typing errors when writing several pages or more of a code in any language) or how they determined the accuracy of the numerical integration method beyond fitting to experimental data, say by varying the time step size over some range or by comparing two different integration algorithms.

Also disappointing is that the authors do not make any predictions to test, except rather weak ones such as that varying a maximum conductance sufficiently (which might be possible by using dynamic clamps) might cause burst propagation to stop or change its properties. Based on their results, the authors do not make suggestions for further experiments or calculations, but they should.

Author response:

eLife Assessment

Birdsong production depends on precise neural sequences in a vocal motor nucleus HVC. In this useful biophysical model, Daou and colleagues identify specific biophysical parameters that result in sparse neural sequences observed in vivo. While the model is presently incomplete because it is overfit to produce sequences and therefore not robust to real biological variation, the model has the potential to address some outstanding issues in HVC function.

We are grateful for the extensive supportive comments from the reviewers, including broad, strong appreciation of the novel aspects of our manuscript. We believe these will be only strengthened in the next submission.

Public Reviews:

Reviewer #1 (Public review):

Summary:

The paper presents a model for sequence generation in the zebra finch HVC, which adheres to cellular properties measured experimentally. However, the model is fine-tuned and exhibits limited robustness to noise inherent in the inhibitory interneurons within the HVC, as well as to fluctuations in connectivity between neurons. Although the proposed microcircuits are introduced as units for sub-syllabic segments (SSS), the backbone of the network remains a feedforward chain of HVC_RA neurons, similar to previous models.

Strengths:

The model incorporates all three of the major types of HVC neurons. The ion channels used and their kinetics are based on experimental measurements. The connection patterns of the neurons are also constrained by the experiments.

Weaknesses:

The model is described as consisting of micro-circuits corresponding to SSS. This presentation gives the impression that the model's structure is distinct from previous models, which connected HVC_RA neurons in feedforward chain networks (Jin et al 2007, Li & Greenside, 2006; Long et al 2010; Egger et al 2020). However, the authors implement single HVC_RA neurons into chain networks within each micro-circuit and then connect the end of the chain to the start of the chain in the subsequent micro-circuit. Thus, the HVC_RA neuron in their model forms a single-neuron chain. This structure is essentially a simplified version of earlier models.

In the model of the paper, the chain network drives the HVC_I and HVC_X neurons. The role of the micro-circuits is more significant in organizing the connections: specifically, from HVC_RA neurons to HVC_I neurons, and from HVC_I neurons to both HVC_X and HVC_RA neurons.

We thank Reviewer 1 for their thoughtful comments.

While the reviewer is correct about the fact that the propagation of sequential activity in this model is primarily carried by HVC_RA neurons in a feed-forward manner, we need to emphasize that this is true only if there is no intrinsic or synaptic perturbation to the HVC network. For example, we showed in Figures 10 and 12 how altering the intrinsic properties of HVC_X neurons or for interneurons disrupts sequence propagation. In other words, while HVC_RA neurons are the key forces to carry the chain forward, the interplay between excitation and inhibition in our network as well as the intrinsic parameters for all classes of HVC neurons are equally important forces in carrying the chain of activity forward. Thus, the stability of activity propagation necessary for song production depend on a finely balanced network of HVC neurons, with all classes contributing to the overall dynamics. Moreover, all existing models that describe premotor sequence generation in the HVC either assume a distributed model (Elmaleh et al., 2021) that dictates that local HVC circuitry is not sufficient to advance the sequence but rather depends upon momentto-moment feedback through Uva (Hamaguchi et al., 2016), or assume models that rely on intrinsic connections within HVC to propagate sequential activity. In the latter case, some models assume that HVC is composed of multiple discrete subnetworks that encode individual song elements (Glaze & Troyer, 2013; Long & Fee, 2008; Wang et al., 2008), but lacks the local connectivity to link the subnetworks, while other models assume that HVC may have sufficient information in its intrinsic connections to form a single continuous network sequence (Long et al. 2010). The HVC model we present extends the concept of a feedforward network by incorporating additional neuronal classes that influence the propagation of activity (interneurons and HVC_X neurons). We have shown that any disturbance of the intrinsic or synaptic conductances of these latter neurons will disrupt activity in the circuit even when HVC_RA neurons properties are maintained.

In regard to the similarities between our model and earlier models, several aspects of our model distinguish it from prior work. In short, while several models of how sequence is generated within HVC have been proposed (Cannon et al., 2015; Drew & Abbott, 2003; Egger et al., 2020; Elmaleh et al., 2021; Galvis et al., 2018; Gibb et al., 2009a, 2009b; Hamaguchi et al., 2016; Jin, 2009; Long & Fee, 2008; Markowitz et al., 2015), all the models proposed either rely on intrinsic HVC circuitry to propagate sequential activity, rely on extrinsic feedback to advance the sequence or rely on both. These models do not capture the complex details of spike morphology, do not include the right ionic currents, do not incorporate all classes of HVC neurons, or do not generate realistic firing patterns as seen in vivo. Our model is the first biophysically realistic model that incorporates all classes of HVC neurons and their intrinsic properties. We tuned the intrinsic and the synaptic properties bases on the traces collected by Daou et al. (2013) and Mooney and Prather (2005) as shown in Figure 3. The three classes of model neurons incorporated to our network as well as the synaptic currents that connect them are based on HodgkinHuxley formalisms that contain ion channels and synaptic currents which had been pharmacologically identified. This is an advancement over prior models that primarily focused on the role of synaptic interactions or external inputs. The model is based on a feedforward chain of microcircuits that encode for the different sub-syllabic segments and that interact with each other through structured feedback inhibition, defining an ordered sequence of cell firing. Moreover, while several models highlight the critical role of inhibitory interneurons in shaping the timing and propagation of bursts of activity in HVC_RA neurons, our work offers an intricate and comprehensive model that help understand this critical role played by inhibition in shaping song dynamics and ensuring sequence propagation.

How useful is this concept of micro-circuits? HVC neurons fire continuously even during the silent gaps. There are no SSS during these silent gaps.

Regarding the concern about the usefulness of the 'microcircuit' concept in our study, we appreciate the comment and we are glad to clarify its relevance in our network. While we acknowledge that HVC_RA neurons interconnect microcircuits, our model's dynamics are still best described within the framework of microcircuitry particularly due to the firing behavior of HVC_X neurons and interneurons. Here, we are referring to microcircuits in a more functional sense, rather than rigid, isolated spatial divisions (Cannon et al. 2015). A microcircuit in our model reflects the local rules that govern the interaction between all HVC neuron classes within the broader network, and that are essential for proper activity propagation. For example, HVC_INT neurons belonging to any microcircuit burst densely and at times other than the moments when the corresponding encoded SSS is being “sung”. What makes a particular interneuron belong to this microcircuit or the other is merely the fact that it cannot inhibit HVC_RA neurons that are housed in the microcircuit it belongs to. In particular, if HVC_INT inhibits HVC_RA in the same microcircuit, some of the HVC_RA bursts in the microcircuit might be silenced by the dense and strong HVC_INT inhibition breaking the chain of activity again. Similarly, HVC_X neurons were selected to be housed within microcircuits due to the following reason: if an HVC_X neuron belonging to microcircuit i sends excitatory input to an HVC_INT neuron in microcircuit j, and that interneuron happens to select an HVC_RA neuron from microcircuit i, then the propagation of sequential activity will halt, and we’ll be in a scenario similar to what was described earlier for HVC_INT neurons inhibiting HVC_RA neurons in the same microcircuit.

We agree that there are no sub-syllabic segments described during the silent gaps and we thank the reviewer to pointing this out. Although silent gaps are integral to the overall process of song production, we have not elaborated on them in this model due to the lack of a clear, biophysically grounded representation for the gaps themselves at the level of HVC. Our primary focus has been on modeling the active, syllable-producing phases of the song, where the HVC network’s sequential dynamics are critical for song. However, one can think the encoding of silent gaps via similar mechanisms that encode SSSs, where each gap is encoded by similar microcircuits comprised of the three classes of HVC neurons (let’s called them GAP rather than SSS) that are active only during the silent gaps. In this case, the propagation of sequential activity is carried throughout the GAPs from the last SSS of the previous syllable to the first SSS of the subsequent syllable. We’ll make sure to emphasize this mechanism more in the revised version of the manuscript.

A significant issue of the current model is that the HVC_RA to HVC_RA connections require fine-tuning, with the network functioning only within a narrow range of g_AMPA (Figure 2B). Similarly, the connections from HVC_I neurons to HVC_RA neurons also require fine-tuning. This sensitivity arises because the somatic properties of HVC_RA neurons are insufficient to produce the stereotypical bursts of spikes observed in recordings from singing birds, as demonstrated in previous studies (Jin et al 2007; Long et al 2010). In these previous works, to address this limitation, a dendritic spike mechanism was introduced to generate an intrinsic bursting capability, which is absent in the somatic compartment of HVC_RA neurons. This dendritic mechanism significantly enhances the robustness of the chain network, eliminating the need to fine-tune any synaptic conductances, including those from HVC_I neurons (Long et al 2010).

Why is it important that the model should NOT be sensitive to the connection strengths?

We thank the reviewer for the comment. While mathematical models designed for highly complex nonlinear biological processes tangentially touch the biological realism, the current network as is right now is the first biologically realistic-enough network model designed for HVC that explains sequence propagation. We do not include dendritic processes in our network although that increases the realistic dynamics for various reasons. 1) The ion channels we integrated into the somatic compartment are known pharmacologically (Daou et al. 2013), but we don’t know about the dendritic compartment’s intrinsic properties of HVC neurons and the cocktail of ion channels that are expressed there. 2) We are able to generate realistic bursting in HVC_RA neurons despite the single compartment, and the main emphasis in this network is on the interactions between excitation and inhibition, the effects of ion channels in modulating sequence propagation, etc. 3) The network model already incorporates thousands of ODEs that govern the dynamics of each of the HVC neurons, so we did not want to add more complexity to the network especially that we don’t know the biophysical properties of the dendritic compartments.

Therefore, our present focus is on somatic dynamics and the interaction between HVC_RA and HVC_INT neurons, but we acknowledge the importance of these processes in enhancing network resiliency. Although we agree that adding dendritic processes improves robustness, we still think that somatic processes alone can offer insightful information on the sequential dynamics of the HVC network. While the network should be robust across a wide range of parameters, it is also essential that certain parameters are designed to filter out weaker signals, ensuring that only reliable, precise patterns of activity propagate. Hence, we specifically chose to make the HVC_RA-to-HVC_RA excitatory connections more sensitive (narrow range of values) such that only strong, precise and meaningful stimuli can propagate through the network representing the high stereotypy and precision seen in song production.

First, the firing of HVC_I neurons is highly noisy and unreliable. HVC_I neurons fire spontaneous, random spikes under baseline conditions. During singing, their spike timing is imprecise and can vary significantly from trial to trial, with spikes appearing or disappearing across different trials. As a result, their inputs to HVC_RA neurons are inherently noisy. If the model relies on precisely tuned inputs from HVC_I neurons, the natural fluctuations in HVC_I firing would render the model non-functional. The authors should incorporate noisy HVC_I neurons into their model to evaluate whether this noise would render the model non-functional.

We acknowledge that under baseline and singing settings, interneurons fire in an extremely noisy and inaccurate manner, although they exhibit time locked episodes in their activity (Hahnloser et al 2002, Kozhinikov and Fee 2007). In order to mimic the biological variability of these neurons, our model does, in fact, include a stochastic current to reflect the intrinsic noise and random variations in interneuron firing shown in vivo (and we highlight this in the Methods). If necessary and to make sure the network is resilient to this randomness in interneuron firing, we will investigate different approaches to enhance the noise representation even further and check its effect on sequence propagation.

Second, Kosche et al. (2015) demonstrated that reducing inhibition by suppressing HVC_I neuron activity makes HVC_RA firing less sparse but does not compromise the temporal precision of the bursts. In this experiment, the local application of gabazine should have severely disrupted HVC_I activity. However, it did not affect the timing precision of HVC_RA neuron firing, emphasizing the robustness of the HVC timing circuit. This robustness is inconsistent with the predictions of the current model, which depends on finely tuned inputs and should, therefore, be vulnerable to such disruptions.

We thank the reviewer for the comment. The differences between the Kosche et al. (2015) findings and the predictions of our model arise from differences in the aspect of HVC function we are modeling. Our model is more sensitive to inhibition, which is a designed mechanism for achieving precise song patterning. This is a modeling simplification we adopted to capture specific characteristics of HVC function. Hence, Kosche et al. (2015) findings do not invalidate the approach of our model, but highlights that HVC likely operates with several, redundant mechanisms that overall ensure temporal precision.Nevertheless, we will investigate further the effects of the degree of inhibition on song patterning.

Third, the reliance on fine-tuning of HVC_RA connections becomes problematic if the model is scaled up to include groups of HVC_RA neurons forming a chain network, rather than the single HVC_RA neurons used in the current work. With groups of HVC_RA neurons, the summation of presynaptic inputs to each HVC_RA neuron would need to be precisely maintained for the model to function. However, experimental evidence shows that the HVC circuit remains functional despite perturbations, such as a few degrees of cooling, micro-lesions, or turnover of HVC_RA neurons. Such robustness cannot be accounted for by a model that depends on finely tuned connections, as seen in the current implementation.

Our model of individual HVC_RA neurons and as stated previously is reductive model that focuses on understanding the mechanisms that govern sequential neural activity. We agree that scaling the model to include many of HVC_RA neurons poses challenges, specifically concerning the summation of presynaptic inputs. However, our model can still be adapted to a larger network without requiring the level of fine-tuning currently needed. In fact, the current fine-tuning of synaptic connections in the model is a reflection of fundamental network mechanisms rather than a limitation when scaling to a larger network. Besides, one important feature of this neural network is redundancy. Even if some neurons or synaptic connections are impaired, other neurons or pathways can compensate for these changes, allowing the activity propagation to remain intact.

The authors examined how altering the channel properties of neurons affects the activity in their model. While this approach is valid, many of the observed effects may stem from the delicate balancing required in their model for proper function.

In the current model, HVC_X neurons burst as a result of rebound activity driven by the I_H current. Rebound bursts mediated by the I_H current typically require a highly hyperpolarized membrane potential. However, this mechanism would fail if the reversal potential of inhibition is higher than the required level of hyperpolarization. Furthermore, Mooney (2000) demonstrated that depolarizing the membrane potential of HVC_X neurons did not prevent bursts of these neurons during forward playback of the bird's own song, suggesting that these bursts (at least under anesthesia, which may be a different state altogether) are not necessarily caused by rebound activity. This discrepancy should be addressed or considered in the model.

In our HVC network model, one goal with HVC_X neurons is to generate bursts in their underlying neuron population. Since HVC_X neurons in our model receive only inhibitory inputs from interneurons, we rely on inhibition followed by rebound bursts orchestrated by the IH and the I_CaT currents to achieve this goal. The interplay between the T-type Ca⁺⁺ current and the H current in our model is fundamental to generate their corresponding bursts, as they are sufficient for producing the desired behavior in the network. Due to this interplay, we do not need significant inhibition to generate rebound bursts, because the T-type Ca⁺⁺ current’s conductance can be stronger leading to robust rebound bursting even when the degree of inhibition is not very strong. We will highlight this with more clarity in the revised version.

Some figures contain direct copies of figures from published papers. It is perhaps a better practice to replace them with schematics if possible.

We will replace the relevant figures with schematic representations where possible.

Reviewer #2 (Public review):

Summary:

In this paper, the authors use numerical simulations to try to understand better a major experimental discovery in songbird neuroscience from 2002 by Richard Hahnloser and collaborators. The 2002 paper found that a certain class of projection neurons in the premotor nucleus HVC of adult male zebra finch songbirds, the neurons that project to another premotor nucleus RA, fired sparsely (once per song motif) and precisely (to about 1 ms accuracy) during singing.

The experimental discovery is important to understand since it initially suggested that the sparsely firing RA-projecting neurons acted as a simple clock that was localized to HVC and that controlled all details of the temporal hierarchy of singing: notes, syllables, gaps, and motifs. Later experiments suggested that the initial interpretation might be incomplete: that the temporal structure of adult male zebra finch songs instead emerged in a more complicated and distributed way, still not well understood, from the interaction of HVC with multiple other nuclei, including auditory and brainstem areas. So at least two major questions remain unanswered more than two decades after the 2002 experiment: What is the neurobiological mechanism that produces the sparse precise bursting: is it a local circuit in HVC or is it some combination of external input to HVC and local circuitry?

And how is the sparse precise bursting in HVC related to a songbird's vocalizations?

The authors only investigate part of the first question, whether the mechanism for sparse precise bursts is local to HVC. They do so indirectly, by using conductance-based Hodgkin-Huxley-like equations to simulate the spiking dynamics of a simplified network that includes three known major classes of HVC neurons and such that all neurons within a class are assumed to be identical. A strength of the calculations is that the authors include known biophysically deduced details of the different conductances of the three major classes of HVC neurons, and they take into account what is known, based on sparse paired recordings in slices, about how the three classes connect to one another. One weakness of the paper is that the authors make arbitrary and not well-motivated assumptions about the network geometry, and they do not use the flexibility of their simulations to study how their results depend on their network assumptions. A second weakness is that they ignore many known experimental details such as projections into HVC from other nuclei, dendritic computations (the somas and dendrites are treated by the authors as point-like isopotential objects), the role of neuromodulators, and known heterogeneity of the interneurons. These weaknesses make it difficult for readers to know the relevance of the simulations for experiments and for advancing theoretical understanding.

Strengths:

The authors use conductance-based Hodgkin-Huxley-like equations to simulate spiking activity in a network of neurons intended to model more accurately songbird nucleus HVC of adult male zebra finches. Spiking models are much closer to experiments than models based on firing rates or on 2-state neurons.

The authors include information deduced from modeling experimental current-clamp data such as the types and properties of conductances. They also take into account how neurons in one class connect to neurons in other classes via excitatory or inhibitory synapses, based on sparse paired recordings in slices by other researchers.

The authors obtain some new results of modest interest such as how changes in the maximum conductances of four key channels (e.g., A-type K⁺ currents or Ca-dependent K⁺ currents) influence the structure and propagation of bursts, while simultaneously being able to mimic accurately current-clamp voltage measurements.

Weaknesses:

One weakness of this paper is the lack of a clearly stated, interesting, and relevant scientific question to try to answer. In the introduction, the authors do not discuss adequately which questions recent experimental and theoretical work have failed to explain adequately, concerning HVC neural dynamics and its role in producing vocalizations. The authors do not discuss adequately why they chose the approach of their paper and how their results address some of these questions.

For example, the authors need to explain in more detail how their calculations relate to the works of Daou et al, J. Neurophys. 2013 (which already fitted spiking models to neuronal data and identified certain conductances), to Jin et al J. Comput. Neurosci. 2007 (which already discussed how to get bursts using some experimental details), and to the rather similar paper by E. Armstrong and H. Abarbanel, J. Neurophys 2016, which already postulated and studied sequences of microcircuits in HVC. This last paper is not even cited by the authors.

We thank the reviewer for this valuable comment, and we agree that we did not clarify enough throughout the paper the utility of our model or how it advanced our understanding of the HVC dynamics and circuitry. To that end, we will revise several places of the manuscript and make sure to cite and highlight the relevance and relatedness of the mentioned papers.

In short, and as mentioned to Reviewer 1, while several models of how sequence is generated within HVC have been proposed (Cannon et al., 2015; Drew & Abbott, 2003; Egger et al., 2020; Elmaleh et al., 2021; Galvis et al., 2018; Gibb et al., 2009a, 2009b; Hamaguchi et al., 2016; Jin, 2009; Long & Fee, 2008; Markowitz et al., 2015; Jin et al., 2007), all the models proposed either rely on intrinsic HVC circuitry to propagate sequential activity, rely on extrinsic feedback to advance the sequence or rely on both. These models do not capture the complex details of spike morphology, do not include the right ionic currents, do not incorporate all classes of HVC neurons, or do not generate realistic firing patterns as seen in vivo. Our model is the first biophysically realistic model that incorporates all classes of HVC neurons and their intrinsic properties.

No existing hypothesis had been challenged with our model, rather; our model is a distillation of the various models that’s been proposed for the HVC network. We go over this in detail in the Discussion. We believe that the network model we developed provide a step forward in describing the biophysics of HVC circuitry, and may throw a new light on certain dynamics in the mammalian brain, particularly the motor cortex and the hippocampus regions where precisely-timed sequential activity is crucial. We suggest that temporally-precise sequential activity may be a manifestation of neural networks comprised of chain of microcircuits, each containing pools of excitatory and inhibitory neurons, with local interplay among neurons of the same microcircuit and global interplays across the various microcircuits, and with structured inhibition as well as intrinsic properties synchronizing the neuronal pools and stabilizing timing within a firing sequence.

The authors' main achievement is to show that simulations of a certain simplified and idealized network of spiking neurons, which includes some experimental details but ignores many others, match some experimental results like current-clamp-derived voltage time series for the three classes of HVC neurons (although this was already reported in earlier work by Daou and collaborators in 2013), and simultaneously the robust propagation of bursts with properties similar to those observed in experiments. The authors also present results about how certain neuronal details and burst propagation change when certain key maximum conductances are varied.

However, these are weak conclusions for two reasons. First, the authors did not do enough calculations to allow the reader to understand how many parameters were needed to obtain these fits and whether simpler circuits, say with fewer parameters and simpler network topology, could do just as well. Second, many previous researchers have demonstrated robust burst propagation in a variety of feed-forward models. So what is new and important about the authors' results compared to the previous computational papers?

A major novelty of our work is the incorporation of experimental data with detailed network models. While earlier works have established robust burst propagation, our model uses realistic ion channel kinetics and feedback inhibition not only to reproduce experimental neural activity patterns but also to suggest prospective mechanisms for song sequence production in the most biophysical way possible. This aspect that distinguishes our work from other feed-forward models. We go over this in detail in the Discussion. However, the reviewer is right regarding the details of the calculations conducted for the fits, we will make sure to highlight this in the Methods and throughout the manuscript with more details.

We believe that the network model we developed provide a step forward in describing the biophysics of HVC circuitry, and may throw a new light on certain dynamics in the mammalian brain, particularly the motor cortex and the hippocampus regions where precisely-timed sequential activity is crucial. We suggest that temporally-precise sequential activity may be a manifestation of neural networks comprised of chain of microcircuits, each containing pools of excitatory and inhibitory neurons, with local interplay among neurons of the same microcircuit and global interplays across the various microcircuits, and with structured inhibition as well as intrinsic properties synchronizing the neuronal pools and stabilizing timing within a firing sequence.

Also missing is a discussion, or at least an acknowledgment, of the fact that not all of the fine experimental details of undershoots, latencies, spike structure, spike accommodation, etc may be relevant for understanding vocalization. While it is nice to know that some models can match these experimental details and produce realistic bursts, that does not mean that all of these details are relevant for the function of producing precise vocalizations. Scientific insights in biology often require exploring which of the many observed details can be ignored and especially identifying the few that are essential for answering some questions. As one example, if HVC-X neurons are completely removed from the authors' model, does one still get robust and reasonable burst propagation of HVC-RA neurons? While part of the nucleus HVC acts as a premotor circuit that drives the nucleus RA, part of HVC is also related to learning. It is not clear that HVC-X neurons, which carry out some unknown calculation and transmit information to area X in a learning pathway, are relevant for burst production and propagation of HVC_RA neurons, and so relevant for vocalization. Simulations provide a convenient and direct way to explore questions of this kind.

One key question to answer is whether the bursting of HVC-RA projection neurons is based on a mechanism local to HVC or is some combination of external driving (say from auditory nuclei) and local circuitry. The authors do not contribute to answering this question because they ignore external driving and assume that the mechanism is some kind of intrinsic feed-forward circuit, which they put in by hand in a rather arbitrary and poorly justified way, by assuming the existence of small microcircuits consisting of a few HVC-RA, HVC-X, and HVC-I neurons that somehow correspond to "sub-syllabic segments". To my knowledge, experiments do not suggest the existence of such microcircuits nor does theory suggest the need for such microcircuits.

Recent results showed a tight correlation between the intrinsic properties of neurons and features of song (Daou and Margoliash 2020, Medina and Margoliash 2024), where adult birds that exhibit similar songs tend to have similar intrinsic properties. While this is relevant, we acknowledge that not all details may be necessary for every aspect of vocalization, and future models could simplify concentrate on core dynamics and exclude certain features while still providing insights into the primary mechanisms.

The question of whether HVC_X neurons are relevant for burst propagation given that our model includes these neurons as part of the network for completeness, the reviewer is correct, the propagation of sequential activity in this model is primarily carried by HVC_RA neurons in a feed-forward manner, but only if there is no perturbation to the HVC network. For example, we have shown how altering the intrinsic properties of HVC_X neurons or for interneurons disrupts sequence propagation. In other words, while HVC neurons are the key forces to carry the chain forward, the interplay between excitation and inhibition in our network as well as the intrinsic parameters for all classes of HVC neurons are equally important forces in carrying the chain of activity forward. Thus, the stability of activity propagation necessary for song production depend on a finely balanced network of HVC neurons, with all classes contributing to the overall dynamics.

We agree with the reviewer however that a potential drawback of our model is that its sole focus is on local excitatory connectivity within the HVC (Kornfeld et al., 2017; Long et al., 2010), while HVC neurons receive afferent excitatory connections (Akutagawa & Konishi, 2010; Nottebohm et al., 1982) that plays significant roles in their local dynamics. For example, the excitatory inputs that HVC neurons receive from Uvaeformis may be crucial in initiating (Andalman et al., 2011; Danish et al., 2017; Galvis et al., 2018) or sustaining (Hamaguchi et al., 2016) the sequential activity. While we acknowledge this limitation, our main contribution in this work is the biophysical insights onto how the patterning activity in HVC is largely shaped by the intrinsic properties of the individual neurons as well as the synaptic properties where excitation and inhibition play a major role in enabling neurons to generate their characteristic bursts during singing. This is true and holds irrespective of whether an external drive is injected onto the microcircuits or not. We will however elaborate on and investigate this more during the next submission.

Another weakness of this paper is an unsatisfactory discussion of how the model was obtained, validated, and simulated. The authors should state as clearly as possible, in one location such as an appendix, what is the total number of independent parameters for the entire network and how parameter values were deduced from data or assigned by hand. With enough parameters and variables, many details can be fit arbitrarily accurately so researchers have to be careful to avoid overfitting. If parameter values were obtained by fitting to data, the authors should state clearly what the fitting algorithm was (some iterative nonlinear method, whose results can depend on the initial choice of parameters), what the error function used for fitting (sum of least squares?) was, and what data were used for the fitting.

The authors should also state clearly the dynamical state of the network, the vector of quantities that evolve over time. (What is the dimension of that vector, which is also the number of ordinary differential equations that have to be integrated?) The authors do not mention what initial state was used to start the numerical integrations, whether transient dynamics were observed and what were their properties, or how the results depended on the choice of the initial state. The authors do not discuss how they determined that their model was programmed correctly (it is difficult to avoid typing errors when writing several pages or more of a code in any language) or how they determined the accuracy of the numerical integration method beyond fitting to experimental data, say by varying the time step size over some range or by comparing two different integration algorithms.

We thank the reviewer again. The fitting process in our model occurred only at the first stage where the synaptic parameters were fit to the Mooney and Prather as well as the Kosche results. There was no data shared and we merely looked at the figures in those papers and checked the amplitude of the elicited currents, the magnitudes of DC-evoked excitations etc, and we replicated that in our model. While this is suboptimal, it was better for us to start with it rather than simply using equations for synaptic currents from the literature for other types of neurons (that are not even HVC’s or in the songbird) and integrate them into our network model. However, we will certainly highlight the details of this fitting process in the new submission. We will also highlight more technical details in the Methods regarding the exact number of ODEs, the initial conditions to run them, etc.

Also disappointing is that the authors do not make any predictions to test, except rather weak ones such as that varying a maximum conductance sufficiently (which might be possible by using dynamic clamps) might cause burst propagation to stop or change its properties. Based on their results, the authors do not make suggestions for further experiments or calculations, but they should.

We agree that making experimental testable predictions is crucial for the advancement of the model. Our predictions include testing whether eradication of a class of neurons such as HVC_X neurons disrupts activity propagation which can be done through targeted neuron elimination. This also can be done through preventing rebound bursting in HVC_X by pharmacologically blocking the I_h channels. Others include down regulation of certain ion channels (pharmacologically done through ion blockers) and testing which current is fundamental for song production (and there a plenty of test based our results, like the SK current, the T-type Ca⁺⁺ current, the A-type K⁺ current, etc). We will incorporate these into the revised manuscript to better demonstrate the model's applicability and to guide future research directions.

为了确保在计算阶段开始之前完成对共享内存的写入，需要同步一次。然后，为了防止在所有的线程完成他们的计算之前覆盖共享内存，在计算阶段之后也需要同步一次。

Weak bullet comparatively -- might request you to drop it

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

DiPeso et al. develop two tools to (i) classify micronucleated (MN) cells, which they call VCS MN, and (ii) segment micronuclei and nuclei with MMFinder. They then use these tools to identify transcriptional changes in MN cells.

The strengths of this study are:

(1) Developing highly specialized tools to speed up the analysis of specific cellular phenomena such as MN formation and rupture is likely valuable to the community and neglected by developers of more generalist methods.

(2) A lot of work and ideas have gone into this manuscript. It is clearly a valuable contribution.

(3) Combining automated analysis, single-cell labeling, and cell sorting is an exciting approach to enrich phenotypes of interest, which the authors demonstrate here.

Weaknesses:

(1) Images and ground truth labels are not shared for others to develop potentially better analysis methods.

We regret this omission and thank the reviewer for pointing it out. Both the images and ground truth labels for VCS MN and MNFinder are now available on the lab’s github page and described in the README.txt files. VCS MN: https://github.com/hatch-lab/fast-mn. MNFinder: https://github.com/hatch-lab/mnfinder.

(2) Evaluations of the methods are often not fully explained in the text.

The text has been extensively updated to include a full description of the methods and choices made to develop the VCS MN and MNFinder image segmentation modules.

(3) To my mind, the various metrics used to evaluate VCS MN reveal it not to be terribly reliable. Recall and PPV hover in the 70-80% range except for the PPV for MN+. It is what it is - but do the authors think one has to spend time manually correcting the output or do they suggest one uses it as is?

VCS MN attempts to balance precision and recall with speed to reduce the fraction of MN changing state from intact to ruptured during a single cell cycle during a live-cell isolation experiment. In addition, we chose to prioritize inclusion of small MN adjacent to the nucleus in our positive calls. This meant that there were more false positives (lower PPV) than obtained by other methods but allowed us to include this highly biologically relevant class of MN in our MN+ population. Thus, for a comprehensive understanding of the consequences of MN formation and rupture, we recommend using the finder as is. However, for other visual cell sorting applications where a small number of highly pure MN positive and negative cells is preferred, such as clonal outgrowth or metastasis assays, we would recommend using the slower, but more precise, MNFinder to get a higher precision at a cost of temporal resolution. In addition, MNFinder, with its higher flexibility and object coverage, is recommended for all fixed cell analyses.

Reviewer #2 (Public review):

Summary:

Micronuclei are aberrant nuclear structures frequently seen following the missegregation of chromosomes. The authors present two image analysis methods, one robust and another rapid, to identify micronuclei (MN) bearing cells. The authors induce chromosome missegregation using an MPS1 inhibitor to check their software outcomes. In missegregation-induced cells, the authors do not distinguish cells that have MN from those that have MN with additional segregation defects. The authors use RNAseq to assess the outcomes of their MN-identifying methods: they do not observe a transcriptomic signature specific to MN but find changes that correlate with aneuploidy status. Overall, this work offers new tools to identify MN-presenting cells, and it sets the stage with clear benchmarks for further software development.

Strengths:

Currently, there are no robust MN classifiers with a clear quantification of their efficiency across cell lines (mIoU score). The software presented here tries to address this gap. GitHub material (tools, protocols, etc) provided is a great asset to naive and experienced computational biologists. The method has been tested in more than one cell line. This method can help integrate cell biology and 'omics' studies.

Weaknesses:

Although the classifier outperforms available tools for MN segmentation by providing mIOU, it's not yet at a point where it can be reliably applied to functional genomics assays where we expect a range of phenotypic penetrance.

We agree that the MNFinder module has limitations with regards to the degree of nuclear atypia and cell density that can be tolerated. Based on the recall and PPV values and their consistency across the majority conditions analyzed, we believe that MNFinder can provide reliable results for MN frequency, integrity, shape, and label characteristics in a functional genomics assay in many commonly used adherent cell lines. We also added a discussion of caveats for these analyses, including the facts that highly lobulated nuclei will have higher false positive rates and that high cell confluency may require additional markers to ensure highly accurate assignment of MN to nuclei.

Spindle checkpoint loss (e.g., MPS1 inhibition) is expected to cause a variety of nuclear atypia: misshapen, multinucleated, and micronucleated cells. It may be difficult to obtain a pure MN population following MPS1 inhibitor treatment, as many cells are likely to present MN among multinucleated or misshapen nuclear compartments. Given this situation, the transcriptomic impact of MN is unlikely to be retrieved using this experimental design, but this does not negate the significance of the work. The discussion will have to consider the nature, origin, and proportion of MN/rupture-only states - for example, lagging chromatids and unaligned chromosomes can result in different states of micronuclei and also distinct cell fates.

We appreciate the reviewer’s comments and now quantify the frequency of other nuclear atypias and MN chromosome content in RPE1 cells after 24 h Mps1 inhibition (Fig. S1). In summary, we find only small increases in nuclear atypia, including multinucleate cells, misshapen nuclei, and chromatin bridges, compared to the large increase in MN formation. This contrasts with what is observed when mitosis is delayed using nocodazole or CENPE inhibitors where nuclear atypia is much more frequent. Importantly, after Mps1 inhibition, RPE1 cells with MN were only slightly more likely to have a misshapen nucleus compared to cells without MN (Fig. S1C).

Interestingly, this analysis showed that the VCS MN pipeline, which uses the Deep Retina segmenter to identify nuclei, has a strong bias against lobulated nuclei and frequently fails to find them (Fig. S2B). Therefore, the cell populations analyzed by RNAseq were largely depleted of highly misshapen nuclei and differences in nuclear atypia frequency between MN+ and MN- cells in the starting population were lost (Fig. S9A, compare to Fig. S1C). This strongly suggests that the transcript changes we observed reflect differences in MN frequency and aneuploidy rather than differences in nuclei morphology.

We agree with the reviewer that MN rupture frequency and formation, and downstream effects on cell proliferation and DNA damage, are sensitive to the source of the missegregated chromatin. In the revised manuscript we make clear that we chose Mps1 inhibition because it is strongly biased towards whole chromosome MN (Fig. S1E), limiting signal from DNA damage products, including chromosome fragments and chromatin bridges. This provides a base line to disambiguate the consequences of micronucleation and DNA damage in more complex chromosome missegregation processes, such as DNA replication disruption and irradiation. 

Reviewer #3 (Public review):

Summary:

The authors develop a method to visually analyze micronuclei using automated methods. The authors then use these methods to isolate MN post-photoactivation and analyze transcriptional changes in cells with and without micronuclei of RPE-1 cells. The authors observe in RPE-1 cells that MN-containing cells show similar transcriptomic changes as aneuploidy, and that MN rupture does not lead to vast changes in the transcriptome.

Strengths:

The authors develop a method that allows for automating measurements and analysis of micronuclei. This has been something that the field has been missing for a long time. Using such a method has the potential to advance micronuclei biology. The authors also develop a method to identify cells with micronuclei in real time and mark them using photoconversion and then isolate them via FACS. The authors use this method to study the transcriptome. This method is very powerful as it allows for the sorting of a heterogenous population and subsequent analysis with a much higher sample number than could be previously done.

Weaknesses:

The major weakness of this paper is that the results from the RNA-seq analysis are difficult to interpret as very few changes are found to begin with between cells with MN and cells without. The authors have to use a 1.5-fold cut-off to detect any changes in general. This is most likely due to the sequencing read depth used by the authors. Moreover, there are large variances between replicates in experiments looking at cells with ruptured versus intact micronuclei. This limits our ability to assess if the lack of changes is due to truly not having changes between these populations or experimental limitations. Moreover, the authors use RPE-1 cells which lack cGAS, which may contribute to the lack of changes observed. Thus, it is possible that these results are not consistent with what would occur in primary tissues or just in general in cells with a proficient cGAS/STING pathway.

We agree with the reviewer’s assessment of the limitations of our RNA-Seq analysis. After additional analysis, we propose an alternative explanation for the lower expression changes we observe in the MN+ and Mps1 inhibitor RNA-Seq experiments. In summary, we find that VCS MN has a strong bias against highly lobulated nuclei that depletes this class of cells from both the bulk analysis and the micronucleated cell populations (Fig. S9A). Based on this result, we propose that our analysis reduces the contribution of nuclear atypia to these transcriptional changes and that nuclear morphology changes are likely a signaling trigger associated with aneuploidy.

We believe that this finding strengthens our overall conclusion that MN formation and rupture do not cause transcriptional changes, as suppressing the signaling associated with nuclei atypia should increase sensitivity to changes from the MN. However, we cannot completely rule out that MN formation or rupture cause a broad low-level change in transcription that is obscured by other signals in the dataset.

As to cGAS signaling, several follow up papers and even the initial studies from the Greenburg lab show that MN rupture does not activate cGAS and does not cause cGAS/STING-dependent signaling in the first cell cycle (see citations and discussion in text). Therefore, we expect the absence of cGAS in RPE1 cells will have no effect in the first cell cycle, but could alter the transcriptional profile after mitosis. Although analysis of RPE1  cGAS+ cells or primary cells in these experiments will be required to definitively address this point, we believe that our interpretation of our RNAseq results is sufficiently backed up by the literature to warrant our conclusion that MN formation and rupture do not induce a transcriptional response in the first cell cycle.

Reviewer #1 (Recommendations for the authors):

I do not recommend additional experimental or computational work. Instead, I just recommend adapting the claims of the manuscript to what has been done. I am just asking for further clarification and minor rewriting.

(1) The manuscript is written like a molecular biology paper with sparse explanations of the authors' reasoning, especially in the development of their algorithms. I was often lost as to why they did things in one way or another.

The revised manuscript has thorough explanations and additional data and graphics defining how and why the VCS MN and MNFinder modules were developed. We hope that this clears up many of the questions the reviewer had and appreciate their guidance on making it more readable for scientists from different backgrounds.

(2) Evaluations of their method are often not fully explained, for example:

"On average, 75% of nuclei per field were correctly segmented and cropped."

"MN segments were then assigned to 'parent' nuclei by proximity, which correctly associated 97% of MN."

Were there ground truth images and labels created? How many? For example, I don't know how the authors could even establish a ground-truth for associating MNs to nuclei if MNs happened to be almost equidistant between two nuclei in their images.

I suggest a separate subsection early in the Results section where the underlying imaging data + labels are presented.

We added new sections to the text and figures at the beginning of the VCS MN and MNFinder subsections (Fig. S2 and Fig. S5) with specific information about how ground truth images and labels were generated for both modules and how these were broken up for training, validation, and testing.

We also added information and images to explain how ground truth MN/nucleus associations were derived. In summary, we took advantage of the fact that 2xDendra-NLS is present at low levels in the cytoplasm to identify cell boundaries. This combined with a subconfluent cell population allowed us to unambiguously group MN and nuclei for 98% of MN, we estimate. These identifications were used to generate ground truth labels and analyze how well proximity defines MN/nuclei groups (Fig.s S1 and S2).

(3) Overall, I find the sections long and more subtitles would help me better navigate the manuscript.

Where possible, we have added subtitles.

(4) Everything following "To train the model, H2B channel images were passed to a Deep Retina neural net ..." is fully automated, it seems to me. Thus, there seems to be no human intervention to correct the output before it is used to train the neural network. Therefore, I do not understand why a neural network was trained at all if the pipeline for creating ground truth labels worked fully automatically. At least, the explanations are insufficient.

We apologize for the initial lack of clarity in the text and included additional details in the revision. We used the Deep Retina segmenter to crop the raw images to areas around individual nuclei to accelerate ground truth labeling of MN. A trained user went through each nucleus crop and manually labeled pixels belonging to MN to generate the ground truth dataset for training, validation, and imaging in VCS MN (Fig. S2A).

(5) To my mind, the various metrics used to evaluate VCS MN reveal it not to be terribly reliable. Recall and PPV hover in the 70-80% range except for the PPV for MN+. It is what it is - but do the authors think one has to spend time manually correcting the output or do they suggest one uses it as is? I understand that for bulk transcriptomics, enrichment may be sufficient but for many other questions, where the wrong cell type could contaminate the population, it is not.

Remarks in the Results section on what the various accuracies mean for different applications would be good (so one does not need to wait for the Discussion section).

One of the strengths of the visual cell sorting system is that any image analysis pipeline can be used with it. We used VCS MN for the transcriptomics experiment, but for other applications a user could run visual cell sorting in conjunction with MNFinder for increased purity while maintaining a reasonable recall or use a pre-existing MN segmentation program that gives 100% purity but captures only a specific subgroup of micronucleated cells (e.g. PIQUE). 

To maintain readability, especially with the expansion of the results sections, we kept the discussion of how we envision using visual cell sorting for other MN-based applications in the discussion section.

(6) I am confused about what "cell" is referring to in much of the manuscript. Is it the nucleus + MNs only? Is it the whole cell, which one would ordinarily think it is? If so, are there additional widefield images, where one can discern cell boundaries? I found the section "MNFinder accurately ..." very hard to read and digest for this reason and other ambiguous wording. I suggest the authors take a fresh look at their manuscript and see whether the text can be improved for clarity. I did not find it an easy read overall, especially the computational part.

After re-examining how “cell” was used, we updated the text to limit its use to the MNFinder arm tasked with identifying MN-nucleus associations where the convex hull defined by these objects is used to determine the “cell” boundary. In all other cases we have replaced cell with “nucleus” because, as the reviewer points out, that is what is being analyzed and converted. We hope this is clearer.

(7) Post-FACS PPVs are not that great (Figure 3c). It depends on the question one wants to answer whether ~70% PPV is good enough. Again, would be good to comment on.

We added discussion of this result to the revision. In summary, a likely reason for the reduced PPV is that, although we maintain the cells in buffer with a Cdk1 inhibitor, we know that some proportion of the cells go through mitosis post-sorting. Since MN are frequently reincorporated into the nucleus after mitosis (Hatch et al, 2013; Zhang et al., 2015), we expect this to reduce the MN+ population. Thus, we expect that the PPV in the RNAseq population is higher than what we can measure by analyzing post-sorted cells that have been plated and analyzed later.

(8) I am thoroughly confused as to why the authors claim that their system works in the "absence of genetic perturbations" and why they emphasize the fact that their cells are non-transformed: They still needed a fluorescent label and they induce MNs with a chemical Mps1 inhibitor. (The latter is not a genetic manipulation, of course, but they still need to enrich MNs somehow. That is, their method has not been tested on a cell population in which MNs occur naturally, presumably at a very low rate, unless I missed something.) A more careful description of the benefits of their method would be good.

We apologize for the confusion on these points and hope this is clarified in the revision. We were comparing our system, which can be made using transient transfection, if desired, to current tools that disambiguate aneuploidy and MN formation by deleting parts of chromosomes or engineering double strand breaks with CRISPR to generate single chromosome-specific missegregation events. Most of these systems require transformed cancer cells to obtain high levels of recombination. In contrast, visual cell sorting can isolate micronucleated cells from any cell line that can exogenously express a protein, including primary cells and non-transformed cells like RPE1s.

Other minor points:

(1) The authors should not refer to "H2B channels" but to "H2B-emiRFP703 channels". It may seem obvious to the authors but for someone reading the manuscript for the very first time, it was not. I was not sure whether there were additional imaging modalities used for H2B/nucleus/chromatin detection before I went back and read that only fluorescence images of H2B-emiRFP703 were used. To put it another way, the authors are detecting fluorescence, not histones -- unless I misunderstood something.

To address this point, we altered the text to read “H2B-emiRFP703” when discussing images of this construct. For MNFinder some images were of cells expressing H2B-GFP, which has also been clarified.

(2) If the level of zoom on my screen is such that I can comfortably read the text, I cannot see much in the figure panels. The features that I should be able to see are the size of a title. The image panels should be magnified.

In the revision, the images are appended to the end at full resolution to overcome this difficulty. Thank you for your forbearance.

Reviewer #2 (Recommendations for the authors):

The methods are adequately explained. The Results text narrating experiments and data analysis is clear. Interpretation of a few results could be clarified and strengthened as explained below.

(1) RNAseq experiments are a good proof of principle. To strengthen their interpretation in Figures 4 and 6, I would recommend the authors cite published work on checkpoint/MPS1 loss-induced chromosome missegregation (PMID: 18545697, PMID: 33837239, PMC9559752) and consider in their discussion the 'origin' and 'proportion' of micronucleated cells and irregularly shaped nuclei expected in RPE1 lines. This will help interpret Figure 6 findings on aneuploidy signature accurately. Not being able to see an MN-specific signature could be due to the way the biological specimen is presented with a mixture of cells with 'MN only' or 'rupture' or 'MN along with misshapen nuclei'. These features may all link to aneuploidy rather than 'MN' specifically.

We appreciate the reviewer’s suggestion and added a new analysis of nuclear atypia after Mps1 inhibition in RPE1 cells to Fig. S1. Overall, we found that Mps1 inhibition significantly, but modestly, increased the proportion of misshapen nuclei and chromatin bridges. Multinucleate cells were so rare that instead of giving them their own category we included them in “misshapen nuclei.” These results are consistent with images of Msp1i treated RPE1 cells from He et al. 2019 and Santaguida et al. 2017 and distinct from the stronger changes in nuclear morphology observed after delaying mitosis by nocodazole or CENPE inhibition.

We also found that the Deep Retina segmenter used to identify nuclei in VCS MN had a significant bias against highly lobulated nuclei (Fig. S2B) that led to misshapen nuclei being largely excluded from the RNAseq analyses. As a result we found no enrichment of misshapen nuclei, chromatin bridges, or dead/mitotic nuclear morphologies in MN+ compared to MN- nuclei in our RNASeq experiments (Fig. S9A).

(2) As the authors clarify in the response letter, one round of ML is unlikely to result in fully robust software; additional rounds of ML with other markers will make the work robust. It will be useful to indicate other ML image analysis tools that have improved through such reiterations. They could use reviews on challenges and opportunities using ML approaches to support their statement. Also in the introduction, I would recommend labelling as 'rapid' instead of 'rapid and precise' method.

We updated the text to reference review articles that discuss the benefit of additional training for increasing ML accuracy and changed the text to “rapid.”

(3) The lack of live-cell studies does not allow the authors to distinguish the origin of MN (lagging chromatids or unaligned chromosomes). As explained in 1, considering these aspects in discussion would strengthen their interpretation. Live-cell studies can help reduce the dependencies on proximity maps (Figure S2).

The revised text includes new references and data (Fig. S1E) demonstrating that Mps1 inhibition strongly biases towards whole chromosome missegregation and that MN are most likely to contain a single centromere positive chromosome rather than chromatin fragments or multiple chromosomes.

(4) Mean Intersection over Union (mIOU) is a good measure to compare outcomes against ground truth. However, the mIOU is relatively low (Figure 2D) for HeLa-based functional genomics applications. It will help to discuss mIOU for other classifiers (non-MN classifiers) so that they can be used as a benchmark (this is important since the authors state in their response that they are the first to benchmark an MN classifier). There are publications for mitochondria, cell cortex, spindle, nuclei, etc. where IOU has been discussed.

We added references to classifiers for other small cellular structures. We also evaluated major sources of error in MNFinder found that false negatives are enriched in very small MN (3 to 9 pixels, or about 0.4 µm² – 3 µm², Fig. S6B). A similar result was obtained for VCS MN (Fig. S3B). Because small changes in the number of pixels identified in small objects can have outsized effects on mIoU scores, we suspect that this is exerting downward pressure on the mIoU value. Based on the PPV and recall values we identified, we believe that MNFinder is robust enough to use for functional genomics and screening applications with reasonable sample sizes.

(5) Figure 5 figure legend title is an overinterpretation. MN and rupture-initiated transcriptional changes could not be isolated with this technique where several other missegregation phenotypes are buried (see point 1 above).

We decided to keep the figure title legend based on our analysis of known missegregation phenotypes in Fig. S1 and S9 showing that there is no difference in major classes of nuclear atypia between MN+ and MN- populations in this analysis. Although we cannot rule out that other correlated changes exist, we believe that the title represents the most parsimonious interpretation.

Minor comments

(1) The sentence in the introduction needs clarification and reference. "However, these interventions cause diverse "off-target" nuclear and cellular changes, including chromatin bridges, aneuploidy, and DNA damage." Off-target may not be the correct description since inhibiting MPS1 is expected to cause a variety of problems based on its role as a master kinase in multiple steps of the chromosome segregation process. Consider one of the references in point 1 for a detailed live-cell view of MPS1 inhibitor outcomes.

We have changed “off-target” to “additional” for clarity.

(2) In Figure 3 or S3, did the authors notice any association between the cell cycle phase and MN or rupture presence? Is this possible to consider based on FACS outcomes or nuclear shapes?

Previous work by our lab and others have shown that MN rupture frequency increases during the cell cycle (Hatch et al., 2013; Joo et al., 2023). Whether this is stochastic or regulated by the cell cycle may depend on what chromosome is in the MN (Mammel et al., 2021) and likely the cell line. Unfortunately, the H2B-emiRFP703 fluorescence in our population is too variable to identify cell cycle stage from FACS or nuclear fluorescence analysis.

(3) Figure 5 - Please explain "MA plot".

An MA plot, or log fold-change (M) versus average (A) gene expression, is a way to visualize differently expressed genes between two conditions in an RNASeq experiment and is used as an alternative to volcano plots. We chose them for our paper because most of the expression changes we observed were small and of similar significance and the MA plot spreads out the data compared to a volcano plot and allowed a better visualization of trends across the population.

(4) Page 7: "our results strongly suggest that protein expression changes in MN+ and rupture+ cells are driven mainly by increased aneuploidy rather than cellular sensing of MN formation and rupture.". This is an overstatement considering the mIOU limits of the software tool and the non-exclusive nature of MN in their samples.

We agree that we cannot rule out that an unknown masking effect is inhibiting our ability to observe small broad changes in transcription after MN formation or rupture. However, we believe we have minimized the most likely sources of masking effects, including nuclear atypia and large scale aneuploidy differences, and thus our interpretation is the most likely one.

Reviewer #3 (Recommendations for the authors):

Overall, the authors need to explain their methods better, define some technical terms used, and more thoroughly explain the parameters and rationale used when implementing these two protocols for identifying micronuclei; primarily as this is geared toward a more general audience that does not necessarily work with machine learning algorithms.

(1) A clearer description in the methods as to how accuracy was calculated. Were micronuclei counted by hand or another method to assess accuracy?

We significantly expanded the section on how the machine learning models were trained and tested, including how sensitivity and specificity metrics were calculated, in both the results and the methods sections. The code used to compare ground truth labels to computed masks is also now included in the MNFinder module available on the lab github page. 

(2) Define positive predictive value.

The text now says “the positive predictive value (PPV, the proportion of true positives, i.e. specificity) and recall (the proportion of MN found by the classifier, i.e. sensitivity)…”.

(3) Why is it a problem to use the VCS MN at higher magnifications where undersegmentation occurs? What do the authors mean by diminished performance (what metrics are they using for this?).

We have included a representative image and calculated mIoU and recall for 40x magnification images analyzed by MNFinder after rescaling in Fig. 2A. In summary, VCS MN only correctly labeled a few pixels in the MN, which was sufficient to call the adjacent nucleus “MN+” but not sufficient for other applications, such as quantifying MN area. In addition, VCS MN did much worse at identifying all the MN in 40x images with a recall, or sensitivity, metric of 0.36. We are not sure why. Developing MNFinder provided a module that was well suited to quantify MN characteristics in fixed cell images, an important use case in MN biology.

(4) The authors should compare MN that are analyzed and not analyzed using these methods and define parameters. Is there a size limitation? Closeness to the main nucleus?

We added two new figures defining what contributes to module error for both VCS MN (Fig. S3) and MNFinder (Fig. S6). For VCS MN, false negatives are enriched in very large or very small MN and tend to be dimmer and farther from the nucleus than true positives. False positives are largely misclassification of small dim objects in the image as MN. For MNFinder, the most missed class of MN are very small ones (3-9 px in area) and the majority of false positives are misclassifications of elongated nuclear blebs as MN.

(5) Are there parameters in how confluent an image must be to correctly define that the micronucleus belongs to the correct cell? The authors discussed that this was calculated based on predicted distance. However, many factors might affect proper calling on MN. And the authors should test this by staining for a cytosolic marker and calculating accuracy.

We updated the text with more information about how the cytoplasm was defined using leaky 2x-Dendra2-NLS signal to analyze the accuracy of MN/nucleus associations (Fig. S2G-H). In addition, we quantified cell confluency and distance to the first and second nearest neighbor for each MN in our training and testing image datasets. We found that, as anticipated, cells were imaged at subconfluent concentrations with most fields having a confluency around 30% cell coverage (Fig. S2E) and that the average difference in distance between the closest nucleus to an MN and the next closest nucleus was 3.3 fold (Fig. S2F). We edited the discussion section to state that the ability of MN/nuclear proximity to predict associations at high cell confluencies would have to be experimentally validated.

(6) The authors measure the ratio of Dendra2(Red) v. Dendra2 (Green) in Figure 3B to demonstrate that photoconversion is stable. This measurement, to me, is confusing, as in the end, the authors need to show that they have a robust conversion signal and are able to isolate these data. The authors should directly demonstrate that the Red signal remains by analyzing the percent of the Red signal compared to time point 0 for individual cells.

We found a bulk analysis to be more powerful than trying to reidentify individual cells due to how much RPE1 cells move during the 4 and 8 hours between image acquisitions. In addition, we sort on the ratio between red and green fluorescence per cell, rather than the absolute fluorescence, to compensate for variation in 2xDendra-NLS protein expression between cells. Therefore, demonstrating that distinct ratios remained present throughout the time course is the most relevant to the downstream analysis.

To address the reviewer’s concern, we replotted the data in Fig. 3B to highlight changes over time in the raw levels of red and green Dendra fluorescence (Fig. S7D). As expected, we see an overall decrease in red fluorescence intensity, and complementary increase in green fluorescence intensity, over 8 hours, likely due to protein turnover. We also observe an increase in the number of nuclei lacking red fluorescence. This is expected since the well was only partially converted and we expect significant numbers of unconverted cells to move into the field between the first image and the 8 hour image.

(7) The authors isolate and subsequently use RNA-sequencing to identify changes between Mps1i and DMSO-treated cells. One concern is that even with the less stringent cut-off of 1.5 fold there is a very small change between DMSO and MPS1i treated cells, with only 63 genes changing, none of which were affected above a 2-fold change. The authors should carefully address this, including why their dataset sees changes in many more pathways than in the He et al. and Santaguida et al. studies. Is this due to just having a decreased cut-off?

The reviewer correctly points out that we observed an overall reduction in the strength of gene expression changes between our dataset of DMSO versus Mps1i treated RPE1 cells compared to similar studies. We suggest a couple reasons for this. One is that the log₂ fold changes observed in the other studies are not huge and vary between 2.5 and -3.8 for He et al., 3.3 and -2.3 for Santaguida et al., and -0.8 and 1.6 for our study. This variability is within a reasonable range for different experimental conditions and library prep protocols. A second is that our protocol minimizes a potential source of transcriptional change – nuclear lobulation – that is present in the other datasets.

For the pathway analysis we did not use a fold-change cut-off for any data set, instead opting to include all the genes found to be significantly different between control and Mps1i treated cells for all three studies. Our read-depth was higher than that of the two published experiments, which could contribute to an increased DEG number. However, we hypothesize that our identification of a broader number of altered pathways most likely arises from increased sensitivity due to the loss of covering signal from transcriptional changes associated with increased nuclear atypia. Additional visual cell sorting experiments sorting on misshapen nuclei instead of MN would allow us to determine the accuracy of this hypothesis.

(8) Moreover, clustering (in Figure 5E) of the replicates is a bit worrisome as the variances are large and therefore it is unclear if, with such large variance and low screening depth, one can really make such a strong conclusion that there are no changes. The authors should prove that their conclusion that rupture does not lead to large transcriptional changes, is not due to the limitations of their experimental design.

We agree with the reviewers that additional rounds of RNAseq would improve the accuracy of our transcriptomic analysis and could uncover additional DEGs. However, we believe the overall conclusion to be correct based on the results of our attempt to validate changes in gene expression by immunofluorescence. We analyzed two of the most highly upregulated genes in the ruptured MN dataset, ATF3 and EGR1. Although we saw a statistically significant increase in ATF3 intensity between cells without MN and those with ruptured MN, the fold change was so small compared to our positive control (100x less) that we believe it is it is more consistent with a small increase in the probability of aneuploidy rather than a specific signature of MN rupture.

(9) The authors also need to address the fact that they are using RPE-1 cells more clearly and that the lack of effect in transcriptional changes may be simply due to the loss of cGAS-STING pathway (Mackenzie et al., 2017; Harding et al., 2017; etc.).

As we discuss above in the public comments section, the literature is clear that MN do not activate cGAS in the first cell cycle after their formation, even upon rupture. Therefore, we do not expect any changes in our results when applied to cGAS-competent cells. However, this expectation needs to be experimentally validated, which we plan to address in upcoming work.

Claude 3.7 Sonnet again excels at coding, as verified by multiple teams

Introduction and Motivation: Term rewriting with Meander offers an intuitive introduction aimed at everyday software engineers.

> "Meander is heavily inspired by the capabilities of term rewriting languages. But sadly, there aren't many introductions to term rewriting aimed at everyday software engineers."

Basic Concept of Term Rewriting: It transforms data by applying rules that match a left-hand-side pattern to produce a right-hand-side output.

> "The goal of Term Rewriting is to take some bit of data and rewrite it into some other bit of data. We accomplish this by writing rules that tell us for a given piece of data what we should turn it into."

Simple Rewrite Rule Example: A basic rule maps a specific input (e.g., :x) to a designated output (:y).

> "Here is the most simple rewrite rule imaginable. If we are given :x we turn it into :y."

Combining Multiple Rewrite Rules: Multiple rules can be defined to handle various inputs simultaneously.

> "Here we've extended our rewrite to have multiple rules."

Utilizing Variables in Patterns: Variables (prefixed with ?) match any value and allow that matched value to be reused in the output.

> "Here we added the variable ?x to our left-hand-side. Variables start with a ? and match any value."

Pattern Matching on Data Structures: Rules can be crafted to operate on vectors, enabling extraction of specific elements such as the first element.

> "Here we can see some really simple rules that work on vectors of various sizes. We can use this to extract the first element from each."

Rewriting Strategies for Control: Strategies allow precise control over how and when rewrite rules are applied during computation.

> "Strategies let us control how our terms are rewritten."

The Attempt Strategy: The attempt strategy tries to apply a rewrite and, if it fails, returns the original value to handle cases where no match is found.

> "We can fix that by using the attempt strategy. It will try to rewrite and if it fails, just return our value."

Iterative Application with (until =) Strategy: The (until =) strategy repeatedly applies rules until the expression stops changing.

> "What we really want to say is to continue applying our rewrite rules until nothing changes. We can do that by using the (until =) strategy."

Traversal Approaches – Bottom-Up vs. Top-Down: Different strategies, such as bottom-up and top-down, affect the order and frequency of rule application in nested expressions.

> "If we look at the top-down approach, we can see that the top-down strategy actually gets called three times... Our bottom-up strategy however is only called twice."

Tracing the Rewriting Process: The trace strategy provides visibility into each step of the rewriting process for debugging and analysis.

> "We can inspect our strategies at any point by using the trace strategy."

General Computation via Term Rewriting: Term rewriting can express any computable function, exemplified by implementing Fibonacci with Peano numbers.

> "Term Rewriting is a general programming technique. Using it we can compute absolutely anything that is computable."

Code and Execution as Data: It enables code to be treated as data, allowing for introspection of intermediate execution steps.

> "Not only can our 'code' be data more than it can in lisp, but we can actually have our execution as data."

Support for Partial Programs: The paradigm facilitates working with incomplete programs by allowing unimplemented parts to be represented without immediate failure.

> "Term Rewriting also gives us an easy basis for talking about partial programs."

A New Programming Paradigm: Term rewriting represents a uniform, pattern-based approach to programming that challenges traditional distinctions between code and data.

> "Term Rewriting represents a distinct way of programming. It offer s us a uniform way of dealing with data."

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-2024-02588

Corresponding author(s): Frederic SALTEL

__1. __Point-by-point description of the revisions

Reviewer #1:

Invadosomes are dynamic, actin-based structures that enable cells to interact with and remodel the extracellular matrix (ECM), playing a crucial role in tumor cell invasion and metastasis. Prior studies by the authors and other groups have established the formation, activation, and appearance of invadosomes. This study demonstrates the following:

1. Key elements of the translation machinery and endoplasmic reticulum (ER) proteins are constituents of the invadosome structure.
2. Specific proteins are associated with distinct invadosome structures.

The researchers utilized two cellular models (NIH3T3-Src and A431 melanoma cell line) and Tks5, a specific invadosome marker, for immunoprecipitation and mass spectrometry, validating the results through fluorescent images, electron microscopy, and time-lapse live imaging.

Major Comments

The manuscript is well-written, with a clear and detailed experimental workflow. Compared to their previous seminal work that first demonstrated invadosomes concentrate mRNA and exhibit translational activity using NIH3T3-Src cells, this study adds details about the specific enrichment of translation proteins for each type of invadosome and the presence of ribosomal and ER proteins. However, the experiments do not further enhance our understanding of the intricate mechanisms linking invadosome structures, function, and translation factors.

Further experiments are needed to better demonstrate the hypothesis of active translation within these structures, including the use of additional cellular models.

To demonstrate the hypothesis of active translation within these structures, we performed the same translation inhibition experiments, using CHX in additional cellular models. Indeed, these experiments were performed on MDA-MB-231 breast cancer cell lines, as well as on Huh6 liver cancer cell lines. Degradation experiments showed the same results as for NIH-3T3-Tks5-GFP and A431-Tks5-GFP, since we were able to observe a significant decrease in the degradation capacities of cells in the absence of translation (see graphs below).

Left: Quantification and representative images of ECM degradation properties of Huh6 cells on gelatin treated (CHX) or not (DMSO) with cycloheximide. Gelatin is stained in green and nuclei in blue. Values represent the mean +/- SEM of n=4 independent experiments (15 images per condition and per replicate) and were analyzed using student t-test.

Right: Left: Quantification and representative images of ECM degradation properties of MDA-MB-231 cells on gelatin treated (CHX) or not (DMSO) with cycloheximide. Gelatin is stained in green and nuclei in blue. Values represent the mean +/- SEM of n=4 independent experiments (15 images per condition and per replicate) and were analyzed using student t-test.

The authors should also investigate the effects of Tks5 silencing on ER-associated translational machinery.

The effects of Tks5 silencing on the ER-associated translation machinery were investigated using a SunSET experiment. We were able to demonstrate that Tks5 silencing had no significant impact on translation in both cellular models since no translation modification was observed between control and siTks5 conditions.

Quantification and relative western blot analysis of the effect of Tks5-targeting siRNA treatment on A431 and NIH-3T3-Src cells by using puromycin quantification. Values represent the mean +/- SEM of n=4 independent experiments and were analyzed using Anova.

How do the authors propose Tks5 is linked to these proteins? Directly or indirectly? Focusing on specific proteins night provide an opportunity to study the molecular mechanisms in greater depth.

Tks5 is a scaffold protein, a multi-domain “bridging molecule” that serve as regulators by simultnneously binding multipe molecular partners. TKs5 contain a PX domain and 5 SAH Domains. Consequently, Tks5 can bind different partners. Moreover, as focal adhesion, invadosome are large macromolecular assemblies. Here, in this study, Tks5 serve as a specific molecular hook, to precipitate partners. At this step, there is no evidence of a direct or indirect link of the translational machineray with Tks5. Even if we can hypothetize un indirect connection. In this version we focused more precisely on a specific and common Tks5 partners, such as EIF4B.

They used chemical inhibitors and siRNA approaches to assess the role of specific players, such as EIF4B, in the proteolytic activity of invadosomes, which can be considered proof of concept. Additional experiments aligning the results with the involved pathways would add molecular details and enhance the manuscript's significance. Resolving these issues is crucial for the manuscript to meet the publication standards for contributing novel and impactful insights to the field.

To better understand the variation of the pathways involved, we first wanted to observe the impact of Eif4b silencing on active translation in both cellular models. To do this, we performed SunSET experiments in both cell models. An experiment was performed for the A431 cell line and the results seem to show little difference between control conditions and conditions in the presence of siEIF4B. Conversely, SunSET experiments in the NIH 3T3 Src cell line show an increase in translation in the presence of siEIF4B.

__ __

Quantification of the effect of cycloheximide (CHX) and EIF4B-targeting siRNA (siEIF4B #1 and #2) treatment on A431 and NIH-3T3-Src cells by using puromycin quantification. Values represent the mean +/- SEM of n=1 independent experiment for A431 or n=2 independent experiments for NIH-3T3-Src.

In order to better understand the variation of the signaling pathways involved, spectrometry experiments were performed to compare the variation of the pathways in control conditions and in the presence of siRNA against EIF4B. These results allowed us to provide a better understanding of the variability of the pathways and therefore of the mechanism of action.

Volcano plot of overexpressed and underexpressed proteins after silencing of the EIF4B protein identified by mass spectrometry analysis.

These mass spectrometry experiments allowed us to highlight that the pathway mainly impacted during Eif4b depletion was the Hras pathway. However, this information is given for information purposes only. It would be necessary to look more closely at the Hras pathway to understand what the link with EIF4B and therefore the link with the formation of invadosomes could be.

Table of translation-related proteins or proteins involved in the formation or function of invadosomes that are overexpressed or underexpressed in at least one siRNA of EIF4B.

These experiments also allowed us to highlight that the depletion of EIF4B directly impacts the translation pathway by modulating translation initiation factors as well as ribosomal proteins but also proteins involved in the formation and function of invadosomes such as ADAM17, ACTR5, IGFBP6 RPL22 and RPS6KA5 proteins (see table below). It will be necessary to validate these data and determine their specificity due to the fact that some other proteins appear under-expressed like IGFBP3 and ADAM19. To conclude, to fully understand the exact impact of EIF4B into this process, additional investigations are necessary.

__ __Minor Comments :

A more detailed discussion of the implications of their findings within the broader context of cancer cell signaling and the potential impact on related cancer research areas would further advance our understanding in this area.

This part was added in the new version of the discussion. Indeed, deregulation of the translation is now a hallmark of cancer. This notion is now present in the manuscript and concluded the discussion (see page 12).

Reviewer #1 (Significance (Required)):

General Assessment:

This study offers novel insights into a new function of the invadosome-specific player Tks5 as a molecular crossroad between ER-related translation proteins and invadosomes. The authors suggest that Tks5 could act as a scaffold, supporting the rapid clustering of translation-related proteins during invadosome formation or proteolytic activity. However, a major limitation is the lack of mechanistic exploration. The results do not elucidate how Tks5 mediates the recruitment of these proteins or the specific molecular mechanisms involved.

Advances: The study extends knowledge in the field by confirming the presence of specific markers linked to different invadosome structures and demonstrating the Tks5 interactome's association with translation machinery.

Audience: This study will primarily interest specialists working on invadosomes and, secondarily, those interested in cancer cell signaling, invasion, and metastasis.

Field of Expertise: Invadosome and related signaling pathways in cancer.

__ __

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

Summary In this work, Normand and her colleagues analyze and compare the interactome of the key invadopodia component, TKS5 (overexpressed as a GFP-tagged protein), in two transformed cell models cultured on different substrates. Potential TKS5 interacting partners are identified including previously known and validated TKS5 interactors, some known to contribute to the mechanism of invadopodia formation and function. Bioinformatic (GSEA) analysis reveals a specific enrichment for proteins related to protein translation and interaction with ER-associated ribosome machinery. Evidence is presented that some of these proteins (RPS6, a component of the 40S ribosomal subunit, and translation factor, EIF4B) localize to TKS5-positive invadopodia in Src-transformed cells. Experiments based on translation inhibitor, cycloheximide, and silencing of EIF4B factor could demonstrate a link between overall protein translation and invadosome formation. Live cell imaging and microscopy analysis of fixed samples could document some proximity between the endoplasmic reticulum network and invadosome rosettes.

Major comments

__ __1- In the Results Section, the IP/proteomics-based pipeline used by Normand and colleagues to identify TKS5 partners is not clearly described and is confusing. Cut-off used to select the proteins in the different classes summarized in Table S1 should be better described. In addition, the nomenclature of the different protein subgroups used in Table S1 is confusing (see minor point#5).

Details have been added in the results section regarding the IP/proteomics section to complete the materials and methods section. As described in the materials and methods section, control versus IP data were quantified by an enrichment ratio ≥ 2. These criteria are the most classically used in the practices analyzed.

For clarity, additional tables have been added for each category (A431/NIH plastic or collagen) and gene names, protein descriptions and abundance ratios have been indicated (Supp table 2, 3, 4 and 5).

2- The effects of cycloheximide treatment or EIF4B silencing on gelatin degradation are clear and convincing. However, these are correlative evidence, and they may reflect a general implication of protein translation in the control of invadopodia function. A direct link between the observed interactions of TKS5 with the protein translation machinery and the formation and/or function of invadopodia is missing.

To demonstrate the direct links between Tks5 and the translation machinery, a fluorophore was used to visualize active translation within invadopodia. We were able to highlight an active translation localized in the rosettes (see figure below). Indeed, we can observe a localized translation within the rosettes. However, these same results were not observed in linear invadosomes where we could not observe any localized translation. We can however hypothesize that it is more difficult to observe a localized translation in linear invadosomes which are much smaller structures than rosettes.

Confocal microscopy images of NIH-3T3-Src cells. The cells were stained for B-actin RNA in green, B-actin in red, nuclei in blue and actin in grey. Scale bar: 20µm, zoom: 5µm.

In order to provide additional elements to show the link between Tks5 and the translation machinery, we performed immunofluorescence experiments by labeling the Sec61 protein. Sec61 is a well-described ER marker that allows the insertion of proteins into the ER but is also a key player in the docking of ribosomes to the ER. We were able to highlight the colocalization between Tks5 and Sec61 in all types of invadosomes, allowing to show the link between the Tks5 protein and the translation machinery. These images were inserted in the manuscript (see Figure 6b).

Confocal microscopy images of NIH-3T3-Src and A431 cells. The cells were seeded on gelatin or type I collagen and stained for Sec61 in red, nuclei in blue and Actin in grey. Scale bar: 20µm, zoom: 5µm.

__ __3- Images showing the interrelations between the ER and the adhesive podosome rosettes are striking (Figure 5). Src-transformed cells forming invadosome rosettes when in contact with the collagen substratum change shape and produce adhesive protrusions towards the substratum. As the ER is a huge compartment that fills the entire cytoplasm, it is maybe not so surprising to observe the ER filling the protrusions and getting close to the rosettes at the tip of these membrane extensions. Again, these observations are essentially correlative and there is no prove of some direct contact between some ER regions and the invadosomes.

For clarity, the contrast of the images has been improved. Thus, time-lapse imaging clearly demonstrate that the ER is not present in all the cytoplasm but is enriched in the destination of the rosettes as well as in the rosettes. Moreover, this is not systematic with all invadosome rosettes (see video 1)

4- Overall, this report is lacking a clear hypothesis or model of what could be the consequence of the interaction of TKS5 and the translation machinery on the formation and/or the activity of the invadosomes in transformed cells.

We performed a sunset experiment to analyze the impact of Tks5 depletion into translation. No variation of global translation was observable in the absence of Tks5 (see results below). Tks5 depletion block invadosome formation. So, the impact on total translation activity cannot be measurable at the cell level, suggesting that invadosome recruit a specific translation machinery. Indeed, even if we obtained a good percentage of Tks5 depletion, around 90%, the impact in total translation activity is not quantifiable. However, we noticed that some specific translation actors are modulated and specifically localized into invadosome structures suggesting that it is more a question of localization and local translation of specific mRNAs, and not a global modification. This is consistent with the fact that Tks5 expression is not altered during tumor cell invasion, and it is just recruited and activated at specific sites to form these invasive structures.

Thus, in this paper, Tks5 only served as an anchor point in order to be able to extract the specific molecular machinery and specific translational actors.

Quantification and relative western blot analysis of the effect of Tks5-targeting siRNA treatment on A431 and NIH-3T3-Src cells by using puromycin quantification. Values represent the mean +/- SEM of n=4 independent experiments and were analyzed using Anova.

Minor comments

1- Discussion Section (page 2). The statement that TKS4 is involved in ECM degradation in podosomes only and not in invadopodia is not correct. TKS4 knock down has been shown to interfere with ECM degradation in Human DLD1 colon cancer cells (Gianni et al. SCIENCESIGNALING Vol 2 Issue 88, 2009) and in in mouse and human melanoma cell lines (Iizuka et al. Oncotarget, Vol. 7, 2016). In addition, an unphosphorylable mutant form of Tks4 blocked invadopodia formation and ECM degradation in Src-transformed DLD1 cells (Gianni et al. Molecular Biology of the Cell Vol. 21, 4287- 4298, 2010). We (this reviewer's team) reported that TKS4 was associated with cortactin-positive invadopodia in MDA-MB-231 and Hs578T triple-negative breast cancer cell lines (Zagryazhskaya-Masson et al. J. Cell Biol. 219, 2020).

The involvement of TKS4 protein in extracellular matrix degradation has been changed in the text (page 2).

2- Discussion Section (page 3). A431 is wrongly referred to as a melanoma cell line; it is a human epidermoid carcinoma cell line.

The text has been modified according to the recommendations, the A431 cell line has been designated as a human epidermoid carcinoma cell line.

3- Results Section (page 4 & 5). The authors compare the proteins they identified as potential TKS5 partners to previously published data by Stilly et al. (based on TKS5 IP like in the present study) and Thuault et al. (TKS5 bioIB). Additionally, authors should mention and discuss previously published data based on TKS5 coIP experiment and Mass Spec analysis similar to the present study, identifying potential TKS5 partners; some of which were similarly found in the present study including proteins involved in translation and ribosome function although these were not the focus of this work (several 40S and 60S ribosomal proteins, see Zagryazhskaya-Masson et al. J. Cell Biol. 219, 2020).

This comparison is now present int the text of the manuscript (page 10).

4- Figure 1b. Matrix degradation is not visible in association with the invadopodia in selected high magnification images in Figure 1a and 1b.

Matrix degradation is indeed not visible in association with invadopodia in the selected high magnification images. Indeed, the imaging techniques used, Interference Refection Microscopy (IRM) do not allow us to observe matrix degradation at the invadosomes, since the reflection also highlights the cells. The aim here was to show only the presence collagen fibers that correspond to inducer of linear invadosome reorganization. It is widely accepted that all these structures are capable of degrading the extracellular matrix.

5- Supplemental table 1. The names of the different lists of proteins in the summary table is not clear and is rather confusing.

For clarity, additional tables have been added for each category (A431/NIH plastic or collagen) and gene names, protein descriptions and abundance ratios have been indicated (Supp table 2, 3, 4 and 5).

6- Supp Figure 1. Please define what is the sample named 'D' (Delta).

The Delta sample corresponds to the material that was not attached to the bead.

7- Results Section (page 5). 'These experiments confirm the correct co-localization between Tks5 and the proteins identified in Tks5 interactome by mass spectrometry analysis.' This statement is too general; in fact, data validate only colocalization between TKS5 and some identified partners, namely CD44 and MAP4.

To be less general, this statement has been modified in the text to show that the data only validate colocalization between TKS5 and certain identified partners, namely CD44 and MAP4.

8- Figure 2e and Figure 3. It would have been nice to show the colocalization of selected proteins and TKS5 in association with collagen fibers to validate that enrichment occurs at matrix/cell contact sites and corresponds to bona fide invadopodia.

As commented above, the reflection highlights the collagen fibers but also the cells. Thus, it is complex in this case to show the colocalization of the selected proteins in association with the collagen fibers with this approach. The other possibility is to stain collagen fibrils, however this kind of approach reduce the quality of interaction between fibers and associated receptors inducing a decrease of linear invadosome formation.

9- Figure 3c (high mag insets). TKS5 and EIF4b do not seem particularly enriched in invadopodia rosettes as compared to the rest of the cytoplasm.

Indeed, we can observe on this image a colocalization of Tks5 and EIF4B in the rosettes without showing an enrichment.

However, the enrichment of EIF4B remains clearly visible in the linear invadosomes and the dots.

10- Figure 4c-f. Treatments (i.e. CHX, siEIF4b) affect gelatin degradation. It would be interesting to assess the capacity of cells to form invadopodia under these conditions.

As demonstrated in this study, the CHX treatment and EIF4B depletion affect the degradation of gelatin. In addition, we were able to show that CHX only impacts the formation of rosettes on gelatin (Figure 4a, 4b and Supp 3).

Moreover, we added in the manuscript the impact of siEIF4B on invadosome formation (Supp Figure 3g). We show that it affects the formation of rosettes as CHX, but also affects the formation of linear invadosomes on collagen by A431 cells.

Quantification of the numbers of invadosomes per cell on gelatin and collagen silencing (siEIF4B) or not (DMSO) for EIF4B in A431-Tks5-GFP and NIH3T3-Src-Tks5-GFP cells. Values represent the mean +/- SEM of n=4 independent experiments (10 images per condition and per replicate) and were analyzed using student t-test.

Reviewer #2 (Significance (Required)):

This study confirms and adds to a previously published report by this research group based on invadosome laser capture microdissection and proteomics revealing that invadosomes contain specific components of the translational machinery, and that protein translation activity is required to maintain invadosome structure and activity (Ezzoukhry et al. Nat Commun 2018). It also adds to a recent study that established a crucial role for ribosome biogenesis in promoting cell invasion in the C. elegans anchor cell invasion model (Development. 2023).

The experimentation presented in this paper is of good quality and convincingly support the authors conclusions of a link between the ER-associated translation machinery and invadosome function in transformed cells. Overall, although this study adds to the emerging idea of an evolutionary-conserved translational control of cell invasion through the extracellular matrix it is mostly correlative and lacking a direct prove that the interaction of TKS5 with components of the translation machinery has a direct contribution to invadopodia function.

__ __

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

Summary: To invade the surrounding extracellular matrix (ECM), cells organize actin-rich cellular membrane structures capable of ECM degradation, called invadosomes. Depending on the composition and organization of the ECM, cells organize their invadosomes differently. The authors aimed to identify specific and common components of different types of invadosomes: rosettes formed by NIH3T3-Src cells seeded on gelatin, dots formed by A431 cells seeded on gelatin, and linear invadosomes formed by NIH3T3-Src and A431 cells when seeded on fibrillar collagen I. For this, they generated cells stably expressing GFP-Tks5, a ubiquitous constituent of invadosomes, and determined its interactome. They identified 88 common proteins, among which the protein translation machinery was enriched. Whereas general protein inhibition impaired only rosette formation and impaired every type of invadosome-associated degradation, EIF4B inhibition inhibited the formation of every type of invadosomes. They then analyzed the impact of the ER on invadosome formation and degradation activity. First, they documented the presence of the ER in the center of the NIH3T3-Src rosettes and correlated ER presence with rosette initiation and persistence. They then demonstrated that chemical inhibition of Sec61 translocon decreased formation of invadosomes in general.

Major comments:

1- The authors use cells overexpressing GFP-Tks5 for their analysis of Tks5 interactome in the different invadosomes (Fig. 2). The impact of GFP-Tks5 overexpression on invadosome formation and degradation activity should be mentioned.

Depending the cell type the TKS5-GFP overexpression do not increase the number of invadosomes but increase the matrix degradation activity (Di Martino et al 2014); or could impact the number of invadosomes as in B16 cell line (Shinji Iizuka et al, 2016). This point was added in the introduction.

However, the Tks5 overexpression was used fo immunoprecipitation and mass spectrometry analysis. The rest of the study and targets validation are done on wild type cells.

2- Concerning the analysis of the mass spectrometry (MS) data, clarifications would be appreciated:

a. The authors first "determined the specific molecular signature associated with each invadosome organization" (p.4). As I understand it, the proteins in each of these signatures correspond to proteins identified only in a particular type of invadosomes, not in the others. Could the authors indicate the percentage of the total proteins identified for each type of invadosomes that corresponds to the specific molecular signature?

The meaning of the sentence has been changed in the paper to provide more understanding. The term "molecular signature" has been replaced by "specific proteins". Percentages have been added to the tables in Figure 1 Supp.

1. __ __ The GSEA pathways related to each of the specific molecular signature were then analyzed and the authors "commonly identified an enrichment in mitochondrial, ER and Golgi proteins" (page 4) (Supp Fig 1c,e,g). Could the authors provide numbers/percentage/statistics? It is not clear to me whether the biological processes (Supp Fig 1b,d,f) are derived from the analysis of the specific molecular signature or of the total proteins identified for each type of invadosomes. Could the authors clarify this point? The percentages of each specific protein category have been added in Figure 1 Supp.

The biological processes (Supp Fig 1b, d, f) arise from the analysis of the molecular signature common to the 4 invadosomes conditions, namely the dots, rosettes and linear invadosomes of A431 and NIH-3T3-Src. Thus, the biological processes arise here from the 88 proteins commonly identified for all types of invadosomes.

1. The authors also identified "translation proteins" enriched in the specific molecular signature of each type of invadosomes (p.4). They commented on this category, indicating that each type of invadosome contains a specific set of translation-related proteins. This is true, but according to my analysis of the provided tables, the same applies to the other categories as well. Could the authors comment this point? Indeed, some proteins involved in translation can appear specific or common depending the type of invadosome. Our comment is at this step, only suggest that some of this protein should be specific for invadosome and some could be associated to only one organization. Of course, the role of each protein needs to be investigated.

2. Would similar categories of proteins (translation, ER, Golgi, mitochondrial) appear as enriched if the Tks5 interactome was analyzed as a whole for each type of invadosomes? (the authors may disregard this comment if comment a. is inaccurate). Protein pathways enriched in the different type of invadosome differ, for example, Protein activity GTPase activity, vs cell adhesion molecule binding or hydrolase activity acting on Acid Anyhdrides. This analysis demonstrates and highlights differences between the different invadosome organization. However, we focus on translational proteins, ER proteins for example and calculated the percentage of protein identified and associated with this different structure. We can notice important difference as 3% of translation proteins for rosette vs 9 % for dots in A431 cells. This point suggests that the part of each element can differ.

3. __ __ The authors identified that "cell adhesion proteins" are specifically enriched in linear invadosomes (page 4) (Supp Fig 1f). This conclusion appears to be based on the analysis of NIH3T3-Src and A431 cells. Could the authors provide more details on how this analysis was performed? Specifically, was the analysis conducted on a mixture of the specific signatures of each of the 2 cell models, or on their shared proteins? Additionally, is this category still enriched if each linear invadosome model is analyzed separately? The analysis was performed on common proteins of linear invadosomes, grouping the two cellular models. The category "cell adhesion protein" is not specifically enriched in linear invadosomes because adhesion proteins are also found in the other groups. However, this category represents a larger percentage in linear invadosomes, thus justifying our choice to highlight it for this category.

4. __ __ The authors identified 88 proteins common to all types of invadosomes (Fig. 2b) and classified them as validated or not in invadosomes. Could the authors give details on the criteria used for this classification? References for the already validated proteins should also be provided. RTN4 has been described as partially localized at invadopodia formed by MDA-MB-231 cells in Thuault et al., yet the authors classified it as not validated in invadosomes. The RTN4 protein has been moved to the category of proteins identified as localized in at least one invadosomes organization, thank you for this precision.

Please find below the list of papers having among the proteins classification as identified in at least one invadosomes organization, based on literature searches.

ADAM15 : Aspartate β-hydroxylase promotes pancreatic ductal adenocarcinoma metastasis through activation of SRC signaling pathway - Ogawa et al 2019

ADAM19 : The Adaptor Protein Fish Associates with Members of the ADAMs Family and Localizes to Podosomes of Src-transformed Cells - Abram et al 2003

ASPH : Aspartate β-hydroxylase promotes pancreatic ductal adenocarcinoma metastasis through activation of SRC signaling pathway - Ogawa et al, 2019

BAG3 : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

CALD1 :

• Caldesmon is an integral component of podosomes in smooth muscle cells - Eves et al, 2006
• Caldesmon is an integral component of podosomes in smooth muscle cells, Gu et al 2007

• Changes in the balance between caldesmon regulated by p21‐activated kinases and the Arp2/3 complex govern podosome formation, Morita et al 2007 CD44 :

• The CD44s splice isoform is a central mediator for invadopodia activity, Zhao et al

• CD147, CD44, and the Epidermal Growth Factor Receptor (EGFR) Signaling Pathway Cooperate to Regulate Breast Epithelial Cell Invasiveness, Grass et al, 2013
• CD44 and beta3 integrin organize two functionally distinct actin-based domains in osteoclasts, Chabadel et al, 2007
• Macrophages podosomes go 3, Goethem et al 2011 CTTN : ERβ promoted invadopodia formation-mediated non-small cell lung cancer metastasis via the ICAM1/p-Src/p-Cortactin signaling pathway - Wang et al, 2023

EIF4B : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

FNBP1L : Transducer of Cdc42-dependent actin assembly promotes breast cancer invasion and metastasis - Chander et al, 2013

FXR1 : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

G3BP1 : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

HNRNPA1 : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

IGF2BP2 : IMP2 and IMP3 cooperate to promote the metastasis of triple-negative breast cancer through destabilization of progesterone receptor - Kim et al, 2018

ITGA5 : Membrane Proteome Analysis of Glioblastoma Cell Invasion, Mallawaaratchy et al, 2015

LAMP1 : Lysosomal cathepsin B participates in the podosome-mediated extracellular matrix degradation and invasion via secreted lysosomes in v-Src fibroblasts - Chun Tu et al, 2008

MAP4 : A proximity-labeling proteomic approach to investigate invadopodia molecular landscape in breast cancer cells, Thuault et al, 2020

MMP14 :

• Receptor-type protein tyrosine phosphatase alpha (PTPα) mediates MMP14 localization and facilitates triple-negative breast cancer cell invasion - Decotret 2021

• Deciphering the involvement of the Hippo pathway co-regulators, YAP/TAZ in invadopodia formation and matrix degradation - Venghateri 2023 MYH9 :

• TRPM7, a novel regulator of actomyosin contractility and cell adhesion 6 Clarck et al, 2006

• Bradykinin promotes migration and invasion of hepatocellular carcinoma cells through TRPM7 and MMP2, Chen et al, 2016 NONO : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

NPM1 : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

PABPC1 : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

PPP1CA : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

PRKAA1 : A proximity-labeling proteomic approach to investigate invadopodia molecular landscape in breast cancer cells, Thuault et al, 2020

PTBP1 : The lncRNA MIR99AHG directs alternative splicing of SMARCA1 by PTBP1 to enable invadopodia formation in colorectal cancer cells - Li et al, 2023

RPL10A : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

RPL34 : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

RPS4X : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

RRBP1 : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

RTN4 : A proximity-labeling proteomic approach to investigate invadopodia molecular landscape in breast cancer cells, Thuault et al, 2020

SSB : The PDGFRα-laminin B1-keratin 19 cascade drives tumor progression at the invasive front of human hepatocellular carcinoma - Govaere 2017

STX7 : Syntaxin 7 contributes to breast cancer cell invasion by promoting invadopodia formation, Parveen et al, 2022

SYNCRIP : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

THBD : VEGF-Induced Endothelial Podosomes via ROCK2-Dependent Thrombomodulin Expression Initiate Sprouting Angiogenesis - Cheng-Hsiang Kuo - 2021

YBX3 : Combining laser capture microdissection and proteomics reveals an active translation machinery controlling invadosome formation, Ezzoukhry et al, 2018

1. __ __ Page 7, "In addition to translation proteins, the MS analysis highlighted the presence of ER-related proteins such as RTN4, LRRC59 or RRBP1 in all invadosomes linked with Tks5 (Figure 2c)". Is the "ER proteins" category enriched among the 88 common proteins? GSEA analysis on the 88 proteins showed an enrichment in proteins related to ribosomes and mRNA binding.

2. __ __ The comparative analysis of the TKS5 interactome from NIH3T3-Src-GFP-TKS5 on gelatin (this study) with the proteome of NIH3T3-Src rosettes from Ezzoukhry et al. (Fig 5a and Supp Table 2) should be included in the analysis of the MS data obtained in this study (Fig 2), rather than in the paragraph "Recruitment of ER into invadosome rosettes". Are "ER proteins" enriched? Comparative analysis of the TKS5 interactome of NIH3T3-Src-GFP-TKS5 on gelatin (this study) with the proteome of NIH3T3-Src rosettes from Ezzoukhry et al. was included in Supp Figure 2.

The proteins related to translation are enriched, but not those of the ER.__ __3- Was the localization of the newly identified Tks5 partners, such as RPS6 and EIF4B, but also MAP4 and CD44, to invadosomes analyzed in cells expressing endogenous levels of Tks5? If not, this should be addressed to rule out the possibility that their localization in invadosomes is linked to Tks5 overexpression. Through the figures, it is important to indicate whether cells overexpressing or not Tks5 were used.

The precision on the overexpression of Tks5 has been added in the figures.

The experiments were also carried out on cells not overexpressing Tks5 (see results below). Clarifications have been added in the article to specify that these experiments were carried out on cell lines overexpressing Tks5 but also on WT cell lines not overexpressing Tks5 (data not shown in the paper).

Confocal microscopy images of A431 and NIH-3T36Src cells. The cells were seeded on gelatin or type I collagen and stained for Tks5 in green, actin in red, nuclei in blue and Eif4b in grey. Scale bar: 40µm, zoom: 10µm.

Confocal microscopy images of A431 and NIH-3T3-Src cells. The cells were seeded on gelatin or type I collagen and stained for Tks5 in green, actin in red, nuclei in blue and RPS6 in grey. Scale bar: 40µm, zoom: 10µm.

Confocal microscopy images of A431 and NIH-3T3-Src cells. The cells were seeded on gelatin or type I collagen and stained for Tks5 in green, actin in red, nuclei in blue and MAP4 in grey. Scale bar: 40µm, zoom: 10µm.

4- EIF4B depletion inhibits ECM degradation (Fig 4e-f). The authors should address the impact of EIF4B depletion on invadosome formation. In other words, does EIF4B depletion corroborate the results obtained with CHX treatment, where only rosette formation is inhibited (Fig. 4a and Supp Fig. 3d).

The impact of EIF4B depletion on invadosome formation was studied. We were able to show that EIF4B depletion partly corroborates with the results obtained with CHX treatment, since rosette formation is also inhibited by EIF4B depletion but linear invadosomes formed on collagen by A431 are also inhibited by EIF4B depletion.

These results have been added to the paper (see Figure 3g).

Quantification of the numbers of invadosomes per cell on gelatin and collagen silencing (siEIF4B) or not (DMSO) for EIF4B in A431-Tks5-GFP and NIH3T3-Src-Tks5-GFP cells. Values represent the mean +/- SEM of n=4 independent experiments (10 images per condition and per replicate) and were analyzed using student t-test.

__ __5- The authors treated NIH3T3-Src-KDEL-GFP and LifeAct-Ruby cells with CHX and conclude that "translation inhibition led to the collapse of the rosette structure (Fig 6a, Video 4)" (page 8): could extra time points be added before T300 to appreciate the collapse of actin before the retraction of ER from the center of the rosette. No video 4 is provided. A video 5 is provided but does not correspond to a rosette collapse. The lifetime/dissociation rate of rosettes with and without CHX treatment should be determined.

Live cell imaging has been performed by recording one image every 2 minutes as described in methods. Graphs represent all recorded points along the experiment however we modified scale of original graph included into the manuscript to better appreciate the dissociation of fluorescence intensity curves revealing the collapse of actin before the retractation of ER. We also added a second graph which confirmed our first interpretation.

For video 4, we submitted the videos to make sure there were no errors. So, we can now clearly see the collapse of the rosette in video 4.

Lifeact-mRuby and KDEL-GFP signals were recorded in NIH-3T3-Src cells treated with cycloheximide (CHX; 35µM)

__ __6- Sec61 translocon inhibition by the chemical inhibitor ES1 decreases formation of dots by A431 and rosettes and linear invadosomes by NIH3T3-Src (Fig. 6b). Sec61 siRNA should be analyzed. Does Sec61 localize at invadosomes?

Immunofluorescence on NIH-3T3-Src and A431 WT cell lines were performed and added in the paper showing the localization of Sec61 in invadosomes (Figure 6b). Currently, we did not test siRNA targeting Sec61.

Confocal microscopy images of NIH-3T3-Src and A431 cells. The cells were seeded on gelatin or type I collagen and stained for Sec61 in red, nuclei in blue and Actin in grey. Scale bar: 20µm, zoom: 5µm.

__ __Minor comments:

1- The data of Figure 1 is not totally new, at least plasticity of NIH3T3-Src invadosomes has already been described in Juin A., MBoC, 2012. References to original work should be mentioned.

Indeed, the reference has been added to the text at Figure 1.

2- Page 4 "We realized immunoprecipitation against GFP in both cell lines on plastic and type I collagen conditions": the authors should show/mention that on plastic, cells behave has on gelatin coating.

A sentence has been added to the text to mention this: "Indeed, on plastic, the cells behave as on a gelatin coating and thus form the same types of invadosomes, i.e. dots for A431 cells and rosettes for NIH-3T3-Src cells." (see page 4).

3- The authors compared their MS data to previously published Tks5 interactomes (page 4) (Supp Fig 2a). A study from Zagryakhskaya-Masson et al (PMID: 32673397) identified Tks5 interactome of MDA-MB-231 cells generating linear invadosomes. Could the authors comment this study?

This study shows that FGD1, a guanine nucleotide exchange factor for the Rho-GTPase CDC42 interacts with Tks5 and plays a role in the formation of linear invadosomes. We have added this reference in the manuscript, but we have not found FGD1 in our data. It is possible that the GEF of Cdc42 varies from one cell type to another. This study has been added to the discussion.

4- The comparison of translation proteins found in this study with the ones found in other studies (Supp. Fig. 3 a) should be combined with the paragraph commenting the 88 common proteins (Fig. 2c-d).

For clarity, we decided to separate these two parts. There is indeed a lot of information, so it seemed clearer to us to keep the structure of the figures in this sense.

5- The table Supp Fig 2c listing the proteins present in each of the functional categories enriched among the 88 common Tks5 partners should be included as main figure or a color code representing the different biological processes should be included in Fig 2c.

A color code has been added between the two tables. A sentence has been added in the legends for clarity: "Color codes are according to Table Supp Figure 2c: orange: translation, green: actin cytoskeleton, and blue: adhesion."

__ __6- The SUnSET assay is not correctly untitled and described in the Material and Methods. Indeed, the paragraph refering to it is entitled "Inhibition of translation machinery present in invadosomes" and is a mixture of immunofluorescence and SUnSET protocols.

The SunSET assay materials and methods were modified in the paper in the "Sunset Assay" section as described below:

Sunset assay

Cells were treated with puromycin (10mg/ml) during 10min at 37°C then washed twice in ice-cold PBS for protein extraction as described above in Western Blot section. For negative control we pre-treated cells with the translation inhibitor cycloheximide (35mM) during 10min at 37°C.

7- Figure 4, the decrease in ECM degradation of A431 (GFP-Tks5) cells seeded on gelatin by CHX is not statistically different. The affirmation that "CHX treatment limited degradation activity by A431 and NIH3T3-Src cells on gelatin and collagen matrices" (page 6) should be modulated.

Indeed, thank you for your observation. We realized that incorrect values had been reported. Statistical tests (t-tests) were redone for each CHX condition, and significant results were found for each condition.

8- Page 8, "These results therefore confirm the presence but also the involvement of the ER in the rosette formation and maintenance over time". At this point in the study, there is a correlation between the presence of the ER and rosette persistence but no direct evidence of ER involvement is provided. The authors should moderate their conclusion.

That's absolutely right, the sentence has been modified accordingly (page 8).

9- Fig 5d: the authors should specify in the figure legend what are the red head arrows.

The red arrows show membranes of the endoplasmic reticulum, present at the level of the invadosome rosette. This point was added in the figure legend.

10- Some references are not correct. For example p.10, "MAP4 and LAMP1 were described in podosomes": ref 23 and 26 are studies on invadopodia, not on podosomes.

Corrections have been made to the text, the term podosomes has been replaced by invadopodia (see section references).

11- The authors indicate p.10, "Thanks to mass spectrometry experiments, we were able to show for the first time the presence of translation proteins in linear invadosomes". In their previous study Ezzoukry et al, they showed the localization of overexpressed Caprin1, eEF2 and eEF1A1 translation machinery components in linear invadosomes formed by NIH3T3-Src seeded on fibrillar collagen I. The authors should modulate their affirmations.

Indeed, this sentence has been modulated in the text (see page 10).

12- Could the authors refer to figures in the Discussion.

References to figures were added in the discussion.

Reviewer #3 (Significance (Required)):

This work extends their previous work, Ezzoukhry et al, in which the proteome of rosettes of NIH3T3-Src was identified after laser microdissection. In this work, they had identified protein translation machinery as components of rosettes and its implication in the degradation activity and/or the formation of rosettes and linear invadosomes.

The present study extends the presence of protein translation machinery to other types of invadosomes and the implication of protein translation in invadosome activity and/or formation. It also confirms the presence of ER in the center of rosettes. It suggests that ER-associated translation is required for invadosomes formation and activity. This knowledge will be of interest for the invadosome researcher community.

My expertise is in: cellular biology, invadopodia, ECM degradation, cancer. I do not have sufficient expertise to evaluate the accuracy of the analysis of mass spectrometry data and the quantification of videomicroscopy experiments.

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 #3

Evidence, reproducibility and clarity

Summary:

To invade the surrounding extracellular matrix (ECM), cells organize actin-rich cellular membrane structures capable of ECM degradation, called invadosomes. Depending on the composition and organization of the ECM, cells organize their invadosomes differently. The authors aimed to identify specific and common components of different types of invadosomes: rosettes formed by NIH3T3-Src cells seeded on gelatin, dots formed by A431 cells seeded on gelatin, and linear invadosomes formed by NIH3T3-Src and A431 cells when seeded on fibrillar collagen I. For this, they generated cells stably expressing GFP-Tks5, a ubiquitous constituent of invadosomes, and determined its interactome. They identified 88 common proteins, among which the protein translation machinery was enriched. Whereas general protein inhibition impaired only rosette formation and impaired every type of invadosome-associated degradation, EIF4B inhibition inhibited the formation of every type of invadosomes. They then analyzed the impact of the ER on invadosome formation and degradation activity. First, they documented the presence of the ER in the center of the NIH3T3-Src rosettes and correlated ER presence with rosette initiation and persistence. They then demonstrated that chemical inhibition of Sec61 translocon decreased formation of invadosomes in general.

Major comments:

1- The authors use cells overexpressing GFP-Tks5 for their analysis of Tks5 interactome in the different invadosomes (Fig. 2). The impact of GFP-Tks5 overexpression on invadosome formation and degradation activity should be mentioned.

2- Concerning the analysis of the mass spectrometry (MS) data, clarifications would be appreciated:

a. The authors first "determined the specific molecular signature associated with each invadosome organization" (p.4). As I understand it, the proteins in each of these signatures correspond to proteins identified only in a particular type of invadosomes, not in the others. Could the authors indicate the percentage of the total proteins identified for each type of invadosomes that corresponds to the specific molecular signature?

b. The GSEA pathways related to each of the specific molecular signature were then analyzed and the authors "commonly identified an enrichment in mitochondrial, ER and Golgi proteins" (page 4) (Supp Fig 1c,e,g). Could the authors provide numbers/percentage/statistics? It is not clear to me whether the biological processes (Supp Fig 1b,d,f) are derived from the analysis of the specific molecular signature or of the total proteins identified for each type of invadosomes. Could the authors clarify this point?

c. The authors also identified "translation proteins" enriched in the specific molecular signature of each type of invadosomes (p.4). They commented on this category, indicating that each type of invadosome contains a specific set of translation-related proteins. This is true, but according to my analysis of the provided tables, the same applies to the other categories as well. Could the authors comment this point?

d. Would similar categories of proteins (translation, ER, Golgi, mitochondrial) appear as enriched if the Tks5 interactome was analyzed as a whole for each type of invadosomes? (the authors may disregard this comment if comment a. is inaccurate)

e. The authors identified that "cell adhesion proteins" are specifically enriched in linear invadosomes (page 4) (Supp Fig 1f). This conclusion appears to be based on the analysis of NIH3T3-Src and A431 cells. Could the authors provide more details on how this analysis was performed? Specifically, was the analysis conducted on a mixture of the specific signatures of each of the 2 cell models, or on their shared proteins? Additionally, is this category still enriched if each linear invadosome model is analyzed separately?

f. The authors identified 88 proteins common to all types of invadosomes (Fig. 2b) and classified them as validated or not in invadosomes. Could the authors give details on the criteria used for this classification? References for the already validated proteins should also be provided. RTN4 has been described as partially localized at invadopodia formed by MDA-MB-231 cells in Thuault et al., yet the authors classified it as not validated in invadosomes.

g. Page 7, "In addition to translation proteins, the MS analysis highlighted the presence of ER-related proteins such as RTN4, LRRC59 or RRBP1 in all invadosomes linked with Tks5 (Figure 2c)". Is the "ER proteins" category enriched among the 88 common proteins?

h. The comparative analysis of the TKS5 interactome from NIH3T3-Src-GFP-TKS5 on gelatin (this study) with the proteome of NIH3T3-Src rosettes from Ezzoukhry et al. (Fig 5a and Supp Table 2) should be included in the analysis of the MS data obtained in this study (Fig 2), rather than in the paragraph "Recruitment of ER into invadosome rosettes". Are "ER proteins" enriched?

3- Was the localization of the newly identified Tks5 partners, such as RPS6 and EIF4B, but also MAP4 and CD44, to invadosomes analyzed in cells expressing endogenous levels of Tks5? If not, this should be addressed to rule out the possibility that their localization in invadosomes is linked to Tks5 overexpression. Through the figures, it is important to indicate whether cells overexpressing or not Tks5 were used.

4- EIF4B depletion inhibits ECM degradation (Fig 4e-f). The authors should address the impact of EIF4B depletion on invadosome formation. In other words, does EIF4B depletion corroborate the results obtained with CHX treatment, where only rosette formation is inhibited (Fig. 4a and Supp Fig. 3d).

5- The authors treated NIH3T3-Src-KDEL-GFP and LifeAct-Ruby cells with CHX and conclude that "translation inhibition led to the collapse of the rosette structure (Fig 6a, Video 4)" (page 8): could extra time points be added before T300 to appreciate the collapse of actin before the retraction of ER from the center of the rosette. No video 4 is provided. A video 5 is provided but does not correspond to a rosette collapse. The lifetime/dissociation rate of rosettes with and without CHX treatment should be determined.

6- Sec61 translocon inhibition by the chemical inhibitor ES1 decreases formation of dots by A431 and rosettes and linear invadosomes by NIH3T3-Src (Fig. 6b). Sec61 siRNA should be analyzed. Does Sec61 localize at invadosomes?

Minor comments:

1- The data of Figure 1 is not totally new, at least plasticity of NIH3T3-Src invadosomes has already been described in Juin A., MBoC, 2012. References to original work should be mentioned.

2- Page 4 "We realized immunoprecipitation against GFP in both cell lines on plastic and type I collagen conditions": the authors should show/mention that on plastic, cells behave has on gelatin coating.

3- The authors compared their MS data to previously published Tks5 interactomes (page 4) (Supp Fig 2a). A study from Zagryakhskaya-Masson et al (PMID: 32673397) identified Tks5 interactome of MDA-MB-231 cells generating linear invadosomes. Could the authors comment this study?

4- The comparison of translation proteins found in this study with the ones found in other studies (Supp. Fig. 3 a) should be combined with the paragraph commenting the 88 common proteins (Fig. 2c-d).

5- The table Supp Fig 2c listing the proteins present in each of the functional categories enriched among the 88 common Tks5 partners should be included as main figure or a color code representing the different biological processes should be included in Fig 2c.

6- The SUnSET assay is not correctly untitled and described in the Material and Methods. Indeed, the paragraph refering to it is entitled "Inhibition of translation machinery present in invadosomes" and is a mixture of immunofluorescence and SUnSET protocols.

7- Figure 4, the decrease in ECM degradation of A431 (GFP-Tks5) cells seeded on gelatin by CHX is not statistically different. The affirmation that "CHX treatment limited degradation activity by A431 and NIH3T3-Src cells on gelatin and collagen matrices" (page 6) should be modulated.

8- Page 8, "These results therefore confirm the presence but also the involvement of the ER in the rosette formation and maintenance over time". At this point in the study, there is a correlation between the presence of the ER and rosette persistence but no direct evidence of ER involvement is provided. The authors should moderate their conclusion.

9- Fig 5d: the authors should specify in the figure legend what are the red head arrows.

10- Some references are not correct. For example p.10, "MAP4 and LAMP1 were described in podosomes": ref 23 and 26 are studies on invadopodia, not on podosomes.

11- The authors indicate p.10, "Thanks to mass spectrometry experiments, we were able to show for the first time the presence of translation proteins in linear invadosomes". In their previous study Ezzoukry et al, they showed the localization of overexpressed Caprin1, eEF2 and eEF1A1 translation machinery components in linear invadosomes formed by NIH3T3-Src seeded on fibrillar collagen I. The authors should modulate their affirmations.

12- Could the authors refer to figures in the Discussion.

Significance

This work extends their previous work, Ezzoukhry et al, in which the proteome of rosettes of NIH3T3-Src was identified after laser microdissection. In this work, they had identified protein translation machinery as components of rosettes and its implication in the degradation activity and/or the formation of rosettes and linear invadosomes.

The present study extends the presence of protein translation machinery to other types of invadosomes and the implication of protein translation in invadosome activity and/or formation. It also confirms the presence of ER in the center of rosettes. It suggests that ER-associated translation is required for invadosomes formation and activity. This knowledge will be of interest for the invadosome researcher community.

My expertise is in: cellular biology, invadopodia, ECM degradation, cancer. I do not have sufficient expertise to evaluate the accuracy of the analysis of mass spectrometry data and the quantification of videomicroscopy experiments.

Reviewer #2 (Public review):

The authors aim to investigate the ability of evolution to create strong transcription factor binding sites (TFBSs) de novo in E. coli. They focus on three global transcriptional regulators: CRP, Fis, and IHF, using a massively parallel reporter assay to evaluate the regulatory effects of over 30,000 TFBS variants. By analyzing the resulting genotype-phenotype landscapes, they explore the ruggedness, accessibility, and evolutionary dynamics of regulatory landscapes, providing insights into the evolutionary feasibility of strong gene regulation. Their experiments show that de novo adaptive evolution of new gene regulation is feasible. It is also subject to a blend of chance, historical contingency, and evolutionary biases that favor some peaks and evolutionary paths.

(1) Strengths of the methods and results:

The authors successfully employed a well-designed sort-seq assay combined with high-throughput sequencing to map regulatory landscapes. The experimental design ensures reliable measurement of regulation strengths. Their system accounts for gene expression noise and normalizes measurements using appropriate controls.

Comprehensive Landscape Mapping:<br /> The study examines ~30,000 TFBS variants per transcription factor, providing statistically robust and thorough maps of the regulatory landscapes for CRP, Fis, and IHF. The landscapes are rigorously analyzed for ruggedness (e.g., number of peaks) and epistasis, revealing parallels with theoretical uncorrelated random landscapes.

Evolutionary Dynamics Simulations:<br /> Through simulations of adaptive walks under varying population dynamics, the authors demonstrate that high peaks in regulatory landscapes are accessible despite ruggedness. They identify key evolutionary phenomena, such as contingency (multiple paths to peaks) and biases toward specific evolutionary outcomes.

Biological Relevance and Novelty:<br /> The author's work is novel in focusing on global regulators, which differ from previously studied local regulators (e.g., TetR). They provide compelling evidence that rugged landscapes are navigable, facilitating de novo evolution of regulatory interactions. The comparison of landscapes for CRP, Fis, and IHF underscores shared topographical features, suggesting general principles of global transcriptional regulation in bacteria.

(2) Weaknesses of the methods and results:

Undersampling of Genotype Space:<br /> While the quality filtering of the data ensures robustness, ~40% of the TFBS space remains uncharacterized. The authors acknowledge this limitation but could improve the analysis by employing subsampling or predictive modeling.

Simplified Regulatory Architecture:<br /> The study considers a minimal system of a single TFBS upstream of a reporter gene. While this may have been necessary for clarity, this simplification may not reflect the combinatorial complexity of transcriptional regulation in vivo.

Lack of Experimental Validation of Simulations:<br /> The adaptive walks are based on simulated dynamics rather than experimental evolution. Incorporating in vivo experimental evolution studies would strengthen the conclusions. Although this is a large request for the paper, that would not prevent publication.

Impact on the Field:<br /> This study advances our understanding of adaptive landscapes in gene regulation and offers a critical step toward deciphering how global regulators evolve de novo binding sites. The findings provide foundational insights for synthetic biology, evolutionary genetics, and systems biology by highlighting the evolutionary accessibility of strong regulation in bacteria.

Utility of Methods and Data:<br /> The sort-seq approach, combined with landscape analysis, provides a robust framework that can be extended to other transcription factors and systems. If made publicly available, the study's data and code would be valuable for researchers modeling transcriptional regulation or studying evolutionary dynamics.

Additional Context:<br /> The study builds on a growing body of work exploring regulatory evolution. For instance, recent studies on local regulators like TetR and AraC have revealed high ruggedness and epistasis in TFBS landscapes. This study distinguishes itself by focusing on global regulators, which are more biologically complex and influential in bacterial gene networks. The observed evolutionary contingency aligns with findings in other biological systems, such as protein evolution and RNA folding landscapes, underscoring the generality of these evolutionary principles.

Conclusion:<br /> The authors successfully mapped the genotype-phenotype landscapes for three global regulators and simulated evolutionary dynamics to assess the feasibility of strong TFBS evolution. They convincingly demonstrate that ruggedness and epistasis, while prominent, do not preclude the evolution of strong regulation. Their results support the notion that gene regulation evolves through a blend of chance, contingency, and evolutionary biases.

This paper makes a significant contribution to the understanding of regulatory evolution in bacteria. While minor limitations exist, the authors' methods are robust, and their findings are well-supported. The work will likely be of broad interest to researchers in molecular evolution, synthetic biology, and gene regulation.

• Why use Kubernetes

  • Best for running multiple processes/servers/jobs with redundancy and load balancing
  • Enables infrastructure-as-code configuration for service relationships
  • Outsourced infrastructure management via cloud providers (e.g., Google Kubernetes Engine)

• What they use

  • Core resources: Deployments (with rolling updates), Services (ClusterIP/LoadBalancer), CronJobs
  • Configuration: ConfigMaps and Secrets via Pulumi (TypeScript) instead of raw YAML
  • Cautious adoptions: StatefulSets for limited persistence, RBAC only when necessary

• What they avoid

  • Hand-written YAML and Helm charts ("fragility for no gain")
  • Operators, custom resources, service meshes, and most third-party controllers
  • Local k8s stack replication (prefer Docker Compose for local dev)

• Key insights

  • "A human should never wait for a pod" - unsuitable for interactive workloads requiring fast startup
  • Use managed databases/storage for critical data instead of k8s volumes
  • Alternatives like Railway/Render may be better for simpler SaaS apps
  • Recently adopted Ingress controllers for Cloud Armor integration despite initial reservations

Reviewer #2 (Public review):

Summary:

The authors provide an open-source graphic user interface (GUI) called Heron, implemented in Python, that is designed to help experimentalists to:

(1) Design experimental pipelines and implement them in a way that is closely aligned with their mental schemata of the experiments<br /> (2) Execute and control the experimental pipelines with numerous interconnected hardware and software on a network.

The former is achieved by representing an experimental pipeline using a Knowledge Graph and visually representing this graph in the GUI. The latter is accomplished by using an actor model to govern the interaction among interconnected nodes through messaging, implemented using ZeroMQ. The nodes themselves execute user-supplied code in, but not limited to, Python.

Using three showcases of behavioral experiments on rats, the authors highlighted four benefits of their software design:

(1) The knowledge graph serves as a self-documentation of the logic of the experiment, enhancing the readability and reproducibility of the experiment,<br /> (2) The experiment can be executed in a distributed fashion across multiple machines that each has different operating system or computing environment, such that the experiment can take advantage of hardware that sometimes can only work on a specific computer/OS, a commonly seen issue nowadays,<br /> (3) The users supply their own Python code for node execution that is supposed to be more friendly to those who do not have a strong programming background,<br /> (4) The GUI can also be used as an experiment control panel for users to control/update parameters on the fly.

Strengths:

(1) The software is light-weight and open-source, provides a clean and easy-to-use GUI,<br /> (2) The software answers the need of experimentalists, particularly in the field of behavioral science, to deal with the diversity of hardware that becomes restricted to run on dedicated systems. It can also be widely adopted in many other experimental settings.<br /> (3) The software has a solid design that seems to be functionally reliable and useful under many conditions, demonstrated by a number of sophisticated experimental setups.<br /> (4) The software is well documented. The authors pay special attention to documenting the usage of the software and setting up experiments using this software.

Comments on revisions: The authors have addressed my concerns from the initial review.

Author response:

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

Public Reviews

Reviewer #1 (Public Review):

Summary:

The authors have created a system for designing and running experimental pipelines to control and coordinate different programs and devices during an experiment, called Heron. Heron is based around a graphical tool for creating a Knowledge Graph made up of nodes connected by edges, with each node representing a separate Python script, and each edge being a communication pathway connecting a specific output from one node to an iput on another. Each node also has parameters that can be set by the user during setup and runtime, and all of this behavior is concisely specified in the code that defines each node. This tool tries to marry the ease of use, clarity, and selfdocumentation of a purely graphical system like Bonsai with the flexibility and power of a purely code-based system like Robot Operating System (ROS).

Strengths:

The underlying idea behind Heron, of combining a graphical design and execution tool with nodes that are made as straightforward Python scripts seems like a great way to get the relative strengths of each approach. The graphical design side is clear, selfexplanatory, and self-documenting, as described in the paper. The underlying code for each node tends to also be relatively simple and straightforward, with a lot of the complex communication architecture successfully abstracted away from the user. This makes it easy to develop new nodes, without needing to understand the underlying communications between them. The authors also provide useful and well-documented templates for each type of node to further facilitate this process. Overall this seems like it could be a great tool for designing and running a wide variety of experiments, without requiring too much advanced technical knowledge from the users.

The system was relatively easy to download and get running, following the directions and already has a significant amount of documentation available to explain how to use it and expand its capabilities. Heron has also been built from the ground up to easily incorporate nodes stored in separate Git repositories and to thus become a large community-driven platform, with different nodes written and shared by different groups. This gives Heron a wide scope for future utility and usefulness, as more groups use it, write new nodes, and share them with the community. With any system of this sort, the overall strength of the system is thus somewhat dependent on how widely it is used and contributed to, but the authors did a good job of making this easy and accessible for people who are interested. I could certainly see Heron growing into a versatile and popular system for designing and running many types of experiments.

Weaknesses:

(1) The number one thing that was missing from the paper was any kind of quantification of the performance of Heron in different circumstances. Several useful and illustrative examples were discussed in depth to show the strengths and flexibility of Heron, but there was no discussion or quantification of performance, timing, or latency for any of these examples. These seem like very important metrics to measure and discuss when creating a new experimental system.

Heron is practically a thin layer of obfuscation of signal passing across processes. Given its design approach it is up to the code of each Node to deal with issues of timing, synching and latency and thus up to each user to make sure the Nodes they author fulfil their experimental requirements. Having said that, Heron provides a large number of tools to allow users to optimise the generated Knowledge Graphs for their use cases. To showcase these tools, we have expanded on the third experimental example in the paper with three extra sections, two of which relate to Heron’s performance and synching capabilities. One is focusing on Heron’s CPU load requirements (and existing Heron tools to keep those at acceptable limits) and another focusing on post experiment synchronisation of all the different data sets a multi Node experiment generates.   

(2) After downloading and running Heron with some basic test Nodes, I noticed that many of the nodes were each using a full CPU core on their own. Given that this basic test experiment was just waiting for a keypress, triggering a random number generator, and displaying the result, I was quite surprised to see over 50% of my 8-core CPU fully utilized. I don’t think that Heron needs to be perfectly efficient to accomplish its intended purpose, but I do think that some level of efficiency is required. Some optimization of the codebase should be done so that basic tests like this can run with minimal CPU utilization. This would then inspire confidence that Heron could deal with a real experiment that was significantly more complex without running out of CPU power and thus slowing down.

The original Heron allowed the OS to choose how to manage resources over the required process. We were aware that this could lead to significant use of CPU time, as well as occasionally significant drop of packets (which was dependent on the OS and its configuration). This drop happened mainly when the Node was running a secondary process (like in the Unity game process in the 3rd example). To mitigate these problems, we have now implemented a feature allowing the user to choose the CPU that each Node’s worker function runs on as well as any extra processes the worker process initialises. This is accessible from the Saving secondary window of the node. This stops the OS from swapping processes between CPUs and eliminates the dropping of packages due to the OS behaviour. It also significantly reduces the utilised CPU time. To showcase this, we initially run the simple example mentioned by the reviewer. The computer running only background services was using 8% of CPU (8 cores). With Heron GUI running but with no active Graph, the CPU usage went to 15%. With the Graph running and Heron’s processes running on OS attributed CPU cores, the total CPU was at 65% (so very close to the reviewer’s 50%). By choosing a different CPU core for each of the three worker processes the CPU went down to 47% and finally when all processes were forced to run on the same CPU core the CPU load dropped to 30%.  So, Heron in its current implementation running its GUI and 3 Nodes takes 22% of CPU load. This is still not ideal but is a consequence of the overhead of running multiple processes vs multiple threads. We believe that, given Heron’s latest optimisation, offering more control of system management to the user, the benefits of multi process applications outweigh this hit in system resources. 

We have also increased the scope of the third example we provide in the paper and there we describe in detail how a full-scale experiment with 15 Nodes (which is the upper limit of number of Nodes usually required in most experiments) impacts CPU load. 

Finally, we have added on Heron’s roadmap projects extra tasks focusing only on optimisation (profiling and using Numba for the time critical parts of the Heron code).

(3) I was also surprised to see that, despite being meant specifically to run on and connect diverse types of computer operating systems and being written purely in Python, the Heron Editor and GUI must be run on Windows. This seems like an unfortunate and unnecessary restriction, and it would be great to see the codebase adjusted to make it fully crossplatform-compatible.

This point was also mentioned by reviewer 2. This was a mistake on our part and has now been corrected in the paper. Heron (GUI and underlying communication functionality) can run on any machine that the underlying python libraries run, which is Windows, Linux (both for x86 and Arm architectures) and MacOS. We have tested it on Windows (10 and 11, both x64), Linux PC (Ubuntu 20.04.6, x64) and Raspberry Pi 4 (Debian GNU/Linux 12 (bookworm), aarch64). The Windows and Linux versions of Heron have undergone extensive debugging and all of the available Nodes (that are not OS specific) run on those two systems. We are in the process of debugging the Nodes’ functionality for RasPi. The MacOS version, although functional requires further work to make sure all of the basic Nodes are functional (which is not the case at the moment). We have also updated our manuscript (Multiple machines, operating systems and environments) to include the above information. 

(4) Lastly, when I was running test experiments, sometimes one of the nodes, or part of the Heron editor itself would throw an exception or otherwise crash. Sometimes this left the Heron editor in a zombie state where some aspects of the GUI were responsive and others were not. It would be good to see a more graceful full shutdown of the program when part of it crashes or throws an exception, especially as this is likely to be common as people learn to use it. More problematically, in some of these cases, after closing or force quitting Heron, the TCP ports were not properly relinquished, and thus restarting Heron would run into an "address in use" error. Finding and killing the processes that were still using the ports is not something that is obvious, especially to a beginner, and it would be great to see Heron deal with this better. Ideally, code would be introduced to carefully avoid leaving ports occupied during a hard shutdown, and furthermore, when the address in use error comes up, it would be great to give the user some idea of what to do about it.

A lot of effort has been put into Heron to achieve graceful shut down of processes, especially when these run on different machines that do not know when the GUI process has closed. The code that is being suggested to avoid leaving ports open has been implemented and this works properly when processes do not crash (Heron is terminated by the user) and almost always when there is a bug in a process that forces it to crash. In the version of Heron available during the reviewing process there were bugs that caused the above behaviour (Node code hanging and leaving zombie processes) on MacOS systems. These have now been fixed. There are very seldom instances though, especially during Node development, that crashing processes will hang and need to be terminated manually. We have taken on board the reviewer’s comments that users should be made more aware of these issues and have also described this situation in the Debugging part of Heron’s documentation. There we explain the logging and other tools Heron provides to help users debug their own Nodes and how to deal with hanging processes.

Heron is still in alpha (usable but with bugs) and the best way to debug it and iron out all the bugs in all use cases is through usage from multiple users and error reporting (we would be grateful if the errors the reviewer mentions could be reported in Heron’s github Issues page). We are always addressing and closing any reported errors, since this is the only way for Heron to transition from alpha to beta and eventually to production code quality.

Overall I think that, with these improvements, this could be the beginning of a powerful and versatile new system that would enable flexible experiment design with a relatively low technical barrier to entry. I could see this system being useful to many different labs and fields. 

We thank the reviewer for positive and supportive words and for the constructive feedbacks. We believe we have now addressed all the raised concerns.  

Reviewer #2 (Public Review):

Summary:

The authors provide an open-source graphic user interface (GUI) called Heron, implemented in Python, that is designed to help experimentalists to

(1) design experimental pipelines and implement them in a way that is closely aligned with their mental schemata of the experiments,

(2) execute and control the experimental pipelines with numerous interconnected hardware and software on a network.

The former is achieved by representing an experimental pipeline using a Knowledge Graph and visually representing this graph in the GUI. The latter is accomplished by using an actor model to govern the interaction among interconnected nodes through messaging, implemented using ZeroMQ. The nodes themselves execute user-supplied code in, but not limited to, Python.

Using three showcases of behavioral experiments on rats, the authors highlighted three benefits of their software design:

(1) the knowledge graph serves as a self-documentation of the logic of the experiment, enhancing the readability and reproducibility of the experiment,

(2) the experiment can be executed in a distributed fashion across multiple machines that each has a different operating system or computing environment, such that the experiment can take advantage of hardware that sometimes can only work on a specific computer/OS, a commonly seen issue nowadays,

(3) he users supply their own Python code for node execution that is supposed to be more friendly to those who do not have a strong programming background.

Strengths:

(1) The software is light-weight and open-source, provides a clean and easy-to-use GUI,

(2) The software answers the need of experimentalists, particularly in the field of behavioral science, to deal with the diversity of hardware that becomes restricted to run on dedicated systems.

(3) The software has a solid design that seems to be functionally reliable and useful under many conditions, demonstrated by a number of sophisticated experimental setups.

(4) The software is well documented. The authors pay special attention to documenting the usage of the software and setting up experiments using this software.

Weaknesses:

(1) While the software implementation is solid and has proven effective in designing the experiment showcased in the paper, the novelty of the design is not made clear in the manuscript. Conceptually, both the use of graphs and visual experimental flow design have been key features in many widely used softwares as suggested in the background section of the manuscript. In particular, contrary to the authors’ claim that only pre-defined elements can be used in Simulink or LabView, Simulink introduced MATLAB Function Block back in 2011, and Python code can be used in LabView since 2018. Such customization of nodes is akin to what the authors presented.

In the Heron manuscript we have provided an extensive literature review of existing systems from which Heron has borrowed ideas. We never wished to say that graphs and visual code is what sets Heron apart since these are technologies predating Heron by many years and implemented by a large number of software. We do not believe also that we have mentioned that LabView or Simulink can utilise only predefined nodes. What we have said is that in such systems (like LabView, Simulink and Bonsai) the focus of the architecture is on prespecified low level elements while the ability for users to author their own is there but only as an afterthought. The difference with Heron is that in the latter the focus is on the users developing their own elements. One could think of LabView style software as node-based languages (with low level visual elements like loops and variables) that also allow extra scripting while Heron is a graphical wrapper around python where nodes are graphical representations of whole processes. To our knowledge there is no other software that allows the very fast generation of graphical elements representing whole processes whose communication can also be defined graphically. Apart from this distinction, Heron also allows a graphical approach to writing code for processes that span different machines which again to our knowledge is a novelty of our approach and one of its strongest points towards ease of experimental pipeline creation (without sacrificing expressivity). 

(2) The authors claim that the knowledge graph can be considered as a self-documentation of an experiment. I found it to be true to some extent. Conceptually it’s a welcoming feature and the fact that the same visualization of the knowledge graph can be used to run and control experiments is highly desirable (but see point 1 about novelty). However, I found it largely inadequate for a person to understand an experiment from the knowledge graph as visualized in the GUI alone. While the information flow is clear, and it seems easier to navigate a codebase for an experiment using this method, the design of the GUI does not make it a one-stop place to understand the experiment. Take the Knowledge Graph in Supplementary Figure 2B as an example, it is associated with the first showcase in the result section highlighting this self-documentation capability. I can see what the basic flow is through the disjoint graph where 1) one needs to press a key to start a trial, and 2) camera frames are saved into an avi file presumably using FFMPEG. Unfortunately, it is not clear what the parameters are and what each block is trying to accomplish without the explanation from the authors in the main text. Neither is it clear about what the experiment protocol is without the help of Supplementary Figure 2A.

In my opinion, text/figures are still key to documenting an experiment, including its goals and protocols, but the authors could take advantage of the fact that they are designing a GUI where this information, with properly designed API, could be easily displayed, perhaps through user interaction. For example, in Local Network -> Edit IPs/ports in the GUI configuration, there is a good tooltip displaying additional information for the "password" entry. The GUI for the knowledge graph nodes can very well utilize these tooltips to show additional information about the meaning of the parameters, what a node does, etc, if the API also enforces users to provide this information in the form of, e.g., Python docstrings in their node template. Similarly, this can be applied to edges to make it clear what messages/data are communicated between the nodes. This could greatly enhance the representation of the experiment from the Knowledge graph.

In the first showcase example in the paper “Probabilistic reversal learning.

Implementation as self-documentation” we go through the steps that one would follow in order to understand the functionality of an experiment through Heron’s Knowledge Graph. The Graph is not just the visual representation of the Nodes in the GUI but also their corresponding code bases. We mention that the way Heron’s API limits the way a Node’s code is constructed (through an Actor based paradigm) allows for experimenters to easily go to the code base of a specific Node and understand its 2 functions (initialisation and worker) without getting bogged down in the code base of the whole Graph (since these two functions never call code from any other Nodes). Newer versions of Heron facilitate this easy access to the appropriate code by also allowing users to attach to Heron their favourite IDE and open in it any Node’s two scripts (worker and com) when they double click on the Node in Heron’s GUI. On top of this, Heron now (in the versions developed as answers to the reviewers’ comments) allows Node creators to add extensive comments on a Node but also separate comments on the Node’s parameters and input and output ports. Those can be seen as tooltips when one hovers over the Node (a feature that can be turned off or on by the Info button on every Node).  

As Heron stands at the moment we have not made the claim that the Heron GUI is the full picture in the self-documentation of a Graph. We take note though the reviewer’s desire to have the GUI be the only tool a user would need to use to understand an experimental implementation. The solution to this is the same as the one described by the reviewer of using the GUI to show the user the parts of the code relevant to a specific Node without the user having to go to a separate IDE or code editor. The reason this has not been implemented yet is the lack of a text editor widget in the underlying gui library (DearPyGUI). This is in their roadmap for their next large release and when this exists we will use it to implement exactly the idea the reviewer is suggesting, but also with the capability to not only read comments and code but also directly edit a Node’s code (see Heron’s roadmap). Heron’s API at the moment is ideal for providing such a text editor straight from the GUI.

(3) The design of Heron was primarily with behavioral experiments in mind, in which highly accurate timing is not a strong requirement. Experiments in some other areas that this software is also hoping to expand to, for example, electrophysiology, may need very strong synchronization between apparatus, for example, the record timing and stimulus delivery should be synced. The communication mechanism implemented in Heron is asynchronous, as I understand it, and the code for each node is executed once upon receiving an event at one or more of its inputs. The paper, however, does not include a discussion, or example, about how Heron could be used to address issues that could arise in this type of communication. There is also a lack of information about, for example, how nodes handle inputs when their ability to execute their work function cannot keep up with the frequency of input events. Does the publication/subscription handle the queue intrinsically? Will it create problems in real-time experiments that make multiple nodes run out of sync? The reader could benefit from a discussion about this if they already exist, and if not, the software could benefit from implementing additional mechanisms such that it can meet the requirements from more types of experiments.

In order to address the above lack of explanation (that also the first reviewer pointed out) we expanded the third experimental example in the paper with three more sections. One focuses solely on explaining how in this example (which acquires and saves large amounts of data from separate Nodes running on different machines) one would be able to time align the different data packets generated in different Nodes to each other. The techniques described there are directly implementable on experiments where the requirements of synching are more stringent than the behavioural experiment we showcase (like in ephys experiments). 

Regarding what happens to packages when the worker function of a Node is too slow to handle its traffic, this is mentioned in the paper (Code architecture paragraph): “Heron is designed to have no message buffering, thus automatically dropping any messages that come into a Node’s inputs while the Node’s worker function is still running.” This is also explained in more detail in Heron’s documentation. The reasoning for a no buffer system (as described in the documentation) is that for the use cases Heron is designed to handle we believe there is no situation where a Node would receive large amounts of data in bursts while very little data during the rest of the time (in which case a buffer would make sense). Nodes in most experiments will either be data intensive but with a constant or near constant data receiving speed (e.g. input from a camera or ephys system) or will have variable data load reception but always with small data loads (e.g. buttons). The second case is not an issue and the first case cannot be dealt with a buffer but with the appropriate code design, since buffering data coming in a Node too slow for its input will just postpone the inevitable crash. Heron’s architecture principle in this case is to allow these ‘mistakes’ (i.e. package dropping) to happen so that the pipeline continues to run and transfer the responsibility of making Nodes fast enough to the author of each Node. At the same time Heron provides tools (see the Debugging section of the documentation and the time alignment paragraph of the “Rats playing computer games”  example in the manuscript) that make it easy to detect package drops and either correct them or allow them but also allow time alignment between incoming and outgoing packets. In the very rare case where a buffer is required Heron’s do-it-yourself logic makes it easy for a Node developer to implement their own Node specific buffer.

(4) The authors mentioned in "Heron GUI’s multiple uses" that the GUI can be used as an experimental control panel where the user can update the parameters of the different Nodes on the fly. This is a very useful feature, but it was not demonstrated in the three showcases. A demonstration could greatly help to support this claim.

As the reviewer mentions, we have found Heron’s GUI double role also as an experimental on-line controller a very useful capability during our experiments. We have expanded the last experimental example to also showcase this by showing how on the “Rats playing computer games” experiment we used the parameters of two Nodes to change the arena’s behaviour while the experiment was running, depending on how the subject was behaving at the time (thus exploring a much larger set of parameter combinations, faster during exploratory periods of our shaping protocols construction). 

(5) The API for node scripts can benefit from having a better structure as well as having additional utilities to help users navigate the requirements, and provide more guidance to users in creating new nodes. A more standard practice in the field is to create three abstract Python classes, Source, Sink, and Transform that dictate the requirements for initialisation, work_function, and on_end_of_life, and provide additional utility methods to help users connect between their code and the communication mechanism. They can be properly docstringed, along with templates. In this way, the com and worker scripts can be merged into a single unified API. A simple example that can cause confusion in the worker script is the "worker_object", which is passed into the initialise function. It is unclear what this object this variable should be, and what attributes are available without looking into the source code. As the software is also targeting those who are less experienced in programming, setting up more guidance in the API can be really helpful. In addition, the self-documentation aspect of the GUI can also benefit from a better structured API as discussed in point 2 above.

The reviewer is right that using abstract classes to expose to users the required API would be a more standard practice. The reason we did not choose to do this was to keep Heron easily accessible to entry level Python programmers who do not have familiarity yet with object oriented programming ideas. So instead of providing abstract classes we expose only the implementation of three functions which are part of the worker classes but the classes themselves are not seen by the users of the API. The point about the users’ accessibility to more information regarding a few objects used in the API (the worker object for example) has been taken on board and we have now addressed this by type hinting all these objects both in the templates and more importantly in the automatically generated code that Heron now creates when a user chooses to create a Node graphically (a feature of Heron not present in the version available in the initial submission of this manuscript).  

(6) The authors should provide more pre-defined elements. Even though the ability for users to run arbitrary code is the main feature, the initial adoption of a codebase by a community, in which many members are not so experienced with programming, is the ability for them to use off-the-shelf components as much as possible. I believe the software could benefit from a suite of commonly used Nodes.

There are currently 12 Node repositories in the Heron-repositories project on Github with more than 30 Nodes, 20 of which are general use (not implementing a specific experiment’ logic). This list will continue to grow but we fully appreciate the truth of the reviewer’s comment that adoption will depend on the existence of a large number of commonly used Nodes (for example Numpy, and OpenCV Nodes) and are working towards this goal.

(7) It is not clear to me if there is any capability or utilities for testing individual nodes without invoking a full system execution. This would be critical when designing new experiments and testing out each component.

There is no capability to run the code of an individual Node outside Heron’s GUI. A user could potentially design and test parts of the Node before they get added into a Node but we have found this to be a highly inefficient way of developing new Nodes. In our hands the best approach for Node development was to quickly generate test inputs and/or outputs using the “User Defined Function 1I 1O” Node where one can quickly write a function and make it accessible from a Node. Those test outputs can then be pushed in the Node under development or its outputs can be pushed in the test function, to allow for incremental development without having to connect it to the Nodes it would be connected in an actual pipeline. For example, one can easily create a small function that if a user presses a key will generate the same output (if run from a “User Defined Function 1I 1O” Node) as an Arduino Node reading some buttons. This output can then be passed into an experiment logic Node under development that needs to do something with this input. In this way during a Node development Heron allows the generation of simulated hardware inputs and outputs without actually running the actual hardware. We have added this way of developing Nodes also in our manuscript (Creating a new Node).

Reviewer #3 (Public Review):

Summary:

The authors present a Python tool, Heron, that provides a framework for defining and running experiments in a lab setting (e.g. in behavioural neuroscience). It consists of a graphical editor for defining the pipeline (interconnected nodes with parameters that can pass data between them), an API for defining the nodes of these pipelines, and a framework based on ZeroMQ, responsible for the overall control and data exchange between nodes. Since nodes run independently and only communicate via network messages, an experiment can make use of nodes running on several machines and in separate environments, including on different operating systems.

Strengths:

As the authors correctly identify, lab experiments often require a hodgepodge of separate hardware and software tools working together. A single, unified interface for defining these connections and running/supervising the experiment, together with flexibility in defining the individual subtasks (nodes) is therefore a very welcome approach. The GUI editor seems fairly intuitive, and Python as an accessible programming environment is a very sensible choice. By basing the communication on the widely used ZeroMQ framework, they have a solid base for the required non-trivial coordination and communication. Potential users reading the paper will have a good idea of how to use the software and whether it would be helpful for their own work. The presented experiments convincingly demonstrate the usefulness of the tool for realistic scientific applications.

Weaknesses:

(1) In my opinion, the authors somewhat oversell the reproducibility and "selfdocumentation" aspect of their solution. While it is certainly true that the graph representation gives a useful high-level overview of an experiment, it can also suffer from the same shortcomings as a "pure code" description of a model - if a user gives their nodes and parameters generic/unhelpful names, reading the graph will not help much. 

This is a problem that to our understanding no software solution can possibly address. Yet having a visual representation of how different inputs and outputs connect to each other we argue would be a substantial benefit in contrast to the case of “pure code” especially when the developer of the experiment has used badly formatted variable names.

(2) Making the link between the nodes and the actual code is also not straightforward, since the code for the nodes is spread out over several directories (or potentially even machines), and not directly accessible from within the GUI. 

This is not accurate. The obligatory code of a Node always exists within a single folder and Heron’s API makes it rather cumbersome to spread scripts relating to a Node across separate folders. The Node folder structure can potentially be copied over different machines but this is why Heron is tightly integrated with git practices (and even politely asks the user with popup windows to create git repositories of any Nodes they create whilst using Heron’s automatic Node generator system). Heron’s documentation is also very clear on the folder structure of a Node which keeps the required code always in the same place across machines and more importantly across experiments and labs. Regarding the direct accessibility of the code from the GUI, we took on board the reviewers’ comments and have taken the first step towards correcting this. Now one can attach to Heron their favourite IDE and then they can double click on any Node to open its two main scripts (com and worker) in that IDE embedded in whatever code project they choose (also set in Heron’s settings windows). On top of this, Heron now allows the addition of notes both for a Node and for all its parameters, inputs and outputs which can be viewed by hovering the mouse over them on the Nodes’ GUIs. The final step towards GUI-code integration will be to have a Heron GUI code editor but this is something that has to wait for further development from Heron’s underlying GUI library DearPyGUI.

(3) The authors state that "[Heron’s approach] confers obvious benefits to the exchange and reproducibility of experiments", but the paper does not discuss how one would actually exchange an experiment and its parameters, given that the graph (and its json representation) contains user-specific absolute filenames, machine IP addresses, etc, and the parameter values that were used are stored in general data frames, potentially separate from the results. Neither does it address how a user could keep track of which versions of files were used (including Heron itself).

Heron’s Graphs, like any experimental implementation, must contain machine specific strings. These are accessible either from Heron’s GUI when a Graph json file is opened or from the json file itself. Heron in this regard does not do anything different to any other software, other than saving the graphs into human readable json files that users can easily manipulate directly.

Heron provides a method for users to save every change of the Node parameters that might happen during an experiment so that it can be fully reproduced. The dataframes generated are done so in the folders specified by the user in each of the Nodes (and all those paths are saved in the json file of the Graph). We understand that Heron offers a certain degree of freedom to the user (Heron’s main reason to exist is exactly this versatility) to generate data files wherever they want but makes sure every file path gets recorded for subsequent reproduction. So, Heron behaves pretty much exactly like any other open source software. What we wanted to focus on as the benefits of Heron on exchange and reproducibility was the ability of experimenters to take a Graph from another lab (with its machine specific file paths and IP addresses) and by examining the graphical interface of it to be able to quickly tweak it to make it run on their own systems. That is achievable through the fact that a Heron experiment will be constructed by a small amount of Nodes (5 to 15 usually) whose file paths can be trivially changed in the GUI or directly in the json file while the LAN setup of the machines used can be easily reconstructed from the information saved in the secondary GUIs.

Where Heron needs to improve (and this is a major point in Heron’s roadmap) is the need to better integrate the different saved experiments with the git versions of Heron and the Nodes that were used for that specific save. This, we appreciate is very important for full reproducibility of the experiment and it is a feature we will soon implement. More specifically users will save together with a graph the versions of all the used repositories and during load the code base utilised will come from the recorded versions and not from the current head of the different repositories. This is a feature that we are currently working on now and as our roadmap suggests will be implemented by the release of Heron 1.0. 

(4) Another limitation that in my opinion is not sufficiently addressed is the communication between the nodes, and the effect of passing all communications via the host machine and SSH. What does this mean for the resulting throughput and latency - in particular in comparison to software such as Bonsai or Autopilot? The paper also states that "Heron is designed to have no message buffering, thus automatically dropping any messages that come into a Node’s inputs while the Node’s worker function is still running."- it seems to be up to the user to debug and handle this manually?

There are a few points raised here that require addressing. The first is Heron’s requirement to pass all communication through the main (GUI) machine. We understand (and also state in the manuscript) that this is a limitation that needs to be addressed. We plan to do this is by adding to Heron the feature of running headless (see our roadmap). This will allow us to run whole Heron pipelines in a second machine which will communicate with the main pipeline (run on the GUI machine) with special Nodes. That will allow experimenters to define whole pipelines on secondary machines where the data between their Nodes stay on the machine running the pipeline. This is an important feature for Heron and it will be one of the first features to be implemented next (after the integration of the saving system with git). 

The second point is regarding Heron’s throughput latency. In our original manuscript we did not have any description of Heron’s capabilities in this respect and both other reviewers mentioned this as a limitation. As mentioned above, we have now addressed this by adding a section to our third experimental example that fully describes how much CPU is required to run a full experimental pipeline running on two machines and utilising also non python code executables (a Unity game). This gives an overview of how heavy pipelines can run on normal computers given adequate optimisation and utilising Heron’s feature of forcing some Nodes to run their Worker processes on a specific core. At the same time, Heron’s use of 0MQ protocol makes sure there are no other delays or speed limitations to message passing. So, message passing within the same machine is just an exchange of memory pointers while messages passing between different machines face the standard speed limitations of the Local Access Network’s ethernet card speeds. 

Finally, regarding the message dropping feature of Heron, as mentioned above this is an architectural decision given the use cases of message passing we expect Heron to come in contact with. For a full explanation of the logic here please see our answer to the 3rd comment by Reviewer 2.

(5) As a final comment, I have to admit that I was a bit confused by the use of the term "Knowledge Graph" in the title and elsewhere. In my opinion, the Heron software describes "pipelines" or "data workflows", not knowledge graphs - I’d understand a knowledge graph to be about entities and their relationships. As the authors state, it is usually meant to make it possible to "test propositions against the knowledge and also create novel propositions" - how would this apply here?

We have described Heron as a Knowledge Graph instead of a pipeline, data workflow or computation graph in order to emphasise Heron’s distinct operation in contrast to what one would consider a standard pipeline and data workflow generated by other visual based software (like LabView and Bonsai). This difference exists on what a user should think of as the base element of a graph, i.e. the Node. In all other visual programming paradigms, the Node is defined as a low-level computation, usually a language keyword, language flow control or some simple function. The logic in this case is generated by composing together the visual elements (Nodes). In Heron the Node is to be thought of as a process which can be of arbitrary complexity and the logic of the graph is composed by the user both within each Node and by the way the Nodes are combined together. This is an important distinction in Heron’s basic operation logic and it is we argue the main way Heron allows flexibility in what can be achieved while retaining ease of graph composition (by users defining their own level of complexity and functionality encompassed within each Node). We have found that calling this approach a computation graph (which it is) or a pipeline or data workflow would not accentuate this difference. The term Knowledge Graph was the most appropriate as it captures the essence of variable information complexity (even in terms of length of shortest string required) defined by a Node.

Recommendations for the authors:  

Reviewer #1 (Recommendations For The Authors):

-  No buffering implies dropped messages when a node is busy. It seems like this could be very problematic for some use cases... 

This is a design principle of Heron. We have now provided a detailed explanation of the reasoning behind it in our answer to Reviewer 2 (Paragraph 3) as well as in the manuscript. 

-  How are ssh passwords stored, and is it secure in some way or just in plain text?  

For now they are plain text in an unencrypted file that is not part of the repo (if one gets Heron from the repo). Eventually, we would like to go to private/public key pairs but this is not a priority due to the local nature of Heron’s use cases (all machines in an experiment are expected to connect in a LAN).  

Minor notes / copyedits:

-  Figure 2A: right and left seem to be reversed in the caption. 

They were. This is now fixed. 

-  Figure 2B: the text says that proof of life messages are sent to each worker process but in the figure, it looks like they are published by the workers? Also true in the online documentation.  

The Figure caption was wrong. This is now fixed.

-  psutil package is not included in the requirements for GitHub

We have now included psutil in the requirements.

-  GitHub readme says Python >=3.7 but Heron will not run as written without python >= 3.9 (which is alluded to in the paper)

The new Heron updates require Python 3.11. We have now updated GitHub and the documentation to reflect this.

-  The paper mentions that the Heron editor must be run on Windows, but this is not mentioned in the Github readme.  

This was an error in the manuscript that we have now corrected.

-  It’s unclear from the readme/manual how to remove a node from the editor once it’s been added.  

We have now added an X button on each Node to complement the Del button on the keyboard (for MacOS users that do not have this button most of the times).

-  The first example experiment is called the Probabilistic Reversal Learning experiment in text, but the uncertainty experiment in the supplemental and on GitHub.  

We have now used the correct name (Probabilistic Reversal Learning) in both the supplemental material and on GitHub

-  Since Python >=3.9 is required, consider using fstrings instead of str.format for clarity in the codebase  

Thank you for the suggestion. Latest Heron development has been using f strings and we will do a refactoring in the near future.

-  Grasshopper cameras can run on linux as well through the spinnaker SDK, not just Windows.  

Fixed in the manuscript. 

-  Figure 4: Square and star indicators are unclear.

Increased the size of the indicators to make them clear.

-  End of page 9: "an of the self" presumably a typo for "off the shelf"?  

Corrected.

-  Page 10 first paragraph. "second root" should be "second route"

Corrected.

-  When running Heron, the terminal constantly spams Blowfish encryption deprecation warnings, making it difficult to see the useful messages.  

The solution to this problem is to either update paramiko or install Heron through pip. This possible issue is mentioned in the documentation.

-  Node input /output hitboxes in the GUI are pretty small. If they could be bigger it would make it easier to connect nodes reliably without mis-clicks.

We have redone the Node GUI, also increasing the size of the In/Out points.

Reviewer #2 (Recommendations For The Authors):

(1) There are quite a few typos in the manuscript, for example: "one can accessess the code", "an of the self", etc.  

Thanks for the comment. We have now screened the manuscript for possible typos.

(2) Heron’s GUI can only run on Windows! This seems to be the opposite of the key argument about the portability of the experimental setup.  

As explained in the answers to Reviewer 1, Heron can run on most machines that the underlying python libraries run, i.e. Windows and Linux (both for x86 and Arm architectures). We have tested it on Windows (10 and 11, both x64), Linux PC (Ubuntu 20.04.6, x64) and Raspberry Pi 4 (Debian GNU/Linux 12 (bookworm), aarch64). We have now revised the manuscript and the GitHub repo to reflect this.

(3) Currently, the output is displayed along the left edge of the node, but the yellow dot connector is on the right. It would make more sense to have the text displayed next to the connectors.  

We have redesigned the Node GUI and have now placed the Out connectors on the right side of the Node.

(4) The edges are often occluded by the nodes in the GUI. Sometimes it leads to some confusion, particularly when the number of nodes is large, e.g., Fig 4.

This is something that is dependent on the capabilities of the DearPyGUI module. At the moment there is no way to control the way the edges are drawn.

Reviewer #3 (Recommendations For The Authors):

A few comments on the software and the documentation itself:

- From a software engineering point of view, the implementation seems to be rather immature. While I get the general appeal of "no installation necessary", I do not think that installing dependencies by hand and cloning a GitHub repository is easier than installing a standard package.

We have now added a pip install capability which also creates a Heron command line command to start Heron with. 

-The generous use of global variables to store state (minor point, given that all nodes run in different processes), boilerplate code that each node needs to repeat, and the absence of any kind of automatic testing do not give the impression of a very mature software (case in point: I had to delete a line from editor.py to be able to start it on a non-Windows system).  

As mentioned, the use of global variables in the worker scripts is fine partly due to the multi process nature of the development and we have found it is a friendly approach to Matlab users who are just starting with Python (a serious consideration for Heron). Also, the parts of the code that would require a singleton (the Editor for example) are treated as scripts with global variables while the parts that require the construction of objects are fully embedded in classes (the Node for example). A future refactoring might make also all the parts of the code not seen by the user fully object oriented but this is a decision with pros and cons needing to be weighted first. 

Absence of testing is an important issue we recognise but Heron is a GUI app and nontrivial unit tests would require some keystroke/mouse movement emulator (like QTest of pytest-qt for QT based GUIs). This will be dealt with in the near future (using more general solutions like PyAutoGUI) but it is something that needs a serious amount of effort (quite a bit more that writing unit tests for non GUI based software) and more importantly it is nowhere as robust as standard unit tests (due to the variable nature of the GUI through development) making automatic test authoring an almost as laborious a process as the one it is supposed to automate.

-  From looking at the examples, I did not quite see why it is necessary to write the ..._com.py scripts as Python files, since they only seem to consist of boilerplate code and variable definitions. Wouldn’t it be more convenient to represent this information in configuration files (e.g. yaml or toml)?  

The com is not a configuration file, it is a script that launches the communication process of the Node. We could remove the variable definitions to a separate toml file (which then the com script would have to read). The pros and cons of such a set up should be considered in a future refactoring.

Minor comments for the paper:

-  p.7 (top left): "through its return statement" - the worker loop is an infinite loop that forwards data with a return statement?  

This is now corrected. The worker loop is an infinite loop and does not return anything but at each iteration pushes data to the Nodes output.

-  p.9 (bottom right): "of the self" → "off-the-shelf"  

Corrected.

-  p.10 (bottom left): "second root" → "second route"  

Corrected.

-  Supplementary Figure 3: Green start and square seem to be swapped (the green star on top is a camera image and the green star on the bottom is value visualization - inversely for the green square).  

The star and square have been swapped around.

-  Caption Supplementary Figure 4 (end): "rashes to receive" → "rushes to receive"  

Corrected.

Voici un sommaire minuté basé sur la transcription du webinaire Solidatech:

• 0:00-0:12: Introduction du webinaire et présentation du programme Solidatech, une initiative de solidarité numérique.

L'objectif est de présenter en détail les offres de Solidatech, incluant logiciels, services, accompagnement et matériel, ainsi que d'expliquer leur fonctionnement.

• 0:26-1:51: Présentation globale de Solidatech par Laurine.

Solidatech est un programme de solidarité numérique visant à aider les associations à renforcer leur impact grâce à une meilleure utilisation du numérique.

Créé en 2008, le programme est géré par une équipe d'une douzaine de personnes réparties entre Paris et les Deux-Sèvres, appartenant aux Ateliers du Bocage, une coopérative d'insertion.

Les Ateliers du Bocage se spécialisent dans la collecte et le réemploi de matériel informatique, de smartphones et de cartouches. Solidatech fait partie du réseau international TechSoup, ce qui lui permet d'offrir des réductions sur divers logiciels.

• 1:58-2:28: Bénéficiaires de Solidatech. Principalement, Solidatech accompagne les associations loi 1901, mais aussi d'autres structures comme les fondations RUP, les fonds de dotation et les bibliothèques publiques.

L'inscription à Solidatech est gratuite via leur site internet, et le programme est accessible quel que soit le secteur d'activité ou le mode de fonctionnement de la structure.

• 2:35-3:25: Modes d'action de Solidatech. Solidatech intervient de trois manières : en facilitant l'accès au numérique par des tarifs réduits sur les logiciels et le matériel informatique, en accompagnant les associations dans le développement de leurs usages numériques (centre de ressources, diagnostic numérique, formations), et en participant à la coproduction et à la diffusion de savoirs sur la transition numérique des associations, notamment à travers des études triennales.

• 3:25-4:25: Logiciels et solutions proposés par Solidatech. Solidatech offre une variété de logiciels couvrant divers besoins des associations, allant de la comptabilité aux antivirus, en passant par les outils collaboratifs.

Le catalogue inclut des acteurs internationaux comme Zoom, Adobe et Microsoft, ainsi que des solutions françaises comme Net Explorer. Solidatech propose également du matériel informatique reconditionné provenant des Ateliers du Bocage, ainsi que du matériel neuf via des partenaires comme Cisco et Dell.

• 4:32-5:27: Accompagnement aux usages numériques.

Solidatech met à disposition des ressources pour une utilisation autonome, comme un centre de ressources et des newsletters.

Des webinaires thématiques sont aussi proposés, comme celui sur le RGPD prévu à la fin du mois. Un accompagnement individuel est également disponible, avec un catalogue de formations et des enjeux de sensibilisation au numérique.

• 5:27-5:53: Chiffres clés et transition vers la présentation du site web. Solidatech accompagne plus de 42 000 associations.

• 5:53-6:23: Transition vers la présentation du site web par Elodie.

• 6:23-9:54: Visite guidée du site Solidatech.fr. Le site Solidatech.fr est la porte d'entrée vers le programme de solidarité numérique, présentant les offres, services et accompagnements.

Le site oriente également vers des plateformes secondaires pour les logiciels à tarif réduit et l'achat de matériel. L'inscription gratuite est nécessaire pour bénéficier de ces offres.

• 9:54-11:32: Processus de connexion et présentation de l'espace utilisateur. Après l'inscription, les utilisateurs peuvent se connecter via le site Solidatech ou directement lorsqu'ils cliquent sur une offre.

La connexion redirige vers la plateforme TechSoup.fr, où les associations accèdent à leur interface utilisateur.

• 11:32-12:16: Présentation de la plateforme TechSoup.fr. TechSoup.fr est la plateforme principale pour commander des logiciels à tarif solidaire et du matériel neuf, grâce aux partenariats internationaux. D'autres partenariats et offres spécifiques sont disponibles sur d'autres plateformes.

• 12:16-15:38: Catalogue de logiciels à tarif réduit. Le catalogue de logiciels peut être consulté sans être connecté, mais la connexion est nécessaire pour commander.

Les utilisateurs peuvent rechercher des logiciels par nom ou filtrer par catégorie pour identifier les solutions répondant à leurs besoins. Le catalogue inclut plus de 260 offres, mais l'éligibilité varie en fonction des critères définis par les éditeurs, tels que le budget annuel ou le secteur d'activité.

• 15:38-17:22: Fiche produit et modalités de l'offre. Chaque fiche produit détaille les caractéristiques du logiciel et les modalités de l'offre Solidatech, incluant les réductions, les critères de restriction et les conditions d'éligibilité. Il est conseillé de lire attentivement l'onglet "Modalités de l'offre" pour comprendre les conditions spécifiques.

• 17:22-19:20: Fonctionnement des coupons de réduction. Pour certaines offres, l'achat d'un coupon de réduction est nécessaire pour bénéficier d'un tarif réduit sur l'abonnement au logiciel, comme pour Adobe Creative Cloud.

Le paiement des frais administratifs à Solidatech permet d'acquérir ce coupon, donnant accès à la réduction sur le site du partenaire. Pour d'autres solutions, comme l'antivirus Bitdefender, le paiement se fait uniquement auprès de Solidatech.

• 19:20-20:36: Équipements neufs. L'offre d'équipement neuf concerne principalement le matériel réseau de la marque Cisco, proposé avec des réductions importantes, ainsi que quelques caméras de surveillance.

Des coupons d'achat permettent d'accéder à des réductions sur le matériel neuf chez Dell. Le matériel réseau Cisco est proposé avec une réduction d'environ 90 % par rapport au prix du marché.

• 20:36-23:33: Matériel reconditionné. Le matériel reconditionné est accessible via la plateforme TechSoup ou directement depuis le site Solidatech.

La boutique de matériel reconditionné propose du matériel informatique reconditionné au sein des ateliers, incluant des ordinateurs portables, des unités centrales, de la téléphonie, des tablettes, ainsi que des accessoires et périphériques. Des articles et conseils sur le reconditionné sont également disponibles.

• 23:33-26:10: Informations sur les produits reconditionnés.

Les produits reconditionnés sont garantis 12 mois et sont effacés et testés. Pendant le mois de février, l'extension de garantie est offerte, passant de 12 à 24 mois.

Des informations détaillées sur les produits, leurs performances et les conditions de garantie sont fournies sur le site. Une demande de devis peut être faite pour des commandes spécifiques ou en grande quantité.

• 26:10-27:50: Processus de commande de matériel reconditionné. Pour la première commande sur la boutique de reconditionné, il est nécessaire de créer un compte client et de renseigner un jeton de validation pour vérifier l'éligibilité de l'association. Ce jeton est unique et peut être généré sur le compte Solidatech.

• 27:50-29:49: Explication du jeton de validation.

Le jeton de validation est un code permettant de vérifier que la structure est bien une association. Il est disponible sur le compte Solidatech et peut être généré autant de fois que nécessaire.

• 29:49-30:42: Utilisation du jeton de validation pour accéder aux offres. Le jeton de validation est souvent demandé pour accéder à des offres logicielles ou à certains services.

• 30:42-31:51: Formations et conseils. Solidatech propose des formations certifiées Qualiopi, destinées aux salariés et bénévoles des associations, couvrant les enjeux du numérique et l'utilisation d'outils. Les formations peuvent être suivies à distance ou en présentiel dans les locaux de l'association.

• 31:51-33:51: Services de migration vers le cloud.

Solidatech propose des services de migration vers le cloud avec des partenaires comme IT for Life et Se Consulting, pour accompagner les associations vers Workspace ou Microsoft 365. De nouveaux services et formations seront proposés prochainement.

• 33:51-35:23: Prestatech.

Prestatech est un annuaire de prestataires de confiance, préqualifiés par Solidatech, offrant des services de conseil, de prestation informatique et de formation. Les prestataires sont classés selon leur domaine d'expertise et leurs modalités de financement.

• 35:23-37:20: Centre de ressources.

Le centre de ressources est un blog contenant des articles conseils, des cas d'usage et des comparatifs d'outils, classés par thématique. Il inclut également les replays des webinaires.

• 37:20-38:35: Autodiagnostic numérique.

L'autodiagnostic numérique est un outil gratuit permettant d'évaluer la maturité numérique de l'association à travers un questionnaire interactif organisé autour de sept piliers stratégiques.

Il aide à prioriser les chantiers à mener et propose une sélection de logiciels et de ressources pertinentes.

• 38:35-38:47: Étude nationale sur le numérique associatif. Solidatech mène une étude tous les trois ans sur la place du numérique dans le projet associatif, en partenariat avec Recherche et Solidarités.
• 38:47-49:23: Questions/réponses.

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

Reply to the Reviewers

We are very grateful to the reviewers for their time and care in reviewing our manuscript. We have tried to incorporate all of their feedback to the best of our ability, and we feel that this has greatly improved the manuscript.

Reviewer #1

This study provides a strong support for the relationship between replication starting point competition and initial factor concentration. However, some predictive conclusions, such as "the origin of high efficiency may not be activated earlier", are still preliminary. Can the author further clarify the scope of these predictions and any potential mechanism in the discussion part to improve the rigor of this study?

__Response: __In the discussion, we now emphasize the complexity of predicting origin firing time distributions, which are influenced by multiple interrelated factors beyond efficiency alone.

The resolution and accuracy of the model prediction are obvious to all, but the specific generalization ability is still unknown, which makes the further promotion slightly insufficient. Does the author consider conducting additional experiments? To detect the replication time and efficiency in yeast cells with changed levels of key initiation factors (such as Cdc45 or Dpb11). The empirical data can be compared with the model prediction by editing CRISPR gene or manipulating the initial factor abundance through overexpression vector.

__Response: __We fully agree that this would be a very interesting direction, but as this is a theoretical study focused on mathematical modelling, conducting further wet lab experiments would be beyond the scope of this work.

The model currently uses single values for the initiation factor number and recycling rate, though these parameters may vary across cell cycles or under different growth conditions. It is suggested that sensitivity analysis should be added to supplementary materials to explore how the changes of these parameters affect the model output, such as replication time distribution and origin efficiency.

__Response: __Sensitivity analysis of how the model fit and validation is affected by using different recycling rates and initial firing factor counts will be conducted.

While the authors use mean absolute error (MAE) to assess model fit, it is suggested to add other statistical methods, such as root mean square error or correlation analysis, to further evaluate the model's accuracy and robustness. In addition, this model lacks comparison with other studies on fitting yeast replication time, and it is difficult to evaluate the effect of this model compared with other models from the specific performance.

__Response: __We have now included the root mean squared error (RMSE) alongside the mean absolute error (MAE) and R-squared value to compare the simulated replication timing profiles with the experimental data. We agree that we could have been more detailed in comparing our model to other approaches. We have now added a lengthened discussion of this. In some cases, a direct comparison of performance is difficult due to fundamental differences between the approaches, but we have highlighted why this is the case.

Although the code is open, it is suggested to provide specific instructions or examples of the running code in supplementary materials, so as to facilitate reproduction and application by other researchers.

__Response: __The GitHub repository will be updated to enable the running of the entire pipeline. This update will include code for processing replication timing data from Müller et al. (2014) and extracting origin positions from the OriDB. Code will also be provided for writing Beacon Calculus scripts with different parameters and origin firing rates. Instructions on the recommended sequence in which scripts should be executed will also be provided. To enable users to run the model locally on their own computers, a smaller version focused on chromosome 2 will be included in the supplementary information and GitHub repository, along with example input data and expected outputs.

In Figure 2(a), compared with other chromosomes, the fitting effect of chromosome 1 seems to be not good. Has the author ever thought about the reason? In addition, what is the guiding significance of this model in practical applications, such as online services, forecasting tools, or experiments? Can the author give relevant application examples in this regard?

__Response: __Potential explanations for the poorer fit of the replication timing profile for chromosome 1 are now discussed. The y-axis range has also now been set as the same for all subplots in Figure 2a to make the replication timing profiles for each chromosome more easily comparable. In the discussion, we highlight how the intuitive and flexible nature of the model places it as a valuable tool which could be adapted to predict the effect of different perturbations on DNA replication dynamics.

Reviewer #2

In figure 5, the authors demonstrate that replication dynamics are robust to an increase in the number of available firing factors. However, experimental data from strains in which these limiting factors are overexpressed indicate that replication dynamics are substantially altered (e.g. PMID 22081107 and 23562327) since dNTPs become limiting. So the conclusions of the analysis in figure 5 are at best an oversimplification and at worst rather misleading. If adding dNTPs as a factor that becomes limiting only at higher firing factor concentrations is not technically feasible, the authors should be more circumspect in their description and discussion of the results in figure 5.

__Response: __We now discuss the interpretation of the effect of increasing the number of firing factors, given that factors such as dNTP availability are not included in the model.

The analysis of replication dynamics appears to exclude origins within the rDNA, which in the average strain account for ~20-25% of all replication origins in S. cerevisiae depending on the origin list chosen. Ignoring this large number of origins likely has a substantial impact on the model: if rDNA origins are intentionally ignored due to the difficulty of modeling repetitive regions or of having multiple identical origins in the competition model, this should be explicitly addressed in the text.

__Response: __We now emphasize that our model restricts initiation to specific sites and note that some low-efficiency origins, such as those in rDNA, have not been included.

Reviewer #3

Can the authors provide some insight into the model's dependency on the Müller, 2014 replication data set? They initialize and converge to this dataset so this paper's findings are highly contingent on treating this data set as ground truth.

__Response: __In the discussion, we now highlight that, despite the model's reliance on the Müller, 2014 replication data set for fitting, its ability to reproduce other features of DNA replication demonstrates its ability to reflect DNA replication dynamics more broadly.

The authors describe their model as one that simplifies the origin firing mechanisms compared to more complex models. Is there a direct comparison available that can quantify this advantage? Likewise, how does their model compare to a naive discriminative model, such as one that performs peak finding on the replication timing data. For example, the replication fork directionality can be estimated, naively, using a peak finding algorithm. This type of analysis will provide a stronger argument for the usage of their model.

__Response: __Quantitative comparisons between our model and other published models are challenging due to differences in underlying assumptions and metrics used to assess goodness of fit. However, we have now added a discussion addressing these challenges and highlighting how our model's design contrasts with that of other models.

Currently the code is available as supplemental data. Ideally, the code should be available and provided to run the entire pipeline beginning with the initialization of the origin firing program from the Müller, 2014 data set.

__Response: __The GitHub repository will be updated to enable the running of the entire pipeline. This update will include code for processing replication timing data from Müller et al. (2014) and extracting origin positions from the OriDB.

The authors mention that origin firing factors and their recycling time to be the basis of how this model is constructed. While also describing the recycle time as a general timing delay that is dependent on a number of reasons such as diffusion and replisome complex formation. Can the authors discuss the limitations of their model towards this simplification?

__Response: __Limitations of our model's assumptions of constant recycling rates of firing factors are now discussed, as well as our assumption that the firing rates of origins and the maximum number of available firing factors remain constant between simulations.

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 #3

Evidence, reproducibility and clarity

Summary:

In this paper, the authors create a model of origin replication in yeast using Beacon calculus and a small set of parameters. The model is described as the relationship between origin firing rate and the abundance and recycling of origin firing factors. Using the (Müller, 2014) replication timing data to initialize and fit their model, the authors show that their model recapitulates known replication-related work such as inter-origin distances, replication fork directionality, and origin efficiency. Next, they utilize their model to make predictions that characterize the broader replication program, such as in the quantification of active replication forks, replicons, and replication timing.

Major comments:

Can the authors provide some insight into the model's dependency on the Müller, 2014 replication data set? They initialize and converge to this dataset so this paper's findings are highly contingent on treating this data set as ground truth.

The authors describe their model as one that simplifies the origin firing mechanisms compared to more complex models. Is there a direct comparison available that can quantify this advantage?

Likewise, how does their model compare to a naive discriminative model, such as one that performs peak finding on the replication timing data. For example, the replication fork directionality can be estimated, naively, using a peak finding algorithm. This type of analysis will provide a stronger argument for the usage of their model.

Currently the code is available as supplemental data. Ideally, the code should be available and provided to run the entire pipeline beginning with the initialization of the origin firing program from the Müller, 2014 data set.

The authors mention that origin firing factors and their recycling time to be the basis of how this model is constructed. While also describing the recycle time as a general timing delay that is dependent on a number of reasons such as diffusion and replisome complex formation. Can the authors discuss the limitations of their model towards this simplification?

Minor comments:

The author describes the prediction of 200 active replication forks 22 minutes into S phase. Please discuss why this peak number of active replication forks may have been reached. Is this related to the model configured for the number of firing factors F = 200?

The recycling parameter appears to be very important for this model. A sensitivity analysis of the value of 0.05 would be helpful to understand why this value was chosen.

It would be helpful to understand the convergence of the model better. Can the authors provide insight or a plot to better understand why the convergence parameter alpha was chosen as 1.2?

The authors comment that simulated origin efficiencies were estimated close to zero (6.2%{plus minus}22%). Can the authors comment on the large variability in this estimation (the {plus minus}22%)?

Significance

General Assessment

The strength of the model is in summarizing the origin efficiency firing mechanism into a small set of parameters. This also relates to its limitations. The model asserts that the origin firing depends solely on the abundance and recycling of origin firing factors. This limits the scope of the interpretation of the mechanisms of origin firing compared to more complex models.

Additionally, the model is fit to, and thus, highly dependent on the quality of the Müller, 2014 dataset.

Improvements:

This work can be improved by comparing and contrasting their results to existing models where they argue the advantages of employing a simpler model for origin firing compared to more complex ones they cite (Arbona, 2018; de Moura, 2010; Retkute, 2014; Brümmer, 2010).

While their modeling and dependency on the Müller, 2024 replication timing data may be sufficient, some of the findings can be naively characterized from this data set, such as in replication fork direction and origin firing times. Thus, the authors can argue the strengths of their model by contrasting theirs to more simpler and naive quantifications.

Currently the paper is very descriptive. A nice addition would be to model the effects of Rpd3 deletion which is thought to either have a direct effect on late origins (advancing their time of replication) or an indirect effect via the rDNA locus which may, in the absence of rpd3) act as sink for limiting replication factors. (Vogelauer et al., Mol Cell, 2002; Yoshida et al.,Mol Cell 2014, He et al., PNAS 2022). Specifically, how does titrating the number of active rDNA origins out of the ~150 available rDNA origins impact global origin usage under this model?

Scope:

Audience: Specialized towards groups modeling and studying replication.

Reviewer's field of expertise: Computer science, computational biology, bioinformatics, and general computational modeling

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

Summary

This study develops a high-resolution stochastic model to explore DNA replication timing regulation in Saccharomyces cerevisiae, specifically focusing on competition between replication origins for limited initiation factors. The model, based on "Beacon Calculus," utilizes an iterative optimization process to fit experimental data, successfully reproducing timing, efficiency, and directionality features of genome replication origins. Additionally, the authors use the model to make predictions on replication dynamics under varying initiation factor concentrations, providing new insights into DNA replication processes that have not yet been observed empirically or experimentally.

Major Comments:

1. This study provides a strong support for the relationship between replication starting point competition and initial factor concentration. However, some predictive conclusions, such as "the origin of high efficiency may not be activated earlier", are still preliminary. Can the author further clarify the scope of these predictions and any potential mechanism in the discussion part to improve the rigor of this study?
2. The resolution and accuracy of the model prediction are obvious to all, but the specific generalization ability is still unknown, which makes the further promotion slightly insufficient. Does the author consider conducting additional experiments? To detect the replication time and efficiency in yeast cells with changed levels of key initiation factors (such as Cdc45 or Dpb11). The empirical data can be compared with the model prediction by editing CRISPR gene or manipulating the initial factor abundance through overexpression vector.
3. The model currently uses single values for the initiation factor number and recycling rate, though these parameters may vary across cell cycles or under different growth conditions. It is suggested that sensitivity analysis should be added to supplementary materials to explore how the changes of these parameters affect the model output, such as replication time distribution and origin efficiency.
4. While the authors use mean absolute error (MAE) to assess model fit, it is suggested to add other statistical methods, such as root mean square error or correlation analysis, to further evaluate the model's accuracy and robustness. In addition, this model lacks comparison with other studies on fitting yeast replication time, and it is difficult to evaluate the effect of this model compared with other models from the specific performance.
5. Although the code is open, it is suggested to provide specific instructions or examples of the running code in supplementary materials, so as to facilitate reproduction and application by other researchers.
6. In Figure 2(a), compared with other chromosomes, the fitting effect of chromosome 1 seems to be not good. Has the author ever thought about the reason? In addition, what is the guiding significance of this model in practical applications, such as online services, forecasting tools, or experiments? Can the author give relevant application examples in this regard?

Minor Comments:

1. Suggestions for Improving Figures: Figures 2 and 3: It is suggested that the differences between experimental data and model fitting data should be clearly marked by using more distinctive colors or symbols with different shapes in these figures, so as to help readers quickly distinguish between simulation results and experimental observation results. Density Plot in Figure 4: The current color gradient is dense, making it difficult to differentiate activation distributions for different origins. Consider using a broader color gradient or adding a slight separation between each origin's curve to improve readability.
2. Model Parameter Table: Adding a table in the Methods section or supplementary materials that summarizes the main model parameters (e.g., number of initiation factors, recycling rate, replication speed) and the basis for each parameter's setting would be helpful. This will allow readers to quickly understand the model setup and provide a reference for future researchers who may wish to use or adjust this model.
3. Citation and Description of Experimental Data: Clarify the origin and characteristics of the experimental data used, such as the specific details of the replication timing dataset applied for model fitting, and indicate whether the data represents single-cell or population-averaged measurements. This information will help readers better understand the comparison between the model and actual data.
4. Background and References: In the Introduction, consider adding a brief explanation of "Beacon Calculus" to aid non-specialist readers in understanding the novelty and applicability of this method. Adding foundational references for Beacon Calculus would further help readers appreciate the advantages of this approach. Additionally, in the discussion of the model's suitability for other biological systems, citing some reviews on high-efficiency replication origin analyses would help demonstrate the model's broader applicability.

Significance

1. Significance of the Research:

This study advances our understanding of DNA replication timing regulation in S. cerevisiae and presents a mathematical modeling approach with theoretical importance. By reconstructing a DNA replication timing framework for yeast, the model also provides a foundation that could be adapted for other systems, potentially advancing modeling techniques in genome replication research. 2. Relation to Existing Literature:

This study builds upon prior research on S. cerevisiae DNA replication initiation and proposes a simplified, reproducible model. Compared to more complex mathematical models or large-scale data analyses, this approach is more interpretable and easier to reproduce. The study's predictions on initiation factor concentration effects provide another perspective for future experimental work. 3. Target Audience:

This work will influence researchers studying DNA replication regulation, yeast genomics, and bioinformatics modeling. Additionally, scholars in microbiology and genetics may also benefit from the innovative modeling methods introduced. 4. Reviewer Expertise:

My expertise includes computational biology and bioinformatics, with a professional knowledge in DNA replication origins and bioinformatics modeling.

Author response:

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

Public Reviews:

Reviewer #1:

The entire study is based on only 2 adult animals, that were used for both the single cell dataset and the HCR. Additionally, the animals were caught from the ocean preventing information about their age or their life history. This makes the n extremely small and reduces the confidence of the conclusions. 

This statement is incorrect.  While the scRNAseq was indeed performed in two animals (n=2), the HCR-FISH was performed in 3-5 animals (depending on the probe used).  These were different animals from those used for the scRNAseq.  The number of animals used has now been included in the manuscript.

All the fluorescent pictures present in this manuscript present red nuclei and green signals being not color-blind friendly. Additionally, many of the images lack sufficient quality to determine if the signal is real. Additional images of a control animal (not eviscerated) and of a negative control would help data interpretation. Finally, in many occasions a zoomed out image would help the reader to provide context and have a better understanding of where the signal is localized. 

Fluorescent photos have been changed to color-blind friendly colors.  Diagrams, arrows and new photos have been included as to guide readers to the signal or labeling in cells. Controls for HCR-FISH and labeling in normal intestines have been included.  

Reviewer #2:

The spatial context of the RNA localization images is not well represented, making it difficult to understand how the schematic model was generated from the data. In addition, multiple strong statements in the conclusion should be better justified and connected to the data provided.

As explained above we have made an effort to provide a better understanding of the cellular/tissue localization of the labeled cells. Similarly, we have revised the conclusions so that the statements made are well justified.

Reviewer #3:

Possible theoretical advances regarding lineage trajectories of cells during sea cucumber gut regeneration, but the claims that can be made with this data alone are still predictive.

We are conscious that the results from these lineage trajectories are still predictive and have emphasized this in the text. Nonetheless, they are important part of our analyses that provide the theoretical basis for future experiments.

Better microscopy is needed for many figures to be convincing. Some minor additions to the figures will help readers understand the data more clearly.

As explained above we have made an effort to provide a better understanding of the cellular/tissue localization of the labeled cells.  

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

-  Page 4, line 70-81: if the reader is not familiar with holothurian anatomy and regeneration process, this section can be complicated to fully understand. An illustration, together with clear definitions of mesothelium, coelomic epithelium, celothelium and luminal cells would help the reader. 

A figure (now Figure 1) detailing the holothurian anatomy of normal and regenerating animals has been added. A figure detailing the intestinal regeneration process has also been included (S1).

-  Page 5 line 92-104: this paragraph could be shortened. It would be more important to explain what the main question is the Authors would like to answer and why single cell would be the best technique to answer it, than listing previous studies that used scRNA-Seq. 

The paragraph has been shortened and the focus has been shifted to the question of cellular components of regenerative tissues in holothurians.

-  Page 6, line 125-127 and line 129-132: this belongs to the method section. 

This information is now provided in the Materials and Methods section.

-  Page 11, line 210-217: this belongs to the discussion. 

This section has now been included in the Discussion.

-  How many mesenteries are present in one animal? 

This has now been included as part of Figure S1.

-  In the methods there are no information about the quality of the dataset and the sequencing and the difference between the 2 samples coming from the 2 animals. How many cells from each sample and which is the coverage? The Authors provided this info only between mesentery and anlage but not between animals. 

We have added additional information about the sequencing statistics in S4 Fig and S15 Table. Description has also been added in the methods in lines 922-926 under Single Cell RNA Sequencing and Data Analysis section.

-  The result section "An in-depth analysis of the various cluster..." is particularly long and very repetitive. I would encourage to Authors to remove a lot of the details (list of genes and GO terms) that can be found in the figures and stressed only the most important elements that they will need to support their conclusions. Having full and abbreviated gene names and the long list of references makes the text difficult to read and it is challenging to identify the main point that the Authors are trying to highlight. 

This section has been abbreviated.

-  Figure 1: I would suggest adding a graph of holothurian anatomy before and after the evisceration to provide more context of the process we are looking at and remove 1C. 

Information on the holothurian anatomy has been included in a new Fig 1 and in supplementary figure S1

-  Figure 2: I would suggest removing this figure that is redundant with Figure 3 and several genes are not cluster specific. Figure 3 is doing a better job in showing similar concepts. 

Figure 2 was removed and placed in the Supplement section. 

- In figure 3 how were the 3 cell types defined? Was this done manually or through a bioinformatic analysis? 

The cell definition was done following the analysis of the highly expressed transcripts and comparisons to what has been shown in the scientific literature.

-  Figure 2O shows that one of the supra-cluster is made of C2, C7, C6 and C10. This contradicts the text page 9, line 195. 

The transcript chosen for this figure gives the wrong idea that these 4 clusters are similar. We have now addressed this in the manuscript.

-  Figure 4A and 4C: if these are representing a subset of Figure 3, they should be removed in one or the other. The same comment is valid also for Figures 5, 6 and 7. In general the manuscript is very redundant both in terms of Figures and text. 

These are indeed subsets of Fig 3 that were added with the purpose of clarifying the findings, however, in view of the reviewer’s comment we have deleted the redundant information from all figures.

-  Figure 9: since the panels are not in order, it is difficult to follow the flow of the figure.  - All UMAP should have the number of the cluster on the UMAP itself instead of counting only on the color code in order to be color-blind friendly. 

The figure has been modified and clusters are now identified in the UMAP by their number.

-  Figure S1F seems acquired in very different conditions compared to the other images in the same figure. 

Fig S1F (now S2 Fig) is an overlay of fluorescent immune-histochemistry (UV light detected) with “classical” toluidine blue labeling (visible light detected).  This has now been explained in the figure legend.

-  Table S7 is lacking some product numbers. 

The toluidine blue product number has now been added to the table.  The antibodies that lack product number correspond to antibodies generated in our lab  and described in the references provided.

-  The discussion is pretty long and partially redundant with the result section. I would encourage the Authors to shorten the text and shorten paragraphs that have repeating information.  - It might be out of the scope of the Authors but the readers would benefit from having a manuscript that focuses more on the novel aspects discovered with the single-cell RNA-Seq and then have a review that will bring together all the literature published on this topic and integrating the single-cell data with everything that is known so far. 

We have tried to shorten the discussion by eliminating redundant text.

Reviewer #2 (Recommendations For The Authors): 

-  An intriguing finding is the lack of significant difference in the cell clusters between the anlage and mesentery during regeneration. This discovery raises important questions about the regenerative process. The authors should provide a more detailed explanation of the implications of this finding. For example, does it suggest that both organs contribute equally to the regenerated tissues? 

The lack of significant differences in the cell clusters between the anlage and the mesentery is somewhat surprising but can be explained by two different facts. First, we have previously shown that many of the cellular processes that take place in the anlage, including cell proliferation, apoptosis, dedifferentiation and ECM remodeling occur in a gradient that begins at the tip of the mesentery where the anlage forms and extends to various degrees into the mesentery.  Similarly, migrating cells move along the connective tissue of the mesentery to the anlage.  Thus, there is no clear partition of the two regions that would account for distinct cell populations associated with the regenerative stage.  Second, the two cell populations that would have been found in the mesentery but not in the regenerating anlage, mature muscle and neurons, were not dissociated by our experimental protocol as to allow for their sequencing.  Our current experiments are being done using single nuclei RNA sequencing to overcome this hurdle. This has now been included in the discussion.

-  Proliferating cells are obviously important to the study of regeneration as it is assumed these form the regenerating tissue. The authors describe cluster 8 as the proliferative cells. Is there evidence of proliferation in other cell types or are these truly the only dividing cells? Is c8 of multiple cell types but the clustering algorithm picks up on the markers of cell division i.e. what happens if you mask cell division markers - does this cluster collapse into other cluster types? This is important as if there is only one truly proliferating cell type then this may be the origin of the regenerative tissues and is important for this study to know this. 

As the reviewer highlights, we also believe this to be an important aspect to discuss. We have addressed this in the manuscript discussion with the following: “Our data suggest that there appears to be a specific population of only proliferative cells (C8) characterized by a large number of cell proliferation genes, which can be visualized by the top genes shown in Fig 3. These cell proliferation genes are specific to C8, with minimum representation in other populations. Interestingly, as mentioned before C8 expresses at lower levels many of the genes of other coelomic epithelium populations. Nevertheless, even if we mask the top 38 proliferation genes (not shown), this cluster is maintained as an independent cluster, suggesting that its identity is conferred by a complex transcriptomic profile rather than only a few proliferation-related genes. Therefore, the identity and potential role of C8 could be further described by two distinct alternatives: (1) cells of C8 could be an intermediate state between the anlage precursor cells (discussed below) and the specialized cell populations or (2) cells of C8 are the source of the anlage precursor populations from which all other populations arise. The pseudotime data is certainly complex and challenging to interpret with our current dataset, yet the RNA velocity analysis showed in Fig 11B would suggests that cells of C8 transition into the anlage precursor populations, rather than being an intermediate state. This is also supported by the Slingshot pseudotime analysis that incorporates C8 (S13 Fig).

Nevertheless, additional experiments are needed to confirm this hypothesis.”

-  The schematic model presented in Fig 10 is essential for clarifying the paper's findings and will provide a crucial baseline model for future research. However, the comparison of the data shown in the HCR figures with the schematic is challenging due to the lack of spatial context in the HCR figures. The authors should find a way to provide better context in the figures, such as providing two-color in situ images to compare spatial relationships of cell types and/or including lower resolution and side-by-side fluorescent and bright field images if possible. 

The figure has been modified to explain the spatial arrangement of the tissues.

The authors make several strong statements in the discussion that weren't well connected to the findings in the data. Specifically: 

“Regardless of which cell population is responsible for giving rise to the cells of the regenerating intestine, our study reveals that the coelomic epithelium, as a tissue layer, is pluripotent.” 

This has now been expanded to better explain the statement.

738 “…we postulate that cells from C1 stand as the precursor cell population from which the rest of the cells in the coelomic epithelium arise”. 

This has now been expanded to better explain the statement.

748 “differentiation: muscle, neuroepithelium, and coelomic epithelium cells. We also propose the presence of undifferentiated and proliferating cell populations in the coelomic epithelia, which give rise to the cells in this layer…”

This has now been expanded to better explain the statement.

777 “amphibians, the cells of the holothurian anlage coelomic epithelium are proliferative undifferentiated cells and originated via a dedifferentiation process…”

This has now been expanded to better explain the statement.

Reviewer #3 (Recommendations For The Authors): 

Specific questions: 

- Is there any way to systematically compare these cells to evolutionarily-diverged cells in distant relatives to sea cucumbers? Or even on a case-by-case basis? For example, is there evidence for any of these transitory cell types to have correlate(s) in vertebrate gut regeneration? 

This is a most interesting question but one that is perhaps a bit premature to answer due to multiple reasons.  First, most of the studies in vertebrates focus on the regeneration of the luminal epithelium, a layer that we are not studying in our system since it appears later in the regeneration process.  Second, there is still too little data from adult echinoderms to fully comprehend which cells are cell orthologues to vertebrates. Third, we are only analyzing one regenerative stage.  It is our hope that this is just the start of a full description of what cell types/stages are found and how they function in regeneration and that this will lead us to identify the cellular orthologues among animal species.

Major revisions: 

- If lineage tracing is within the scope of this paper, it would provide more definitive evidence to the conclusions made about the precursor populations of the regenerating anlage. 

Response:  This is certainly one of the next steps, however at present, it is not possible due to technical limitations.

Minor revisions: 

- Line 47: "for decades" even longer! Could the authors also cite some other amphibians, such as other salamanders (newts) and larval frogs? 

References have been added.

- Line 85: "specially"-could authors potentially change to "specifically" 

Corrected

- Line 122: Authors should add the full words of what these abbreviations stand for in the caption for Figure 1 or in Figure 1A itself. 

Corrected

- Lines 153: What conclusions are the authors trying to make from one type of tubulin presence compared to the others? It's unclear from the text. 

The authors are not trying to reach any particular conclusion.  They are just stating what was found using several markers, and the possibility that what might be viewed first hand as a single cell population might be more heterogenous.  Although the tubulin-type information might not be relevant for the conclusions in the present manuscript, it might be important for future work on the cell types involved in the regeneration process.

- Line 226: Could the authors clarify if "WNT9" is "WNT9a". Figure 3 lists WNT9a but authors refer to WNT9 in the text. 

The gene names in Fig 3 are based on the human identifiers. H. glaberrima only has one sequence of Wnt9 (Auger et al. 2023) and this sequence shares the highest similarity to human Wnt9a, thus the name in the list. We have now identified the gene as Wnt9 to avoid confusion.

- Lines 236-237: Can authors rule out that some immune cells might infiltrate the mesenchymal population? 

No, this cannot be ruled out.  In fact, we believe that most of the immune cells found in our scRNA-seq are indeed cells that have infiltrated the anlage and are part of the mesenchyma.  This has been reported by us previously (see Garcia-Arraras et al. 2006). We have now included this in the text.

- Line 452-453: The over-representation of ribosomal genes not shown. Would it be possible to show this information in the supplementary figures? 

The sentence has been modified, the data is being prepared as part of a separate publication that focuses on the ribosomal genes.

- Line 480: Could authors clarify if it's WNT9a or just WNT9?

It is indeed Wnt9. See previous response above.

- Line 500: In future experiments, it would be interesting to compare to populations at different timepoints in order see how the populations are changing or if certain precursors are activated at different times. 

We fully agree with the reviewer. These are ongoing experiments or are part of new grant proposals.

- Line 567-568: Choosing 9-dpe allowed for 13 clusters, but do authors expect a different number of clusters at different timepoints as things become more terminally differentiated? 

Definitely, we believe that clusters related to the different regenerative stages of cells can be found by looking at earlier or later regeneration stages of the organ.  A clear example is that if the experiment is done at 14-dpe, when the lumen is forming, cells related to luminal epithelium populations will appear. It is also possible that different immune cells will be associated with the different regeneration stages.

- Line 653: References Figure 10D (not in this manuscript). Are authors referring to only 1D or 9D or an old draft figure number? 

As the reviewer correctly points out, this was a mistake where the reference is to a previous draft. It has now been corrected.

- Line 701: "our study reveals that the coelomic epithelium, as a tissue layer, is pluripotent." Phrasing may be better as referring to the cell population making up the tissue layer as pluripotent/multipotent or that the cells it contains would likely be pluripotent or multipotent. Additionally, lineage tracing may be needed to definitively demonstrate this. 

This has been modified.

- Line 808: The authors may make a more accurate conclusion by saying that the characteristics are similar to blastemas or behave like a blastema rather than it is blastema. There is ambiguity about the meaning of this term in the field, but most researchers seem to currently have in mind that the "blastema" definition includes a discrete spatial organization of cells, and here these cells are much more spread out. This could be a good opportunity for the authors to engage in this dialogue, perhaps parsing out the nuances of what a "blastema" is, what the term has traditionally referred to, and how we might consider updating this term or at least re-framing the terminology to be inclusive of functions that "blastemas" have traditionally had in the literature and how they may be dispersed over geographical space in an organism more so than the more rigid, geographically-restricted definition many researchers have in mind. However, if the authors choose to elaborate on these issues, those elaborations do belong in the discussion, and the more provisional terminology we mention here could be used throughout the paper until that element of the revised discussion is presented. We would welcome the authors to do this as a way to point the field in this direction as this is also how we view the matter. For example, some of the genes whose expression has been observed to be enriched following removal of brain tissue in axolotls (such as kazald2, Lust et al.), are also upregulated in traditional blastemas, for instance, in the limb, but we appreciate that the expression domain may not be as localized as in a limb blastema. Additionally, since there is now evidence that some aspects of progenitor cell activation even in limb regeneration extend far beyond the local site of amputation injury (Johnson et al., Payzin-Dogru et al.), there is an opportunity to connect the dots and make the claim that there could be more dispersion of "blastema function" than previously appreciated in the field. Diving a bit more into these nuances may also enable better conceptual framework of how blastema function may evolve across vast evolutionary time and between different injury contexts in super-regenerative organisms. 

We have followed the reviewer’s suggestion and stated that the holothurian anlage behaves as a blastema. Though we would love to elaborate on the blastema topic, as suggested by the reviewer, we believe that it would extend the discussion too much and that the topic might be better served in a different publication.

- In the discussion, it would be important not to leave the reader with the impression that all amphibian blastema cells originate via dedifferentiation. This is not the case. For example, in axolotls (Sandoval-Guzman et al.) and in larval/juvenile newts, muscle progenitors within the blastema structure have been shown to originate from muscle satellite cells, a kind of stem cell, in stump tissues (while adult newts use dedifferentiation of myofibers to generate muscle progenitors in the blastema). Most cell lineages simply have not been evaluated in the level of detail that would be required to definitively conclude one way or the other, and the door is open for a more substantial contribution from stem cell populations than previously appreciated especially because new tools exist to detect and study them. Providing the reader with a more nuanced view of this situation will not negatively impact the findings in this paper, but it will show that there is biological complexity still waiting to be discovered and that we don't have all the answers at this point. 

This has now been corrected. 

Figures: Overall, the figures need minor work. 

- Figure 1A: Can the authors draw a smaller, full-body cartoon and feature the current high-mag cartoon as an inset to that? Can they label the axes and make it clear how the geometry works here?

Fig 1 has been re-done and now is split into Fig 1 and Fig 2.

- Figure 1B: Can the authors label the UMAP with cluster identities on the map itself? This will make it easier to identify each cluster (especially to make sure cluster 11 is easier to find). 

This has been corrected.

- Figure 2: Could the authors put boxes/clearly distinguish panel labels around each cluster (AO), so that there are clear boundaries? 

Fig 2 has been moved to Supplement, following another reviewer recommendation.

- "Gene identifiers starting with "g" correspond to uncharacterized gene models of H. glaberrima." - The sentence is from another figure caption but this figure would benefit from having this sentence in the figure caption as well. 

This has been added to other figures as suggested.

- Figure 3A: Can the authors potentially bold, highlight, or underline genes you discuss in text, so it's easier for the reader to reference? 

This has been added as suggested.

- Figure 3C: Can the authors please label the cell types directly on the UMAP here as well? 

The changes were made following the reviewer’s recommendation.

- Figure 4D-E: There's not much context here to determine if this HCR-FISH validation can tell us anything about these cells besides some of them appear to be there. Do authors expect the coelomocyte morphology to look different in regenerating/injured tissue versus normal animals? Can the authors provide some double in situs, as well as some lower-magnification views showing where the higher-magnification insets are located? Is there any spatial pattern to where these cells are found? Counter stains would be helpful. 

- Figure 6C: If clusters C5, C8, C9 are part of the coelomic epithelium, then authors could show a smaller diagram above with blue and grey to show types and then show clusters separately to help get their point across better. 

- Figure 6G: This image appears to have high background- would it be possible for authors to repeat phalloidin stain or reimage with a lower exposure/gain. Additionally, imaging with Zstacks would help to obtain maximum intensity projections. It would greatly aid the reader if each image was labeled with HCR probes/antibodies that have been applied to the sample. 

- Figure 7E: The cells appear to be out of focus and have high background. Additionally, they are lacking the speckled appearance expected to be seen with HCR-FISH. Would it be possible for authors to collect another image utilizing z-stacks? 

HCR-FISH figures identifying the gene expression characteristic of cell clusters have been modified following the reviewer’s concerns.  The changes include:

(1) Additional clusters have been verified with probes to gene identifiers. These include clusters 8, 9 and 12.

(2) Redundant information has been removed.

(3) Colors have been changed to make figures friendlier to color-impaired readers.

(4) Spatial context has been added or identified.

(5) In some cases, improved photos have been added

(6) Better labels have been included

(7) When necessary individual photos used for the overlay have been included.

- Figure 9A: Could authors add cluster labels onto UMAP directly? 

This change was made to Fig 2A. UMAP in Fig 9A is the same and used just as reference of the subset.

- Figure 10: It could be useful if authors put a small map of the sea cucumber like in other images so that readers know where in the anlage this zoomed in model represents. 

Added as suggested by the reviewer.

- Supplementary figure 1F: Could authors add an arrow to the dark cell that's being pointed out? 

Changed made as suggested by the reviewer.

- Supplementary figure 1: Could authors label clearly what color is labeled with what marker? 

Changed made as suggested by the reviewer.