On 2017 May 29, Michael Goard commented:

We thank the Janelia Neural Circuit Computation Journal Club for taking the time to review our paper. However, we wish to clarify a few of the points brought up in the review.

1) Off-target effects of inactivation. The authors of the review correctly point out that off-target effects can spread laterally from an inactivated region, potentially complicating the interpretation of the V1/PPC inactivation experiments. We have since carried out (not yet published) electrophysiology experiments in V1 during PPC photoinactivation and find there is some suppression (though not silencing) of V1 excitatory neurons through polysynaptic effects. The suppression is moderate and the V1 neurons maintain stimulus selectivity, so it is unlikely off-target suppression in V1 is responsible for the PPC inactivation effects, but the results do need to be interpreted with some caution.

Notably, the suppression effect is not distance-dependent; it instead appears heterogeneous and is likely dependent on connectivity, as has been recently demonstrated in other preparations (Otchy et al, Nature, 2015). Given these findings, describing off-target effects as a simple function of power and distance is likely misleading. Indeed, even focal cortical silencing is likely to have complex effects on subcortical structures in addition to the targeted region. Instead, we suggest that while photoinactivation experiments are still useful for investigating the role of a region in behavior, the results need to be interpreted carefully (e.g., as demonstrating an area as permissive rather than instructive; per Otchy et al., 2015).

2) Silencing of ALM in addition to M2. The photoinactivation experiments were designed to discriminate between sensory, parietal, and motor contributions to the task, rather than specific regions within motor cortex. We did not intend to suggest that ALM was unaffected in our photoinactivation experiments (this is the principal reason we used the agnostic term “fMC” rather than referring to a specific region). Although the center of our window was located posterior and medial to ALM, we used a relatively large window (2 x 2.5 mm), so ALM was likely affected.

3) Rebound activity contributing to fMC photoinactivation effects. Rebound effects are not likely to be responsible for the role of fMC during the stimulus epoch. First, our photostimulus did not cause consistent rebound excitation (e.g., Figure 8B). This is likely due to the use of continuous rather than pulsed photoinactivation (see Figure 1G in Zhao et al., Nat Methods, 2011). Second, we did run several inactivation experiments with a 100-200 ms offset ramp (as in Guo et al., 2014), and found identical results (we did not include these experiments in the publication since we did not observe rebound activity). We suspect the discrepancy with Guo et al. is due to the unilateral vs. bilateral photoinactivation (Li, Daie, et al., 2016), as the reviewers suggest.

On 2017 Apr 07, Janelia Neural Circuit Computation Journal Club commented:

Highlight/Summary

This is one of several recent papers investigating cortical dynamics during head-restrained behaviors in mice using mostly imaging methods. The questions posed were:

Which brain regions are responsible for sensorimotor transformation? Which region(s) are responsible for maintaining task-relevant information in the delay period between the stimulus and response?

These questions were definitely not answered. However, the study contains some nice cellular calcium imaging in multiple brain regions in a new type of mouse behavior.

The behavior is a Go / No Go behavioral paradigm. The S+ and S- stimuli were drifting horizontal and vertical gratings, respectively. The mouse had to withhold licking during a delay epoch. During a subsequent response epoch the mouse responded by licking for a reward on Go trials.

Strengths

Perhaps the greatest strength of the paper is that activity was probed in multiple regions in the same behavior (all L2/3 neurons, using two-photon calcium imaging). Activity was measured in primary visual cortex (V1), ‘posterior parietal cortex’ (PPC; 2 mm posterior, 1.7 mm lateral), and fMC. ‘fMC' overlaps sMO in the Allen Reference Atlas, posterior and medial to ALM (distance approximately 1 mm) (Li/Daie et al 2016). This location is analogous to rat 'frontal orienting field’ (Erlich et al 2011) or M2 (Murakami et al 2014). Folks who work on whiskers refer to this area as vibrissal M1, because it corresponds to the part of motor cortex with the lowest threshold for whisker movements.

In V1, a large fraction (> 50 %) of neurons were active and selective during the sample epoch. One of the more interesting findings is that a substantial fraction of V1 neurons were suppressed during the delay epoch. This could be a mechanism to reduce ‘sensory gain’ and ’distractions' during movement preparation. Interestingly, PPC neurons were task-selective during the sample or response epochs; consistent with previous work in primates (many studies in parietal areas) and rats (Raposo et al 2014), individual neurons multiplexed sensory and movement selectivity. However, there was little activity / selectivity during the delay epoch. This suggests that their sequence-like dynamics in maze tasks (e.g. Harvey et al 2012) might reflect ongoing sensory input and movement in the maze tasks, rather than more cognitive variables. fMC neurons were active and selective during the delay and response epoch, consistent with a role in movement planning and motor control, again consistent with many prior studies in primates, rats (Erlich et al 2011), and mice (Guo/Li et al 2014).

Weaknesses

Delayed response or movement tasks have been used for more than forty years to study memory-guided movements and motor preparation. Typically different stimuli predict different movement directions (e.g. saccades, arm movements or lick directions). Previous experiments have shown that activity during the delay epoch predicts specific movements, long before the movement. In this study, Go and No Go trials are fundamentally asymmetric and it is unclear how this behavioral paradigm relates to the literature on movement preparation. What does selectivity during the delay epoch mean? On No Go trials a smart mouse would simply ignore the events post stimulus presentation, making delay activity difficult to interpret.

The behavioral design also makes the interpretation of the inactivation experiments suspect. The paper includes an analysis of behavior with bilateral photoinhibition (Figure 9). The authors argue for several take-home messages (‘we were able to determine the necessity of sensory, association, and frontal motor cortical regions during each epoch (stimulus, delay, response) of a memory-guided task.'); all of these conclusions come with major caveats.

1.) Inactivation of both V1 and PPC during the sample epoch abolishes behavior, caused by an increase in false alarm rate and decrease in hit rate (Fig. 9d). The problem is that the optogenetic protocol silenced a large fraction of the brain. The methods are unlikely to have the spatial resolution to specifically inactivate V1 vs PPC. The authors evenly illuminated a 2 mm diameter window with 6.5mW/mm² light in VGat-ChR2 mice. This amounts to 20 mW laser power. According to the calibrations performed by Guo / Li et al (2014) in the same type of transgenic mice, this predicts substantial silencing over a radius (!) of 2-3 mm (Guo / Li et al 2014; Figure 2). Photoinhibiting V1 will therefore silence PPC and vice versa. It is therefore expected that silencing V1 and PPC have similar behavioral effects.

2.) Silencing during the response window abolished the behavioral response (licking). Other labs labs have also observed total suppression of voluntary licking with frontal bilateral inactivation (e.g. Komiyama et al 2010; and unpublished). However, the proximal cause of the behavioral effect is likely silencing of ALM, which is more anterior and lateral to ‘fMC’. ALM projects to premotor structures related to licking. Low intensity activation of ALM, but not more medial and posterior structures such as fMC, triggers rhythmic licking (Li et al 2015) The large photostimulus used here would have silenced ALM as well as fMC.

3.) Somewhat surprisingly, behavior is perturbed after silencing fMC during the sample (stimulus) and delay epochs. In Guo / i et al 2014, unilateral silencing of frontal cortex during the sample epoch (in this case ALM during a tactile decision task, 2AFC type) did not cause a behavioral effect (although bilateral silencing is likely different; see Li / Daie et al 2016). The behavioral effect in Goard et al 2016 may not be caused by the silencing itself, but by the subsequent rebound activity (an overshoot after silencing; see for example Guo JZ et al eLife 2016; Figure 4—figure supplement 2). Rebound activity is difficult to avoid, but can be minimized by gradually ramping down the photostimulus, a strategy that was not used here. The key indication that rebound was a problem is that behavior degrades almost exclusively via an increase in false alarm rate -- in other words - mice now always lick independent of trial type. Increased activity in ‘fMC’, as expected with rebound, is expected to promote these false alarms. More experiments are needed to make the inactivation experiments solid.

On 2017 Apr 07, Janelia Neural Circuit Computation Journal Club commented:

Highlight/Summary

This is one of several recent papers investigating cortical dynamics during head-restrained behaviors in mice using mostly imaging methods. The questions posed were:

Which brain regions are responsible for sensorimotor transformation? Which region(s) are responsible for maintaining task-relevant information in the delay period between the stimulus and response?

These questions were definitely not answered. However, the study contains some nice cellular calcium imaging in multiple brain regions in a new type of mouse behavior.

The behavior is a Go / No Go behavioral paradigm. The S+ and S- stimuli were drifting horizontal and vertical gratings, respectively. The mouse had to withhold licking during a delay epoch. During a subsequent response epoch the mouse responded by licking for a reward on Go trials.

Strengths

Perhaps the greatest strength of the paper is that activity was probed in multiple regions in the same behavior (all L2/3 neurons, using two-photon calcium imaging). Activity was measured in primary visual cortex (V1), ‘posterior parietal cortex’ (PPC; 2 mm posterior, 1.7 mm lateral), and fMC. ‘fMC' overlaps sMO in the Allen Reference Atlas, posterior and medial to ALM (distance approximately 1 mm) (Li/Daie et al 2016). This location is analogous to rat 'frontal orienting field’ (Erlich et al 2011) or M2 (Murakami et al 2014). Folks who work on whiskers refer to this area as vibrissal M1, because it corresponds to the part of motor cortex with the lowest threshold for whisker movements.

In V1, a large fraction (> 50 %) of neurons were active and selective during the sample epoch. One of the more interesting findings is that a substantial fraction of V1 neurons were suppressed during the delay epoch. This could be a mechanism to reduce ‘sensory gain’ and ’distractions' during movement preparation. Interestingly, PPC neurons were task-selective during the sample or response epochs; consistent with previous work in primates (many studies in parietal areas) and rats (Raposo et al 2014), individual neurons multiplexed sensory and movement selectivity. However, there was little activity / selectivity during the delay epoch. This suggests that their sequence-like dynamics in maze tasks (e.g. Harvey et al 2012) might reflect ongoing sensory input and movement in the maze tasks, rather than more cognitive variables. fMC neurons were active and selective during the delay and response epoch, consistent with a role in movement planning and motor control, again consistent with many prior studies in primates, rats (Erlich et al 2011), and mice (Guo/Li et al 2014).

Weaknesses

Delayed response or movement tasks have been used for more than forty years to study memory-guided movements and motor preparation. Typically different stimuli predict different movement directions (e.g. saccades, arm movements or lick directions). Previous experiments have shown that activity during the delay epoch predicts specific movements, long before the movement. In this study, Go and No Go trials are fundamentally asymmetric and it is unclear how this behavioral paradigm relates to the literature on movement preparation. What does selectivity during the delay epoch mean? On No Go trials a smart mouse would simply ignore the events post stimulus presentation, making delay activity difficult to interpret.

The behavioral design also makes the interpretation of the inactivation experiments suspect. The paper includes an analysis of behavior with bilateral photoinhibition (Figure 9). The authors argue for several take-home messages (‘we were able to determine the necessity of sensory, association, and frontal motor cortical regions during each epoch (stimulus, delay, response) of a memory-guided task.'); all of these conclusions come with major caveats.

1.) Inactivation of both V1 and PPC during the sample epoch abolishes behavior, caused by an increase in false alarm rate and decrease in hit rate (Fig. 9d). The problem is that the optogenetic protocol silenced a large fraction of the brain. The methods are unlikely to have the spatial resolution to specifically inactivate V1 vs PPC. The authors evenly illuminated a 2 mm diameter window with 6.5mW/mm² light in VGat-ChR2 mice. This amounts to 20 mW laser power. According to the calibrations performed by Guo / Li et al (2014) in the same type of transgenic mice, this predicts substantial silencing over a radius (!) of 2-3 mm (Guo / Li et al 2014; Figure 2). Photoinhibiting V1 will therefore silence PPC and vice versa. It is therefore expected that silencing V1 and PPC have similar behavioral effects.

2.) Silencing during the response window abolished the behavioral response (licking). Other labs labs have also observed total suppression of voluntary licking with frontal bilateral inactivation (e.g. Komiyama et al 2010; and unpublished). However, the proximal cause of the behavioral effect is likely silencing of ALM, which is more anterior and lateral to ‘fMC’. ALM projects to premotor structures related to licking. Low intensity activation of ALM, but not more medial and posterior structures such as fMC, triggers rhythmic licking (Li et al 2015) The large photostimulus used here would have silenced ALM as well as fMC.

3.) Somewhat surprisingly, behavior is perturbed after silencing fMC during the sample (stimulus) and delay epochs. In Guo / i et al 2014, unilateral silencing of frontal cortex during the sample epoch (in this case ALM during a tactile decision task, 2AFC type) did not cause a behavioral effect (although bilateral silencing is likely different; see Li / Daie et al 2016). The behavioral effect in Goard et al 2016 may not be caused by the silencing itself, but by the subsequent rebound activity (an overshoot after silencing; see for example Guo JZ et al eLife 2016; Figure 4—figure supplement 2). Rebound activity is difficult to avoid, but can be minimized by gradually ramping down the photostimulus, a strategy that was not used here. The key indication that rebound was a problem is that behavior degrades almost exclusively via an increase in false alarm rate -- in other words - mice now always lick independent of trial type. Increased activity in ‘fMC’, as expected with rebound, is expected to promote these false alarms. More experiments are needed to make the inactivation experiments solid.

On 2017 May 29, Michael Goard commented:

We thank the Janelia Neural Circuit Computation Journal Club for taking the time to review our paper. However, we wish to clarify a few of the points brought up in the review.

1) Off-target effects of inactivation. The authors of the review correctly point out that off-target effects can spread laterally from an inactivated region, potentially complicating the interpretation of the V1/PPC inactivation experiments. We have since carried out (not yet published) electrophysiology experiments in V1 during PPC photoinactivation and find there is some suppression (though not silencing) of V1 excitatory neurons through polysynaptic effects. The suppression is moderate and the V1 neurons maintain stimulus selectivity, so it is unlikely off-target suppression in V1 is responsible for the PPC inactivation effects, but the results do need to be interpreted with some caution.

Notably, the suppression effect is not distance-dependent; it instead appears heterogeneous and is likely dependent on connectivity, as has been recently demonstrated in other preparations (Otchy et al, Nature, 2015). Given these findings, describing off-target effects as a simple function of power and distance is likely misleading. Indeed, even focal cortical silencing is likely to have complex effects on subcortical structures in addition to the targeted region. Instead, we suggest that while photoinactivation experiments are still useful for investigating the role of a region in behavior, the results need to be interpreted carefully (e.g., as demonstrating an area as permissive rather than instructive; per Otchy et al., 2015).

2) Silencing of ALM in addition to M2. The photoinactivation experiments were designed to discriminate between sensory, parietal, and motor contributions to the task, rather than specific regions within motor cortex. We did not intend to suggest that ALM was unaffected in our photoinactivation experiments (this is the principal reason we used the agnostic term “fMC” rather than referring to a specific region). Although the center of our window was located posterior and medial to ALM, we used a relatively large window (2 x 2.5 mm), so ALM was likely affected.

3) Rebound activity contributing to fMC photoinactivation effects. Rebound effects are not likely to be responsible for the role of fMC during the stimulus epoch. First, our photostimulus did not cause consistent rebound excitation (e.g., Figure 8B). This is likely due to the use of continuous rather than pulsed photoinactivation (see Figure 1G in Zhao et al., Nat Methods, 2011). Second, we did run several inactivation experiments with a 100-200 ms offset ramp (as in Guo et al., 2014), and found identical results (we did not include these experiments in the publication since we did not observe rebound activity). We suspect the discrepancy with Guo et al. is due to the unilateral vs. bilateral photoinactivation (Li, Daie, et al., 2016), as the reviewers suggest.