Layer-specific integration of locomotion and sensory information in mouse barrel cortex.

During navigation, rodents continually sample the environment with their whiskers. How locomotion modulates neuronal activity in somatosensory cortex, and how it is integrated with whisker-touch remains unclear. Here, we compared neuronal activity in layer 2/3 (L2/3) and L5 of barrel cortex using calcium imaging in mice running in a tactile virtual reality. Both layers increase their activity during running and concomitant whisking, in the absence of touch. Fewer neurons are modulated by whisking alone. Whereas L5 neurons respond transiently to wall-touch during running, L2/3 neurons show sustained activity. Consistently, neurons encoding running-with-touch are more abundant in L2/3 and they encode the run-speed better during touch. Few neurons across layers were also sensitive to abrupt perturbations of tactile flow during running. In summary, locomotion significantly enhances barrel cortex activity across layers with L5 neurons mainly reporting changes in touch conditions and L2/3 neurons continually integrating tactile stimuli with running.

Clarifications 1) P3. The general statement "whisking behavior has been the main focus of studies on sensorimotor integration in vibrissal primary sensory cortex (S1 or 'barrel cortex')10-19 should also include the early work "Phase-to-rate transformations encode touch in cortical neurons of a scanning sensorimotor system. J. C. Curtis and D. Kleinfeld, Nature Neuroscience (2009)" and review "Neuronal basis for object location in the vibrissa scanning sensorimotor system. D. Kleinfeld and M. Deschênes. Neuron (2011)" on this topic.
2) P3) Introduction. The statement "... whisking was found ... to increase thalamic activity should include reference to the works "Vibrissa self-motion and touch are reliably encoded along the same somatosensory pathway from brainstem through thalamus J. D. Moore, N. Mercer Lindsay, M. Deschênes and D. Kleinfeld, Public Library of Science: Biology (2015)" and "Parallel thalamic pathways for whisking and touch signals in the rat. C. Yu C, Derdikman, S. Haidarliu S and E. Ahissar E. PLoS Biology (2006)".
3) P8, "39% of L2/3 and 45% of L5 neurons were running speed modulated". On page 45 line 955 the number of neurons modulated by run speed is presented as "441/705 for L2/3 neurons and 193/233 L5 neurons", which is 63% and 83%. These two group numbers do not match. Please clarify the claims 4) Figure S3 shows three types of running neurons, i.e., MI, MD and BP, as above. Please further analysis and discussion locomotion and wall-touch integration by these different cell types. 5) Please further analyze and show the relationship between whisk angle and running speed. 6) Can the animal run only forward or also backward on the treadmill? If it can run backward, what is the nature of neuronal encoding od velocity? 7) P16, "Neurons integrating self-motion and touch are more …", here "self-motion" can be replaced by "locomotion". As "Self-motion" can also refer to the whisker's self sweeping during active sensing, which is not involved in this study. Figure 1 (c), please use "L5A" to replace "L5" according to the scale bar. 9) P7, the authors write "Whisking barely modulated the mean activity in L2/3 and L5". As they imaged down to 66 4μm depth, which is mainly L5A, it will be more appropriate to use "L5A" rather that "L5" here and throughout the manuscript.

Ayaz et al., NCOMMS-18-21652-T, point-by-point responses to reviewers' comments
We thank the reviewers for their critical comments, which have triggered further experiments and analysis that are now included in the revised manuscript (we have highlighted major changes in red text). We believe that these additions have improved the manuscript and that we have addressed all points raised by the reviewers. Below we list major changes and additions: 1) We now provide histological evidence of low expression of R-CaMP1.07 in inhibitory neurons ( Supplementary Fig. 1) 2) We have conducted additional experiments in brain slices that confirm similar action potential-evoked R-CaMP1.07 signals in L2/3 and L5 pyramidal neurons ( Supplementary Fig. 3)

3)
We now present touch onset responses for all cells to further highlight the transient character of L5 responses in comparison to sustained L2/3 responses ( Fig. 3d and Fig. 4b,e)

4)
We compared onset responses of all neurons to locomotion and touch, highlighting that transiency of L5 responses is limited to touch onset but not to locomotion onset ( Supplementary Fig. 7)

5)
We added a presentation of the differences of perturbation responses during running vs resting state (Fig  5e,f,g) Fig. 9)

9)
We tracked single whiskers using DeepLabCut and compare the contribution of various whisker parameters to calcium signal encoding ( Figure R1)

10)
We compare touch modulations and sensory-motor integration properties of different classes of runmodulated cells ( Figure R2)

11)
We present the relationship between run speed and whisking angle ( Figure R3)

Responses to Reviewer #2:
Ayaz et al report on modulation of neural activity in L2/3 and L5 S1 barrel cortex neurons during whisking, whisking + locomotion, and touch under both conditionsadditionally in an open-loop and closed-loop configuration with respect to locomotion. I have few complaints with the methodological approachthese minor points are addressed below. My major issue is the lack of substantive new insight provided by this study. The idea of locomotion-associated changes of neural activity has been well established in V1 and A1, and anecdotal reports exist in S1although the authors are correct this has not been carefully quantified.

Answer1.
We do consider several of our findings substantive new insights. Not only do we provide a more careful quantification of locomotion-modulation of S1 L2/3 activity, beyond previous anecdotal evidence, in addition our study goes beyond the state-of-the-art (also considering V1 and A1 studies) by exploring how different layers of the sensory areas alter their responses to sensory stimuli in the presence or absence of locomotion. We provide clear evidence for differential processing of sensory inputs in layer 2/3 and layer 5 and we particularly find that superficial neurons are more integrative-co-processing motor and sensory information-compared to deep layer neurons. To our knowledge these are novel findings that have not been described previously for similar behavioral conditions and will be relevant for the broad community interested in cortical function. In the revised manuscript we have edited the main text to better highlight these salient novel aspects and to better convey our ideas of their meaning regarding sensory coding (see below). We also adapted Figs. 3 and 4 to more clearly show the L2/3-L5 differences.
Their major finding is a statistically significant, although not particularly large difference in how L5 and L2/3 neurons 'encode' locomotion and touch under the various conditions.

A2.
We are not entirely sure what the reviewer is referring to. One of our major findings is that L2/3 neurons display a sustained response to continuous touch during running whereas L5 neurons respond only transiently after touch onset (Fig. 3). This difference is statistically highly significant. In the revised manuscript, we have now also added several additional data and amended the text to convince the reviewer that this difference is real (see below). A second major finding is that a higher fraction of L2/3 neurons compared to L5 neurons shows integrative features, i.e., highest activation when sensation is combined with motor behavior. Again, this result is statistically highly significant at p<0.01. Nonetheless, we have performed additional analyses to corroborate this finding with further evidence. We have compared mutual information between calcium signals and run speed in the absence and presence of texture touch. L2/3 neurons increased their information content in the presence of texture touch while the presence of texture touch decreased information content in L5 neurons (Supplementary Fig. 10d).This result indicates better coding of a behavioral variable by superficial neurons when two input streams are combined. In addition mutual information analysis led to distinct outcomes for different cell categories which were defined independently in Fig 6. The main failing of this study is to connect these results to something meaningful for sensory coding or behavior.

A3.
In our view, the results of our study prompt new conceptual ideas regarding sensory coding. First, they highlight the large modulation of neuronal population activity in a primary somatosensory area in the locomotion state and the additional modulation of activity if then sensory input arrives. Our study therefore emphasizes the necessity to further investigate sensory coding under naturalistic conditions, when the body is actively engaged in sensory sampling. The novel virtual tactile environment we present in our study should be a useful tool for further studies of this kind. Second, our results point to an interesting laminar difference in sensory coding that is contingent on behavioral state. Based on our finding of sustained versus transient touch responses in L2/3 and L5, respectively, and the stronger sensory-motor integrative features of L2/3 neurons, we speculate that the neuronal population in superficial layers 'stay online' during ongoing sensorymotor sampling-essentially continually monitoring the world and presumably matching it with continuous expectations-while neurons in deeper layers, with their connections to subcortical nuclei such as thalamus, striatum, and brain stem, may react to salient, unexpected events in order to convey this information to relevant subcortical regions, in the end to adapt the animal's behavior. While in the old Discussion we had listed multiple possible mechanisms of how differential laminar processing might be implemented, we perhaps failed to clearly describe these more general ideas regarding the meaning of our findings. We therefore now expanded this part in the Discussion (pgs. 20-24) before going into the Discussion of possible mechanisms.
The encoding and decoding models do not provide insight. The little bit of data on closed-loop versus openloop begins to get at this, but the data is still scant.

A4.
We believe our open-loop experiments provide valuable data in understanding how S1 encodes motor and sensory variables. Figure 6 conveys the main finding regarding differences in representation of these variables. The majority of L2/3 neurons were most responsive when both sensory stimulation and running occurred jointly. As explained in A2 we now also provide mutual information analysis, which further supports our statement that L2/3 neurons are more integrative. See also our answers A14,15.
Overall, this study has the sense of being either highly preliminary or secondary to some other data the authors collected. This work is from a highly respected lab that frequently publishes highly impactful and informative studies about neural circuits and coding mechanisms. There is very little of that in this report.

A5.
We disagree with the reviewer. We present a full data set using a novel approach and we present highly significant and relevant data regarding locomotion-induced modulation of sensory processing in barrel cortex. See also our comments above about novelty.
It's possible that further analysis of the presented dataset might yield something more interestingone thing that was not done here is to carefully quantify the amount of touches/force of touches under the various conditions and try to explain the encoding of these variables differentially.

A6.
A more detailed analysis of the whisker touches to the texture surface certainly would be interesting. However, from our videos (e.g. Supplementary Video 2) it is hardly possible to extract touch forces on individual whiskers, especially because we kept the full set of whiskers intact in order to have a naturalistic setting and not to induce plasticity effects. That being said we made further efforts to track single whisker from our movies using the newly available DeepLabCut method (Nat. Neurosci. 21:1281, 2018) Figure R1). This finding was valid for parameters computed from single whisker tracking and all whisker tracking. On the other hand same analysis showed that both run speed and texture speed contributed significantly to predicting calcium signals (Supplementary Fig. 10a-c). For the updated manuscript we have not included Figure R1 as a supplement as it involves limited data set and we may not have been imaging the corresponding whisker barrel in S1.
Another way to strengthen this study would be to selectively label sub-populations of L2/3 or L5 neurons with defined projection targets and relate the potential differential coding of touch or motor variables to their projection targets.

A7. Subdividing anatomically distinct subpopulations certainly is a very interesting suggestion. However, such
investigation is well beyond the scope of this study. We believe the current study presents a strong ground work for such future studies, which can explore the specific mechanisms in further detail. ) that was used throughout the study. We only considered time periods where texture was in contact with whiskers as continuous tracking of single whiskers was not feasible in the absence of a texture contact as single whiskers came in and out of the field of view. We could track single whiskers only for 10 sessions (4 L2/3 and 6 L5 imaging sessions). (c) We predicted calcium signals of each neuron during texture touch using a random forest algorithm given run speed, texture speed and various whisking parameters (amplitude of single-whisker envelope, frequency of single-whisker stick-slip events, single--whisker angular speed, amplitude of all-whisker envelope) as predictors. Models were trained and evaluated on separate parts of the data set. Then we shuffled each whisking related parameter one by one while keeping other parameters intact and compared quality of predictions to understand contribution of each parameter in predicting calcium signals. 1.
The authors claim they are only labeling L2/3 and L5 pyramidal neurons. There is no direct evidence of this in the study (they cite a previous study of theirs that used a similar vector), and I am skeptical. The promoter in their vector is a pan-cellular driver, and is used widely in AAV vectors for expression in interneurons with the DIO construct. They might be correct, but this should be shown through immunohistochemistry or RNA analysis that all labeled cells are glutamatergic. This is important because the diversity they observe in their data could in part be due to labeling some inhibitory neurons that are known to act differently. A8. In our experience expression of calcium indicators in inhibitory subpopulations under EF1a promoter is difficult (we tried this approach for other projects). We now provide further evidence of low expression rates in inhibitory neurons. We injected viral construct AAV2.1-EFα1-R-CaMP1.07 into barrel cortex of VGAT-CHR2-EYFP transgenic mice, which express EYFP in GABAergic cell population. Histological analysis revealed that only 3.3% of R-CaMP1.07 expressing neurons were GABAergic neurons in L2/3 and only 6.2% in L5 of S1 cortical slices (Supplementary Fig. 1). This is consistent with only about a third of interneurons showing expression and confirms that our findings mainly represent pyramidal cell populations in L2/3 and L5.

2.
Why did the authors use R-CaMP? This should at least be justified, when the green reporters have much better SNR. Is it for depth? Plenty of groups have imaged L5 with GCaMP6.

3.
The total data set is not on that many neurons for two photon calcium imaging. Still a lot compared to ephys, but it still suggests a preliminary study. Fig. 3i-l, 5, 6) consider all imaging sessions independently and report results from up to 1800 neuronal measurements. Hence our study is not preliminary at all. Cleary, there are ongoing developments to expand the field-of-views and to increase the number of imaged neurons. However, the respective publications so far mostly are technical demonstrations with weak biological and behavioral aspects.

4.
The authors report on a transient response in L5 vs. a sustained response in L2/3. Is this real or an artifact of how calcium is handles differentially between these cells, or because the indicator dye acts different in these two cell groups? A11. There is no evidence for calcium handling is different in L2/3 versus L5 neurons. All information that is available from >20 years of experiments in vitro and in vivo (e.g. Helmchen et al. 1999, Nature Neuroscience;Svoboda et al., 1999 Nature Neuroscience) Fig. 3a-h). Moreover, we also compared baseline noise and signal-to-noise ratio (SNR) of each neuron in our in vivo recordings during 'no touch' and 'closed-loop' sessions (Supplementary Fig. 3i-l; same analysis as in Masamizu et al., 2014). Both L2/3 and L5 neurons showed similar distributions. We provide the relevant descriptions and discussion of this new data set in the revised manuscript (Results lines 116-120 and 209-215, Methods lines 616-662).

indicates that action-potential evoked somatic calcium dynamics in neocortical pyramidal neurons is similar in L2/3 and L5 pyramidal neurons in terms of amplitudes and decay times (relating to calcium influx and buffering properties). Most recently Masamizu et al., 2014 directly compared calcium transients in L2/3 and L5 neurons in M1 of mice expressing GCaMP3 and reported similar calcium dynamics. Since we agree with the reviewer that his is an important issue, we now provide further evidence from additional experiments. We performed simultaneous patch-clamp recordings and two-photon calcium imaging in acute cortical slices of wild type mice. We compared calcium dynamics using either the synthetic calcium indicator dye Cal-520 or R-CaMP1.07 and confirmed similar (and fast) action potentialevoked calcium transients in L2/3 and L5 neurons (Supplementary
Another major argument why differences in sustained vs. transient responses are not due to differences in intrinsic cell properties comes from the locomotion-onset responses. Here, responses after locomotion-onset displayed similar dynamics with sustained increases over seconds for both L2/3 and L5 neurons. This is shown in Fig. 2i of the manuscript and in addition in Supplementary Fig. 7 Supplementary Fig. 7b. Fig. 3a, right: what's going on with the L5 Reponses? The 'transient' response seems typically preceded by a big increase in calcium signal for most of the bottom neurons that is not clearly locked to any touch or behavioral variable.

6.
Previous electrophysiological recordings in L5 have not always shown such transient responses during touch with objects, but typically more sustained responses following a brief high frequency response. Does the dye just not capture the smaller sustained response in these neurons?
A13. It would be helpful to know, which previous studies the reviewer is referring to. Again, we do not think that there is any major difference regarding dye sensitivity. Our finding is not conflicting with what the reviewer is describing here either. On average L5 neurons show about 20% and 10% increase in ΔF/F (compared to pre-touch activity) during early and late phase after touch respectively, which indicates low sustained response following a brief high frequency response. The relative nature of the presented ΔF/F values (compared to either pre-locomotion-start or pre-touch) should be acknowledged, meaning that L5 may very well retain a low continuous firing rate during running. To better present this now we also plotted actual values of calcium signals (pre-activity not subtracted) in Supplementary Fig. 7a. This figure clearly shows how cells increase their activity with locomotion and how later texture touch augments this activity, in a sustained manner in L2/3 and transiently in L5.

Reviewer #3 (Remarks to the Author):
By using recently developed red-sensitive calcium indicator, R-CaMP1.07, Ayaz and colleagues measured locomotion and whisker-related activity of L2/3 and L5 neurons in barrel cortex. They conclude that L2/3 cells are sustained relative to L5 cells' activity that transient. They show that wall touching differently affected L5 cells whose response was transient whereas L2/3 cells had a sustained response.

Comments:
The animals were not preforming a task. Thus the functions of the different responses are analyzed with a model. The results as shown in supplemental Figs. 7&8 are only marginally supporting different encoding/decoding of neurons in L2/3 vs. L5.

A14.
We would like to highlight again (as in our answer A4) that the main finding regarding differences in representation of sensory and motor variables is shown in Figure 6.To strengthen our encoding/decoding analysis we used mutual information between neuronal responses and run speed as an alternative analysis (Supplementary Fig. 10). Comparing L2/3 and L5, we found that superficial neurons' information content was higher during touch, in line with a better ability to integrate sensory-motor information (see also our answer A2 and A15). Fig. 7D: the running vs. resting difference between L2/3 and L5 is marked as significant for L2/3 but not significant for L5. Given the small number of observations and their large variability, this conclusion seems weak at best. Supplementary Fig. 10 Fig. 10d)

revealed two interesting observations: (1) 'integrative cells' increased their mutual information about run-speed in the presence of wall-touch whereas for 'run cells' this variable decreased both in L2/3 and L5. (2) The overall increase in information content upon touch in L2/3 can be explained by the larger fraction of integrative cells in superficial layers compared to L5.
3. In Figs. 2,3&4 responses of L2/3 and L5 neurons are superimposed. This gives the impression that the recordings were done simultaneously. It should be stated that this was not the case (or describe the method used to record simultaneously across layers).

A16.
We apologize for the confusion. Imaging in L2/3 and L5 was done in separate sessions (now clarified in lines 119-120). Fig. 4B&C, please add interval of confidence (or other statistic) along with the response time course of L2/3 and L5 cells. Fig. 4d&c (old Fig 4b,c) we show ± s.e.m. as shading around the population average responses as we have done in Fig. 2d,g, and Fig. 3e. 5. Was the objective lens tilted? If so, please describe. What was the angle of the glass window relative to the animal axis?

A17. In new
A18. The objective was not tilted but the head holder was implanted in a way to slightly tilt the animals' head when head-fixed. We now specify this in line 557-558 of the text. 7. What could be the mechanisms that generated selective expression L5&L2/3? Please discuss. Supplementary Fig. 2). To our best knowledge, the mechanism is still unknown.

A20. It is a well-known phenomenon that AAV constructs typically do not infect L4 neurons (for which we provide evidence in
8. Was the imaging obtained from a specific barrel (or between barrels)? A21. We did not restrict our imaging windows to any specific barrel column. However, we did select our field of views in the region of the barrels that were touching the texture stimulus, which were identified by instrinsic signal imaging (Methods, lines 562-565).
Is there a way to record spikes and image calcium from the same cells in L2/3 and L5 to calibrate the system? This may be important if, for example, the system detects single spikes reliably in L2/3 but is less sensitive in L5. Some discussion of this potential issue should be included.
A22. For L2/3 neurons expressing R-CaMP1.07 we have previously performed juxtacellular recordings in vivo simultaneously with two-photon imaging and thereby characterized the sensitivity of R-CaMP1.07 (Bethge et al., 2017 (ref. 44); see also the relevant Methods section, page 32). In addition, we have now performed simultaneous patch-clamp recording and two-photon imaging in acute S1 (Supplementary Fig. 3). As explained in our response in A12 the action potential-evoked calcium transients in the somata of L5 neurons are similar to L2/3 neurons. In addition Supplementary Fig. 3i-l show that in our in vivo experiments baseline noise levels as well as SNR are similar for L2/3 and L5 neurons reflecting similar sensitivity for detecting action potentials.
Further to the above point, is the integration of locomotion and wall-touch is dependent on different neuron types, i.e., the MI, MD or BP cell types. Fig. 5) we compared their touch onset responses during 'Closed-loop' condition. All three classes qualitatively showed similar touch responses, with L5 neurons showing transient touch responses. (Figure R2a-c). In addition, we compared the integrative features of these classes. MI and BP run-modulated L5 cells mostly fell into the 'run cell' and 'integrative cell' classes with almost no 'texture cells' (Figure R2d-f  Clarifications 1) P3. The general statement "whisking behavior has been the main focus of studies on sensorimotor integration in vibrissal primary sensory cortex (S1 or 'barrel cortex')10-19 should also include the early work "Phase-to-rate transformations encode touch in cortical neurons of a scanning sensorimotor system. J. C. Curtis and D. Kleinfeld, Nature Neuroscience (2009)" and review "Neuronal basis for object location in the vibrissa scanning sensorimotor system. D. Kleinfeld and M. Deschênes. Neuron (2011)" on this topic.

A25.
Thank you for the suggestions. We added these references.
2) P3) Introduction. The statement "... whisking was found ... to increase thalamic activity should include reference to the works "Vibrissa self-motion and touch are reliably encoded along the same somatosensory pathway from brainstem through thalamus J. D. Moore, N. Mercer Lindsay, M. Deschênes and D. Kleinfeld, Public Library of Science: Biology (2015)" and "Parallel thalamic pathways for whisking and touch signals in the rat. C. Yu C, Derdikman, S. Haidarliu S and E. Ahissar E. PLoS Biology (2006)".

A26.
Thank you for the suggestions. We added these references.
3) P8, "39% of L2/3 and 45% of L5 neurons were running speed modulated". On page 45 line 955 the number of neurons modulated by run speed is presented as "441/705 for L2/3 neurons and 193/233 L5 neurons", which is 63% and 83%. These two group numbers do not match. Please clarify the claims.

A27.
We thank the reviewer for pointing out this mistake. We corrected the Supplementary Fig. 5 Figure S3 shows three types of running neurons, i.e., MI, MD and BP, as above. Please further analysis and discussion locomotion and wall-touch integration by these different cell types.