Anticipatory responses along motion trajectories in awake monkey area V1

Article activity feed

1. 

   eLife

   May 26, 2020

   ###This manuscript is in revision at eLife

   The decision letter after peer review, sent to the authors on May 25, 2020, follows.

   Summary

   Using a combination of single and array recordings, and voltage sensitive dye (VSD) imaging, the authors demonstrate that neurons in awake macaque V1 show anticipatory responses to a smoothly moving object outside their classical receptive field (RF). This anticipatory activity builds up slowly and can lead to spiking. Combining theoretical modeling, VSD imaging and LFP recordings, they further demonstrate that the spatio-temporal properties of these anticipatory responses are consistent with the hypothesis that they are generated by intra-V1 horizontal and inter-areal feedback connections. These results are important because they challenge classical models of motion integration that are largely based …

   ###This manuscript is in revision at eLife

   The decision letter after peer review, sent to the authors on May 25, 2020, follows.

   Summary

   Using a combination of single and array recordings, and voltage sensitive dye (VSD) imaging, the authors demonstrate that neurons in awake macaque V1 show anticipatory responses to a smoothly moving object outside their classical receptive field (RF). This anticipatory activity builds up slowly and can lead to spiking. Combining theoretical modeling, VSD imaging and LFP recordings, they further demonstrate that the spatio-temporal properties of these anticipatory responses are consistent with the hypothesis that they are generated by intra-V1 horizontal and inter-areal feedback connections. These results are important because they challenge classical models of motion integration that are largely based on feedforward mechanisms. In contrast to these models, the work presented here demonstrates a role for horizontal and feedback mechanisms in motion processing, and that motion integration starts at the lowest level of cortical processing, within V1 itself. The authors use a variety of methodologies to corroborate this result.

   Appreciation

   The reviewers in general made positive comments about the work and found the findings interesting. They also made many critical remarks that require substantial and essential revision, data analysis and experimentation.

   Revisions

   1. The reviewers made many critical point about receptive field quantification, and the interpretation of anticipatory firing related to this.

   (i) The anticipatory responses in the manuscript are assumed to be due to the smooth motion of the stimulus rather than an extra-classical RF effect. This is discussed by the authors, but not truly demonstrated in the manuscript. It is possible that the authors might have observed responses to bars placed outside the RF that are flashed rather than moving? The sparse-noise mapping that they used to delineate the RF might help to distinguish these possibilities as the authors could look at the responses to noise-flashes that fell on the trajectory of the bar (but outside the RF) to determine if these drove the cell.

   (ii) The quantification of RF-sizes is not well explained.

   • The reader would appreciate a plot of the (classical) RF boundaries and the starting positions of the 3 bar sweeps in the example cells shown in Fig. 1B-D.
   • Since the conclusions of the study rely heavily on estimation of RFs, it would be important to show some examples of RF mapping with flashed squares, and to plot the activation profile with flashed squares for the same neuron as a function of DVA in Figure 1. The RF mapping is described quite briefly in the Methods and it is not entirely clear what it amounts to in terms of neural activation.
   • Can the authors indicate at which distance from the RF the short bar sweep typically started?
   • What is the latency of the response if aligned on the start of the short bar sweep? If it is close to 40ms, this might indicate that the bar actually started in the RF of some of the neurons.
   • The finding that some neurons do not have anticipatory responses does not provide a control (in contrast to what is stated in line 184) because these neurons might have had smaller RFs. However, the data in Figure S2D-E might address this point, and could be presented in the main paper.
   • We would like to see a distribution of RF sizes for the single units, the pixels of the VSD measurements and the spectral components for the MEA recordings.

   (iii) The spatial extent of thalamic inputs arriving from the M- and P-pathway going into V1 differs (Lund et al., 2003, Figure 7). In particular M-pathway inputs have a wider termination zone. It is not clear whether this may account for a discrepancy between RF sizes mapped with moving stimuli and flashing stimuli. Moreover, since the RFs were mapped initially with flashing squares, it is possible that eye movements exhibited less variability in that condition, and that this leads to effectively larger RF sizes with moving stimuli. The finding of anticipatory finding might thus be explained by these factors without requiring recurrent connections. It is important to discuss this possibility and to address it with analyses.

   (iv) If the bar would proceed to move after going out of the RF, is there also a widening observed there? This would be congruent with generally larger RFs for the moving stimuli.

   (v) How did the authors compute the "time to peak" (line 191)?

   • Line 244: time 0 is when the RF crosses the RF center and the peak response happens before time zero as shown in Fig 2C. This is worrisome, because the peak response for a moving bar is actually expected after the bar reaches the RF center, given the delay between retina and cortex (see e.g. Fig. 3 in Supèr and Roelfsema, 2005 Prog. Brain Res. 147, 263-282). Do the authors correct for the delay between the retina and V1 to compute time zero? How? Please specify this.
   • Reviewers were confused in the methods section by lines 875-877. Is this where a correction for the response latency is described? If yes, please clarify this text (also in the main text) because such adjustments may have an large impact on the main result.
   • There is a similarly confusion section in lines 893-898. It is not clear what happens here.
   • Same in lines 907-910. What is "probability of anticipation"?
   • Same in lines 912-921 what is "skp timing - 50"? What is the aim?

   (vi) Is it possible to also show the retinotopic maps obtained with VSD imaging?

   (vii) Overall, the authors should quantify RF size with the same methods used for flashes and bars and compare these directly with the same quantification; in addition correct for eye movements and delays.

   (2) The reviewers made several critical remarks w.r.t. the relationship of the findings to trajectory prediction.

   A main concern is whether the build-up activity demonstrated here is related in anyway to predictions of smooth motion trajectories or whether it is a passive spread of activity in cortex. Points 2.i and 2.ii are related to this concern. The spread of horizontal activity has been demonstrated previously and would reduce the novelty of the findings demonstrated here.

   The reviewers agree that there are three aspects here that require further experimentation:

   (i) The authors link their findings to psychophysical studies suggesting that we can use smooth motion to predict the upcoming location of the stimulus and improve perception on the leading edge of the stimulus. However, throughout the manuscript the activity triggered by the stimulus entering the classical RF is largely identical. Furthermore, there is no behavioral manipulation in the manuscript or any manipulation of the predictability of the motion path. This makes it difficult to determine if the build-up activity they observe has any functional significance or whether it is simply a passive spread of activity around the moving stimulus.

   (ii) The lack of build-up activity for stimuli activating the ipsilateral cortex is an interesting finding which supports the authors' claims that these results are due to the spread of activity in (unmyelinated) horizontal connections. But doesn't this result also severely limit the functionality of this effect? If the prediction is unable to 'jump' across the vertical meridian, then this suggests it is more of a passive spread of activity around the stimulus rather than an active process providing cortex with a prediction of an upcoming moving stimulus.

   • Is there psychophysical evidence that the effects of motion prediction on behavior (mentioned in e.g. lines 94-100) has a discontinuity at the vertical meridian?

   (iii) The implication of these findings is that V1 neurons start responding to a moving stimulus before the stimulus reaches their receptive field. However, objects do not always move smoothly, and sudden changes in trajectories occur. Would V1 neurons, in this case, signal the "wrong" trajectory?

   (3) The quantification of anticipatory firing needs to be substantially improved.

   (i) Line 259: how was the "first significant change" estimated? Please specify here or refer to the relevant Methods section. In general, the data analysis section of the Methods is presented as a long list of metrics and statistical analyses without clear reference to which part of the results and figures each refers to. Vice versa there is no reference to any specific Methods section in the description of the Results or figure legends. This makes the manuscript somewhat difficult to read.

   (ii) The representation of the data in Figure 1 is somewhat problematic, because the population average is shown only for neurons with anticipatory responses (n=26). In Figure Supplement 2, the number of cells is 22 (why the difference?). Are these now only the anticipatory neurons? Why did the authors not show the average population responses across all neurons before splitting into anticipatory and non-anticipatory neurons? It would be good to see the average PSTH.

   (iii) Fig S2: The curves look less asymmetric there, and seems to show a general widening for the long bar condition.

   (iv) Normalization is a concern for the group average plots, because an average PSTH can be biased by a few cells with high firing rates.

   (v) line 386: there are no statistics for the anticipatory response here.

   (4) The relationship of direction tuning to anticipatory activity requires further data analysis.

   (i) It is unclear why the authors chose not to optimize the stimulus trajectory to the direction preference of the cells under study, at least in the single cell recording experiments.

   (ii) The authors describe the relationship with direction tuning on line 276. They find that the anticipatory response is considerably stronger in cells where the direction of motion of the bar is aligned with the preferred direction of the cell. This interesting effect isn't quantified statistically and seems under-explored in the manuscript as a whole. It seems to be an extremely strong effect, to the extent that the reviewers wonder if the build-up effect is significant for cells with a non-aligned direction tuning? The effect is also not included in any of the later models and would not be predicted by their proposed model. The reviewers wondered whether the authors could also use the bar sweeps that they use for RF mapping, which move in 12 different directions, to further explore the relationship between direction-tuning and anticipatory responses?

   (iii) Lines 274-281. Did neurons with preferred direction aligned to the stimulus trajectory also show shorter latencies of anticipatory responses? The authors speculate this could be the case in the following sentence, and present this as a result in the discussion section, but they never really showed any analysis addressing this.

   (5) Motion speed:

   The manuscript does not manipulate motion speed which limits the interpretation of the findings. One simple way to test the proposed model would be to vary the speed of the bar. At high speeds the feedforward drive would 'catch-up' with the horizontal and feedback spread and the anticipatory response should disappear. Do the authors have any data on this and would they agree with this prediction?

   (6) Reviewers had concerns about smoothing in the data analysis.

   Reviewers worried about the smoothing that took place in the analysis (e.g. line 856, 863) and the sliding window used for the computation of a power spectrum (line 928). Something similar may happen with detrending VSD signal in line 942. First smoothing the data in time can cause "responses" at earlier time points, but would be artifactual. Do the results also hold up if the data is not smoothed?

   (7) Laminar differences:

   Lines 260-266. The observed variability could depend on laminar differences. Do the authors have any record of laminar location of the recordings? More importantly, the variability could also reflect differences in the direction preference of the neurons. The authors should check this. It is possible that neurons with the direction preference matching the stimulus trajectory are facilitated while those with direction preferences unlike the stimulus trajectory are suppressed, or vice versa. This further analysis would provide additional insights into the underlying mechanisms.

   (8) Analyses of frequency bands:

   (i) Did the authors measure the RFs of the different frequency bands of the LFP? Low-frequency bands tend to be sensitive to changes in visual information over very wide regions of the visual field. The spatial precision of this effect may be very low and is not shown by the authors (did the low-frequency power drop simultaneously across the whole array at bar onset for example). It is hard to imagine that a coarse effect, which may be more related to arousal or attention, is related to a prediction about the precise location of an upcoming stimulus.

   (ii) Reviewers would like to see the average power spectrum in Fig. 5.

   (iii) line 455 Even though previous studies suggested that feedback influences have a spectral signature, this does not mean that changes in the power spectrum can provide causal evidence for the involvement of feedback connections. It can well be that subcortical inputs also decrease/increase power in particular frequency bands.

   (iv) What is shown in Fig. 5B? Is this the evoked potential or some spectral measure?

   (v) The authors should clarify in the main text how they normalized the power and determined statistical significance.

   (vi) Line 475. The decrease in the low frequency band is actually preceded by a slight increase, as evident in both Fig. 5C and D. In the discussion, the authors only emphasize the decrease in power (which indeed is stronger), but they do not provide any rationale for why one would see a decrease, rather than an increase, in power. One would have predicted that an increase in feedback excitation would result in an increase in low frequency band power (and this would be consistent with published results indicating an increase in alpha/beta power when feedback processing increases). On lines 628-630 of the Discussion, the authors attempt to provide some explanation for such decrease in power, but the sentence is rather obscure. The authors need to clarify and expand this idea. Also, should the earlier increase in low frequency power be ignored?

   (9) Results on the ipsilateral hemifield stimulation.

   Reviewers did not understand the rationale for, and the interpretation of, this result. Surely callosal connections exist within about 2 deg of the midline and they would activate V1 neurons in the contralateral hemifield which, in turn, send horizontal connections within the contralateral hemifield. Also, in contrast to what stated in the discussion, reviewers pointed out that callosal connections are topographically organized.

   (10) Discussion lines 677-678. The authors need to expand on these two concepts by clarifying to the broad readership of the journal what "diffusion of motion information" means, and how their results could underlie this phenomenon as well as enhance motion discriminability. Same comment applies to the sentence on line 688.

   (11) Critique of the model:

   (i) The model has many parameters that do not appear well justified. Such a model can provide support for horizontal spread but the authors should acknowledge that other models without horizontal connections in V1 could give rise to similar predictions. E.g. horizontal spread could happen in a higher areas that feeds back to V1. Arbitrary parameters appear e.g. in line 971; where does the estimate of 0.41mm come from? The same question can be asked about the choice of the model in lines 973-1006 and its parameters. Model choices, such as the function in equation 1 and the equation in line 993 are not well explained and make an ad hoc impression.

   • Note that the equation in line 993 is not connected to equation 1, because the reader expects k_h to reappear in the equation in line 993 but it does not. It is also unclear why "h" appears twice in the equation in line 993 (within and after the brackets). Is h=k_h?
   • Is there an equation of how the feedback influences k_h or V1?

   (ii) It is unclear whether the speed of propagation beyond 2deg from the RF, which the authors attribute to feedback, is in fact consistent with feedback conduction velocities. On line 415, the authors seem to imply 0.02m/s is consistent with feedback (but feedback is much faster than this).

   line 407: the authors arrive at a horizontal propagation speed of 0.06 m/s but the calculations that gave rise to this estimate are lacking. How strongly does it depend on the model with its many arbitrary assumptions ? Can they also provide a 95% confidence interval for this estimate?

   (iii) Line 331. The model incorporates isotropic horizontal and feedback connections, but in real cortex these are anisotropic, and therefore only contact neurons having similar orientation and direction along the anisotropic axis. Incorporating real-life functional connectivity may help the model account for some of the variability in surround effects observed in the data.

   (iv) Lines 978-979. What parameters values were used for this, and on what basis were these selected? More in general, all the parameter values used in the model (e.g. dh= 6 mm) need to be justified and appropriate references cited.

   (12) Details about data analysis:

   In general there are many details in the data analysis that are difficult to understand. The authors may first wish to consult with another specialist (not involved in the study) about the clarity of their descriptions because the reviewers found them insufficiently clear. Some of these have been summarized in the other points listed above and below. There are further points here:

   (i) Line 273. Why was a 2-sample t test used here, given it is said that 3 trajectories are statistically compared? This should be an ANOVA. The T-test is used throughout the manuscript. From the Methods, it appears this may be justified by the fact that the authors pooled medium and long trajectories into a single group. IF so, this should be clearly stated in the Results and/or a reference to the relevant Methods section should be added to the Results. Moreover, what was the rationale for grouping the 2 longer trajectories?

   Read the original source
2. 

   Version published to 10.1101/2020.03.26.010017 on bioRxiv

   Mar 29, 2020