NEURAL ENCODING OF FELT AND IMAGINED TOUCH WITHIN HUMAN POSTERIOR PARIETAL CORTEX

  1. Srinivas Chivukula
  2. Carey Zhang
  3. Tyson Aflalo
  4. Matiar Jafari
  5. Kelsie Pejsa
  6. Nader Pouratian
  7. Richard A. Andersen

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

In the human posterior parietal cortex (PPC), single units encode high-dimensional information with partially mixed representations that enable small populations of neurons to encode many variables relevant to movement planning, execution, cognition, and perception. Here we test whether a PPC neuronal population previously demonstrated to encode visual and motor information is similarly selective in the somatosensory domain. We recorded from 1423 neurons within the PPC of a human clinical trial participant during objective touch presentation and during tactile imagery. Neurons encoded experienced touch with bilateral receptive fields, organized by body part, and covered all tested regions. Tactile imagery evoked body part specific responses that shared a neural substrate with experienced touch. Our results are the first neuron level evidence of touch encoding in human PPC and its cognitive engagement during tactile imagery which may reflect semantic processing, sensory anticipation, and imagined touch.

Article activity feed

  1. eLife

    ###Reviewer #3:

    This manuscript reports a series of unique experiments with a single human participant, using an electrode array implanted in the left posterior parietal cortex several years after high-level spinal cord injury. There is a small but increasing number of groups capable of performing this type of research in humans. Most of this work has been focused on the motor system, but studies like this one, characterizing the somatosensory system (touch, in particular), have been increasingly common in the past five years. However, this is the only group focusing on this higher-level, multimodal association area of the cortex.

    Most of the recorded neurons were activated bilaterally, which is consistent with earlier monkey work from this lab. Probably the most important component of the work is the analysis of the modest activation in this area that occurs simply when the participant imagines different places on her body being touched - even the insensate arm. This work is virtually impossible to do in monkeys. There are extensive and overlapping analyses of the relation between actual and imagined activation, and the activation arising from inputs (or imagined inputs) from the two sides of the body. Eliminating a number of these and clarifying the remainder may improve the impact.

    1. 63: in a tetraplegic human subject recorded with an electrode array implanted in the left PPC I am curious why the array was placed in the left PPC, given the clinical evidence for the greater role of the right side in the formation of internal, multi-modal maps. Some comments would be useful.

    2. Fig 1: It would be good to show a panel of representative spikes, perhaps with their single-trial raster responses. This could be in a new figure that includes panel 1D, which is presented in a bit of an odd order as it now stands, coming in the midst of higher-level analyses. Indicate how many trials went into the averages in 1D.

    3. 146: we computed a cross-validated coefficient of determination (R^2 within) to measure the strength of neuronal selectivity for each body side. Even after reading the methods (further comments below) it is difficult to figure out what all these related measures reveal. At this point in the text it is very difficult to intuit how R^2 would measure selectivity.

    4. Fig 4: Several panels would be more effective if plotted as a function of distance rather than a category. 4E: This panel is borderline too small 4F: definitely too small. Enlarge, perhaps with fewer examples The curves drawn on the panels do not appear to be Gaussian, but neither are they just connected points. Show whatever it was you actually used. The Gaussian assumption does not appear to be very good for the edge cases (first two, last two) which is not terribly surprising.

    5. What is added by including both classification and Mahalanobis distance?

    6. 354: information coding evolves for a single unit. Two complementary analyses were then performed. In what sense are they complementary? What is added (besides complexity) by including both cluster analysis and PCA?

    7. Fig 8C: Despite my best efforts, I have no idea what this is showing

    8. 753: Classification was performed using linear discriminant analysis with the following assumptions:

    One, the prior probability across tested task epochs was uniform; It is not clear what prior probability this refers to. Just stimulus site?

    Two, the conditional probability distribution of each unit on any epoch was normal; Is this a reference to firing rate probability conditioned on stimulus site?

    Three, only the mean firing rates differ for unit activity during each epoch (covariance of the normal distributions are the same for each);

    Four, firing rates for each input are independent (covariance of the normal distribution is diagonal).

    Does this refer to independent firing rates of neurons across stimulus sites? This seems very unlikely, given everything we know about dimensionality of cortex. Perhaps it refers to something else. Cannot all of these assumptions be tested? Were they?

    1. 768: we computed the cross-validated coefficient of determination (R2 within) to measure how well a neuron's firing rate could be explained by the responses to the sensory fields. This needs a better description, and I may be missing the point entirely. I assume it is an analysis of mean firing rate (which should be stated explicitly) and that it uses something like the indicator variable of the linear analysis of individual neuron tuning above. In this case is this a logistic regression? As it is computed for each side independently, it would appear that there are only four bits to describe the firing of any given neuron. This would seem to be a pretty impoverished statistic, even if the statistical model is accurate.

    2. 786: The purpose of computing a specificity index was to quantify the degree to which a neuron was tuned to represent information pertaining to one side of the body over the other. This is all pretty hard to follow. The R2 metric itself is a bit mysterious, as noted above. Within and across R2 is fairly straightforward, but adds to the complexity, as does SI, which makes comparisons of three different combinations of these measures across sides. Aside from R2 itself, the math is pretty transparent. However, a better high-level description of what insight all the different combinations provide would help to justify using them all. As is, there is no discussion and virtually no description of the difference across these three scatter plots. The critical point apparently, is that, "nearly all recorded PC-IP neurons demonstrate bilateral coding". There should be much a more direct way to make this point.

    3. Computing response latency via RF discrimination is rather indirect and assumes that there is significant classification in the first place. I suspect it will add at least some delay beyond more typical tests. Why not a far simpler and more direct test of means in the same sliding window? Alternatively, a change point analysis?

  2. eLife

    ###Reviewer #2:

    General assessment:

    The study by Chivukula et al., explored a unique (n=1) dataset of multi-unit neuron recordings collected in the postcentral-intraparietal area (PC-IP) of a tetraplegic human subject taking part in a brain machine interface clinical trial. The recordings were collected across a set of tasks designed to investigate neuronal responses to both experienced and imagined touch.

    Overall I found the manuscript to be well-written, the study to be interesting, and the analysis reasonable. I do, however, think the manuscript would benefit by addressing two main, and a number of minor, issues.

    Major comments:

    1. The methods would benefit from additional rationale / supporting references throughout. Whereas it is generally clear what was done, it is sometimes less clear why certain choices were made. Perhaps some of the choices are "standard practice" when working with single unit recordings, but I was left in want of a bit more reasoning (or at least direction to relevant literature). Some examples are below:

    For the population correlation (line 723): why was the correlation computed 250 times or why were the two distributions shuffled together 2000 times?

    For the decode analysis (line 752): consider providing a reference for those interested in better understanding the "peeking" effects mentioned.

    Response latency (line 798): how were window parameters determined (for both visualization and the latency calculation). And what was the rationale for them being different - especially given that the data used for the response latency calculation was still visualized (at least in part)? Relatedly, I'd be curious to see the entire time-course for that data rather than just the shaded region of the "visualization" data. Also, it would be nice if a comment (or some data) could be provided regarding how much the latency estimates change based on these parameter choices.

    Temporal dynamics of population activity (line 830): why use a 500 ms window, stepped at 100 ms intervals instead of something else?

    Temporal dynamics of single unit activity (line 887): it is stated that the neurons were restricted to those whose 90th percentile accuracy was at least 50% to ensure only neurons with some degree of significant selectivity were used for the cluster analysis. But why these particular values? Are the results sensitive to this choice? In this section, I'd also suggest providing references for those interested in better understanding the use of Bayesian information criteria. Similarly, it is stated that PCA is a "standard method for describing the behavior of neural populations" - as such it would be nice to provide some relevant references for the reader.

    1. The manuscript would benefit from additional context in the intro as well as a more thorough discussion - particularly with respect to the imagination aspect of the experiment.

    Intro: The second paragraph did well in establishing why one might be interested in examining somatosensory processing in the PPC. It was however, less clear why the particular questions at the end of the paragraph were being posed. Perhaps an extra paragraph could be added to bridge the notion that a sizeable body of literature has been developed around somatosensory representation within the PPC and the several "fundamental" questions remaining that are of interest here.

    Discussion: The manuscript would benefit from a more thorough discussion of "imagination per se" and the various top-down processes that might be involved - as well as better positioning with respect to previous studies investigating top-down modulation of the somatosensory system. The authors state that the cognitive engagement during the tactile imagery may reflect semantic processing, sensory anticipation, and imagined touch per se - which I would not argue. But I would also expect some explicit mention of processes like attention and prediction. Perhaps these are intended to be captured by "sensory anticipation" - but, for example, attention can be deployed even if no sensation is anticipated. Importantly, it seems that imagining a sensation at a particular body site might well involve attending to that body part. That is, one may first attend to a body part before "imagining" a sensation there - and then even continue to attend there the entire time the imagining is being done. Because of this, perhaps the authors are considering attention to be a part of "imagination per se". But since attention has been shown to modulate somatosensory cortex without imagination, how can one exclude the possibility that the neuronal activity measured here simply reflects this attention component? Regardless, I think the discussion would benefit from a more explicit treatment of these top-down processes - especially given the number of previous studies showing that they are able to modulate activity throughout the somatosensory system. Some literature that may be of interest include:

    Roland P (1981) Somatotopical tuning of postcentral gyrus during focal attention in man. A regional cerebral blood flow study. Journal of Neurophysiology 46 (4):744-754

    Johansen-Berg H, Christensen V, Woolrich M, Matthews PM (2000) Attention to touch modulates activity in both primary and secondary somatosensory areas. Neuroreport 11 (6):1237-1241

    Hamalainen H, Hiltunen J, Titievskaja I (2000) fMRI activations of SI and SII cortices during tactile stimulation depend on attention. Neuroreport 11 (8):1673-1676. doi:10.1097/00001756-200006050-00016

    Puckett AM, Bollmann S, Barth M, Cunnington R (2017) Measuring the effects of attention to individual fingertips in somatosensory cortex using ultra-high field (7T) fMRI. Neuroimage 161:179-187. doi:10.1016/j.neuroimage.2017.08.014

    Yu Y, Huber L, Yang J, Jangraw DC, Handwerker DA, Molfese PJ, Chen G, Ejima Y, Wu J, Bandettini PA (2019) Layer-specific activation of sensory input and predictive feedback in the human primary somatosensory cortex. Sci Adv 5 (5):eaav9053. doi:10.1126/sciadv.aav9053

  3. eLife

    ###Reviewer #1:

    In this study Chivukula, Zhang, Aflalo et al. report on an extensive set of neural recordings from human PPC. It is found that many neurons are responsive to touch in specific locations. Interestingly, a considerable fraction of the neurons displayed symmetric bilateral receptive fields. Furthermore, these neurons also became active during imagined touches. The study paves the way for a deeper understanding of the role of the human PPC.

    The paper presents a wealth of analysis on an extensive set of recordings. It is generally well written and the analyses are well thought out. My main concerns are regarding missing information and unclear descriptions of some of the analyses undertaken, which are detailed below.

    1. At the start of the results section it is stated that the recordings were from "well-isolate and multi-unit neurons". This seems to contradict the Methods section, which only talks about "sorted" neurons. This needs to be clarified, and if multi-units were included, it should be stated which sections this concerns as it will have implications for the results (e.g. for selectivity for different body parts). In any case, the number of neurons included in different analyses should be evident. There are some numbers in the Methods and sprinkled throughout the Results section, but for some of the analyses (e.g. clustering analysis, which was run only on a responsive subset of neurons) no numbers are provided.

    2. The linear analysis section needs further details. The coefficients are matched to "conditions" but it is not explained how. I am assuming that each touch location is assigned to a condition c, however the way the model is described suggests that the vector X can in principle have multiple conditions active at the same time. Additionally, could the authors confirm whether it is the significance of the coefficients that determined whether a neuron was classed as responsive as shown in Figure 1? This analysis states a p-value but does give no further information on which test was run and on what data.

    3. Figure 1 C could be converted into a matrix that lists all combinations of RF numbers on either side of the body to highlight whether larger RFs on one side of the body generally imply larger RFs on the other side.

    4. I am confused about the interpretation of the coefficient of determination as shown in Figure 2A. In the text this is described as testing the "selectivity" of the neurons. To clarify, I am assuming that the "regression analysis" is referring to the linear model described in a previous section. The authors then presumably took the coefficients from this model for a single side only and tested how well they could predict the responses to the opposite side, as assessed by R^2 (Fig 2C,E). Before that in Fig 2A, they tested how well each single-side model could predict the responses. This is all fine, but the "within" comparison then simply tests how well a linear model can explain the observed responses, and has nothing to do with the selectivity of the neuron. For example, the neuron might be narrowly or broadly selective, but the model might fit equally well.

    5. Regarding the timing analysis, it is not clear to me how the accuracy can top out at 100% as shown in the figure, when the control conditions were included. Additionally, the authors should state the p value and statistic for the comparison of latencies.

    6. Spatial analysis. Could the authors provide the size of the paintbrush tip that was used in this analysis. Furthermore, as stimulation sites were 2 cm apart, it is not appropriate to specify receptive fields down to millimeter precision.

    7. Imagery: how many neurons were responsive to both imagery and real touch? Were all neurons that were responsive to imagery also responsive to actual touch? This is left vague and Figure 5 only includes the percentages per condition, but gives no indication of how many neurons responded to several conditions. Whether and how many neurons were responsive to both conditions also determines the ceiling for the correlation analysis in Figure 5D (e.g. if the most neurons are responsive only to actual but not imaginary touch, this will limit the population correlation).

  4. eLife

    ##Preprint Review

    This preprint was reviewed using eLife’s Preprint Review service, which provides public peer reviews of manuscripts posted on bioRxiv for the benefit of the authors, readers, potential readers, and others interested in our assessment of the work. This review applies only to version 1 of the manuscript. Tamar R Makin (University College London) served as the Reviewing Editor.

    ###Summary:

    Chivukula and colleagues report an extensive set of multi-unit neural recordings from PPC of a tetraplegic patient taking part in a brain machine interface clinical trial. The recordings were collected across a set of tasks, designed to investigate neuronal responses to both experienced and imagined touch. It was found that many neurons are responsive to touch in specific locations. Most of the recorded neurons were activated bilaterally, which is consistent with earlier monkey work from this lab. Probably the most important component of the work is the analysis of the modest activation in this area that occurs simply when the participant imagines different places on her body being touched - even the insensate arm. This work is virtually impossible to do in animals, and as such offers a unique opportunity to describe neural properties for higher-level representation of touch. The study therefore paves the way for a deeper understanding of the role of the human PPC in the cognitive processing of somatosensation.

    Overall, we found the manuscript to be well-written, the study to be interesting, and for the most part the analyses are well thought out. But at the same time, the reviewers raised multiple main concerns regarding missing information and unclear descriptions of some of the analyses undertaken, which are detailed below over many major and minor comments. In addition, it was felt that there was unnecessary overlap across analyses - the first part especially contains a number of analyses that seem to make very similar points repeatedly or where it is not entirely clear what the point is in the first place. As such, there is a need to identify and cut a lot of the duplicative analyses/results and explain both the essential methods and the interpretation of the remaining results more succinctly and clearly. The key analyses could then be streamlined and better justified, ideally with an eye towards a consistent approach in both parts of the paper. here are also some major considerations regarding the contextualisation and interpretation of the key imagery results, as detailed in the first major comment below.

  5. Version published to 10.1101/2020.07.27.223636 on bioRxiv