Article activity feed

  1. eLife

    Author Response:

    Reviewer #1:

    Assuming the "trend-level" responses related to pain facial expressions are reliable, there are several other interesting characteristics that emerged from the analyses. The analyses suggested overlapping, but separable, distributions of insular locations that encode pain from hands, faces, or both. This is consistent with work on population coding in other areas, and suggests (as the authors argue) that signals at many locations cannot be reduced to "salience" in general as they code for pain inferred from specific stimulus types. These results add to the literature, and appear to correspond with other fMRI studies that have examined intensity-coding of perceived pain. For example, Krishnan et al. 2016, eLife found that among individual brain areas that predict intensity of perceived pain from pictures of hands and feet, the insula was among the most strongly predictive. (They also found that a distributed network including other brain regions as well was much more strongly predictive). Zhou et al. 2020 eLife studied perceived pain from both facial expressions and pictures of body parts. They identified an overlapping area of the mid- and anterior insula that predicted perceived pain across both stimulus types. That area may be similar to the locations with overlapping encoding observed here, and the distribution across the insula of differentially predictive signals for body parts and faces may be similar to the distribution observed here. Both these studies analyzed relationships between brain activity and trial-by-trial ratings of perceived pain, and so are directly comparable.

    (...)

    1. There do not seem to be significant brain correlations within faces alone that survive correction for multiple comparisons. Fig 3 shows a "trend"-level result, not significant. Could this indicate a "hand vs. face" effect that appears as a correlation in Fig 2? If hands are rated higher than faces, and hands produce greater BBP in the insula, then any electrode that responds to hands more than faces will show up as a correlation between brain and rated intensity. The way to test this would be to test and show correlations within hands and faces separately, but these were apparently not significant for faces.

    The aim of our study was to analyze the overall broadband power as a proxy for neural activity, and our interpretations are based on differences across conditions in this overall power in the broadband range (20-190Hz). In this range, we do find significant coding of intensity for both hand and face stimuli, as can be seen in Figure 3g-h and Figure 5. Indeed, in our initial submission, the time-frequency decompositions were only included in supplementary materials (https://www.biorxiv.org/content/10.1101/2021.06.23.449371v2). Because eLife does not encourage supplementary materials, we were asked to move all supplementary materials into the main manuscript. We therefore also show the time-frequency decompositions separately for face and hand Figure 3b-f. However, these time-frequency decompositions suffer from low sensitivity due to the explosion of multiple comparisons when considering the large number of frequencies within the broadband range separately. These panels were thus only meant to illustrate how the power is distributed across frequencies, but were never meant as the basis for assessing which locations encode intensity. For such statistical inference, the overall broadband power analyses, that do not suffer from correcting for many frequencies, are much more sensitive and appropriate. In revising the manuscript we will include these illustrations as child figure, and focus on the main analyses, using the broadband power overall.

  2. eLife

    Evaluation Summary:

    Soyman and colleagues investigate intensity coding for the "pain of others" in the human insula with intracranial human recordings. Additional data of a related fMRI study is analyzed and discussed in the context of the intracranial data. The paper addresses an important research question of broad interest, with extremely unusual data which is investigated in considerable detail.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)

  3. eLife

    Reviewer #1 (Public Review):

    This study concerns insular activity recorded from 7 human participants while they viewed depictions of others in pain, evidenced by movies of both facial expressions made by a model or a hand being hit by a belt. Overall, the study has much to recommend it. The iEEG data are rare and difficult to acquire, and the authors performed a number of interesting and creative analyses. These include analyses of correlations with subjective rating, timing of pain rating-correlated iEEG signal relative to facial and motion information in the movies (and assessments of independent raters), and analyses of functional connectivity leveraging Neurosynth and a previously unpublished fMRI study. The evidence for insular encoding of perceived pain from hand movies seems to be strong, but the evidence for encoding from facial expressions is weak. It may be, however, that some electrodes encode pain intensity from facial expressions reliably. The main signal was a broad-band (20-170 Hz) signal that occurred in the first second after hand presentation, and perhaps somewhat later for faces, corresponding generally to when pain-related information appeared in the stimuli.

    Assuming the "trend-level" responses related to pain facial expressions are reliable, there are several other interesting characteristics that emerged from the analyses. The analyses suggested overlapping, but separable, distributions of insular locations that encode pain from hands, faces, or both. This is consistent with work on population coding in other areas, and suggests (as the authors argue) that signals at many locations cannot be reduced to "salience" in general as they code for pain inferred from specific stimulus types. These results add to the literature, and appear to correspond with other fMRI studies that have examined intensity-coding of perceived pain. For example, Krishnan et al. 2016, eLife found that among individual brain areas that predict intensity of perceived pain from pictures of hands and feet, the insula was among the most strongly predictive. (They also found that a distributed network including other brain regions as well was much more strongly predictive). Zhou et al. 2020 eLife studied perceived pain from both facial expressions and pictures of body parts. They identified an overlapping area of the mid- and anterior insula that predicted perceived pain across both stimulus types. That area may be similar to the locations with overlapping encoding observed here, and the distribution across the insula of differentially predictive signals for body parts and faces may be similar to the distribution observed here. Both these studies analyzed relationships between brain activity and trial-by-trial ratings of perceived pain, and so are directly comparable.

    The present results are also consistent with earlier studies that did not test the relationship with perceived pain, but tested multiple types of stimuli related to pain or other emotions. Corradi-Dell'Acqua et al. 2011 studied fMRI responses to pain-related, non-painful but threatening images, and neutral images, and found responses in the insula to both types of negative images. Later, Corradi-Dell'Acqua et al. 2016 extended this overlap analysis to local multivariate patterns of activity in response to shock pain and disgusting tastes administered to self and other, and to perceived unfairness in the Ultimatum Game. They found evidence for common patterns, particularly in the right anterior insula.

    Based on these studies, it would be interesting to see whether the iEEG signals recorded in this study would respond similarly to other types of aversive stimuli, including somatic pain. That would inform the field on whether they are related to pain perception or to another, correlated affective state. Though the authors rightly argue that the differential encoding of perceived pain implies that the entire insula cannot simply be encoding "salience". However, neurons respond to complex configurations of properties, and it may be possible to find signals in the insula or elsewhere that respond to many different combinations of stimulus properties, including conditional ones (e.g., only aversive stimuli delivered to the hand) without truly "representing" or encoding the perception of pain per se. Conclusively identifying a representation of perceived pain, or any other construct, is a noble but difficult challenge, however, and this work takes a step in this direction.

    Another interesting comparison would be the comparison with somatic pain. While early studies identified common patterns for observed and experienced pain in the anterior insula (Singer et al. 2004, Lamm 2011, and the Corradi-Dell'Acqua et al. papers), studies that used multivariate patterns to predict trial-by-trial pain experience (both observed and experienced) found distinct predictive patterns with little evidence for overlap (Krishnan et al. 2016, Lopez-Sola et al. 2017), though some evidence for an observed-pain predictive pattern transferring to somatic pain was found by Zhou et al. 2020. It's not yet clear whether the iEEG recordings observed here generalize to pain experience.

    There are a number of additional limitations to keep in mind:

    1. Notably, the authors do not intend to interpret the generalizabllity to direct pain experience or other affective states, or specificity to pain compared with other affective conditions; but without more information about these, it is difficult to tell what the signals in the insula actually represent. It leaves open the possibility that there is another, better explanation for activation of insular neurons than perceived pain. e.g., It could be about representation of the body more generally, rather than perceived pain specifically. Could they be coding for choices (ratings) themselves? Baliki found that insular activity correlates with rating intensity of a simple visual stimulus, and other studies (e.g., Grinband et al., Neuron) have found that the insula correlates with simple perceptual magnitude decisions.This will be an ongoing project for future work.

    2. A nice feature is the comparison of insular electrodes random electrodes "throughout the brain", but what electrode locations were available, in how many individuals, and what is their distribution? Surely there are not electrode placements *everywhere* in the brain.

    3. The stimulus set chosen was limited, and appears to relate to one specific type of painful stimulus on one hand model, and one set of facial expressions made by one female model. As it's well known in general that neurons often have complex receptive fields, it's unclear whether other types of painful hand stimuli or facial expressions made by other models would yield similar findings. Perhaps the insula would respond more strongly to other faces - or perhaps not at all. Other studies have shown that women's pain is discounted (Zhang et al. 2021). The need for diverse sets of stimuli to establish relationships with brain activity that are not stimulus-specific is becoming increasingly recognized (e.g., Yarkoni 2020, Westfall and Yarkoni).

    4. There do not seem to be significant brain correlations within faces alone that survive correction for multiple comparisons. Fig 3 shows a "trend"-level result, not significant. Could this indicate a "hand vs. face" effect that appears as a correlation in Fig 2? If hands are rated higher than faces, and hands produce greater BBP in the insula, then any electrode that responds to hands more than faces will show up as a correlation between brain and rated intensity. The way to test this would be to test and show correlations within hands and faces separately, but these were apparently not significant for faces.

    5. The iEEG signals are compared with fMRI indirectly, via Neurosynth and another new fMRI study. It would be useful to compare the maps with the cross-modal (body parts and faces) insula patterns predictive of observed pain from Zhou et al. 2020 in particular, which are available for download. How do the locations of activity patterns in this fMRI study correspond to the present iEEG locations, both in terms of common encoding across hands and faces and differential encoding.

  4. eLife

    Reviewer #2 (Public Review):

    Taking advantage of the high spatio-temporal resolution of intracerebral EEG, the aim of this study was to characterize the electrophysiological responses of the human insula that may be elicited by viewing pain being experienced by another.

    Patients were presented with two kinds of video clips: short videos of a hand being slapped by a belt and short videos of a face expressing pain while the hand was exposed to painful electrical stimuli. The authors found that both types of stimuli elicited broadband activity within the insula which correlated with reported intensity ratings.

    A strength of the manuscript is that, to better understand the time course of the elicited responses and their relation to perceived intensity and take full advantage of the temporal resolution of iEEG recordings, the authors conducted an in-depth analysis of the motion and shape information contained in each video clip, to understand how variations in these stimulus features as a function of time could be used by the participants to decode intensity of the viewed experience. Most importantly, the authors also characterized the relationship between the temporal dynamics of these stimulus features (motion and shape) and the time course of the correlation between insular activity and pain ratings.

    A question that one could raise - especially for the hand condition - is whether part of the trial-by-trial correlation between insular activity and ratings was the consequence of a relationship between that activity and low-level motion/shape features of the video clips, rather than a reflection of the insula being involved in "intensity coding for the pain of others". Fortunately, this question was at least partly addresses with the available data, by showing that the recorded insular activity correlated with motion energy when the motion was associated with pain (during the slapping of the belt), but not while the motion was innocuous (during the initial lifting of the belt).

    Nevertheless, as mentioned by the authors themselves, future experiments using additional control stimuli such as stimuli matched in terms of motion content but differing in the emotions they convey are critical to better understand the specificity of the described insular responses.

    While the broad-band cluster correlating with pain ratings for hand stimuli appears convincing - both in terms of strength and spatial distribution across electrode contacts - , the correlation between insular activity and pain ratings to face stimuli appeared more marginal and spatially scattered. For this reason, it is important to interpret cautiously the observed differences in the topographical distribution of the insular responses in the hand and face conditions. This is already acknowledged by the authors in the discussion section.

  5. eLife

    Reviewer #3 (Public Review):

    The study contains a very large number of analyses and tests. In some sense, this is both a strength and a weakness. It is a strength because it offers a more comprehensive description of the responses in the insula and how they might contribute to intensity of coding for pain-of-others. It is a potential weakness because each analysis involved decisions that can be somewhat questioned in terms of approach, appropriateness and validity. A thoroughly written review of the paper would therefore lead to many, many pages of detailed questions and comments, which I will not pursue here. I will give the authors the benefit of the doubt and assume that most of them are reasonable.

    However, because the study involved recording across 85 electrodes across time, the authors need to provide a more detailed description of the several treatments of multiple comparisons. They mention "Corrections for multiple comparisons are performed when repeated testing is done across timepoints using either FDR corrections, or by calculating a null distribution of cluster statistics." but this is insufficient and merits a subsection explaining the overall logic and necessary information for all instances involving multiple comparisons. For example, the method described by Maris and Oostenvald (2007) is excellent but not described clearly enough.

    One-tailed testing is controversial but again I won't quibble with its use in some of the analyses. But I simply don't understand the motivation in some cases. For example: "Averaging BBP power over the entire pain period revealed that out of 85 macro contacts, 27 (32%) showed a significant positive correlation (assessed as p1<0.05, Fig. 2c)." What justifies a one-sided test of this correlation? I think in this case (and possibly others) two-tailed tests would be more appropriate.

    Two out of the 9 patients were excluded due to "poor behavioral performance" but further information about the exclusion criteria were not sufficiently discussed.

    Data from 28 unit recordings were provided as additional evidence of pain-of-other coding. This is a very low number of units across only 3 patients, especially given that the results are based on 13 units that showed higher firing during the pain period compared to the baseline period. My own take is that these results probably should not take part of the manuscript, but other reviewers and the editor may disagree (besides, I don't agree that reviewers should tell authors what to include or not in their manuscripts!). In any case, the wording related to these findings should be revised to reflect the exploratory nature of these results. For example, I do not believe these results warrant labels such as "hand/face specific".

    The authors report timing effects in the range of 40-320 ms. They cite the well-known study by Krolak-Salmon and colleagues. But it is difficult to know what to make of the current results. On the one hand, latencies on the range of 200-300 ms (across the brain) would be compatible with some studies (but not others), how does one interpret latencies as low as 40-60 ms in the insula, which presumably are not really feasible. On the other hand, the results in the 560-1000 ms range are consistent with many studies but are discussed as "too slow". I don't know what the editor would suggest here, but I'll defer to their opinion about how to handle this.

    Despite familiarity with fMRI I found section 2.7 very difficult to follow. Based on the other reviews/editor, if the authors decide to keep this section, I would very much welcome a careful rewrite of this section.

    My final comment is about how the Discussion (and other parts of the paper) talks about "selectivity for Hand or Hand stimuli". The study makes use of a fantastic stimulus set from the research group. But if I understand the set correctly, the same actor is used for different levels of pain (face- and hand-based). While this is an excellent set to study some aspects of coding related to stimulus "preference" I don't believe it's general enough to establish "selectivity". Accordingly, I think it would be more reasonable to avoid the claims of "selectivity".

  6. Version 3 published on bioRxiv
  7. Version 2 published on bioRxiv
  8. Version 1 published on bioRxiv