Non-shared coding of observed and executed actions prevails in macaque ventral premotor mirror neurons

Article activity feed

1. 

   Version published to 10.7554/elife.77513 on eLife

   Jul 17, 2023
2. 

   eLife

   Jul 4, 2023

   Author Response

   Many thanks for the detailed and sometimes sharp, yet appropriate criticism of our study. It was an incentive for us to carry out additional analyses and to devote more effort to an elaboration of concepts. The outcome is that the results have changed slightly and that we now give more space to a discussion of concepts. We first address here the points raised by more than one reviewer before responding to comments contributed by individual reviewers.

   The points raised can be divided into three thematic groups, 1) conceptual issues, 2) experimental and analytical questions, and 3) comments challenging the novelty of our results. On the first theme, we think it is essential to make a clear distinction between the conceptual and observational domains. As such, the criteria defining a “mirror neuron” and what is meant by the …

   Author Response

   Many thanks for the detailed and sometimes sharp, yet appropriate criticism of our study. It was an incentive for us to carry out additional analyses and to devote more effort to an elaboration of concepts. The outcome is that the results have changed slightly and that we now give more space to a discussion of concepts. We first address here the points raised by more than one reviewer before responding to comments contributed by individual reviewers.

   The points raised can be divided into three thematic groups, 1) conceptual issues, 2) experimental and analytical questions, and 3) comments challenging the novelty of our results. On the first theme, we think it is essential to make a clear distinction between the conceptual and observational domains. As such, the criteria defining a “mirror neuron” and what is meant by the term "mirror mechanism" belong to the conceptual domain. This understanding of terms requires agreement among scientists, but is not experimentally testable. Unfortunately, there is no agreement on how to define a “mirror neuron” and what is meant by “mirror mechanism”. Thus, for the present work, the only option is to refer to specific definitions or to use our own, definitions which try to capture what others, and here most importantly Rizzolatti and colleagues, probably meant. We have adjusted the introduction in an attempt to convey our understanding and usage of the two terms in a hopefully comprehensible manner. Briefly, we use a definition for "mirror neuron" that we take from the first paragraph of the results section of Gallese et al. (Brain, 1996). We do not consider the "properties of mirror neurons" described in that paper as defining a mirror neuron (MN). Classifying neurons as MNs only on the basis of the presence of a modulation of discharge rate during an executed and an observed action compared with a baseline is a common practice also in other single neuron studies on MNs, consistent with this definition. Regarding "mirror mechanism", we refer to Rizzolatti and Sinigaglia (2016) and make a distinction between a broad and a strict definition. Given our finding that there are almost no F5 MNs whose activity during observation is a motor representation according to our strict definition of a mirror mechanism, and also given the problem that the term “mirror mechanism” itself is not uniformly understood, the question arises whether and how the term "mirror neuron" should be used in the future. The answer to this may vary and belongs to the conceptual domain. We briefly address this question at the end of the discussion of the revised manuscript.

   From that understanding of terms, conceptual hypotheses are to be distinguished, which of course must allow experimental predictions, i.e., must be falsifiable. We now distinguish more clearly between a "representation hypothesis" and an "understanding hypothesis". Both hypotheses focus on F5 MNs and are based on the strictly defined mirror mechanism. We test the “representation hypothesis” in our study, and just because it is the basis for the “understanding hypothesis”, falsifying the “representation hypothesis” would allow us to conclude that the “understanding hypothesis” is not valid. In contrast, confirmation of the “representation hypothesis” would not, of course, allow us to conclude that the “understanding hypothesis” holds. That would really be circular reasoning (this conclusion was drawn by some and rightly criticized). However, support for the “representation hypothesis” would be the necessary prerequisite for the “understanding hypothesis” to be true. These two hypotheses take up the original argument that a certain understanding of observed actions could follow from an equality of action-specific F5 MN activity during execution and observation. Because we considered the data on equality of action- specific F5 MN activity to be insufficient, we designed this study. Since our result largely argues against the "representation hypothesis" and thus against the "understanding hypothesis," we now discuss alternative concepts for the function of F5 MNs in more detail. It should be noted here that our fourth concept ("goal-pursuit-by-actor") could well represent the observed action without contradiction to our broad definition of a mirror mechanism, which in principle could also serve a subjective experience (which could be conceived as a kind of understanding). The way we structure the concepts in the discussion of this revised manuscript is, in our opinion, a useful overview of the concepts. The third concept is new in this context. We would like to emphasize that we focus on F5 MNs and intentionally avoid a discussion of mirror neurons beyond F5 in this paper. With the data from this study, we cannot say anything about MNs outside of F5.

   Regarding the key question of how the "understanding hypothesis" is testable, or whether it may not be testable at all, we agree, of course, that for the conclusion of whether F5 MNs contribute to perception, only a manipulation of F5 MNs can clarify it. We now say that explicitly in the introduction. We agree with reviewer #2 that "understanding" here is not limited to "action recognition" or "action categorization”, which in principle could be implemented by purely sensory processing. Therefore, we also do not believe that the approach proposed by reviewer #3, which builds on the distinction of actions, would allow for a critical examination of the "understanding hypothesis”. But we disagree that the "understanding hypothesis" is not testable at all. Operationalization is necessary. If we accept that we can measure certain visual or auditory perceptions of an animal by operationalization (e.g., the subjective visual vertical, see for example Khazali et al., PNAS, 2020), then we must also accept that we can, in principle, measure other subjective experiences by operationalization, such as pain or aiming at a goal or even the co- experience of pain. An example of how to approach this is the study by Carrillo et al. (Curr Biol, 2019), which reviewer #2 and colleagues discussed in a recent review article (Bonini et al., TCS, 2022).

   With regard to the second theme, experimental and analytical questions, we noticed while reading the comments that in our first version we did not distinguish clearly enough between statements about single neurons and statements about populations of neurons. Therefore, we now clearly separate single neuron analysis and population code analysis in the structure of the article. In view of the fact that statements about mirror neurons in the literature mostly refer to single neurons, we added extensive single neuron analyses, so that only now statistically reliable statements about single neurons are possible. This has led to the realization that the number of neurons with exclusively shared code is so small that these neurons should be considered a rare exception. Given the small number of time periods with shared code, we additionally tested against a hypothesis already rightly proposed as an alternative explanation by G. Csibra in 2005 (Mirror neurons and action observation: Is simulation involved? In: What do mirror neurons mean? Interdisciplines Web Forum 2005). We were able to reject this hypothesis based on two of three methods for testing for a shared code. This is the second piece of evidence besides the clustering of time periods with shared code already described in the first version that time periods with shared code cannot be considered random.

   We discuss in more detail the question of whether neurons that exhibit a shared code at least at times support the representation hypothesis. To this end, we additionally examined whether certain action segments are more frequently represented with a shared than with a non-shared code, whether neurons with shared code differ from those with non-shared code in anatomical location, and whether an accuracy can be achieved with a time bin-wise selection of neurons with shared code by population cross-task classifiers as with within-task classifiers in the whole population.

   Another issue was how to test for shared code and how to decide if a code has enough sharing. To answer the question, the exact hypothesis we intended to test here is crucial. The representation hypothesis states that the representation of the observed actions in F5 MNs corresponds to the representation as it occurs during the execution of the same actions. Therefore, the relationship between discharge rate and actions that holds during execution should also hold during observation, which is measurable with a classifier trained on execution trials and tested on observation trials. Moreover, the actions should not be more distinguishable during observation with a classifier other than the execution-trained classifier, because if that were so, it would mean that the representation of observed actions is different from that of executed actions. The detection of a cluster of time bins for which both conditions are satisfied confirms that it is possible to discover in this way the shared codes postulated by the representation hypothesis.

   With respect to concerns that the monkey may not have used the cue at all when the action was executed, we added a comparison with control trials with a non-informative cue and also compared the duration of the approach phase between the three actions. Regarding oculomotor behavior, we verified that the monkey had actually directed his gaze toward the action during action observation for all three actions.

   On the third issue, concerning the novelty of our results, we have now explained in more detail in the introduction why we felt it necessary to conduct a study we considered fundamental. As a result of our study, it can be clearly stated now that representations of observed actions as predicted by the strictly defined mirror mechanism are rare in F5 MNs, but nevertheless cannot be dismissed as random. This dispels the objection rightly raised by Csibra in 2005 and contradicts the currently prevailing view that such a representation can only be found at a population level. Even if these representations are ultimately explained by a concept other than the strictly defined mirror mechanism, their existence must be accounted for by any theory of the function of F5 neurons. Moreover, it is also shown that the observed actions are well discriminated with a non- shared code, at times even optimally. This contradicts the notion – which has been widespread for a long time since the work of Gallese et al. (Brain, 1996) – that mapping to motor representations in terms of broad congruence is simply not perfect. The applied cross-task decoding approach seems promising to test also in the future for a shared action code. Finally, reconsideration of alternative concepts has led us to highlight the possibility of a representation of a goal pursuit by the observer.

   Reviewer #1 (Public Review):

   The authors set out to investigate the hypothesis that mirror neurons in ventral premotor area F5 code actions in a common motor representation framework. To achieve this, they trained a linear discriminant classifier on the neural discharge of three types of action trials and test whether the thus trained classifier could decode the same categories of actions when observed. They showed that codes were fully matched for a small subset of neurons during the action epoch, while a wider set of "mirror neurons" showed only poorly matched codes for different epochs.

   This is one of the descriptions of our results, where we realized that in our first version we did not distinguish clearly enough between statements about single neurons and statements about populations of neurons. This prompted us to perform a detailed single neuron analysis.

   The authors controlled for potential visual object confounds by having identical objects be manipulated in three different ways and by having the animal carry out the motor execution in the dark. The main strength of the study lies in the clever decoding approach testing the matched tuning to behavioural categories in a model-free way. The central result is in the identification of the small sub-group of mirror neurons that show true matching during the execution epoch, which can dissociate the three types of action almost perfectly. This aligns well with some previous work while offering a novel avenue to identify and investigate those neurons. The underlying neuronal mechanism and behavioural relevance of these neurons remain an open question. It would have been interesting to understand better whether the specific motor representations at a recording site, for instance identified through microstimulation prior to recording (see Methods), the reaction times on individual trials or the specific gaze targets (object/hand) had a bearing on the decoding performance for a neuron/trial.

   We agree that these are interesting questions.

   In this study, the focus is on testing for a shared code according to a strictly defined mirror mechanism. We have now compared the anatomical locations of neurons with only time bins in which observed actions were discriminated with a shared code (according to one of the methods) to the locations of neurons with only time bins with non-shared code (see last paragraph in Results). We did not find any relevant difference and this is why one cannot expect topographically specific effects of microstimulation.

   We do not expect the reaction time (i.e., the time interval between LED onset and start button release, or the duration of the approach epoch) during execution or observation to have any effect on our results on shared coding as the analysis was based on relative time bins. The observed actions were predominantly distinguished late in the approach epoch, but especially in the manipulation epoch. At this time, reaction time is not expected to have a relevant influence.

   The relationship between gaze/eye position and the activity of mirror neurons, during execution or observation, is an interesting topic in itself. However, for testing for a shared code according to a strictly defined mirror mechanism, it is only relevant that the observing monkey actually observes the action. We have ensured this in our experiment by a fixation window and have now also confirmed that the monkey actually looked into the area of the object during all three actions (see Results, lines 209-219 in the manuscript with tracked changes).

   Ultimately, the uncovered matched mirror representations should in future experiments be tested with causal interventions and linked trial-by-trial to action selection performance.

   The authors put the focus of their discussion on the wider, less well-matched neuronal pool to support an action selection framework, which is of course a valid view and well established in motor representations. From a sensory perspective, sparse coding, as suggested by the small group of "true" mirror neurons identified with the decoding approach, should also be considered as the basis for a possible neuronal mechanism. A particular strength of the paper is that it could give new data and impetus to the important discussion about how motor and sensory coding frameworks come together in cortical processing.

   We have expanded the discussion considerably and also address the possibility of sparse coding.

   Reviewer #2 (Public Review):

   The paper by Pomper and coworkers is an elegant neurophysiological study, generally sound from a methodological point of view, which presents extremely relevant data of considerable interest for a broad audience of neuroscientists. Indeed, they shed new light on the mirror mechanism in the primate brain, trying to approach its study with a novel paradigm that successfully controls for some important factors that are known to impact mirror neuron response, particularly the target object. In this work, a rotating device is used to present the very same object to the monkey or the experimenter, in different trials, and neurons are recorded while the monkey (motor response) or the experimenter (visual response) performed a different action (twist, shift, lift) cued by a colored LED.

   The results show that there is a small set of neurons with congruent visual and motor selectivity for the observed actions, in line with classical mirror neuron studies, whereas many more cells showed temporally unstable matched or even completely non-matched tuning for the observed and executed actions. Importantly, the population codes allow to accurately decode both executed and observed actions and, to some extent, even to cross-decode observed actions based on the coding principles of the executed ones.

   In my view, however, the original hypothesis that an observer understands the actions of others by the activation of his/her motor representations of the observed actions constitutes circular reasoning that cannot be challenged or falsified, as the author may want to claim. Indeed, 1) there is no causal evidence in the paper favoring or ruling out this hypothesis (and there couldn't be), 2) there is no independent definition (neither in this paper nor in the literature) of what "action understanding" should mean (or how it should be measured). Instead, the findings provide important and compelling evidence to the recently proposed hypothesis that observed actions are remapped onto (rather than matched with) motor substrates, and this recruitment may primarily serve, as coherently hypothesized by the authors, to select behavioral responses to others (at least in monkeys).

   1. One of the main problems of this manuscript is, in my view, a theoretical one. The authors follow a misleading, though very influential, proposal, advanced since the discovery of mirror neurons: if there are (mirror) neurons in the brain of a subject with an action tuning that is matched between observation and execution contexts, then the subject "understands" the observed action. This is clearly circular reasoning because the "understanding" hypothesis uniquely derives from the neuron firing features, which are what the hypothesis should explain. In fact, there is no independent, operational definition of the term "understanding". Not surprisingly there is no causal evidence about the role of mirror neurons in the monkey, and the human studies that have claimed to provide causal evidence of "action understanding" ended up using, practically, operational definitions of "recognition", "match-to-sample", "categorization", etc. Thus, "action understanding" is a theoretical flaw, and there is no way "to challenge" a theoretical flaw with any methodologically sound experiment, especially when the flaw consists of circular reasoning. It cannot be falsified, by definition: it must simply be abandoned. On these bases, I strongly encourage the authors to rework the manuscript, from the title to the discussion, by removing any useless attempt to falsify or challenge a circular concept and, instead, constructively shed new light on how mirror neurons may work and which may be their functional role.

   Please see the response to all.

   1. An important point to be stressed, strictly related to the previous one, concerns the definition of "mirror neuron". I premise that I am perfectly fine with the definition used by the authors, which is in line with the very permissive one adopted in most studies of the last 20 years in this field. However, it does not at all fulfill the very restrictive original criteria of the study in which "action understanding" concept was proposed (see Gallese et al. 1996 Brain): no response to object, no response to pantomimed action or tool actions, activation during execution in the dark and during the observation of another's action.

   We do not agree that the enumerated "very restrictive original criteria" emerge from the Gallese et al. (Brain, 1996) study. Except for the first paragraph in the results section, there is no clear statement on how mirror neurons should be defined.

   If the idea (which I strongly disagree with) was to simply challenge a (very restrictive) definition of mirroring (a very out-of-date one, indeed, and different from the additional implication of "action understanding"), the original definition of this concept should be at least rigorously applied. In the absence of additional control conditions, only the example neuron in Figure 2A could be considered a mirror neuron according to Gallese et al. 1996.

   We have the impression that the question does not distinguish clearly enough between the definition of "mirror neuron" and the definition of "mirror mechanism". In defining "mirror mechanism", we refer to the work of Rizzolatti and Sinigaglia (Nat Rev Neurosci, 2016). We do not think that this definition is out-of-date (see for example the 2018 article by Rizzolatti and Rozzi in Handbook of Clinical Neurology). If the term "mirror mechanism" is to be defined differently, then another term should be used for a new definition or an annotation should be added (such as "version 2"). This would be necessary to avoid unnecessary confusion resulting from unclear terms.

   Permissive criteria implies that more "non-mirror" neurons are accepted as "mirror": simply because they are permissively named "mirror", does not imply they are mirroring anything as initially hypothesized

   Even for a neuron that would be classified as a "mirror neuron" according to your previously stated "very restrictive original criteria”, it does not follow that it "mirrors” according to a mirror mechanism. And, of course, it is quite possible that more neurons do not "mirror” according to a mirror mechanism if one tests more neurons.

   (Example neuron in Fig 2B, for example, could be related to mouth, rather than hand, movements, since it responds strongly and similarly around the reward delivery also during the observation task, when the monkey should be otherwise still).

   We agree, it is not excluded that this neuron has a relation to mouth movements. However, since the neuron meets the conditions to be classified as a "mirror neuron", an additional relation to mouth movements would not be relevant. If mouth movements are to be an exclusion criterion, then this would have to be included and justified in the definition of a "mirror neuron".

   Clearly, these concerns impact all the action preference analyses. To practically clarify what I mean, it should be sufficient to note that 74% (reported in this study) is the highest percentage ever reported so far in a study of neurons with "mirror" properties in F5 (see Kilner and Lemon 2013, Curr Biol) and it is similar to the 68% recently reported by these same authors (Pomper et al. 2020 J Neurophysiol) with very similar criteria. Clearly, there is a bias in the classification criteria relative to the original studies: again, no surprise if by rendering most of the recorded neurons "mirror by definition" then they don't "mirror" so much. I suggest keeping the authors' definition but removing the pervasive idea to challenge the (misleading) concept of understanding.

   We think that it is very important to clearly separate "mirror neuron" from "mirror mechanism". And the question arises whether one should not include a mirroring criterion, which is derived from a definition of a mirror mechanism, in the definition of mirror neurons. We address this briefly in the discussion. Ultimately, the point of our study is to find out how many of the - if you want to put it that way - "permissively defined" mirror neurons actually “mirror”. And the answer depends on how one defines “mirror mechanism”. We provide an answer by resorting to a “strictly defined mirror mechanism”. We have now also given throughout the results section the percentages of neurons with certain properties with respect to all measured F5 neurons. This is a reference that allows comparisons among studies, provided that no neurons were directly discarded during recording, which we avoided in our study.

   1. It would be useful to provide more information on the task. Panel B in Figure 1 is the unique information concerning the type of actions performed by the monkey and the experimenter. Although I am quite convinced of the generally low visuomotor congruence, there are no kinematics data nor any other evidence of the statement "the experimental monkey was asked to pay attention to the same actions carried out by a human actor". First, although the objects were the same, the same object cannot be grasped or manipulated in the same way by a human and a macaque, even just because of the considerable difference in the size of their hands; this certainly changes the way in which monkeys' and experimenter's hands interact with the same object, and this is a quantifiable (but not quantified) source of visuomotor difference between observed and executed actions and a potential source of reduced congruency.

   We agree, of course, that there are kinematic differences in how a monkey and how a human manipulate the same object. We have not measured the kinematics and thus cannot make a systematic statement about this. We now report in the results section the rather incidental observation that already the reaching trajectories for the three actions differed and show corresponding differences in the timing of the approach epoch. However, for the question of this study, how many neurons are eligible to represent observed actions according to a strictly defined mirror mechanism, the kinematic repertoire of the observed actor is irrelevant. The reference is the F5 mirror neuron activity during the monkey's own action, i.e., how the monkey approaches the object with his hand, how he grasps it, and how he brings it to a certain target position and holds it there. The observed action, according to the strictly defined mirror mechanism, is to be mapped to this reference. Therefore, we did not collect kinematic data. But it is of course a possible explanation for a non-shared code if the strictly defined mirror mechanism does not apply.

   Second, there is little information about monkey's oculomotor behavior in the two conditions, which is known to affect mirror neuron activity when exploratory eye movements are allowed (Maranesi et al. 2013 Eur J Neurosci), potentially influencing the present findings: a {plus minus}7 (vertical) and {plus minus}5 (horizontal) window at 49 cm implies that the monkey could explore a space larger than 10 cm horizontally and 14 cm vertically, which is fine, but certainly leaves considerable freedom to perform different exploratory eye movements, potentially different among observed actions and hence capable to account for different "attention" paid by the monkey to different conditions and hence a source of neural variability, in addition to action tuning.

   We agree that the topic of the relationship between F5 MNs activity and eye movements is interesting. And we know from the work of Maranesi et al. (2013) that at least larger eye movements during action observation are related to the activity of F5 MNs. In our study, we ensured that the observing monkey was actually observing the action. For this purpose, we used a fixation window. We now additionally verified that the monkey really looked into the area of the object during all three actions (see Results, lines 209-219 in the manuscript with tracked changes). In our study, the fixation window was so small that the monkey could not see the face of the human actor, in contrast to the study of Maranesi et al. (2013). It was mainly the face that attracted the monkey's attention in that study (measured by gaze position). In our study, the risk that the gaze of observing monkey was out of the fixation window was high when he looked at the human actor's hand above the wrist. The execution of the action by the monkey took place in darkness. We did not use a fixation window because the monkey's own execution of the action can be assumed to direct his attention to the action.

   We cannot rule out the possibility that smaller eye movements during observation, larger eye movements during execution in darkness, covert shifts of spatial attention, or more generally attentional fluctuations have an influence on F5 MNs that might have counteracted a shared action code in our study. However, if this were the case, then the investigated hypothesis that the activity of F5 MNs during action observation is a motor representation according to the strictly defined mirror mechanism would also have to be rejected.

   1. Information about error trials and their relationship with action planning. The monkey cannot really "make errors" because, despite the cue, each object can be handled in a unique way. The monkey may not pay attention to the cue and adjust the movement based on what the object permits once grasped, depending on online object feedback. From the behavioral events and the times reported in Table 1, I initially thought that "shift" action was certainly planned in advance, whereas "lift" and "twist" could in principle be obtained by online adjustments based on object feedback; nonetheless, from the Methods section it appears that these times are not at all informative because they seem to depend on an explicit constraint imposed by the experimenters (in a totally unpredictable way). Indeed, it is stated that "to motivate the monkey even more to use the LED in the execution task, another timeout was active in 30% (rarely up to 100%) of trials for the time period between touch of object to start moving the object: 0.15 (rarely 0.1) for a twist and shift, 0.35 (rarely 0.3s) for a lift". This is totally confusing to me; I don't understand 1) why the monkey needed to be motivated, 2) how can the authors be sure/evaluate that the monkeys were actually "motivated" in this way, and 3) what kind of motor errors the monkey could actually do if any. If there is any doubt that the monkeys did actually select and plan the action in advance based on the cue, there is no way to study whether the activity during action execution truly reflects the planned action goal or a variety of other undetermined factors, that may potentially change during the trials. Please clarify.

   It is true that the three actions could in principle be performed without using the LED as an informative cue. While this is unlikely under the assumption that a monkey prefers the easiest and fastest way to get reward, it remains a possibility. For this reason, we introduced time constraints in a part of the trials. The selection of time constraints and the proportion of trials in which they were applied, was a pragmatic compromise between a time limit, at which the LED must be used as an informative cue for action selection in order to comply with the task, and a time span that allows the task to be completed even when overall motivation is low. The latter takes into account the general experimental experience that a monkey's engagement or motivation in such experiments varies across trials, sessions, and days. To evaluate whether the LED color was, indeed, used as a cue for action planning in the execution task, we randomly interleaved trials with a different LED, non-informative regarding the type of object, as a control in 5% of the trials. We compared the behavioral responses in trials with informative cues and those with a non-informative cue. The behavioral analysis established that both monkeys indeed used the informative cues to guide their choices (see Fig. 1D).

   Further evidence that the monkey used the cue for action selection and planning is the finding that the type of action was encoded before the release of the start button and then further during the approach phase, i.e., much earlier than somatosensory feedback about the manipulability of the object was available (see Fig. 3A and Fig. 6A).

   Regarding the question, which "motor errors" were possible: The answer can be found in the description of the cases in which a trial was aborted (see Material and methods): releasing the start button too early (< 100 ms after turning on the LED), manipulating the object too slowly after touching it (the time constraints mentioned), not holding the object until the reward was given, or not performing the task at all (10 s timeout).

   1. Classification analysis. There seems to be no statistical criterion to establish where and when the decoding is significantly higher than chance: the classifier performance should be formally analyzed statistically. I would expect that, in this way, both the exe-obs and the obs-exe decoding may be significant. Together with the considerations of the previous point 2 about the permissive inclusion criteria for mirror neurons, this is a remarkable (even quite unexpected) result, which would prove somehow contrary to what the authors claim in the title of the paper. The fact that in any classification the "within task" performance is significantly better than the "between task" performance does not appear in any way surprising, considering both the inclusive selection criteria for "mirror neurons" and the unavoidably huge different sources of input (e.g. proprioceptive, tactile, top-down, etc. afferences) between execution and observation. So, please add a statistical criterion to establish and show in the figures when and where the classifications are significantly above chance.

   We have added - in addition to the statistics already performed in the first version (Fig. 3A in the previous version, now Fig. 6A) - a number of analyses including statistics. This mainly concerns the analyses regarding a shared code at the single neuron level, in which we additionally tested against the null hypothesis proposed by Csibra in 2005 using permutation tests. And we have now also calculated confidence intervals for the population classifications that allow the comparison with chance level. We re-performed the classification analyses using eight-fold cross-validation. We also added a statistical analysis to the finding of clustering of time periods with shared code (Fig. 4). In Figure 5, we additionally compared the frequency of action segments with shared and non-shared codes, which is a descriptive, exploratory analysis. For this reason, it does not make sense to perform inferential statistics. Overall, these analyses represent a significant expansion of the analyses in the first version. We have done this primarily to arrive at statistically sound conclusions at the single neuron level.

   Regarding the comparison between within-task classification (o2o) and cross-task classification (e2o), it is important to keep in mind that the goal was to test the hypothesis that the activity of F5 MNs during action observation is a motor representation of the observed action according to the strictly defined mirror mechanism. This hypothesis requires both, 1) an above chance level accuracy of the e2o classifier and 2) no better accuracy of the o2o classifier as compared to the e2o classifier. If the o2o classifier were better, then the actions would not be represented as they are executed. And the reference in this hypothesis is the motor representation, that is, the code at execution. Thus, the direction e2o classification is the crucial one, not the reverse direction (o2e). One explanation for the fact that o2o shows better accuracy in the population may be the different sensory inputs mentioned above. In this case, the tested hypothesis has to be rejected and replaced by another one, which should then have a different name.

   Nevertheless, we also show the result of the o2e cross-task classification in Fig. 6 (yellow curve), which was already included in Fig. 3 of the first version. However, we do not address it in more detail in the main text because it is not relevant for the hypothesis to be tested. It is only a reportable additional result.

   1. "As the concept of a mirror mechanism posits that the observation performance can be led back to an activation of a motor representation, we restricted this analytical step to a comparison of the exe-obs and the obs-obs discrimination performance". I don't understand the rationale of this choice. The so-called "concept" of mirror mechanism in classical terms posits that mirror neurons have a motor nature and hence their functioning during observation should follow the same principle as during action execution. But this logical consideration has never been demonstrated directly (it is indeed costated by several papers), and when motor neurons are concerned (e.g. pyramidal tract neurons, see Kraskov et al. 2009) their behavior during action observation is by far more complex (e.g. suppression vs facilitation) than that hypothesized for classical "mirror neurons". Furthermore, when across-task decoding for execution and observation code has been used, both in neurophysiological (e.g. Livi et al. 2019, PNAS) and neuroimaging (Fiave et al. 2018 Neuroimage) data, the visual-to-motor direction typical produce better performance than the opposite one. Thus, I don't see any good reason not to show also (if not even just) the obs-exe results. Furthermore, I wonder whether it is considered the possible impact of a rescaling in the single neuron firing rate across contexts, as the observation response is typically less strong than the execution response in basically all brain areas hosting neurons with mirror properties, and this should not impact on the matching if the tuning for the three actions remains the same (e.g. see Lanzilotto et al. 2020 PNAS). The analysis shown in Figures 4 and 5 is, for the rest, elegant and very convincing - somehow surprising to me, as the total number of "congruent" neurons (7.5%) is even greater than in the original study by Gallese et al. (5.4%).

   As to the rationale of our approach, please see our response to the previous point.

   On the issue of rescaling: the hypothesis tested here requires that the F5 MNs activity on observation is a motor representation of the observed action. Hence, from the activity during observation the action should be just as readable as from the execution-related activity. If we had to use rescaling to find a shared code, then observed actions would not be represented in F5 MNs in the same way as on execution. Additional information on whether the action is being executed or observed would be needed. This would of course be possible in principle, but would contradict the hypothesis. And we then not only have the difficulty of which readout is the physiological one (here we make a parsimonious assumption with a linear readout), but we would have to make an additional assumption about rescaling. For this study, we have now chosen the solution of performing the action preference analysis on a single neuron level in a statistically clean way. This represents a very liberal form of rescaling, as it only tests whether the action with the highest or lowest discharge rate is the same when executed and observed. That is, if the result here is not fundamentally different, which is the case, then it can also be assumed that one does not get qualitatively different results for other forms of rescaling.

   1. The discussion may need quite deep revision depending on the authors' responses and changes following the comments; for sure it should consider more extensively the numerous recent papers on mirror neurons that are relevant to frame this work and are not even mentioned.

   The discussion has been thoroughly revised considering the comments raised and suggestions of this and the other two reviewers.

   Reviewer #3 (Public Review):

   Mirror neurons are a big deal in the neuroscience literature and have been for thirty years. I (and many others) remain skeptical of whether they serve the functions often attributed to them - specifically, whether they are motor planning neurons that contribute to understanding the actions of others. Testing their functions, therefore, is of great interest and importance. The present study, however, is not a cogent or convincing test. I do not think this study helps to answer the questions surrounding mirror neurons. It purports to provide a crucial test, that comes out mostly against the mirror neuron hypothesis, but the test has too many weaknesses to be convincing.

   Thank you for the clear words. We take from it, first of all, that in the first version of the manuscript we failed to convey the relevance of our study for the discussion of mirror neuron function. The concerns of this reviewer are in line with those of the others and are addressed in our response to all three reviewers.

   First, consider that the motor tuning and the visual tuning match "poorly." How poor or good must the match be before the mirror neuron hypothesis is rejected? I do not know, and the study does not help here. Even a "poor" match could contribute significantly to a social perception function.

   The specific hypothesis tested here assumes that an action-specific activity of F5 MNs evoked by observed actions corresponds to an action-specific activity of these actions if executed. The approach taken here to compare cross-task classification accuracy (execution-trained, tested in observation) with within-task classification accuracy (observation-trained, tested in observation) tests this hypothesis. The fact that we found a cluster of time periods of single neurons in which both accuracies are almost equal supports this approach and also the hypothesis for these time periods. In principle, of course, the decision for the presence of a difference or equality is always only a statistical statement and contains assumptions. For example, the assumption that a linear readout has physiological relevance enters here. But this problem exists in all studies that ultimately try to understand biological neuronal networks in order to explain perceptions and behavior. However, it is such studies that attempt to elucidate what information is contained in which neurons that set the stage for experiments that, in the optimal case, manipulate certain neurons in a particular way in order to then measure the behavior of an animal that is just right for those neurons.

   Second, the results remind me in some ways of other multi-modal responses in the brain. For example, in the visual area MST, neurons are tuned to optic flow fields that imply specific directions of self-motion. Many of the same neurons are tuned to vestibular signals that also imply specific directions of self-motion. But the optic flow tuning and the vestibular tuning are not perfectly matched. There is considerable slop and complexity in how the two tunings compare within individual neurons. That complexity is not evidenced against multi-modal tuning. Instead, it suggests a hidden-layer complexity that is simply not fully understood yet. Just so here, the fact that the apparent motor tuning and apparent visual tuning match "poorly" is not evidence against both a motor planning and a visual encoding function.

   We hope that it is now clearer, in contrast to the first version, that we tested a specific hypothesis that is only a prerequisite for the hypothesis of a very specific form of understanding. Referring to the example, the hypothesis analogous to ours would be that the representation of self-motion direction due to optic flow ("observation") corresponds to the representation of self-motion direction due to vestibular stimulation ("execution"). If it were then found that the self-motion direction due to optic flow cannot be predicted from a classifier trained on vestibular stimulation, and that another classifier trained on optic flow performs better, then the hypothesis would have to be rejected. This is then a reason to realize that "everything is a bit more complex" and to search for better explanations.

   Third, the animals are massively over-trained in three actions. They perform these actions and see them performed thousands of times toward the same object. Surely, if I were in the place of the monkey, every time I saw the object, I'd mentally imagine all three actions. As I saw a person act on the object, I'd mentally imagine the alternative two actions at the same time. Even if the mirror neuron hypothesis is strictly correct, this experiment might still find a confusion of signals, in which neurons that normally might respond mainly to one action begin to respond in a less predictable way during all three trial types.

   In our study, we tested a specific hypothesis related to the time an action is observed. Here, you suggest an alternative hypothesis. The question is whether this alternative hypothesis better explains the result of our study. The alternative hypothesis can be formulated as follows: the F5 MNs activity elicited by an observed action in this experiment corresponds to a mixture of the activities that occur when the other two actions are executed. This hypothesis is to be rejected because it fails to explain why a shared code occurs in single neurons and why cross-task population classifiers show an accuracy above chance level. A modified alternative hypothesis, which states that what is represented in the experiment during observation is a mixture of all three actions, cannot explain why the three actions are very well represented in the population and are optimally represented exactly when the target position of the object is reached.

   Fourth, the experiment relies on a colored LED that acts as an instructional cue, telling the monkey which action to perform. What is to stop the neurons from developing a cue-sensitive response, as in classic studies from Steve Wise and others in the premotor cortex? Perhaps the neuronal signal that the experimenters are trying to measure is partly obscured by other, complex responses influenced in some manner by the instructional cue?

   In principle, there is the possibility that purely sensory information is also represented in area F5, at least in some neurons or at certain points in time. We take your suggestion and discuss this as one of the alternative concepts (we call it "sensory concept"). However, several findings argue against this concept. For example, neural responses to cues usually represent the subsequent action, but not sensory information of the cue such as the color of the cue. In our study, it is evident from Figure 3A, 6A and 6B that during action execution, actions are discriminated even before the start button is released. Since this discrimination of actions occurs with a time delay after the cue and then increases continuously, this is evidence that the action to be executed is represented, but not the cue itself.

   Fifth, finally, and most importantly, the fundamental problem with this study is that it is correlational. Studies that purport to test the function of a set of neurons, and do so by use of correlational measurements, cannot provide strong answers. There are always half a dozen different interpretations and caveats, such as the ones I raised here. Both sides of a debate can always spin the results, and the arguments are never resolved. To test the mirror neuron hypothesis properly would require a causal study. For example, lesion area F5 and test if the monkey is less able to discriminate the actions of others. Or, electrically microstimulate in area F5 and test if the stimulation interferes (either constructively or destructively) with the task of discriminating the actions of others. Only in this way will it be possible to answer the question: do mirror neurons functionally participate in understanding the actions of others? The present study does not answer that question.

   We would like to reiterate that studies aimed at elucidating what information is contained in which neurons or areas are necessary to understand neural network processes and are a prerequisite for conducting well-considered experiments that measure behavioral effects through specific manipulation of the neural network. Without the work of Gallese, Rizzolatti and colleagues, the idea of associating F5 neurons with action understanding would not have occurred in the first place. The current tricky question is whether at all, and if so, to what understanding, to what perception, to what behavior that uses information about mental states of another, F5 MNs might be able to contribute. And for this, it helps to have a clearer idea of what information is contained in F5 MNs during action observation.

   Read the original source
3. 

   eLife

   Jun 30, 2023

   eLife assessment

   The mechanisms underlying mirror neurons are a topic of wide interest for all who study the workings of the brain. The authors use an elegant decoding approach to test whether mirror neurons encode action categories in the same framework regardless of whether actions are executed in the dark or observed in the light. This new approach identifies only a small subset of mirror neurons with fully matched coding among a larger set showing partial matches. The thought-provoking study opens up new principled avenues to probe the mechanics of matching action and perception.

   Read the original source
4. 

   eLife

   Jun 30, 2023

   Reviewer #1 (Public Review):

   The authors set out to investigate the hypothesis that mirror neurons in ventral premotor area F5 code actions in a common motor representation framework. To achieve this, they trained a linear discriminant classifier on the neural discharge of three types of action trials and test whether the thus trained classifier could decode the same categories of actions when observed. They showed that codes were fully matched for a small subset of neurons during the action epoch, while a wider set of "mirror neurons" showed only poorly matched codes for different epochs.

   The authors controlled for potential visual object confounds by having identical objects be manipulated in three different ways and by having the animal carry out the motor execution in the dark. The main strength of the study lies in the clever …

   Reviewer #1 (Public Review):

   The authors set out to investigate the hypothesis that mirror neurons in ventral premotor area F5 code actions in a common motor representation framework. To achieve this, they trained a linear discriminant classifier on the neural discharge of three types of action trials and test whether the thus trained classifier could decode the same categories of actions when observed. They showed that codes were fully matched for a small subset of neurons during the action epoch, while a wider set of "mirror neurons" showed only poorly matched codes for different epochs.

   The authors controlled for potential visual object confounds by having identical objects be manipulated in three different ways and by having the animal carry out the motor execution in the dark. The main strength of the study lies in the clever decoding approach testing the matched tuning to behavioural categories in a model-free way. The central result is in the identification of the small sub-group of mirror neurons that show true matching during the execution epoch, which can dissociate the three types of action almost perfectly. This aligns well with some previous work while offering a novel avenue to identify and investigate those neurons.

   The underlying neuronal mechanism and behavioural relevance of these neurons remain an open question. It would have been interesting to understand better whether the specific motor representations at a recording site, for instance identified through microstimulation prior to recording (see Methods), the reaction times on individual trials or the specific gaze targets (object/hand) had a bearing on the decoding performance for a neuron/trial. Ultimately, the uncovered matched mirror representations should in future experiments be tested with causal interventions and linked trial-by-trial to action selection performance.

   The authors put the focus of their discussion on the wider, less well-matched neuronal pool to support an action selection framework, which is of course a valid view and well established in motor representations. From a sensory perspective, sparse coding, as suggested by the small group of "true" mirror neurons identified with the decoding approach, should also be considered as the basis for a possible neuronal mechanism. A particular strength of the paper is that it could give new data and impetus to the important discussion about how motor and sensory coding frameworks come together in cortical processing.

   Read the original source
5. 

   eLife

   Jun 30, 2023

   Reviewer #2 (Public Review):

   The paper by Pomper and coworkers is an elegant neurophysiological study, generally sound from a methodological point of view, which presents extremely relevant data of considerable interest for a broad audience of neuroscientists. Indeed, they shed new light on the mirror mechanism in the primate brain, trying to approach its study with a novel paradigm that successfully controls for some important factors that are known to impact mirror neuron response, particularly the target object. In this work, a rotating device is used to present the very same object to the monkey or the experimenter, in different trials, and neurons are recorded while the monkey (motor response) or the experimenter (visual response) performed a different action (twist, shift, lift) cued by a colored LED.

   The results show that there …

   Reviewer #2 (Public Review):

   The paper by Pomper and coworkers is an elegant neurophysiological study, generally sound from a methodological point of view, which presents extremely relevant data of considerable interest for a broad audience of neuroscientists. Indeed, they shed new light on the mirror mechanism in the primate brain, trying to approach its study with a novel paradigm that successfully controls for some important factors that are known to impact mirror neuron response, particularly the target object. In this work, a rotating device is used to present the very same object to the monkey or the experimenter, in different trials, and neurons are recorded while the monkey (motor response) or the experimenter (visual response) performed a different action (twist, shift, lift) cued by a colored LED.

   The results show that there is a small set of neurons with congruent visual and motor selectivity for the observed actions, in line with classical mirror neuron studies, whereas many more cells showed temporally unstable matched or even completely non-matched tuning for the observed and executed actions. Importantly, the population codes allow to accurately decode both executed and observed actions and, to some extent, even to cross-decode observed actions based on the coding principles of the executed ones.

   In my view, however, the original hypothesis that an observer understands the actions of others by the activation of his/her motor representations of the observed actions constitutes circular reasoning that cannot be challenged or falsified, as the author may want to claim. Indeed, 1) there is no causal evidence in the paper favoring or ruling out this hypothesis (and there couldn't be), 2) there is no independent definition (neither in this paper nor in the literature) of what "action understanding" should mean (or how it should be measured). Instead, the findings provide important and compelling evidence to the recently proposed hypothesis that observed actions are remapped onto (rather than matched with) motor substrates, and this recruitment may primarily serve, as coherently hypothesized by the authors, to select behavioral responses to others (at least in monkeys).

   1. One of the main problems of this manuscript is, in my view, a theoretical one. The authors follow a misleading, though very influential, proposal, advanced since the discovery of mirror neurons: if there are (mirror) neurons in the brain of a subject with an action tuning that is matched between observation and execution contexts, then the subject "understands" the observed action. This is clearly circular reasoning because the "understanding" hypothesis uniquely derives from the neuron firing features, which are what the hypothesis should explain. In fact, there is no independent, operational definition of the term "understanding". Not surprisingly there is no causal evidence about the role of mirror neurons in the monkey, and the human studies that have claimed to provide causal evidence of "action understanding" ended up using, practically, operational definitions of "recognition", "match-to-sample", "categorization", etc. Thus, "action understanding" is a theoretical flaw, and there is no way "to challenge" a theoretical flaw with any methodologically sound experiment, especially when the flaw consists of circular reasoning. It cannot be falsified, by definition: it must simply be abandoned.
      On these bases, I strongly encourage the authors to rework the manuscript, from the title to the discussion, by removing any useless attempt to falsify or challenge a circular concept and, instead, constructively shed new light on how mirror neurons may work and which may be their functional role.

   2. An important point to be stressed, strictly related to the previous one, concerns the definition of "mirror neuron". I premise that I am perfectly fine with the definition used by the authors, which is in line with the very permissive one adopted in most studies of the last 20 years in this field. However, it does not at all fulfill the very restrictive original criteria of the study in which "action understanding" concept was proposed (see Gallese et al. 1996 Brain): no response to object, no response to pantomimed action or tool actions, activation during execution in the dark and during the observation of another's action. If the idea (which I strongly disagree with) was to simply challenge a (very restrictive) definition of mirroring (a very out-of-date one, indeed, and different from the additional implication of "action understanding"), the original definition of this concept should be at least rigorously applied. In the absence of additional control conditions, only the example neuron in Figure 2A could be considered a mirror neuron according to Gallese et al. 1996. Permissive criteria implies that more "non-mirror" neurons are accepted as "mirror": simply because they are permissively named "mirror", does not imply they are mirroring anything as initially hypothesized (Example neuron in Fig 2B, for example, could be related to mouth, rather than hand, movements, since it responds strongly and similarly around the reward delivery also during the observation task, when the monkey should be otherwise still). Clearly, these concerns impact all the action preference analyses. To practically clarify what I mean, it should be sufficient to note that 74% (reported in this study) is the highest percentage ever reported so far in a study of neurons with "mirror" properties in F5 (see Kilner and Lemon 2013, Curr Biol) and it is similar to the 68% recently reported by these same authors (Pomper et al. 2020 J Neurophysiol) with very similar criteria. Clearly, there is a bias in the classification criteria relative to the original studies: again, no surprise if by rendering most of the recorded neurons "mirror by definition" then they don't "mirror" so much. I suggest keeping the authors' definition but removing the pervasive idea to challenge the (misleading) concept of understanding.

   3. It would be useful to provide more information on the task. Panel B in Figure 1 is the unique information concerning the type of actions performed by the monkey and the experimenter. Although I am quite convinced of the generally low visuomotor congruence, there are no kinematics data nor any other evidence of the statement "the experimental monkey was asked to pay attention to the same actions carried out by a human actor". First, although the objects were the same, the same object cannot be grasped or manipulated in the same way by a human and a macaque, even just because of the considerable difference in the size of their hands; this certainly changes the way in which monkeys' and experimenter's hands interact with the same object, and this is a quantifiable (but not quantified) source of visuomotor difference between observed and executed actions and a potential source of reduced congruency. Second, there is little information about monkey's oculomotor behavior in the two conditions, which is known to affect mirror neuron activity when exploratory eye movements are allowed (Maranesi et al. 2013 Eur J Neurosci), potentially influencing the present findings: a {plus minus}7 (vertical) and {plus minus}5 (horizontal) window at 49 cm implies that the monkey could explore a space larger than 10 cm horizontally and 14 cm vertically, which is fine, but certainly leaves considerable freedom to perform different exploratory eye movements, potentially different among observed actions and hence capable to account for different "attention" paid by the monkey to different conditions and hence a source of neural variability, in addition to action tuning.

   4. Information about error trials and their relationship with action planning. The monkey cannot really "make errors" because, despite the cue, each object can be handled in a unique way. The monkey may not pay attention to the cue and adjust the movement based on what the object permits once grasped, depending on online object feedback. From the behavioral events and the times reported in Table 1, I initially thought that "shift" action was certainly planned in advance, whereas "lift" and "twist" could in principle be obtained by online adjustments based on object feedback; nonetheless, from the Methods section it appears that these times are not at all informative because they seem to depend on an explicit constraint imposed by the experimenters (in a totally unpredictable way). Indeed, it is stated that "to motivate the monkey even more to use the LED in the execution task, another timeout was active in 30% (rarely up to 100%) of trials for the time period between touch of object to start moving the object: 0.15 (rarely 0.1) for a twist and shift, 0.35 (rarely 0.3s) for a lift". This is totally confusing to me; I don't understand 1) why the monkey needed to be motivated, 2) how can the authors be sure/evaluate that the monkeys were actually "motivated" in this way, and 3) what kind of motor errors the monkey could actually do if any. If there is any doubt that the monkeys did actually select and plan the action in advance based on the cue, there is no way to study whether the activity during action execution truly reflects the planned action goal or a variety of other undetermined factors, that may potentially change during the trials. Please clarify.

   5. Classification analysis. There seems to be no statistical criterion to establish where and when the decoding is significantly higher than chance: the classifier performance should be formally analyzed statistically. I would expect that, in this way, both the exe-obs and the obs-exe decoding may be significant. Together with the considerations of the previous point 2 about the permissive inclusion criteria for mirror neurons, this is a remarkable (even quite unexpected) result, which would prove somehow contrary to what the authors claim in the title of the paper. The fact that in any classification the "within task" performance is significantly better than the "between task" performance does not appear in any way surprising, considering both the inclusive selection criteria for "mirror neurons" and the unavoidably huge different sources of input (e.g. proprioceptive, tactile, top-down, etc. afferences) between execution and observation. So, please add a statistical criterion to establish and show in the figures when and where the classifications are significantly above chance.

   6. "As the concept of a mirror mechanism posits that the observation performance can be led back to an activation of a motor representation, we restricted this analytical step to a comparison of the exe-obs and the obs-obs discrimination performance". I don't understand the rationale of this choice. The so-called "concept" of mirror mechanism in classical terms posits that mirror neurons have a motor nature and hence their functioning during observation should follow the same principle as during action execution. But this logical consideration has never been demonstrated directly (it is indeed costated by several papers), and when motor neurons are concerned (e.g. pyramidal tract neurons, see Kraskov et al. 2009) their behavior during action observation is by far more complex (e.g. suppression vs facilitation) than that hypothesized for classical "mirror neurons". Furthermore, when across-task decoding for execution and observation code has been used, both in neurophysiological (e.g. Livi et al. 2019, PNAS) and neuroimaging (Fiave et al. 2018 Neuroimage) data, the visual-to-motor direction typical produce better performance than the opposite one. Thus, I don't see any good reason not to show also (if not even just) the obs-exe results. Furthermore, I wonder whether it is considered the possible impact of a rescaling in the single neuron firing rate across contexts, as the observation response is typically less strong than the execution response in basically all brain areas hosting neurons with mirror properties, and this should not impact on the matching if the tuning for the three actions remains the same (e.g. see Lanzilotto et al. 2020 PNAS). The analysis shown in Figures 4 and 5 is, for the rest, elegant and very convincing - somehow surprising to me, as the total number of "congruent" neurons (7.5%) is even greater than in the original study by Gallese et al. (5.4%).

   7. The discussion may need quite deep revision depending on the authors' responses and changes following the comments; for sure it should consider more extensively the numerous recent papers on mirror neurons that are relevant to frame this work and are not even mentioned.

   Read the original source
6. 

   eLife

   Jun 30, 2023

   Reviewer #3 (Public Review):

   Mirror neurons are a big deal in the neuroscience literature and have been for thirty years. I (and many others) remain skeptical of whether they serve the functions often attributed to them - specifically, whether they are motor planning neurons that contribute to understanding the actions of others. Testing their functions, therefore, is of great interest and importance. The present study, however, is not a cogent or convincing test. I do not think this study helps to answer the questions surrounding mirror neurons. It purports to provide a crucial test, that comes out mostly against the mirror neuron hypothesis, but the test has too many weaknesses to be convincing.

   First, consider that the motor tuning and the visual tuning match "poorly." How poor or good must the match be before the mirror neuron …

   Reviewer #3 (Public Review):

   Mirror neurons are a big deal in the neuroscience literature and have been for thirty years. I (and many others) remain skeptical of whether they serve the functions often attributed to them - specifically, whether they are motor planning neurons that contribute to understanding the actions of others. Testing their functions, therefore, is of great interest and importance. The present study, however, is not a cogent or convincing test. I do not think this study helps to answer the questions surrounding mirror neurons. It purports to provide a crucial test, that comes out mostly against the mirror neuron hypothesis, but the test has too many weaknesses to be convincing.

   First, consider that the motor tuning and the visual tuning match "poorly." How poor or good must the match be before the mirror neuron hypothesis is rejected? I do not know, and the study does not help here. Even a "poor" match could contribute significantly to a social perception function.

   Second, the results remind me in some ways of other multi-modal responses in the brain. For example, in the visual area MST, neurons are tuned to optic flow fields that imply specific directions of self-motion. Many of the same neurons are tuned to vestibular signals that also imply specific directions of self-motion. But the optic flow tuning and the vestibular tuning are not perfectly matched. There is considerable slop and complexity in how the two tunings compare within individual neurons. That complexity is not evidenced against multi-modal tuning. Instead, it suggests a hidden-layer complexity that is simply not fully understood yet. Just so here, the fact that the apparent motor tuning and apparent visual tuning match "poorly" is not evidence against both a motor planning and a visual encoding function.

   Third, the animals are massively over-trained in three actions. They perform these actions and see them performed thousands of times toward the same object. Surely, if I were in the place of the monkey, every time I saw the object, I'd mentally imagine all three actions. As I saw a person act on the object, I'd mentally imagine the alternative two actions at the same time. Even if the mirror neuron hypothesis is strictly correct, this experiment might still find a confusion of signals, in which neurons that normally might respond mainly to one action begin to respond in a less predictable way during all three trial types.

   Fourth, the experiment relies on a colored LED that acts as an instructional cue, telling the monkey which action to perform. What is to stop the neurons from developing a cue-sensitive response, as in classic studies from Steve Wise and others in the premotor cortex? Perhaps the neuronal signal that the experimenters are trying to measure is partly obscured by other, complex responses influenced in some manner by the instructional cue?

   Fifth, finally, and most importantly, the fundamental problem with this study is that it is correlational. Studies that purport to test the function of a set of neurons, and do so by use of correlational measurements, cannot provide strong answers. There are always half a dozen different interpretations and caveats, such as the ones I raised here. Both sides of a debate can always spin the results, and the arguments are never resolved. To test the mirror neuron hypothesis properly would require a causal study. For example, lesion area F5 and test if the monkey is less able to discriminate the actions of others. Or, electrically microstimulate in area F5 and test if the stimulation interferes (either constructively or destructively) with the task of discriminating the actions of others. Only in this way will it be possible to answer the question: do mirror neurons functionally participate in understanding the actions of others? The present study does not answer that question.

   Read the original source
7. 

   Version published to 10.1101/2022.02.01.478703v1 on bioRxiv

   Feb 4, 2022