Congruent neuronal modulation across competing actions challenges the role of the substantia nigra in action selection

  1. Lingfeng Hou
  2. Derin Timuçin
  3. Olivia Sato
  4. Sofia Grijalva Torres
  5. Daniel Svedberg
  6. John Assad

  • Curated by eLife

    eLife logo

    eLife Assessment

    This important study records basal ganglia output neurons in the substantia nigra pars reticulata of head-fixed mice performing self-timed reaching and licking tasks. Contrary to simple predictions of action selection and inhibitory gating by basal ganglia output, the authors report that most substantia nigra pars reticulata neurons modulate congruently across reaching, licking, and reaches to different targets. Optogenetic identification of substantia nigra pars reticulata neurons projecting to the central and lateral, orofacial superior colliculus shows that these congruent signals are routed to both output zones. Evidence from the recordings is convincing, while that from the projection-specific experiments is solid but rests on few animals and would be strengthened by increased power and clearer criteria for unit identification.

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

The basal ganglia are involved in the control of movement, but their exact role is unclear. Paradoxically, most of the inhibitory projection neurons in the main output nuclei increase firing around the time of movement; only a small fraction decrease firing. This antagonistic activity pattern could subserve action selection, with the small “decrease” population selectively disinhibiting the desired movement, and the larger “increase” population inhibiting competing movements. The action-selection hypothesis makes an implicit assumption: neurons that decrease firing to disinhibit a specific action should increase firing to inhibit that action when a different action is desired. To test this hypothesis, we recorded projection neurons in the substantia nigra pars reticulata (SNr) of mice trained to alternate between two different types of movements. Many SNr neurons showed a “ramping” pattern of pre-movement firing-rate modulation, with most neurons increasing firing, consistent with previous findings. However, contrary to the action-selection model, the overwhelming majority of SNr neurons exhibited congruent modulation between the competing actions, either increasing or decreasing their firing rates for both actions; only a small fraction of neurons exhibited opposite signs of modulation. Similar results were observed among SNr neurons identified as projecting to the arm or orofacial zones of motor superior colliculus. In addition, brief pauses in SNr firing outside explicit operant tasks were not accompanied by stereotyped movements. Our results are not easily reconciled with simple antagonistic mechanisms for action selection in the basal ganglia output nuclei. We also found that ramping activity in SNr neurons typically began hundreds of ms before self-timed and spontaneous movements, in contrast to previous findings suggesting that basal ganglia output is modulated too late to be involved in movement initiation. Our findings suggest constraints – and raise new questions – about the role of the basal ganglia in movement initiation and action selection.

Article activity feed

  1. Version published to 10.7554/elife.112772.1 on eLife
  2. Version published to 10.7554/elife.112772 on eLife
  3. eLife

    eLife Assessment

    This important study records basal ganglia output neurons in the substantia nigra pars reticulata of head-fixed mice performing self-timed reaching and licking tasks. Contrary to simple predictions of action selection and inhibitory gating by basal ganglia output, the authors report that most substantia nigra pars reticulata neurons modulate congruently across reaching, licking, and reaches to different targets. Optogenetic identification of substantia nigra pars reticulata neurons projecting to the central and lateral, orofacial superior colliculus shows that these congruent signals are routed to both output zones. Evidence from the recordings is convincing, while that from the projection-specific experiments is solid but rests on few animals and would be strengthened by increased power and clearer criteria for unit identification.

  4. eLife

    Reviewer #1 (Public review):

    Summary:

    The manuscript describes results from a series of experiments designed to test different facets of the gating model and potential alternative explanations. In most respects, the results are squarely inconsistent with the gating model. Some of the authors' attempts to test the model force the authors to define the model in greater detail than even the model's proponents typically do. That is a revealing exercise in itself, which is well worth bringing to the field's attention. The manuscript is well written for the most part, and figures illustrate results clearly.

    Strengths:

    The gating model states that pauses in the activity of a select subset of SNr neurons disinhibit (i.e., "open the gate" to) a specific desired action, whereas a larger population of SNr neurons increase their firing so as to suppress other potentially conflicting actions. Selection of different actions is performed by the generation of pauses and increases in activity in different subsets of SNr neurons. Contrary to that prediction, the authors report that SNr neurons that show a reduction in firing for one action tend to show reductions in activity for other actions as well and, similarly, increase-type neurons tend to show increases for all of the actions examined. The authors tested the generality of these observations across a series of experiments that compared activity between markedly different actions (moving forelimb vs. tongue), similar actions (different directions of forelimb movement) and several forms of self-directed and spontaneous movement. Moreover, the authors examined the activity of anatomically identified subsets of SNr neurons that innervate forelimb and orofacial sub-regions of the superior colliculus. Combined, the series of experiments described here provide overwhelming evidence against any straightforward "simple" form of the gating model of basal ganglia function.

    Weaknesses:

    The most important limitations of the manuscript actually arise from additional ambiguities of the gating model - ambiguities that may be worth mentioning in the manuscript. For example, how granular is the action selection process supposed to be? Perhaps the SNr circuits examined in the present study are selecting "actions" defined at a level that the authors did not examine. This is not a major limitation, however, given the concerted effort invested in testing different levels of granularity and contexts. This relates to another ambiguity of the model in that basal ganglia output does not affect actions directly, but only indirectly via its influences on downstream multi-synaptic neural circuits.

    Another ambiguity of the model is how to interpret the activity of neurons that don't respond. In other words, are comparisons of decrease- and increase-type responses the appropriate comparisons to make to test the model? The present study compares decrease- vs. increase-type responses, but perhaps no-response and increase-type neurons should be lumped together. After all, the simplest version of the action selection hypothesis (cf. Hikosaka) considers the high baseline tonic activity of SNr neurons to be action suppressive by itself, without an additional task-related increase. A comprehensive action selection model should include an explanation of the relative roles of non-response and increase-type responses. The current manuscript would benefit from a discussion of this ambiguity and a rationale for the decisions the authors made to make the model testable.

    A third ambiguity of the model is how to interpret graded changes in SNr firing rate. It comes as no surprise that the authors observed slow ramp-like changes in SNr firing rate prior to movement initiation in the absence of an immediate sensory trigger. The authors interpret these ramp-like changes as being equivalent (from the perspective of the gating model) to the abrupt changes in firing rate observed around the time of initiation for movements that are sensory-triggered. But the simple gating model presents basal ganglia output as having only three states: no-change, pause, and increase. Implied in the model is that these changes in firing rate influence not only what action is performed but also when it is performed. It is not clear that slow ramping responses qualify as "open the gate" disinhibitory temporally-abrupt signals. The manuscript could provide more explanation concerning the position the authors take to make ramping responses fit into the gating model.

  5. eLife

    Reviewer #2 (Public review):

    The manuscript by Hou et al. tests the hypothesis that SNr neurons' activity is inhibited during specific movements, allowing the selection of specific movements. They do so by performing electrophysiological recordings in SNr, in head-fixed mice trained to perform self-paced reaching or self-paced licking. They first show that SNr neurons present ramping activity hundreds of milliseconds before reaching, with the vast majority of cells presenting positive increases in activity in these rampings. Then they show that the activity of these cells is related to the velocity (and ramping) of movements, independently of the timing to initiate. They verify these findings in spontaneous reaching. To then test the activity of the same cells in self-paced reaching vs. licking, they found that the vast majority of cells present congruent firing between conditions (not opposite firing as expected). Further, they ran controls asking whether dips in activity are related to specific movements, checked if the specific cells that are correlated with lick cycles respond differently in self-paced reaching vs. licking, asked whether there could be differences in activity in reaching to different targets, to finally "opto-tag" SNr- orofacial vs SNr-arm cells and asked whether these differentiate their spiking in reaching vs. licking, concluding that a very minority of SNr neurons distinguished between movements while the vast majority presented congruent modulations among movements.

    Strengths:

    The study is a tour de force, testing the general assumption of action selection being reflected in SNr spiking.

    Weakness:

    Even though the authors set up an amazing set of experiments to test the mentioned hypothesis, their final opto-tagging experiment needs to be screened to better support their conclusion.

  6. eLife

    Reviewer #3 (Public review):

    Summary:

    The authors aimed to answer the question of whether the basal ganglia, specifically neurons in the substantia nigra pars reticulata (SNr), directly select actions. This was done by recording the activity of SNr neurons, as well as neurons in the superior colliculus (a major recipient of SNr projections), during cued and self-directed actions, specifically recording the same SNr units during clearly distinct actions.

    Strengths:

    As the major output nucleus of the rodent basal ganglia, the SNr is a key location to determine if single-cell activity within the basal ganglia is specific to the selection or suppression of specific actions. This choice also provides a natural transition to the activity/roles of downstream motor nuclei that receive basal ganglia output. Multiple definitions of "action" were addressed throughout this manuscript, which is essential for an ambiguous concept like "action selection." They convincingly demonstrate that a "simple action selection hypothesis" in which there is a 1:1 correspondence between SNr single-unit activity and specific actions is incompatible with their data. The manuscript was clearly written and well-organized.

    Weaknesses:

    While the authors systematically address multiple definitions of an "action," all of the actions in this work are variations of either licking or reaching. The similar SNr activity during different types of reaches has several potential interpretations, which the authors enumerate and present arguments for why their interpretation (SNr does not implement action selection) is the most plausible. However, the study does not quite fully eliminate these possibilities. Finally, while the authors essentially eliminate the possibility of a 1:1 correspondence between SNr units and specific actions, it remains possible that SNr selects actions through a more complex population coding scheme. Indeed, they demonstrate that they can distinguish lick and reach actions by examining population SNr activity.

    Appraisal of whether authors achieved their aims/evidence supports conclusions:

    The authors achieved their aims of determining whether individual SNr neurons select for specific actions. This includes multiple definitions of "actions," with some actions differentiated by muscle group and body part (licking vs reaching), target, or timing. The evidence provided largely supports the authors' claims, though the low number of subjects in the causal optogenetic experiments makes interpretation of those results more difficult.

    Discussion of likely impact of the work in the field:

    This work will provide a foundation for future research into more nuanced dissections of basal ganglia circuitry, specifically between the SNr and its downstream target nuclei.

  7. eLife

    Author response:

    Thanks to the reviewers, reviewing editor and senior editor for their assessment of our manuscript. We summarize and address reviewer comments and suggestions below:

    (1) The granularity of action selection is ambiguous (Reviewers 1 & 3)

    (2) Consider neurons not modulated by movement (Reviewer 1)

    (3) How to interpret ramp-like vs. abrupt changes in pre-movement firing rates (Reviewer 1)

    (4) Population encoding could be considered a form of action selection (Reviewer 3)

    (5) Improve the sample-size and screening of optotagging results (Reviewers 2 & 3)

    (1) The granularity of action selection is ambiguous (Reviewers 1 & 3)

    The reviewers astutely pointed out a critical ambiguity of the original action selection model: on what level are actions differentiated in the basal ganglia? Is it activation of specific muscles, or concerted movements of multiple body parts, or even complex sequences of behavior? Anatomically, the organization of striatonigral input as well as nigrocollicular pathways appear to be highly somatotopic (Foster et al., 2021; Lee et al., 2020), in the sense that there are parallel circuits that separately project to orofacial or skeletomotor regions of the superior colliculus (Benavidez et al., 2021). This might suggest an upper bound for the granularity of action selection in SNr, manifest at the level of tongue/jaw vs. limbs, different directions of reaching, or reaching vs. retraction. We tested these permutations and found little evidence for sign reversals in SNr responses. Nonetheless, one could always parse “action” into even finer components, or combine them into more complex ensembles or sequences of movements. It was exactly this logic that led us to test more and more movements, starting with different directions of reaching and culminating in the unbiased examination of spontaneous movement (Fig. 5, 6, 7). Nonetheless, we of course could not exhaustively exclude all movement possibilities/permutations, and our arguments are limited to the sets of actions that we tested. But we also point out that the granularity argument could be extended ad infinitum, to the point of being practically unfalsifiable! In the end, we will lay this logic out more clearly in the revised Discussion. (On the other hand, it is exactly this “granularity” conundrum that led us away from trying yet another movement paradigm, and toward the idea of examining functional connectivity between the SNr and SC, as outlined below.)

    (2) Consider neurons not modulated by movement (Reviewer 1)

    Reviewer 1 pointed out that unmodulated SNr neurons could be grouped with increase neurons, since they could maintain or increase inhibitory output in order to inhibit unwanted movements. Upon further inspection, we found 56/427 (13%) SNr units decreased firing for reaching but not for licking; conversely, 12/427 (3%) units decrease firing for licking but not for reaching; while 334/427 (78%) units increased firing or were unmodulated for both reaching and licking. In our original analysis excluding unmodulated SNr units, the proportions of units with congruent (86%) vs. incongruent (14%) signs of modulation for reaching and licking did not differ drastically when we included unmodulated units (78% vs. 16%). Nonetheless, this is a valid point, and we will include this in the revised Results and Discussion.

    (3) How to interpret ramp-like vs. abrupt changes in pre-movement firing rates (Reviewer 1)

    Reviewer 1 questioned how the slow ramp-like modulation of SNr firing relates to the disinhibitory gating model. We adopted the self-timed/spontaneous movement task not only to separate sensory vs. motor activity, but also due to clinical observations in Parkinson’s disease that patients are more debilitated for self-initiated rather than cued movements. We observed early pre-movement neural modulation in SNr during self-initiated movements, whereas Mink and Thach (1991) argued that pallidal activation was too late for movement initiation – albeit only in the context of cued movements. Slow ramp-like changes SNr firing does not contradict action selection per se, since modulation of SNr firing might need to reach some threshold level for a desired movement to occur, even in the context of action selection. On the other hand, these ramp-like changes might also represent a generic movement initiation signal without specifying which action to take. We pointed out the existence of ramp-like modulation during self-initiated movements, but the ramp-like nature of SNr modulation is not fundamentally incompatible with the gating model. We will clarify this point in the revised Discussion.

    (4) Population encoding could be considered a form of action selection (Reviewer 3)

    The reviewers also pointed out that while action selection is largely absent in single SNr neurons, licking vs. reaching could be efficiently decoded from population activity. Reviewer 3 reasonably asked: could population encoding be a form of action selection?

    The answer, of course, is yes – but by the same token, any motor area that displays selectivity (motor cortex, thalamus, cerebellum, brainstem, etc.) could be argued to play a role in action selection. So what’s special about the basal ganglia? In this regard, it’s important to re-examine the basis for the action-selection hypothesis. The original model was put forward by Mink & Thach in 1991 to explain the paradoxical observation that most inhibitory pallidal output neurons increase rather than decrease firing during movement. The authors speculated that this increase population could inhibit unwanted movements, while the smaller decrease population could disinhibit the desired movement. The model has been tweaked here and there through the years, but the fundamental core of the idea has persisted, with little supporting evidence, presumably because it provides an explanation for the paradoxical increase neurons – an interesting explanation to be sure. However, if neurons that increase firing do so to inhibit an unwanted movement, then the simplest “other side of the coin” explanation is that they should decrease firing when that same movement is desired. Yet this sign-reversal motif was exceedingly rare among the SNr neurons in our study. Thus the critical question remains: why do so many pallidal output neurons increase firing with movement? We hope that one useful outcome of our study will be to stimulate others to think about pallidal responses outside of the action-selection framework.

    Nonetheless, one could certainly argue that action selectivity is a prerequisite for action selection, and it may be worth asking how the SNr selectivity could contribute to this role. The SNr projects broadly to multiple brainstem and thalamic motor nuclei, which in turn affect specific movements. One could certainly devise wiring schemes whereby SNr outputs is parsed out among those downstream targets to mediate action selection. But these schemes are not exclusive to the SNr; they could pertain to any motor area with selectivity, nor do they provide an explanation for the paradoxical increase neurons. Nonetheless, R3’s thoughts likely reflect those of other readers, and we will try to lay out our logic more clearly in the revised Introduction and Discussion.

    (5) Improve the sample-size and screening of optotagging results (Reviewers 2 & 3)

    We used an optotagging technique to identify SNr neurons projecting to arm vs. orofacial SC and demonstrated that – during both reaching and licking – SNr neurons predominantly increased their inhibitory output regardless of their putative projection target. We think these results have important implications for the role of SNr output. However, the reviewers pointed out some critical limitations of our optotagging approach. Typically, to rule out polysynaptic confounds, light-evoked spikes are expected to have very low latency (typically a few milliseconds, but highly dependent on light power and opsin efficacy). However, it is unclear where the threshold for spiking latency should be drawn. Instead, to rule out polysynaptic confounds in our study, we relied on the fact we observed extremely few (<5%) silenced SNr neurons during optogenetic stimulation of a predominantly GABAergic region. Because SNr neurons send collateral GABAergic projections to other SNr neurons, optogenetic activation of a neighboring neuron should synaptically inhibit (rather than excite) a neuron of interest. Another possible source of contamination would have come from light leaking to into SC, which could have polysynaptically excited SNr neurons through the collicular-thalamic-striatal pathway. In this case, direct pathway striatal projection neurons should also suppress SNr firing. However, we did not observe any light-evoked inhibition among hundreds of putative SNr neurons, perhaps because viral labeling of neurons was sparse enough that polysynaptic inhibition was rare and largely unable to overcome optogenetic excitation. Therefore we did not suspect polysynaptic contamination among our results. Nonetheless, even after we had restricted our analysis to SNr units with low (<median) optogenetic responses latencies, the fraction of SNr units negatively modulated for reaching or licking was still extremely low (4% for reach among putative arm-SC projecting neurons; 8% for lick among putative orofacial-SC projecting neurons). We plan to collect additional data to address concerns regarding the smaller sample size.

    References:

    Benavidez, N. L., Bienkowski, M. S., Zhu, M., Garcia, L. H., Fayzullina, M., Gao, L., Bowman, I., Gou, L., Khanjani, N., Cotter, K. R., Korobkova, L., Becerra, M., Cao, C., Song, M. Y., Zhang, B., Yamashita, S., Tugangui, A. J., Zingg, B., Rose, K., . . . Dong, H.-W. (2021). Organization of the inputs and outputs of the mouse superior colliculus. Nature Communications, 12(1), 4004. https://doi.org/10.1038/s41467-021-24241-2

    Foster, N. N., Barry, J., Korobkova, L., Garcia, L., Gao, L., Becerra, M., Sherafat, Y., Peng, B., Li, X., Choi, J.H., Gou, L., Zingg, B., Azam, S., Lo, D., Khanjani, N., Zhang, B., Stanis, J., Bowman, I., Cotter, K., . . . Dong, H.-W. (2021). The mouse cortico–basal ganglia–thalamic network. Nature, 598(7879), 188194. https://doi.org/10.1038/s41586-021-03993-3

    Lee, J., Wang, W., & Sabatini, B. L. (2020). Anatomically segregated basal ganglia pathways allow parallel behavioral modulation. Nature Neuroscience, 23(11), 1388-1398. https://doi.org/10.1038/s41593-020-00712-5

  8. Version published to 10.1101/2025.02.18.638935v5 on bioRxiv
  9. Version published to 10.1101/2025.02.18.638935v4 on bioRxiv
  10. Version published to 10.1101/2025.02.18.638935v3 on bioRxiv
  11. Version published to 10.1101/2025.02.18.638935v2 on bioRxiv
  12. Version published to 10.1101/2025.02.18.638935v1 on bioRxiv