Dynamic top-down biasing implements rapid adaptive changes to individual movements
Curation statements for this article:-
Curated by eLife
eLife assessment
By recording simultaneously from premotor and primary motor cortical nuclei in singing birds, this paper provides compelling evidence that premotor activity covaries with primary activity with the temporal specificity necessary to promote learning and drive adaptive vocal variation. As the first study to record from two distant sites at once in singing birds, this study also provides exceptional evidence for temporally precise coordination between two motor areas in the service of vocal learning.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Neuroscience (eLife)
Abstract
Complex behaviors depend on the coordinated activity of neural ensembles in interconnected brain areas. The behavioral function of such coordination, often measured as co-fluctuations in neural activity across areas, is poorly understood. One hypothesis is that rapidly varying co-fluctuations may be a signature of moment-by-moment task-relevant influences of one area on another. We tested this possibility for error-corrective adaptation of birdsong, a form of motor learning which has been hypothesized to depend on the top-down influence of a higher-order area, LMAN (lateral magnocellular nucleus of the anterior nidopallium), in shaping moment-by-moment output from a primary motor area, RA (robust nucleus of the arcopallium). In paired recordings of LMAN and RA in singing birds, we discovered a neural signature of a top-down influence of LMAN on RA, quantified as an LMAN-leading co-fluctuation in activity between these areas. During learning, this co-fluctuation strengthened in a premotor temporal window linked to the specific movement, sequential context, and acoustic modification associated with learning. Moreover, transient perturbation of LMAN activity specifically within this premotor window caused rapid occlusion of pitch modifications, consistent with LMAN conveying a temporally localized motor-biasing signal. Combined, our results reveal a dynamic top-down influence of LMAN on RA that varies on the rapid timescale of individual movements and is flexibly linked to contexts associated with learning. This finding indicates that inter-area co-fluctuations can be a signature of dynamic top-down influences that support complex behavior and its adaptation.
Article activity feed
-
-
eLife assessment
By recording simultaneously from premotor and primary motor cortical nuclei in singing birds, this paper provides compelling evidence that premotor activity covaries with primary activity with the temporal specificity necessary to promote learning and drive adaptive vocal variation. As the first study to record from two distant sites at once in singing birds, this study also provides exceptional evidence for temporally precise coordination between two motor areas in the service of vocal learning.
-
Reviewer #1 (Public Review):
Tian et al impressively record from two motor areas at once in singing birds to test if a premotor cortical area, LMAN, covaries with activity in a primary one, RA, in a way that would support learning. They find that LMAN activity covaries with RA activity at a lag consistent with driving a premotor bias and, moreover, that this covariation is significantly increased in the specific time window of the song where bias is being most strongly driven. Disruptive microstimulation of LMAN in this window reduced learning-associated bias. Though the main results in this paper are consistent with dominant models of birdsong production and learning going back decades (e.g. Kao et al, 2005; Olveczky et al, 2005; Andalman et al, 2008; Charlesworth et al, 2009; Fee and Goldberg, 2011), these results provide …
Reviewer #1 (Public Review):
Tian et al impressively record from two motor areas at once in singing birds to test if a premotor cortical area, LMAN, covaries with activity in a primary one, RA, in a way that would support learning. They find that LMAN activity covaries with RA activity at a lag consistent with driving a premotor bias and, moreover, that this covariation is significantly increased in the specific time window of the song where bias is being most strongly driven. Disruptive microstimulation of LMAN in this window reduced learning-associated bias. Though the main results in this paper are consistent with dominant models of birdsong production and learning going back decades (e.g. Kao et al, 2005; Olveczky et al, 2005; Andalman et al, 2008; Charlesworth et al, 2009; Fee and Goldberg, 2011), these results provide methodologically impressive confirmation that LMAN drives RA activity to drive adaptive bias. It's also meaningful that these covariations were strong enough to be picked up by a pair of randomly targeted LMAN and RA sites. This feature of their dataset is not emphasized by the authors but invites more attention, consideration, and discussion, as detailed below.
(1) Song is complex with many varying acoustic parameters, such as amplitude, entropy, and pitch. It is thought that pitch is controlled by only a subset of the syringeal muscles, and also that there is topography in the LMAN-RA-MN-muscle pathway. Thus, one might expect only a small fraction of neurons/sites in the LMAN-RA pathway to be associated with pitch with enough strength that one would pick it up in single unit recordings from order ~100 syllables. Indeed a past study (Sober et al, ) found that activity in a small fraction of RA neurons was weakly correlated with pitch variation. So it's really surprising that a pitch-contingent learning paradigm produced significant co-variance changes in the LMAN-RA pathway that could be picked up in the present study. Three possible explanations are with consideration. First, one wonders if they were recording specifically from pitch-associated sites. Can the authors please elaborate on what fraction of LMAN and RA recording sites in the present study exhibited significant covariance with pitch? Second, if an LMAN-RA recording site pair does not exhibit a significant correlation with pitch but nonetheless exhibits enhanced co-variation in the pre-target window in a pitch shift paradigm, this would support a different interpretation of the results. For example, it's possible that the extent by which LMAN can drive RA is gated by cholinergic inputs from VP, which might signal predicted uncertainty in the song in the precise moment preceding the target time (e.g. Chen and Goldberg, 202; Chen et al, 2019; Puzerey et al, 2018). No new experiments are required here, but these possibilities (or other considerations of the mechanisms by which LMAN-RA covariance is temporally gated) could, in the discussion or elsewhere, motivate future studies. (For example, if Ach-mediated predicted uncertainty is the key to promoting LMAN-RA covariance, then photoactivation of Ach inputs to RA at a moment in song might increase LMAN drive and, secondly, non-pitch-contingent DAF that does not drive explicitly learning associated bias would also be sufficient to promote temporally precise increases in LMAN-RA covariance. Finally, related to these questions, can the authors please check if their recording sites exhibited correlations with pitch (e.g. as in Sober et al, 2008)?
2. Multiple recording sites in both RA and LMAN provide sensible internal controls for the cross-covariance and its increase in the window before bias production in pitch-shift experiments. Can the authors analyze at LMAN-LMAN co-variance in the same way to test for LMAN-RA co-variance to examine intra-LMAN activity co-flucutations? A negative result would support the specificity of the LMAN-RA covariance but a positive result would indicate that within-LMAN dynamics also exhibit interesting learning-related changes.
-
Reviewer #2 (Public Review):
In studying the neural control of action generation there is a presumption that different nodes within a connected neural control circuit contribute differentially to the production of a given gesture. In many cases, these circuits also receive inputs that can bias ongoing motor commands to alter output and therefore the motor gesture itself. Showing the specific role that each of the different areas play in motor control and how inputs might bias motor output is challenging. Taking advantage of a precisely controlled error-correction learning task of adult birdsong, Tian et al. perform simultaneous neural recording in both the primary forebrain song motor output nucleus (RA) as well as in an input structure (LMAN) known to be necessary for biasing motor output during such learning tasks. By comparing the …
Reviewer #2 (Public Review):
In studying the neural control of action generation there is a presumption that different nodes within a connected neural control circuit contribute differentially to the production of a given gesture. In many cases, these circuits also receive inputs that can bias ongoing motor commands to alter output and therefore the motor gesture itself. Showing the specific role that each of the different areas play in motor control and how inputs might bias motor output is challenging. Taking advantage of a precisely controlled error-correction learning task of adult birdsong, Tian et al. perform simultaneous neural recording in both the primary forebrain song motor output nucleus (RA) as well as in an input structure (LMAN) known to be necessary for biasing motor output during such learning tasks. By comparing the activity pattern and timing between recorded activity in both structures, they show that LMAN activity leads RA activity for each of the song syllables but that there is a preferential gain in activity level in LMAN after learning only during the precise time window (10 - 50 msec) associated with the specific syllable that is targeted during the error-correction paradigm. They then follow these recordings with short focal electrical stimulation in LMAN targeted to the precise time window that shows increased gain in the dual recording paradigm. This stimulation is intended to scramble the bias signal and they show that such manipulation, in a temporally specific manner, does indeed eliminate the acoustic bias imposed by LMAN.
The precise combination of dual recording and targeted stimulation, in my opinion, convincingly shows that LMAN provides a temporally precise command that can bias motor output in RA. It is assumed that LMAN inputs onto RA are mapped with some level of functional topography, especially given that RA is thought to have some degree of motor mapping. The more dorsal areas, for example, likely contribute more to respiratory control while the more ventral portion contributes to acoustic control with a possible acoustic motor map within that region. Unfortunately, the spatial precision of the recording electrodes in both RA as well as LMAN is rather coarse and a careful functional spatial mapping of spike timing correlation is not possible. Hopefully in future studies, more precise spatial mapping will provide correlations within these two structures that might be able to target subareas that encode the signal bias for subcomponents of the specific acoustic features that are being targeted in this error-correction learning paradigm.
-
Reviewer #3 (Public Review):
The present paper uncovers evidence of the coordination of two brain areas involved in a two-step learning process in birdsong plasticity. Indeed, songbirds can modify their song based on an error-correction mechanism that involves a motor bias expressed by a basal ganglia-thalamo-cortical loop. After training (hundreds or a few thousands of renditions), the motor bias necessary to correct vocal errors becomes independent of the BG-thalamo-cortical loop and is transferred into the long-term motor program stored in a primary motor network. Current understanding claims that the output nucleus of the BG-thalamo-cortical loop, LMAN, trains the primary motor networks (in area RA) to drive the learning transfer. However, no clear evidence for such entrainment was available until now. In the present study, the …
Reviewer #3 (Public Review):
The present paper uncovers evidence of the coordination of two brain areas involved in a two-step learning process in birdsong plasticity. Indeed, songbirds can modify their song based on an error-correction mechanism that involves a motor bias expressed by a basal ganglia-thalamo-cortical loop. After training (hundreds or a few thousands of renditions), the motor bias necessary to correct vocal errors becomes independent of the BG-thalamo-cortical loop and is transferred into the long-term motor program stored in a primary motor network. Current understanding claims that the output nucleus of the BG-thalamo-cortical loop, LMAN, trains the primary motor networks (in area RA) to drive the learning transfer. However, no clear evidence for such entrainment was available until now. In the present study, the authors elegantly show that correlations in trial-by-trial fluctuations in the premotor activity in LMAN and RA are present spontaneously (in multi-unit electrophysiological recordings) and are increased during a lab-induced plasticity protocol. The change in correlation is specific to the syllable that undergoes plasticity. Moreover, perturbing LMAN activity through low-intensity and spatially broad electrical stimulation of LMAN during the premotor window prevents behavioral adaptation. Altogether, their results convincingly show that the entrainment of RA neural populations by LMAN neurons is present during baseline, strengthened during plasticity in a syllable-specific manner, and necessary for song plasticity.
This study thus provides important validation of the current model for the 2-step learning process underlying song learning and plasticity, where a BG-thalamo-cortical network drive motor bias to correct vocal errors based on a reinforcement learning mechanism, while the song motor engram is updated slowly through the adjustment of song-related activity in the primary motor areas. Beyond the songbird field, these results will be of importance to all studying sensorimotor learning and adaptation, and more broadly the formation of memory through a two-step learning process.
The authors present the context for their hypothesis clearly, state their hypothesis precisely, and conduct a thorough investigation of the posed question. The conclusions are well supported by data.
In particular, the statistical evaluation of the covariance of LMAN and RA activity in the premotor window is adequate and the interpretation of the results is therefore well backed by their analysis. The methods used here to assess covariation between LMAN and RA activity during singing set the ground for future studies looking at the coordination between brain areas during behavior.
-
