Cyclic, condition-independent activity in primary motor cortex predicts corrective movement behavior

Abstract

SummaryCyclic neural activity occurred not only during initial reaching movements but also subsequent corrective movements as subjects performed a precision center-out task. The cyclic trajectories identified from the condition-independent neural activity were similar between initial and corrective movements. For both, instantaneous cycle phase predicted when contact with the target occurred, even when the subject made multiple corrective movements. Moreover, the phase of the cyclic neural activity correlated with instantaneous movement speed. Neural activity during these corrective movements thus reveals encoding of discrete sub-movements rather than smooth and continuous updating of movement towards a target.

Rotational dynamics in motor cortex are consistent with a feedback controller

“…Critically, similar rotational dynamics were generated whether the networks received relatively simple inputs or inputs that simulated online sensory feedback. Although both studies leave open the question about how these recurrent dynamics are generated, whether it is through intrinsic (e.g., local synapses) or extrinsic factors (e.g., sensory feedback), these studies have been interpreted as evidence that these dynamics are generated from intrinsic connectivity in MC ( Bizzi and Ajemian, 2020 ; Collinger et al, 2018 ; Diedrichsen and Kornysheva, 2015 ; Kalaska, 2019 ; Khanna et al, 2021 ; Pandarinath et al, 2018a ; Perich et al, 2020 ; Rouse et al, 2018 ; Suminski et al, 2015 ; Suresh et al, 2020 ; Vyas et al, 2020 ).…”

“…Thus, activity was generated solely by the intrinsic connections between neurons, and online feedback about the generated muscle patterns was not necessary after training ( Sussillo et al, 2015 ). Collectively, the findings that MC dynamics are well described by a deterministic dynamical system and that RNNs with dynamics dominated by intrinsic connections can approximate MC dynamics have led to a common interpretation that these dynamics reflect a pattern generator for muscle activity and that this real-time process is done largely autonomously from other brain structures and sensory feedback ( Bizzi and Ajemian, 2020 ; Collinger et al, 2018 ; Diedrichsen and Kornysheva, 2015 ; Kalaska, 2019 ; Khanna et al, 2021 ; Pandarinath et al, 2018a ; Perich et al, 2020 ; Rouse et al, 2018 ; Suminski et al, 2015 ; Suresh et al, 2020 ; Vyas et al, 2020 ).…”

M1 dynamics share similar inputs for initiating and correcting movement

“…We hypothesized that these transients might correspond to corrective submovements, which are common in the RTP task (Fig. 6A) and which have been suggested to coarsely share M1 activity patterns with initial movements (Rouse, 2018).…”

“…One approach to understanding the pattern generation aspects of this activity has been through the lens of dynamical systems analysis Seely et al, 2016;Perich et al, 2020;Rouse, 2018;Sauerbrei et al, 2020;Vyas et al, 2020) particularly in reaching primates. In the context of brief and largely open-loop reaching, dynamical systems have been used to show that motor cortex can be approximated well as having rotational dynamics or a variant thereof (Hennequin et al, 2014;Russo et al, 2018;Sabatini & Kaufman, 2021), and is initialized by movement-specific inputs (Churchland et al, , 2010 and then triggered by a large, external, condition-invariant signal (Kaufman et al, 2016;Michaels et al, 2015).…”