Self-organizing recruitment of compensatory areas maximizes residual motor performance post-stroke

Abstract

Whereas the orderly recruitment of compensatory motor cortical areas after stroke depends on the size of the motor cortex lesion affecting arm and hand movements, the mechanisms underlying this reorganization are unknown. Here, we hypothesized that the recruitment of compensatory areas results from the motor system's goal to optimize performance given the anatomical constraints before and after the lesion. This optimization is achieved through two complementary plastic processes: a homeostatic regulation process, which maximizes information transfer in sensory-motor networks, and a reinforcement learning process, which minimizes movement error and effort. To test this hypothesis, we developed a neuro-musculoskeletal model that controls a 7-muscle planar arm via a cortical network that includes a primary motor cortex and a premotor cortex that directly project to spinal motor neurons, and a contra-lesional primary motor cortex that projects to spinal motor neurons via the reticular formation. Synapses in the cortical areas are updated via reinforcement learning and the activity of spinal motor neurons is adjusted through homeostatic regulation. The model replicated neural, muscular, and behavioral outcomes in both non-lesioned and lesioned brains. With increasing lesion sizes, the model demonstrated systematic recruitment of the remaining primary motor cortex, premotor cortex, and contra-lesional cortex. The premotor cortex acted as a reserve area for fine motor control recovery, while the contra-lesional cortex helped avoid paralysis at the cost of poor joint control. Plasticity in spinal motor neurons enabled force generation after large cortical lesions despite weak corticospinal inputs. Compensatory activity in the premotor and contra-lesional motor cortex was more prominent in the early recovery period, gradually decreasing as the network minimized effort. Thus, the orderly recruitment of compensatory areas following strokes of varying sizes results from biologically plausible local plastic processes that maximize performance, whether the brain is intact or lesioned.

Figure 1.

Model Architecture. A. Anatomical diagram showing M1 (blue), PM (green), CM1 (yellow), as sites undergoing reinforcement learning, and MN (purple) as the site undergoing homeoplastic regulation. Lesions (red arrow) of different sizes are simulated in M1 alone or in both M1 and PM. B. Network diagram illustrating the flow of information in the model. Desired force targets are sent to M1, PM, and CM1, which are RBF networks, with fewer RBF neurons in PM than in M1 (see Methods). Neuronal activations from M1 and PM are focally transmitted to MNs via the CST. CM1 projects to the RF, which further projects via synergistic weights to MNs via the RST. Inputs from CST and RST sum in MN sigmoidal neurons, which transmit activity to the data-driven arm musculoskeletal system, implemented with the linear transformation M, which generates the realized force fr. The synaptic weights within PM, M1, and CM1 (shown as dotted lines) are updated via reinforcement learning (RL). The reward depends on force error and effort. MN neurons undergo homeoplasticity that adjusts the sigmoid gain and threshold. M1: primary motor cortex. PM: premotor cortex. CM1: contralesional primary motor cortex. CST: corticospinal tract. RF: reticular formation. RST: reticulospinal tract. MN: motor neurons. M: linear musculoskeletal system. RL: reinforcement learning. RBF: radial basis functions.

Figure 2.

Performance outcomes of the intact baseline model during training (A) and following training (B, C, and D). A. Changes in outcome measures (strength, directional error, and muscle synergies) with training. The traces and shaded areas indicate the mean and standard deviation of the outcome across 50 simulations. B. Target forces (dashed lines) and realized forces (solid lines), produced by the trained model in the isometric arm reaching task to 24 targets (left), and for comparison, forces produced by a representative neurotypical individual (right) in an isometric arm reaching task to 8 targets. C. Forces associated with muscle synergies generated by the model after training (left), and for comparison, muscle synergy forces generated for the representative neurotypical individual (right). D. Muscle tuning curves. Solid blue and red lines indicate the muscle tuning curves of the model after training and those of the representative neurotypical individual, respectively. The segments show the preferred direction of each muscle – all data were collected in the study by Barradas et al. (2020).

Figure 3.

Recovery of motor function after stroke for three different lesion sizes: 50% M1 lesion (yellow), 100% M1 lesion (blue), and 100% M1 and PM lesion (red). A. Recovery of the three outcome measures: strength, directional errors, and muscle synergies. The traces and shaded areas indicate the mean and standard deviation of the outcome across all simulations. The purple trace shows the outcome measures with no lesion. B, C, D. Forces associated with extracted muscle synergies and realized forces at the end of the recovery period. B. 50% M1 lesion. Performance recovered to pre-stroke levels, corresponding to stage 4. C. 100% M1 lesion. Abnormal synergies were still present at the end of recovery, corresponding to stage 3. D. 100% M1 and PM lesion. Weakness, large directional errors, and abnormal synergies were still present at the end of recovery, corresponding to stage 2.

Figure 4.

Output activity of M1, PM, CM1, and MN neurons following a 50% M1 lesion, a 100% M1 Lesion, and a 100% M1 & PM Lesion. The output activity is quantified as the sum of all outputs of each region. The traces and shaded areas indicate the mean and standard deviation of the summed output activity across all simulations. Stroke occurred at 4*10 trials, the end of training of the baseline model (Figure 2)

Figure 5:

Projection of average sum of neural activity during the isometric task on a brain template in the intact brain (A), and in the early (B, D) and late (C, E) recovery phases following a 50% M1 lesion and a 100% M1 lesion. Early recovery is taken at 10⁴ trials post-stroke and late recovery at 10⁵ trials. Maximum activation is scaled to the maximum recorded. Compare with Figure 2 in Murata et al. 2015.

Figure 6.

Results of learning blocking experiments. A. Recovery of motor function after stroke with all plastic processes except PM. B. Recovery of motor function after stroke with all plastic processes except CM1. C. Recovery of motor function after stroke with all plastic processes except MN. Data is presented in the same format as in Figure 3A.