Post stroke rehabilitation improves motor performance
To understand the neuronal circuit underlying functional recovery induced by rehabilitation, we developed a rehabilitation paradigm in the mouse that engages repetitive, skilled forelimb use by the affected limb, imitating rehabilitation for human stroke patients12,32 (Fig. 1a and Supplementary Fig. 1). Such rehabilitation enables intense reach-to-grasp training ( > 800 reaches/day). We tested the effect of this rehabilitation paradigm on functional recovery after stroke in primary motor cortex (M1, caudal forelimb area, CFA, Supplementary Fig. 2) using two motor tests, the skilled forelimb reaching test and the grid walk test, which assess motor performance in reach-to-grasp behavior and innate walking precision, (Fig. 1b). We found that stroke significantly decreased the success rate of skilled reaching (Fig. 1c) and diminished gait function (Fig. 1d). Skilled reach rehabilitation completely recovers the motor performance in the skilled reaching task (Fig. 1c, e) and also improved functional recovery in gait (grid walk) (Fig. 1d,f), indicating that intensive rehabilitative therapy promotes motor recovery at or near non-stroke motor performance.
Motor recovery after stroke is associated with plasticity in brain areas anatomically connected to the stroke site. The ipsilesional premotor cortex and the isotopic area to the stroke site in the contralesional hemisphere are regions well-known for contributing to motor recovery33,34. To identify the brain area most important to motor recovery, we tested the functional significance of the rostral forelimb area (RFA) within ipsilesional premotor cortex and the contralesional CFA by inducing a second stroke in these sites after successful rehabilitation (Supplementary Fig. 3a). A second stroke in ipsilesional RFA eliminated rehabilitation-induced behavioral gains in both tasks, while a contralesional stroke did not cause significant changes (Supplementary Fig. 3b, e). These results indicate that intracortical circuits in RFA (premotor cortex) are a site of rehabilitation-induced recovery.
Post stroke rehabilitation restores functional connectivity in specific circuits
Execution of precise motor tasks requires neuronal population activity in the motor and premotor cortex35,36. Stroke causes not only neuronal death in the infarct core but also cortical network uncoupling characterized by less synchronized firing activity. This uncoupling, quantified as decreased functional connectivity, is closely associated with behavioral deficits28,37,38. We assessed neuronal activity and functional connectivity in the main neurons with intracortical connections (layer 2/3 neurons) in the RFA, employing a two-photon microscope equipped with a rotating grid wheel (Fig. 2a, b). Previous studies demonstrated that walking behavior on a rotating grid sensitively detects motor deficits and network uncoupling after a stroke28,39. Rehabilitation-induced stroke recovery is present in both skilled reach and grid-walking tests. However, the rehabilitation activity in these studies is repetitive skilled reach. Using the same behavioral paradigm for training and assessment could introduce confounding factors, as post-stroke training can lead to compensatory movements that affect outcomes40,41,42. Therefore, we selected the grid walk paradigm over the reaching task for the calcium imaging studies. Stroke reduced the number of active neurons, calcium transient frequency, connection number, and connection density in these neurons, though they are distant from the infarct (Fig. 2c–f, the distance from the infarct to the center of the imaging field is 925 ± 25 µm. n = 6). This reduction signifies a deterioration in neuronal activity and functional connectivity. While the number of active neurons spontaneously increased over time, both the frequency and connection density remained depressed 28 days post-stroke (Fig. 2d, f). Rehabilitation yielded a significant increase in the number of active neurons and functional connections. Rehabilitated animals showed no significant decrease in frequency or connection density compared to sham animals. To further characterize the functional connectivity changes, we calculated the connection probability of a single neuron as the ratio of functional connections to the theoretical maximum connection number in the population. We found that rehabilitation reduced the fraction of neurons exhibiting sparse connectivity in RFA and increased the neurons with higher connection probability compared to non-treated stroke animals (Fig. 2g). These results suggest that rehabilitation enhances functional connectivity in severely affected neurons. Our data also revealed a significant correlation between connection density and motor performance in the grid waking (Fig. 2i).
Post stroke rehabilitation restores synaptic inputs to stroke-projecting neurons
The significant neuronal activity and functional connectivity changes after post-stroke rehabilitation suggest synaptic alterations in the RFA as a direct effect of stroke, and as a substrate for rehabilitation. We next investigated whether anatomical connectivity also changes in the RFA with rehabilitation. In both healthy and stroke animals, skilled reach training induces formation of new dendritic spines that serve as the primary sites of synaptic connections14,43. We measured dendritic spine density in the RFA and further identified the source of synaptic inputs by monosynaptic rabies virus tracing. Using an intersectional virus approach, we targeted two types of cortical projection neurons implicated in motor control after stroke: corticospinal neurons (Supplementary Fig. 4a) and stroke-projecting neurons (Fig. 3a, b). Corticospinal neurons represent the direct cortical outputs to the spinal cord and play a pivotal role in motor control44; the integrity of the corticospinal tract is a strong predictor of stroke outcomes45. Conversely, stroke-projecting neurons are defined by their axonal projection to the stroke site, detected by retrograde labeling from the future stroke site (retrograde AAV injection onto the CFA before the stroke in this study). Stroke-projecting neurons lose their projection target and receive retrograde injury signals, which impact neuronal excitability, morphology, and gene expression46, and may also possess unique plasticity. Before conducting synaptic measurements, we labeled these neuron types by injecting retrograde AAV expressing GFP and tdTomato into the M1 and the cervical spinal cord (C7) to determine whether corticospinal neurons and stroke-projecting neurons (M1-projecting neurons) are distinct. We found that whereas both neuron types were predominantly localized in layer 5 in the RFA, these neurons were rarely co-labeled (Supplementary Fig. 4b), suggesting corticospinal and stroke-projecting neurons are anatomically distinct. Then, we confirmed the specificity of the rabies virus labeling by injecting helper AAVs and rabies virus in the absence of Cre. We found that the rabies virus causes little leak and labeling in the injection site and no label in the distance brain areas without Cre, indicating minimum leakage by non-specific labeling (Supplementary Fig. 5). In dendritic spine analysis, we observed that neither stroke or rehabilitation affected dendritic spine density in corticospinal neurons in the RFA (Supplementary Fig. 4c, d). In corticospinal neurons, rabies tracing showed decreased synaptic inputs from the CFA (near the stroke site) and the somatosensory cortex; rehabilitation did not cause significant changes in any synaptic input sources (Supplementary Fig. 4e–g). In contrast, stroke-projecting neurons lost a substantial proportion of dendritic spines after stroke in both layers 2/3 and 5 (Fig. 3c–e and Supplementary Fig. 6). Monosynaptic tracing revealed that synaptic inputs were lost in major brain areas projecting to the stroke-projecting neurons (Fig. 3f–h and Supplementary Fig. 7). Rehabilitation restored dendritic spine density and a portion of synaptic inputs to the stroke-projecting neurons from some brain areas including the contralesional RFA, peri-infarct CFA, and thalamus. (Fig. 3d–h and Supplementary Figs. 6, 7). These data identify stroke-projecting neurons in the RFA as a population uniquely affected by loss of synaptic inputs after stroke and uniquely responsive in their input connections to rehabilitation, in a brain region integral to rehabilitation-induced recovery.
Stroke-projecting neurons mediate motor performance
The restoration of synaptic inputs to stroke-projection neurons after rehabilitation suggests a functional role in motor recovery. To determine if stroke-projecting neurons are causally involved in rehabilitation-induced recovery, we manipulated the stroke-projecting neurons with designer receptors exclusively activated by designer drugs (DREADDs) after rehabilitation (Fig. 4a, b). The inhibitory DREADD receptor, hM4D(Gi) or mCherry as a non-DREADD control were virally transfected in stroke-projecting neurons in the ipsilesional RFA before the stroke (Fig. 4a). Immunohistochemical analysis (Fig. 4b and Supplementary Fig. 8) revealed that virus expression co-localized with a cortical projecting neuron marker, Satb2 (86.1 ± 1.5%), while showing rare co-localization with an interneuronal marker, parvalbumin (0.38 ± 0.08%). Parvalbumin (PV) interneurons primarily form local cortical connections, indicating effective retrograde labeling without diffusive viral infection. The layer distribution analysis revealed that 30.4 ± 1.3% and 61.1 ± 1.5% of stroke-projecting neurons localized in layers 2–3 and 5, respectively (Supplementary Fig. 8). We also found that their axons densely projected to the contralateral cortex, striatum, thalamus, and cerebral peduncle, but the peri-infarct projection was sparse, indicating thorough destruction of the axonal projection to the CFA and limited axon regeneration (Supplementary Fig. 8). We assessed motor recovery 28 days after stroke and tested stroke-projecting neuron inhibition 3 days later. We assessed motor performance 15 min after the injection of the chemogenetic ligand, clozapine N‐oxide (CNO). Whereas neither hM4D(Gi) or mCherry expression affects baseline motor performance or recovery (Fig. 4c, d: Pre, 7d and 28d), chemogenetic inhibition of stroke-projecting neurons significantly deteriorates motor performance in sham control and rehabilitation-stroke animals in the skilled reaching test (Fig. 4c, e: +CNO). Chemogenetic inhibition also affects the motor performance of the rehabilitation-treated animals in the grid walk test (Fig. 4d, f). These results indicate that stroke-projecting neurons in rehabilitation animals play a significant role in motor recovery.
Rehabilitation selectively enhances synapse formation from parvalbumin interneurons to stroke-projecting neurons
Temporal patterns of neuronal activity are regulated by both local cortical dynamics and external inputs from distant areas. In addition to external inputs to stroke-projecting neurons, local connections in inhibitory neurons may mediate circuit connectivity and dynamics through mechanisms of synaptic plasticity47. Thus, we next examined the composition of local inhibitory synaptic inputs to stroke-projecting neurons by immunohistochemical labeling combined with monosynaptic rabies tracing (Fig. 5a). We found that rehabilitation significantly increased synaptic input from PV interneurons to stroke-projecting neurons (Fig. 5b, c), but not from the other major classes of inhibitory neurons: somatostatin, or 5HT3a interneurons (Fig. 5d, e). Unlike stroke-projecting neurons, this selective synaptic change was not observed in corticospinal neurons after stroke (Supplementary Fig. 4h–j). Importantly, the density of cortical interneurons did not change by stroke or rehabilitation (Supplementary Fig. 9a–e), and increased PV interneuron inputs were not correlated with PV interneuron density (Supplementary Fig. 9f). Because rabies virus labeling could be affected by virus toxicity and neuronal activity, we confirmed the rehabilitation-induced PV synapse formation using a synapse-specific labeling technique, GFP Reconstitution Across Synaptic Partner48 (GRASP: Fig. 5f-i and Supplementary Fig. 10). GRASP utilizes split GFP, with pre- and post-synaptic components (pre- and post-GRASP), which emit a green fluorescent signal exclusively when expressed at both pre- and post-synaptic terminals. To specifically detect synapses formed by presynaptic PV interneuron and postsynaptic stroke-projecting neuron, we induced presynaptic GRASP in PV interneurons using the PV interneuron-specific enhancer, S5E249 (PV specificity: 91.3 ± 1.64%) and postsynaptic GRASP in the stroke-projecting neurons using retrograde AAV-Cre vector. Consistent with the findings of the monosynaptic tracing study, we noted a marked increase in PV synapses on stroke-projecting neurons in stroke animals treated with rehabilitation (Fig. 5j, k, and Supplementary Fig. 10e). This increase in PV synapses was not accompanied by changes in total GABAergic synapse inputs (vGAT, Supplementary Fig. 10d). Interestingly, stroke significantly decreased the proportion of PV synapses within the total GABAergic synapse inputs and rehabilitation restored it (Fig. 5l). These data indicate that rehabilitation selectively increases PV interneuron input to stroke-projecting neurons and adjust the composition of GABAergic synapse inputs.
Parvalbumin interneurons mediate functional recovery
Given the selective synapse formation from PV interneurons to stroke-projecting neurons in rehabilitation, we tested whether post-stroke rehabilitation would activate this neuronal circuit. To this end, we assessed whether post stroke rehabilitation induces activity-dependent gene expression and plasticity changes in the PV/stroke-projecting neuron circuit in RFA by the detection of immediate early genes Zif268 and FosB, and of perineuronal nets, a determinant of PV neuron plasticity50 (Supplementary Fig. 11a, g). We observed that rehabilitation significantly increased Zif268 expression in both types of neurons (Supplementary Fig. 11c, d) and FosB expression in the stroke-projecting neurons (Supplementary Fig. 11e,f). Rehabilitation also decreased the ratio of PV interneurons surrounded by perineuronal nets (Supplementary Fig. 11i,j). These data indicate activation and enhanced plasticity of these neuronal circuits by rehabilitation.
We further tested whether the activation of these neuronal circuits is necessary for functional recovery by rehabilitation. We induced inhibiting hM4D(Gi) or mCherry control in either stroke-projecting neurons or PV interneurons using retrograde AAV-Cre (Supplementary Fig. 12a) and a PV-Cre mouse line (Fig. 6a). We injected the chemogenetic ligand, DCZ, 15 min before every rehabilitation session through the recovery period (Fig. 6b). We validated the efficacy of DREADD inhibition through the treatment period using 2 photon microscope calcium imaging in PV-Cre mice injected with Cre-dependent AAVs expressing hM4D(Gi) and GCaMP8s (Supplementary Fig. 13). We found this DREADD protocol significantly inhibited the movement-induced PV interneuron activation at 10 and 28 days after the stroke (Supplementary Fig. 13). We limited access to the rehabilitation apparatus for 2 h during which DCZ exerts maximum effects51. Compared to the previous studies, this shorter training period decreased the total reaches (millet seed consumption) but remained effective in improving motor performance after stroke (Fig. 6c–e and Supplementary Fig. 12c–e: mCherry). While the inhibition of stroke-projecting neurons influenced rehabilitation-induced motor recovery exclusively in the grid walk task (Supplementary Fig. 12f,g), inhibiting PV interneurons had a significant and broader impact, diminishing the effectiveness of recovery in both skilled reaching and the grid walk (Fig. 6f,g). This indicates that PV interneuron activation has an essential role in rehabilitation-induced recovery in both reach-to-grasp and precision gait function.
Stroke decreases the fraction of fast-rising large amplitude IPSCs, and rehabilitation restores it
Our histological analysis revealed that stroke-projecting neurons undergo structural synaptic alterations following stroke and rehabilitation. To investigate whether these neurons also exhibit changes in synaptic inputs, we conducted patch-clamp recordings of stroke-projecting neurons. We labeled the stroke-projecting neurons with retrograde AAV expressing tdTomato and recorded spontaneous EPSC/IPSC in the tdTomato positive stroke-projecting neurons (CFA-projecting neurons in Sham animals) in the RFA brain slices. We also validated external synaptic inputs by recording evoked EPSCs in response to optogenetic stimulation of thalamocortical axons (Fig. 7a).
In the EPSC recordings, stroke induced a significant increase in peak EPSC amplitude (Fig. 7d), while EPSC frequency remained unchanged (Fig. 7c). In contrast, rehabilitation significantly increased EPSC frequency compared to stroke animals (Fig. 7c). In optogenetic recordings, stroke reduced the EPSC probability to optogenetic stimulation of thalamic afferents (Fig. 7f) without significant differences in peak amplitude (Fig. 7g), suggesting that excitation of thalamic afferents to the CFA produces less reliable responses in stroke-projecting neurons in both stroke and rehabilitation animals. Our histological data, including dendritic spine analysis and rabies virus tracing (e.g., increased spine density in rehabilitation animals compared to untreated stroke animals and reduced synaptic input from distant brain areas), are primarily consistent with EPSC frequency/probability data but not EPSC amplitude. These findings suggest that EPSC frequency/probability is more closely related to structural synaptic alterations (e.g., synapse loss or formation), whereas EPSC amplitude is influenced by additional factors such as intrinsic excitability, postsynaptic receptor expression, presynaptic vesicle release probability, and phasic/tonic inhibitory inputs mediated by transcriptional regulation. (see Discussion for details).
In the IPSC recordings, we found that stroke increased the IPSC frequency (Fig. 7i) without changing the IPSC peak amplitude (Fig. 7j), indicating increased inhibitory synaptic inputs after stroke. Conversely, rehabilitation showed no significant changes. In the GRASP study, stroke decreases the fraction of PV interneuron synapses without total inhibitory synapse number, and rehabilitation restores these abnormalities. Thus, stroke-projecting neurons may also show fractional changes in IPSCs. We further analyzed the IPSC data to determine whether our IPSC recordings capture PV interneuron functionality. Since PV interneurons target the α1 subunit-containing GABAA receptors, they predominantly generate fast-rising IPSCs52,53,54. Also, the somatic location of PV interneuron inputs results in large-amplitude IPSCs. We have previously shown that the loss of fast-rising large amplitude IPSCs in a mouse model of Alzheimer’s disease mirrors the functional and physical loss of PV interneurons55. Using as thresholds the mean rate of rise and peak amplitude of all recorded IPSCs, we estimated the fraction of large amplitude fast-rising events in each group, most likely originating from PV interneuron inputs. The data show a reduction in the fraction of large amplitude fast rate of rise IPSCs following stroke, with recovery observed after rehabilitation (Fig. 7k), indicating that stroke reduces, and rehabilitation recovers fractional synaptic inputs from PV interneurons. Additionally, these findings are consistent with the idea that the increased IPSC frequency observed in stroke animals is primarily attributable to small-amplitude or slow-rising inhibitory synaptic inputs from interneurons other than PV interneurons.
Parvalbumin interneurons regulate functional connectivity after stroke
To study how PV interneurons might enhance functional recovery after stroke, we next asked whether PV interneurons could modulate neuronal connectivity associated with motor control. Previous studies had shown that PV interneurons regulate synchronous neuronal activity in the perception of sensory information56,57. To gain insights into the causal role of PV interneurons in motor function, we conducted a chemogenetic inhibition study combined with calcium imaging. The hM4D(Gi) receptor was induced in PV interneurons and neuronal activity was recorded in the excitatory neurons in RFA (Fig. 8a). Given the distinct roles of PV interneurons depending on their activity state56,57, we recorded neuronal activity in both stationary (stop epoch) and active (running epoch) states in voluntary running on a grid wheel (Fig. 8b). Similar to the previous forced running experiment, the stroke animals exhibited severe motor deficits (Fig. 8d) accompanied by a persistent reduction in connection density in both running and stop epochs Supplementary Fig. 14c-j). Four weeks after the stroke, we assessed the effects of inhibiting PV interneurons on motor performance and neuronal activity. We found that inhibition of PV interneurons impaired motor performance during grid running in sham and rehabilitation animals but had no effect on stroke animals (Fig. 8d, e). Notably, inhibition of PV interneurons also decreased functional connectivity during the running epoch in all groups (Fig. 8f, g). Consistent with the correlation between motor function and functional connectivity in the previous calcium imaging (Fig. 2i), the changes in motor performance and functional connectivity showed a significant correlation (Fig. 8h). As expected from the known PV interneuron functions of feedback and feedforward inhibition to excitatory neurons, chemogenetic inhibition of PV interneurons resulted in a significant increase in active neuron count and calcium transient frequency during the stop epoch (Supplementary Fig. 15). Surprisingly, we observed a dramatic increase in functional connectivity during the stop epoch in stroke animals, while sham animals exhibited decreased functional connectivity (Fig. 8i, j). Collectively, these results indicate PV interneuron inhibition decreases the connection density associated with motor impairment in normal animals but induces distinctive effects in stroke animals. This implies that PV interneurons are involved in the regulation of neural network connections, and stroke disrupts circuit control mediated by PV interneurons.
Gamma power changes after stroke in the mouse stroke model and human patients
The precise timing of neuronal activity and population synchrony are coordinated by network oscillations58. PV interneurons play a pivotal role as key cellular elements in the generation of such network oscillations, especially gamma waves, which establish oscillatory envelopes of increased neuronal activity onto neuronal networks59. To determine whether neuronal oscillations are associated with stroke recovery, we assessed network oscillation in the premotor cortex/RFA of stroke animals. Recording electrodes were implanted in the RFA, and a transparent acrylic column was placed on the CFA (Fig. 9a), enabling stroke induction without disturbing the implanted electrodes. This setup allows direct comparison of pre-and post-stroke oscillation in the same animals. We recorded the local field potential (LFP) in freely moving mice in their home cage 3 weeks after stroke (Fig. 9a). Stroke causes an immediate global spectral power decrease ranging over delta, theta and gamma frequencies (Supplementary Fig. 16). Recovery of spectral power varied according to the spectral frequency band and the vigilance state. Low gamma power (30–60 Hz) exhibited notable recovery following stroke (Fig. 9b–d), and this was significantly enhanced in animals undergoing rehabilitation (Fig. 9c,d).
Next, we further assessed the association of gamma waves with functional recovery in human stroke patients. We studied 27 patients with stroke admitted to an inpatient rehabilitation facility (IRF) a median of 12 [8–17] days post-stroke (Visit 1, Supplementary Table 1a). Multiple EEGs were recorded during time in the IRF admission, and an additional EEG was obtained approximately 3 months (86 – 103 days) after stroke onset. The EEG was a 3-min recording taken at rest, and relative low gamma power (30–40 Hz) was analyzed in ipsi (i)- and contralesional (c) hemispheres within five motor regions: M1 (iM1 and cM1), dorsal premotor areas (iPMD and cPMD) and midline supplementary motor area (SMA). 17 patients were available for repeat EEG and functional assessment at Visit 5 (median 92 [86 – 103] days). We found that the arm motor Fugl-Meyer (FM) score significantly improved from Visit 1 to Visit 5 (Fig. 9f: p = 0.0004). At Visit 1, the relative low gamma power did not differ between patients and 27 age-matched healthy control subjects within any of 5 motor regions (Supplementary Table 1b). Subsequently, low gamma oscillations significantly increased in iM1, cM1 and SMA from Visit 1 to Visit 5 (Supplementary Table 1c). Notably, the increase of gamma power in ipsilesional M1 was significantly correlated with the FM scores in moderate to severe stroke patients during the recovery period (V3-5, V1 FM score <46, Fig. 9g, h). Thus, gamma oscillations increase in stroke patients during rehabilitation recovery after stroke, as they do in the mouse. Together, these data support the hypothesis that rehabilitation enhances functional recovery through PV interneuron-mediated mechanisms, and an oscillatory signal of increased neuronal activity by the gamma rhythm.
Activation of PV interneuron improves functional recovery
Finally, we assessed whether there is a pharmacological approach to simulate rehabilitation-induced recovery from stroke, through activation of PV interneuron circuits. To this end, we tested two compounds: AUT00201, a selective positive modulator of Kv3.1 ion channels predominantly found on PV interneurons60 and DDL-920, a selective negative modulator of the γ-aminobutyric acid type A receptors with α1β2δ subunits (GABAARδ) responsible for the tonic inhibition of PV interneurons61. The α1β2δ GABAAR is uniquely expressed by PV interneurons as opposed to the α4β2/3δ GABAAR of dentate gyrus granule cells, thalamo-cortical, cortical/hippocampal pyramidal and medium spiny neurons, and the α6β2/3δ GABAAR predominantly expressed by cerebellar granule cells62. Positive modulation of Kv3.1 causes faster activating kinetics and increased firing frequency in fast-spiking GABAergic interneurons63. On the other hand, negative modulation of GABAARδ reduces tonic inhibition and enhances the excitability of PV interneurons64. These alterations in activity subsequently modulate gamma oscillations64,65. We orally administered the drugs, to enhance the applicability of the formulation for potential clinical translation (Fig. 10a). We confirmed that Kv3.1 ion channels and GABAARδ expressed mainly in PV interneurons in both intact and stroke animals (Supplementary Fig. 17). Activation of PV interneurons was tested by single dosing. As expected from the PV interneuron-selective effects of these drugs, both AUT00201 (20 mg/kg66, Supplementary Fig. 18) and DDL-920 (10 mg/kg64, Supplementary Fig. 19) increased the expression of the immediate early gene Zif268 in PV interneurons without changing the density of cells expressing Zif268. Only DDL-920 showed a statistically significant increase (Fig. 10b). In a stroke-recovery study, we started the drug treatment 3 days after the stroke and evaluated the recovery of forelimb motor function with skilled reach (pasta matrix) and gait (grid walk) tests. We did not observe any adverse effects such as weight loss or motor deficits in either sham or stroke animals. Stroke animals treated with the vehicle and AUT00201 exhibited prolonged disability in precisely retrieving pasta pieces (Fig. 10c, d). In contrast, DDL-920 treatment led to a complete recovery of motor function after stroke (Fig. 10c, d). AUT00201 and DDL-920 treatment also produced faster recovery in the grid walk test (Supplementary Fig. 20). These data establish the principle that pharmacological agents can drive beneficial cellular effects seen in rehabilitation-induced stroke recovery and promote behavioral recovery equivalent to that seen in rehabilitation-induced stroke recovery.
