Parvalbumin interneurons regulate rehabilitation-induced functional recovery after stroke and identify a rehabilitation drug

Abstract

Motor disability is a critical impairment in stroke patients. Rehabilitation has a limited effect on recovery; but there is no medical therapy for post-stroke recovery. The biological mechanisms of rehabilitation in the brain remain unknown. Here, using a photothrombotic stroke model in male mice, we demonstrate that rehabilitation after stroke selectively enhances synapse formation in presynaptic parvalbumin interneurons and postsynaptic neurons in the rostral forelimb motor area with axonal projections to the caudal forelimb motor area where stroke was induced (stroke-projecting neuron). Rehabilitation improves motor performance and neuronal functional connectivity, while inhibition of stroke-projecting neurons diminishes motor recovery. Stroke-projecting neurons show decreased dendritic spine density, reduced external synaptic inputs, and a lower proportion of parvalbumin synapse in the total GABAergic input. Parvalbumin interneurons regulate neuronal functional connectivity, and their activation during training is necessary for recovery. Furthermore, gamma oscillation, a parvalbumin-regulated rhythm, is increased with rehabilitation-induced recovery in animals after stroke and stroke patients. Pharmacological enhancement of parvalbumin interneuron function improves motor recovery after stroke, reproducing rehabilitation recovery. These findings identify brain circuits that mediate rehabilitation-recovery and the possibility for rational selection of pharmacological agents to deliver the first molecular-rehabilitation therapeutic.

Fig. 1. Rehabilitative training recovers motor performance after stroke.

a Rehabilitation box used in the study. Mice stably engage in reach-to-grasp training for 3 weeks, 5 days a week during recovery. n = 12. The error bars are smaller than the symbols. b Timeline of rehabilitation and behavioral testing. The blue triangles indicate behavioral tests. Pretraining for behavior tests and rehabilitation periods are indicated as blue and red rectangles. c, d Skilled reaching test (c): success rate (time by group, F (6, 64) = 10.35, P < 0.0001) and grid walk test (d): foot fault rate (time by group, F (6, 64) = 26.49, P < 0.0001). Two-way repeated measure ANOVA with Sidak’s multiple comparison test. **P < 0.01, ***P < 0.001. e, f Functional improvement in the recovery period in the skilled reaching test (e) and the grid walk test (f). Two-tailed paired t-test. (a–f) n = 8 (Sham), 8 (Sham + Rehab), 10 (Stroke) or 9 (Stroke + Rehab). (b) Created in BioRender.

Fig. 2. Rehabilitative training recovers functional connectivity after stroke.

a Timeline of calcium imaging study. The green triangles indicate calcium imaging sessions. b Representative trace of calcium transient obtained from active (upper) and inactive (lower) neurons. c Active neuron number (time by group, F (4, 28) = 5.504, P = 0.0021). d calcium transient event frequency (time by group, F (4, 28) = 3.850, P = 0.0129), e functional connection number (time by group, F (4, 28) = 8.518, P = 0.0001). f connection density (time by group, F (4, 26) = 2.936, P = 0.0397). Mixed-effects model, **P < 0.01, ****P < 0.0001: Sham (n = 4) vs Stroke (n = 7), # P < 0.05, ## P < 0.01, ### P < 0.001: Sham vs Stroke + Rehab (n = 6), + P < 0.05, ++ P < 0.01: Stroke vs Stroke + Rehab, Tukey’s multiple comparisons test. Two-sided. Tukey’s HSD correction for multiple comparison. g Fractional distribution of connection probability in individual neurons. Red, blue, and black areas indicate the ranges where the fraction increases in stroke, stroke + rehab, and sham groups. Two-way ANOVA (F (26, 196) = 3.789, P < 0.0001), *P < 0.05, **P < 0.01, ***P < 0.001: Sham vs Stroke, # P < 0.05: Sham vs Stroke + Rehab, + P < 0.05, ++ P < 0.01: Stroke vs Stroke + Rehab, Tukey’s multiple comparisons test. (c–g) n = 4 (Sham), 7 (Stroke) or 6 (Stroke + Rehab). h Representative connection maps 28 days after the stroke. Scale bar, 100 μm. i Correlation between motor performance in the grid walking and the connection density. Pearson correlation (n = 49, r = 0.671, P < 0.0001). Two-sided. Data are presented as means ± sem. (a) Created in BioRender.

Fig. 3. Rehabilitation restores synaptic input to stroke-projecting neurons.

a Approach for synaptic inputs in stroke-projecting neurons. b Timeline, virus vectors, and virus injection locations in dendritic spine analysis (upper) and monosynaptic tracing (lower). The dashed lines in the brain illustration indicate future stroke sites (CFA). c Representative images showing dendritic spines (upper: scale bar 5 μm) and G-deleted rabies virus (RVdG) labeled cells (middle and lower: scale bar 1 mm). cRFA: contralesional rostral forelimb area, iRFA: ipsilesional RFA, TH: thalamus, SS: somatosensory area. d, e Spine density in layer 5 stroke-projecting neurons; basal dendrite (d), n = 19 (Sham), 20 (Sham + Rehab), 27 (Stroke) or 22 (Stroke + Rehab), apical dendrite (e), n = 15 (Sham), 15 (Sham + Rehab), 22 (Stroke) or 16 (Stroke + Rehab). Kruskal-Wallis test. 4–5 animals per group. f–h The number of RVdG-labeled cells normalized by the starter cells in the whole brain (f), RVdG injection neighbor: RFA (g), and distant brain areas (h). ORB: orbital area, CFA: caudal forelimb area, SS: somatosensory area, CLA: claustrum, AC: anterior cingulate area, CNU: cerebrum nuclei (striatum and pallidum), TH: thalamus. Kruskal-Wallis test. n = 9 (Sham), 8 (Sham + Rehab), 9 (Stroke) or 11 (Stroke + Rehab). (a, b) Created in BioRender.

Fig. 4. Stroke-projecting neurons are necessary for rehab-induced functional recovery.

a Timeline, virus vectors, and injection sites in the chemogenetic inhibition targeting stroke-projecting neurons. b Representative images of the stroke-projecting neurons with hM4D(Gi)-mCherry in the RFA layer 5. The stroke-projecting neurons express a cortical projection neuron marker, Satb2 (right), but not PV (left). scale bar 50 μm. Similar staining was confirmed in the sections from 9 mice. See supplemental material for the quantification. c, d Motor performance in the skilled reaching test (c time by group, F (15, 204) = 9.930, P < 0.0001) and the grid walk test (d time by group, F (15, 204) = 17.86, P < 0.0001). Two-way repeated measure ANOVA, Sidak’s multiple comparison test. e, f Motor performance changes induced by the chemogenetic inhibition in the skilled reaching test (e) and the grid walk test (f). Two-tailed paired t-test. (c–f) n = 11 (Sham/mCherry), 11 (Sham/hM4D), 12 (Stroke/mCherry), 13 (Stroke/hM4D), 13 (Stroke+Rehab/mCherry) or 14 (Stroke+Rehab/hM4D) Data are presented as means ± sem. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. (a) Created in BioRender.

Fig. 5. Rehabilitation induces synapse formation from PV interneurons to stroke-projection neurons.

a Schematic illustration of interneuron marker labeling combined with monosynaptic tracing. PV: parvalbumin, SOM: somatostatin, 5HT3a: serotonin receptor 3a. b Representative images of RVdG labeled PV interneurons. The green channel shows the GFP labeled starter cells. Scale bar 50 μm c–e The ratio of PV (c), SOM (d), 5HT3a (e) in total local inputs. Kruskal-Wallis test, n = 6. f Schematic illustration of GRASP labeling of synapse formed by stroke-projecting neuron and PV interneuron. g Timeline, virus vectors, and injection sites in the GRASP study. h, i Representative images of GRASP labeled dendrite (H, scale bar 5 μm) and soma (I, scale bar 10 μm). j Example of reconstruction of dendritic tree and GRASP+vGAT puncta. Scale bar 30 μm. GRASP expression patterns were consistent across the staining sessions and the animals k, l The number of GRASP+vGAT puncta (k) and the proportion of GRASP positive synapse in the total vGAT positive synapse. (l) on the stroke projecting neurons. Kruskal-Wallis test, n = 26 (Sham), 26 (Sham + Rehab), 23 (Stroke) or 24 (Stroke + Rehab). (a, f, g) Created in BioRender.

Fig. 6. PV interneuron activation is necessary for recovery.

a Timeline, virus vectors, and injection sites in the chronic chemogenetic inhibition targeting PV interneurons. b Procedure for rehabilitation with chemogenetic inhibition. c Millet seed consumption with 2-h rehabilitation. n = 6. d, e Motor performance in the skilled reaching test (d time by group, F (6, 68) = 7.887, P < 0.0001) and the grid walk test (e time by group, F (6, 68) = 19.69, P < 0.0001). Two-way repeated measure ANOVA: Stroke + Rehab/mCherry vs Stroke + Rehab/hM4D(Gi), Sidak’s multiple comparison test. f, g Functional recovery by rehabilitation in the skilled reaching test (f) and the grid walk test (g). Two-tailed paired t-test. (d–g) n = 9 (Sham/mCherry), 8 (Sham/hM4D), 10 (Stroke+Rehab/mCherry) or 11 (Stroke+Rehab/hM4D). All data are presented as means ± sem. *P < 0.05, **P < 0.01, ***P < 0.001. (a, b) Created in BioRender.

Fig. 7. Stroke reduces the fraction of fast-rising large amplitude IPSCs and rehabilitation restores it.

a Virus injection and time course for patch-clamp recording. b Representative spontaneous EPSC traces. c, d Frequency (c), and peak amplitude (d) of spontaneous EPSC. Kruskal-Wallis test, n = 11 (Sham), 12 (Stroke), or 13 (Stroke + Rehab). e representative traces of EPSCs evoked by optogenetic stimulation targeting thalamic axons. f, g Response probability (f), and peak amplitudes (g) responding to 50 sequential optogenetic stimulations. Generalized linear mixed model (Fixed effect: Stimulation number; P < 0.0001, Group; P < 0.0001). Two-sided. Tukey’s HSD correction for multiple comparison. n = 11 (Sham), 12 (Stroke), or 13 (Stroke + Rehab). h Representative spontaneous IPSC traces. i, k Frequency (i), and peak amplitude (j) of spontaneous IPSC. Kruskal-Wallis test, n = 12 (Sham), 9 (Stroke), or 10 (Stroke + Rehab). *P < 0.05, ***P < 0.001,****P < 0.0001. k Distribution of IPSC rise rate and peak amplitude. The red dashed lines indicate the mean peak amplitude and rate of rise of all IPSC events from all recordings. The large amplitude with fast rate of rise events above both mean values (upper right quadrant) amount to 16.5%, 11.7%, and 19.7% of all IPSC events in the Sham, Stroke, and Stroke + Rehab groups, respectively. Chi-squared test; X-squared = 73.4, df = 2, P = 2.2e-16, Bonferroni post hoc test, two-sided; Sham vs Stroke: P < 0.0001, Sham vs Stroke + Rehab: P = 0.0092, Stroke vs Stroke + Rehab; P < 0.0001. Data are presented as means ± sem. (a) Created in BioRender.

Fig. 8. PV interneurons regulate functional connectivity in healthy and stroke animals.

a Timeline, virus vectors, and imaging field for calcium imaging with chemogenetic inhibition. b Representative motion trace and color-mapped calcium transient. The orange line above the motion trace indicates binary detection of the running epoch. c Protocol for calcium imaging and DCZ injection. d Foot fault rate during the calcium imaging. Two-way repeated measure ANOVA, time by group, F (6, 69) = 11.04, P < 0.0001Sidak’s multiple comparison test. e Foot fault rate change by DCZ injection. Two-tailed Wilcoxon test (d, e) n = 6 (Sham), 9 (Stroke) or 11 (Stroke + Rehab). f Examples of connection map before and after DCZ injection in Sham and Stroke groups. g Connection density changes by DCZ injection in the run epoch. Two-tailed Wilcoxon test. n = 11 (Sham), 16 (Stroke) or 15 (Stroke + Rehab). h Correlation between the motor performance change in the grid walking and the connection density change after the DCZ injection. Pearson correlation, two-sided (n = 42, r = 0.324, P = 0.0361). i Connection density changes by DCZ injection in the stop epoch. Two-tailed Wilcoxon test. n = 12 (Sham), 17 (Stroke) or 18 (Stroke + Rehab). j Representative recordings before and after the DCZ injection in stroke animals. c *P < 0.05, **P < 0.01, ***P < 0.001. Data are presented as means ± sem. (a, b) Created in BioRender.

Fig. 9. Gamma oscillation in the stroke mouse model and human stroke patients.

a Electrode placement and timeline for the EEG recording. b Representative LFP traces in stroke animals. c Normalized power spectra of network oscillation in ipsilesional premotor cortex in awake period mice 21 days after the stroke (means ± sem). The orange rectangle indicates the lower gamma frequency band. d Normalized spectrum power change in low gamma frequency. Mixed-effects model, time by group, F (10, 62) = 11.87, P < 0.0001. *P < 0.05, ***P < 0.001, ****P < 0.0001: Sham vs Stroke, ++ P < 0.01, ++++ P < 0.0001: Sham vs Stroke + Rehab, ## P < 0.01: Stroke vs Stroke + Rehab, Tukey’s multiple comparisons test. Two-sided. Tukey’s HSD correction for multiple comparison. n = 4 (Sham), 7 (Stroke) or 5 (Stroke+Rehab). e Illustration of human EEG. f Arm motor Fugl-Meyer (FM) score after stroke. g Correlation matrix of the FM score and the relative gamma power during the recovery period (V3-5) in the ipsi (i) and contralesional (c) motor related areas. M1: primary motor area, PMD: dorsal premotor area, SMA: supplementary motor area in moderate to severe stroke patients (V1 FM < 46). h Correlation between the FM score and the normalized gamma power. Spearman correlation, two-sided (n = 25, r = 0.443, P = 0.026). e Created in BioRender.

Fig. 10. PVIN targeting drug therapy reproduces rehabilitation effects.

a Timeline and procedure for the drug administration of AUT00201 (‘AUT’, 20 mg/kg p.o.) or DLL-920 (‘DDL’, 10 mg/kg p.o.) and behavioral test. b Ratio of Zif268 positive PV interneurons (Left, F (2, 12) = 4.332, P = 0.0383) and density of Zif268 positive cells (right, F (2, 12) = 0.1324, P = 0.8773). One-way ANOVA, Tukey’s multiple comparisons test. n = 5. c Normalized motor performance in the pasta matrix test. Two-way repeated measure ANOVA, time by group, F (15, 189) = 3.890, P < 0.0001. *P < 0.05, **P < 0.01: Sham + vehicle vs Stroke + Vehicle, #P < 0.05, ##P < 0.01: Sham + vehicle vs Stroke + AUT, ++P < 0.01: Sham + vehicle vs Stroke + DDL, & &P < 0.01: Stroke + vehicle vs Stroke + DDL, Sidak’s multiple comparison test. d Functional recovery from day 3 to day 14. Two-tailed paired t-test for comparisons of 3 d and 14 d. One-way ANOVA and Tukey’s multiple comparison for comparisons of functional gains. F (2, 28) = 8.967, P = 0.0010. **P < 0.01, ****P < 0.0001. c, d n = 12 (Sham+Vehicle), 12 (Sham+AUT), 14 (Sham+DDL), 8 (Stroke+Vehicle), 12 (Stroke+AUT), 11 (Stroke+DDL). Data are presented as means ± sem.