Animals

All experimental procedures were approved by the Ethics Committee of the Laboratory Animal Science Department, Fudan University (Approval No. 202310008 S) and conducted in accordance with ARRIVE guidelines [20]. Male Sprague-Dawley rats (6–7 weeks old, 200–250 g) were obtained from Shanghai JieSiJie Laboratory Animal Co., Ltd. They were housed under standard conditions: a 12-hour light/dark cycle, controlled temperature and humidity, and ad libitum access to food and water. The experimental design is illustrated in Figure S1.

ICH model establishment

Although the collagenase model allows for modeling of prolonged inflammation, the autologous blood injection model was selected for its reproducibility and ability to simulate the mass effect of hematoma, enabling a controlled evaluation of acute functional and structural recovery mechanisms. Therefore, the autologous blood model offers better hematoma volume control and mimics the acute mechanical effects of hemorrhage, which is aligned with our study.

Rats were fasted for 8 h before surgery and anesthetized using 5% isoflurane in an inhalation anesthesia machine (RWD Life Science, Shenzhen, China). After induction, they were placed in a prone position in a stereotactic frame with the head secured, and anesthesia was maintained with 2% isoflurane. A midline incision exposed the skull, and a 1.0 mm burr hole was drilled 3.0 mm lateral and 0.5 mm anterior to the bregma on the left side. Using a microinjector, 60 µL of autologous blood drawn from the tail artery was injected into the basal ganglia at a depth of 6.0 mm from the dura at 10 µL/min. The needle was retained for 10 min post-injection to prevent backflow. The burr hole was sealed with bone wax, and the incision was sutured. Sham animals underwent burr hole creation without blood infusion. Post-operative care included applying neomycin ointment to prevent infection and single housing for 3 days. Modified Neurological Severity Scores (mNSS) were assessed upon recovery from 30 min [21]. Rats with mNSS ≥ 8 and confirmed hematoma formation via susceptibility-weighted imaging (SWI) on day 3 were considered successfully modeled.

Intervention protocol

The intervention combined taVNS with treadmill training [22]. The taVNS group received electrical stimulation of the left auricular concha using preclinically optimized parameters (current: 0.8 mA; pulse width: 100 µs; frequency: 30 Hz) [13, 23]. Stimulation was applied during passive treadmill sessions (10 m/min, 30 min/session, twice daily). The sham and control groups received treadmill training without taVNS.

Neurological function assessment

Neurological deficits were assessed using the mNSS at 1, 2, and 4 weeks post-ICH. The mNSS evaluates motor, sensory, balance, and reflex functions on an 18-point scale, with higher scores indicating more severe deficits [24, 25].

MRI data acquisition

Brain imaging was performed using an 11.7 T MRI system (Bruker Medizintechnik, Germany) at Fudan University. Rats were anesthetized with 5% isoflurane for induction, followed by maintenance with 1.5% isoflurane and dexmedetomidine (0.05 mg/kg). Body temperature was maintained at ~ 37.5 °C, and vital signs were continuously monitored. SWI was conducted on day 3 post-ICH to confirm hematoma formation. Longitudinal functional imaging, including resting-state fMRI (rs-fMRI), T2, and diffusion tensor imaging (DTI), was performed at weeks 1 and 4 in the modulation and control groups.

SWI parameters: TR = 400 ms, TE = 12 ms, FOV = 30 × 30 mm, FA = 40°, slice thickness = 0.5 mm, 17 slices.

rs-fMRI parameters: TR = 2000 ms, TE = 12.8 ms, FOV = 30 × 30 mm, FA = 90°, slice thickness = 0.5 mm, 50 slices, 300 volumes.

T2 structural imaging parameters: TR = 4200 ms, TE = 25 ms, FOV = 30 × 30 mm, FA = 40°, slice thickness = 0.5 mm, 50 slices.

DTI parameters: TR = 3000 ms, TE = 22 ms, FOV = 30 × 30 mm, FA = 90°, slice thickness = 0.5 mm, 35 slices, b0 = 5, 30 non-collinear diffusion gradients (b-value = 1000 s/mm²).

rs-fMRI data processing

rs-fMRI data were preprocessed using an SPM 12-based pipeline with automated scripts. Steps included removing non-brain tissue, creating whole-brain masks, and rescaling voxel size by a factor of 10. The first 10 time points were discarded to ensure signal stabilization, followed by slice-timing correction using the 50th slice as the reference. Head motion correction, spatial normalization to a standardized rat brain template, detrending to remove linear drifts, and band-pass filtering (0.01–0.1 Hz) were performed. Spatial smoothing was applied with a 4 × 4 × 4 mm³ Gaussian kernel.

For spontaneous brain activity indices, smoothing was performed before filtering. For degree centrality (DC), filtering was performed first, followed by smoothing. Metrics were calculated using REST 1.8 software (https://rfmri.org/REST), excluding hematoma-overlapping regions to prevent confounding effects. Fractional amplitude of low-frequency fluctuation (fALFF) assessed spontaneous activity in the sensorimotor cortex, while DC evaluated core network node distribution in this region. Z-score transformation was applied for normalization, providing a macroscopic functional perspective on cortical remodeling.

DTI data processing

DTI data were processed using Q-space diffeomorphic reconstruction (QSDR, 0.5 mm voxel size) and aligned to the Waxholm Space Standard Rat Brain Atlas (DSI Studio). Regions of interest (ROIs) included the left primary motor cortex and left basal ganglia, excluding contralateral structures, to reconstruct the left corticospinal tract (CST) and generate a reference CST map (Figure S2A). Fractional anisotropy (FA) values were extracted from the left basal ganglia ROI to quantify fiber tract remodeling. These findings were cross-validated with mesoscale myelin remodeling analyses. To further investigate CST integrity post-taVNS, DTI-based tractography was integrated with viral tracer mapping (Figure S2).

Viral tracing of the rat CST

Viral tracing (AAV2/9-hSyn-GFP injected) was conducted using the Allen Rat Brain Atlas to reconstruct the right CST, with the right primary motor cortex defined as the ROI. Point-projection mapping delineated CST pathways, generating a high-resolution reference atlas for subsequent analyses (Figure S2B).

Luxol fast blue (LFB) staining

LFB staining assessed myelin repair and remodeling in the basal ganglia at the mesoscale structural level. Paraffin-embedded sections were deparaffinized, rehydrated, and incubated in 95% ethanol before staining in LFB solution for 12–18 h at room temperature. Excess stain was removed with ethanol, and sections were rinsed in distilled water. Differentiation was performed using 0.05% lithium carbonate for 15 s, followed by 70% ethanol until gray and white matter distinction was clear. Sections were then treated with 95% ethanol, absolute ethanol, and xylene before mounting with neutral resin. This method enabled precise myelin repair evaluation.

Immunofluorescence assay

Immunofluorescence was used to assess myelin and axonal protein expression, including myelin basic protein (MBP) and non-phosphorylated neurofilament protein (SMI-32), in the ipsilateral basal ganglia. Tissue sections were deparaffinized and rehydrated through xylene, a graded ethanol series, and double-distilled water. Antigen retrieval was performed using citrate buffer heated to 95–100 °C, followed by cooling and PBS washes. To block non-specific binding, sections were incubated with 5% goat serum. Primary antibodies (anti-MBP, Abcam, Ab11159; anti-SMI-32, BioLegend, 801702) were applied and incubated overnight at 4 °C. After washing, fluorescently labeled secondary antibodies were applied and incubated in the dark at room temperature for 1–2 h. Nuclei were counterstained with DAPI, and sections were mounted with anti-fade medium. Fluorescence microscopy was used to visualize MBP and SMI-32 expression, providing insights into myelin and axonal remodeling.

Transmission electron microscopy (TEM) for structural remodeling

TEM was used to evaluate myelin repair and synaptic remodeling in the basal ganglia after ICH, integrating findings across micro-, meso-, and macro-scales. Tissue samples were fixed in 2.5% glutaraldehyde at 4 °C, post-fixed in 1% osmium tetroxide, and dehydrated through a graded ethanol series and acetone. Samples were embedded in resin and cured at 70 °C. Ultrathin sections (70–90 nm) were prepared using a Reichert ultramicrotome, stained with lead citrate and uranyl acetate, and imaged under TEM to provide high-resolution structural evidence of remodeling.

Statistical analysis

Group comparisons were performed using independent-samples t-tests.

rs-fMRI data analysis: A mixed-effects model was applied to analyze zfALFF and zDC across the whole brain, with a focus on sensorimotor cortex function. In the 28-day cohort, comparisons included: (1) longitudinal differences between weeks 1 and 4 within each group, (2) cross-sectional differences between modulation and control groups, and (3) time-group interaction effects. Interaction effects in the sensorimotor cortex were identified using AlphaSim correction (p < 0.01, cluster size > 19), capturing functional remodeling and taVNS efficacy. ROIs in the sensorimotor cortex were selected based on zfALFF and zDC metrics, and signals from these ROIs were correlated with mNSS scores using Pearson correlation.

DTI data analysis: The correlational module was used to assess CST reconstruction differences. At week 1, DTI data from the modulation group were labeled as 1 and the control group as 0. Permutation testing (10,000 iterations) was performed to identify group differences in left CST reconstruction (FDR-corrected p < 0.05). The same analysis was conducted at week 4 to assess longitudinal CST remodeling differences.