Potential participants were recruited from the Affiliated Hospital of Hangzhou Normal University from April 2022 to August 2022. All patients had unilateral ischemic stroke observed on a diffusion-weighted MRI scan. The inclusion criteria were: (1) unilateral ischemic stroke in the acute and subacute phase of stroke (< 6 months) (Lefaucheur et al., 2014); and (2) with somatosensory deficits caused by ischemic stroke; and (3) 18–80 years old; and (4) were on regular stroke medicines. The exclusion criteria were: (1) TMS contradictions such as current or a history of seizure or implanted devices (pacemaker, medical pumps) (Rossi et al., 2011); or (2) severe mental disorders assessed with HAMD or HAMA (Hamilton, 1959, 1967); or (3) aphasia or cognitive disorders assessed with MMSE (Folstein et al., 1975); or (4) not able to communicate with a doctor; or (5) severe disorders caused by other conditions such as tumour or severe heart or lung malfunctioning; or (6) mRS > 2; or (7) somatosensory deficits not caused by ischemic stroke, e.g., diabetes pain. mRS was used to include mostly mild impairment as somatosensory deficits are more prominent for these patients compared to motor deficits in patients with severe stroke. The withdrawal criteria were: (1) changes in medication after allocation; or (2) newly diagnosed stroke or other neurological lesions after allocation; or (3) that patients decided to withdraw from the study.

We conducted an open label and randomized trial registered in the Chinese Clinical Trials Registry (ChiCTR2200059098). Patients were randomly assigned to either contralesional 1-Hz or ipsilesional 10-Hz rTMS treatment according to a centrally stratified computer-generated randomization protocol by ZG. Allocation concealment was performed independently by a staff member (JG). Patients received 10 daily sessions in two consecutive weeks delivered by a trained neurologist (ZH). Clinical assessments were performed at pre-and post-treatment by a blinded neurologist (ZG). All participants provided written informed consent before the study commencement. This study was approved by the Ethics Committee in the Affiliated Hospital of Hangzhou Normal University (2021-E2-HS-029) and was conducted in accordance with the Declaration of Helsinki.

Somatosensory functioning was evaluated using the modified Fugl-Meyer and Lindmark Assessment (Fugl-Meyer et al., 1975; Lindmark and Hamrin, 1988), which included superficial sensation (pain, temperature, and touch), deep sensation (proprioception, motion perception, and vibration), cortical sensation (two-point discrimination, stereognosis), and subjective sensation. The subjective sensation was added due to the fact that patients tended to report subjective somatosensory deficits that could not be characterized by the three sensory dimensions (superficial, deep, or cortical sensation) (Gao et al., 2023). The assessment of somatosensory function systematically covered the trunk, limbs, and head with 29 items (superficial = 10; deep = 13, cortical = 3, and subjective = 3), and had a total score of 58 (three-step scale: 0, 1, 2). The score of somatosensory functioning assessed both the ipsilesional and contralesional sides, whereby deficit scores (determined by deficit item and level) were deducted from the total score of 58 (Chen et al., 2019). It is noted that motor functioning was not systematically evaluated with Fugl-Meyer Assessment (FMA) as these patients generally had no complaints about motor functioning compared to somatosensory deficits during the visit to our hospital.

The resting motor threshold (RMT) over the ipsilesional motor cortex was assessed. RMT was defined as the minimum intensity to induce motor-evoked potentials (MEPs) > 0.05 mV of the first dorsal interosseous (FDI) muscle in 5/10 trials. Single pulses to the hand region of the motor cortex (45° to the midline, handle pointing backward) at 5 s ± 10% jitter intervals were sent by a figure-eight coil connected to a Magstim Rapid2 system (Magstim Company Ltd., UK).

Corticospinal excitability was measured with MEP amplitude and CSP latency at rest and during a sustained voluntary FDI muscle contraction, respectively, (Hupfeld et al., 2020; Liu et al., 2022). A total of 40 single pulses (20 of MEP, 20 of CSP) were consecutively delivered to the hand region of the motor cortex at 110% RMT (45° to the midline, handle pointing backward). It is worth noting that CSP was evaluated following MEP as the muscle contraction during CSP may have an impact on resting MEP amplitude (Conforto et al., 2004). Corticospinal excitability was evaluated from the ipsilesional motor cortex.

Patients received ten daily sessions of either contralesional 1-Hz or ipsilesional 10-Hz rTMS treatment in two weeks. Each rTMS session delivered 1,500 pulses over the S1 at 80% RMT (Conforto et al., 2012). The 10-Hz protocol delivered 5-s trains with 25-s intervals, while consecutive 1-Hz pulses were delivered in the contralesional 1-Hz group. The S1 was located 2 cm lateral and 0.5 cm posterior to the M1-FDI scalp location. This localisation method was proposed by a recent study across a series of experimental investigations (Holmes et al., 2019). Coil position was measured relative to the nasion and inion to facilitate consistent re-positioning of the coil between sessions (Chung et al., 2019; Ye et al., 2022; Wang et al., 2023).

The total score of somatosensory functioning was analysed as the primary outcome measure, while the subdimensions of somatosensory functioning and corticospinal excitability were set as the secondary outcome measures. Intent-to-treat (ITT) analysis was set to include all randomized patients. This method preserves the benefits of randomization and allows to draw unbiased conclusions regarding the effectiveness of an intervention. Multiple imputation algorithm was initially performed, which is a highly recommended methodology for dealing with missing data (McCoy, 2017). Specifically, we initially created 10 sets of imputations for the missing values. We then perform the desired analysis (i.e., ANOVA) separately for each dataset. We then combined the results from the separate analyses to obtain a single set of estimates.

MEP amplitude was calculated by peak-to-peak amplitude. The calculation of CSP latency was based on the Mean Consecutive Difference (MCD) (Garvey et al., 2001; Wang et al., 2024), which was highly recommended by a recent expert review (Hupfeld et al., 2020). The MCD methodology is briefly described here: (1) All silent period trials were rectified using the absolute value and then were averaged; (2) The MCD of 100 ms EMG data before a TMS pulse was calculated, in which the MCD is the mean successive difference between individual data points; (3) Thresholds were set at: ± MCD × 2.66 (i.e., 3 standard deviations), which covers 99.76% of possible pre-TMS EMG data points; (4) Silent period onset was determined as the time point at which the post-TMS EMG falls below the variation threshold for three consecutive data points, while the silent period offset was defined as the point at which the post-TMS EMG returns above the variation threshold for three consecutive data points.

Data were analysed in SPSS (v.25.0 Chicago, Illinois, United States). Demographic variables were initially examined with a two-sample t-test or χ² test. A series of tests were performed to check the assumptions of using a mixed-design ANOVA. Specifically, the Shapiro–Wilk test was performed to check the normality of the outcome measures in different combinations of our two factors. Levene’s test for homogeneity of variances and Mauchly’s Test of Sphericity were also performed. Results validated the use of mixed ANOVA.

Two-way mixed-design ANOVAs were then performed to examine the main and interaction effects of Group (2 levels: contralesional 1-Hz, ipsilesional 10-Hz) and Time (2 levels: pre, post). Post-hoc pairwise comparisons were conducted to further explore the significant main and interaction effects, with the α-level set to 0.05 and Bonferroni corrected.

A total of 86 patients were screened, of which 71 were excluded due to noting meeting the inclusion criteria (n = 48) or not willing to participate (n = 23, Figure 1). Fifteen participants were enrolled and equally randomized to the two groups. One patient withdrew from each group due to discharge from the hospital (both in week 2). Data from 15 patients in each group were analysed with ITT methodology.

The demographics and clinical information of patients were presented in Table 1. Patients were mainly old adults with a mean age of 62. Lesions of the brain were mainly distributed in the thalamus, followed by damage to the brainstem and cortical regions such as the temporal and frontoparietal cortex. Somatosensory functioning had a mean score of 47.53 out of a total score of 58. Patients had normal (5-degree, n = 11/15) or slightly impaired (4-degree, n = 4/15) muscle strength as indexed by the muscle strength grading scale (0–5). The two groups were comparable in all these variables. RMT was also comparable between the two groups (t = −0.42, p = 0.679). Regular stroke medicines included a daily dose of Aspirin (100 mg) and Atorvastatin (20 mg) for each patient.

In terms of the primary outcome, a two-way ANOVA indicated a significant main effect of time on somatosensory function (F_1,13 = 5.05, p = 0.043, η p 2 = 0.28 ) (Figure 2A). Results indicated that both contralesional 1-Hz and ipsilesional 10-Hz stimulation improved somatosensory function from pre-to post-treatment (Mean_pre = 47.52, Mean_post = 50.17, P_Bonferroni = 0.043). No other main or interaction effect was observed (all P_Bonferroni > 0.05). Correlation analyses indicated that RMT was negatively associated with somatosensory function at post-treatment across the two groups (r = −0.63, p = 0.012, n = 15) (Figure 2B).

When the somatosensory function was split into subscales, a two-way ANOVA revealed no significant main or interaction effect in superficial sensations or cortical sensations (all P_Bonferroni > 0.05) (Figures 3A,C). Meanwhile, there was a main effect of time on both deep sensations (F_1,13 = 6.97, p = 0.020, η p 2 = 0.35 ) (Figure 3B) and subjective sensations (F_1,13 = 9.29, p = 0.009, η p 2 = 0.42 ) (Figure 3D). Results indicated that both treatments improved deep sensations (Mean_pre = 23.65, Mean_post = 24.98, P_Bonferroni = 0.020) and subjective sensations from pre-to post-treatment (Mean_pre = 3.47, Mean_post = 4.16, P_Bonferroni = 0.009).

MEP data indicated a significant interaction (F_1,13 = 11.94, p = 0.004, η p 2 = 0.48 ) (Figures 4A,B). Post-hoc comparisons indicated that ipsilesional 10-Hz stimulation increased MEP amplitude from pre-to post-treatment (Mean_pre = 0.70, Mean_post = 0.73, P_Bonferroni = 0.004) while contralesional 1-Hz did not (P_Bonferroni > 0.05). In terms of CSP latency, no significant main (F_1,13 = 0.01, p = 0.971) or interaction effect (F_1,13 = 0.21, p = 0.658) was observed.

There was no serious adverse effect by monitoring patients’ vitality, physical and mental health. There was a slight chance to experience mild headache (0 and 0 in the 1-Hz and 10-Hz group) and/or mild scalp discomfort (0 and 1 in the 1-Hz and 10-Hz group), but these sensations dissolved within minutes or hours. Overall, the treatments were safe and well-tolerated.