The study was approved by the Ethics Committee of Charité-Universitätsmedizin Berlin and conducted in accordance with the Declaration of Helsinki. All participants approved written informed consent. The study was registered prospectively at the German Clinical Trials Register (DRKS00023357). We included patients that were diagnosed with Primary or Secondary Progressive MS and gait deficits due to spasticity in the legs. Exclusion criteria for the study comprised active lumbosacral nerve root compression, condition after surgery of the vertebral bodies, metal implants in the stimulation area, pacemaker implant, MS relapse event in the past six months, or current pregnancy. As this was a pilot study, the sample size was based on a similar previously published study (23) in this clinical population and by the available funding and recruitment capacity. The participants maintained their medication regime during the study.

The study design was comprised of two conditions, (1) a 30 min tSCS therapy session and (2) a sham stimulation in randomized order for each patient (Figure 1A). Study examiners implemented a quasi-randomization using Microsoft Excel's random number generator, ensuring balanced group sizes. Participants and outcome raters of the clinical spasticity score were blinded to the assigned sequence. Allocation concealment was maintained until after the trial. Study examiners were responsible for enrolling participants and assigning them to the intervention sequences. Between the two conditions, we ensured a washout phase of at least seven days. Figure 1B displays the study protocol at each study appointment. Assessments were performed prior (pre-therapy, pre-sham) and immediately after each intervention (post-therapy, post-sham) in a fixed order. The assessments consisted of a clinical examination of spasticity and three gait tests: the 10 Meter Walk Test (10MW), the Timed-Up-and-Go Test (TUG) and the 2 Minute Walk Test (2MinW). The total duration of each assessment battery was approximately 20 min. We determined the required time for the 10MW and TUG through recorded videos. The patients completed the gait tests with their preferred walking aid. The walking aid used in each test remained the same on both study days. The examiner instructed the gait speed as “brisk but safe”. For the 2MinW, we mounted IMUs (WaveTrack, Cometa srl, Italy) on the feet, shanks, thighs and hip to record 3D acceleration and gyroscopic data at 286 Hz for kinematic analysis.

During spasticity ratings, a blinded rater determined the Modified Ashworth Scale (MAS) for hip, knee, and ankle joints (24). Ratings were entered for twelve movements on each side: hip flexion, extension, abduction, adduction, internal and external rotation; knee flexion and extension; ankle dorsiflexion with hip and knee in a flexed position, and dorsiflexion, plantar flexion, and pronation of the ankle with hip and knee in an extended position. Each joint movement was scored as 0, 1, 1+, 2, 3, or 4 according to the severity of spasticity ranging from “no increase in muscle tone” (0) to “rigid in flexion/extension” (4). The maximum possible bilateral of the MAS sum score is 96 points.

For tSCS, we placed a self-adhesive hydrogel electrode (9 x 5 cm, axion GmbH, Leonberg, Germany) over the participant's lumbosacral spinal cord (Figure 2A), with the bottom edge 0–5 cm cranial to vertebral level L3–L4 covering the vertebrae T11-T12. The exact position depended on the individual tSCS tuning response. Two interconnected counter electrodes (8 x 13 cm) were located on the patient's lower abdomen. Throughout the interventions, the participants remained in a relaxed supine position on a daybed with a knee roll under their knees. For the stimulation, we used a current-controlled stimulator (RehaMove3, Hasomed GmbH, Germany) that delivered symmetrical biphasic rectangular pulses with 1 ms pulse-width per phase. The electrode on the patient's back served as the anode for the first phase and as the cathode for the second phase of the biphasic pulse. This configuration has been shown to elicit strong depolarization on the transition between the phases (25). Before and after the tSCS therapy or sham intervention, each patient underwent a tSCS tuning procedure to find the target therapy intensity and to investigate recruitment characteristics of leg muscles. Here, double pulses with an inter-pulse-interval of 50 ms were applied with increasing intensities starting at 5 mA and a pause of 5 s between each double pulse. To verify effective stimulation of afferent fibers, electromyographic sensors (EMG) (1,000 Hz; MuscleLab; Ergotest, Porsgrunn, Norway) recorded the responses bilaterally in two leg muscle groups (Triceps Surae (TS) and Quadriceps (Q) muscle group) (Figure 2A). A post activation depression of the Posterior-Root-Muscle-Reflex (PRM-reflex) following the second stimulation pulse confirmed the selective recruitment of posterior root afferents (26) (Figure 2B). A current level of 90% of the PRM-reflex threshold defined as the first muscle response with a peak-to-peak amplitude greater than 50 μV served as the target stimulation intensity for tSCS therapy (23). For the therapy condition, we chose a continuous stimulation of 50 Hz with a total duration of 30 min. At the beginning of the stimulation, we ramped up the current over 1-15 minutes with a median of 6 minutes depending on the participants individual comfort. The therapy application was considered successful when reaching the target therapy intensity or, if the target intensity could not be reached, but the patient reported paresthesia in the lower extremities during stimulation (16, 27, 28). For the sham condition, we applied a time limited current at maximum 20% of the therapy target for a total of 3 min. The examiner then turned off the stimulation but informed the patient that the stimulation was decreased to a lower level. For the residual 27 min of the sham therapy, the patients remained in supine position. This choice of the sham condition was intended (1) to evoke a transient mild tingling on the back, resembling the stimulation condition, (2) to be time-limited to a short period of few minutes (29, 30), and (3) to not evoke a direct recruitment of posterior root afferents at current levels far away from PRM-reflex threshold (30, 31). Before the first intervention, the examiner informed the participant that two different stimulation settings would be tested, and that the stimulation may or may not be perceived.

From the IMU recordings during the 2MinW we derived kinematic variables. The kinematic evaluation is based on orientation estimation using quaternions. All body turns were excluded from the kinematic analysis. We processed the foot IMU data according to Laidig et al. (32) and determined the average of the maximum pitch angle (ankle dorsiflexion) observed in each step, calculated across all steps for each participant (in the following referred to as max pitch). The max pitch is defined as the maximum angle between the foot sole and the floor in the swing phase. To further extract the average Range of Motion (ROM) in each step of hip and knee joint across all steps (in the following referred to as ROM hip and ROM knee), we employed the data of the two adjunctive IMUs and used a heading drift correction (33). Only the ROM for flexion-extension movement during gait was considered, therefore we assumed the joints as 1D hinge joints. Due to data quality issues, we only included a subset of the patient population in the analysis for each kinematic parameter.

The primary outcome was defined as the bilateral MAS sum score of both legs. Since most patients showed asymmetric deficits, we additionally evaluated the MAS sum score and kinematics separately for the more and less severely affected leg. The more affected leg was determined separately for each reported parameter, and was defined as the leg with the higher average baseline MAS sum score, smaller average baseline ROM of the knee and hip, or smaller average baseline max pitch, respectively. Secondary outcomes included the unilateral MAS sum scores, gait test results (time, distance), kinematic gait parameters, and electrophysiological analysis of the tuning procedure. For this analysis, we calculated the maximum EMG response during the tuning procedure after the interventions and normalized it to the maximum EMG response observed before the interventions. This normalized value was then averaged across all active muscles and will be referred to as the EMG tuning ratio. We hypothesized that a reduction in leg spasticity has an influence on the excitability of spinal reflexes. Toward the end of the 30 min tSCS intervention, the participants responded to a pain and sensation questionnaire with items derived from the TES Comfort Questionnaire (34) and McGill Pain Questionnaire (35, 36) (Supplementary material 1).

We compared the change after the interventions (post-pre) between the two conditions for each participant and parameter. The difference between the short-term effect of tSCS- vs. sham-treatment was statistically evaluated using Generalized Estimation Equation (GEE) models (37) on the outcome parameters. We adjusted for baseline values, the order of intervention and appointment number. Only for the EMG tuning ratio the baseline adjustment was redundant. A two-sided significance level of α = 0.05 was used for the primary outcome (bilateral MAS sum score). We further calculated a standardized effect size r based on the model's result. The effect size was calculated using the treatment estimate (GEE_{coeff(therapy)}) from our model, the standard deviation (SD) of the binary condition mapping (x) and the SD of the change in the respective outcome parameter (y_{post − pre}).

This value shows how strong the difference between the condition is, relative to the natural variation in the outcome parameter. The interpretation of the effect size aligns with Cohen's d (38), where values of 0.2 and above indicate a small effect, 0.5 and above a moderate effect, and 0.8 and above a large effect. All signal processing operations were implemented in Python (3.9). Statistical calculations were done in R (4.2.2). We further categorized each participant's individual treatment response in the MAS and gait test performance into different levels of improvement. For the MAS sum score, we distinguished between an deterioration, no change, a small improvement defined as a decrease in the MAS sum score of less than 1, a medium improvement defined as a decrease of at least 1 but less than 2, and a large improvement defined as a decrease of at least 2. For the gait test, we defined the level of improvements in the gait tests as stated in (23) with a large improvement defined as an increase in velocity of at least 0.05 m/s in the 10 MW (39), a decrease in time of at least 15% in the TUG (40, 41), and an increase in distance of at least 6.8 m in the 2MinW (42).

The bilateral MAS sum score varied strongly among the patients (e.g., pre-therapy: mean 24, ranging from 10 to 63). The effect size of the change in bilateral MAS sum score was small (r = −0.25; 95% Confidence Interval (CI): −5.68–0.63; p = 0.12) (Figure 4A). The effect was statistically not significant. Although not significant, the largest numerical reduction was seen at the ankle joint (Supplementary material 3). The number of patients that improved in the MAS sum score was larger in the therapy (63%) than in the sham condition (44%). In the therapy condition, 44% of participants improved by at least 2 points, compared to 31% in the sham condition.

Our results agree with the small effect sizes for spasticity reduction reported in one previous single-armed clinical study (23). The specific reduction in MAS with tSCS therapy may be of clinical interest especially for the treatment of patients showing intolerance to antispastic medication. On the other hand, we did not observe any relevant gait improvements when comparing tSCS to sham. Overall, there was also no significant correlation of the change in MAS with a change in gait speed across modalities (Supplementary material 4). Therefore, changes in leg spasticity may not immediately translate into gait improvement in patients with progressive MS. Future studies would need to address whether repeated interventions of tSCS could increase therapy effects and improve gait execution when combined with individualized physiotherapy (43), and might act on additional impairment related phenomena, such as delayed muscle soreness (44). We did observe small improvements in gait speed during the three gait tests when comparing therapy and sham conditions to their respective daily baselines. These changes were most likely explained by placebo effects and experimental factors, such as lying for 30 min in a relaxed supine position.

Defining optimal sham conditions for the design of neuromodulation studies remains a conceptual challenge; especially in the case of transcutaneous stimulation methods such as tSCS or Transcutaneous Electric Nerve Stimulation (TENS) that can be consciously felt by the study participants. Here, we decided to apply a sham condition that provided a minimal amount of current for a short duration of 3 min. We reasoned that this intervention is appropriate because it would reduce the ability of patients to distinguish between the sham and stimulation condition. In both conditions, the patient felt a mild sensation in the back and abdomen at the beginning of the intervention. We also informed all patients that two different stimulation settings would be tested during the study, and the sensations for each intervention may differ. As described in Estes et al. (27), we informed the patient that the stimulation would be decreased to a lower level after 3 min when it was in fact turned off. These measures were influenced by previous sham-controlled studies on tSCS (29, 30, 45) or TENS (46, 47). Also, we incorporated ideas from previous studies that used sham stimulation currents restricted in time and at currents several factors under sensory threshold (31, 46, 48, 49). In our case, the sham condition was useful for detecting specific effects of tSCS on the reduction of the MAS sum score.

A further challenge for tSCS application is to find optimal stimulation parameters in different disease modalities and patients. For spasticity treatment, previous studies in MS and SCI patients have used various frequencies (e.g. 30 Hz, 50 Hz, 5 Hz), various amplitudes, stimulation durations (30-120 min), and waveform characteristics (monophasic or biphasic) (10, 50, 51). In our case, we chose a set of parameters at 50 Hz with biphasic pulses that has previously been reported to reduce leg spasticity (16, 23). For movement control or stimulation during motor tasks the frequency is typically set to a lower value of around 30 Hz (30, 50, 52). An accompanying reduction in spasticity has been reported for these frequencies concerning the upper extremities (50). Some studies have incorporated high-frequency carrier frequencies (at 10 kHz), but their benefits for spasticity reduction remain unconfirmed (51, 53). Keesey et al. (54) even found an undesirable recruitment shift toward efferent fibers with carrier frequencies. In summary, further studies are needed to investigate optimal tSCS settings. MS patients typically show substantial heterogeneity in lesion location and symptom manifestations. Future characterizations of patient phenotypes in relation to stimulation effects could lead to the development of better personalized treatment strategies.

Current guidelines on non-medication treatments for spasticity include TENS. In comparison, tSCS is a rather new intervention and has not entered common medical guidelines. At present, the question remains how tSCS and TENS compare. Both interventions target sensory afferents (55). TENS electrodes are typically located on the affected peripheral muscle or nerve, and therefore only activate few nerve fibers (56). In comparison, Danner et al. (57) computationally predicted that lumbar tSCS activates the roots of several spinal segments at once. The reported effect of TENS on spasticity in MS patients varies in literature. Shaygannejad et al. (58) reported a greater decrease in MAS after repeated TENS treatment compared to baclofen. Another study did not find a significant short-term effect after TENS therapy (48). Transcutaneous Spinal Cord Stimulation may present a treatment method, when several muscle groups are affected by leg spasticity or other treatment modalities are not effective.

Here, we used patient reported outcome measures to assess potential side effects of tSCS. Similar to previous reports (10, 13, 50, 51), our study participants reported no severe side effects. Mild side effects such as discomfort and skin irritations have been described for SCI. Six MS patients in our study reported a sensation of pain. This effect may be more prominent in MS patients with intact body sensations. In the depth of the spinal cord, tSCS primarily targets large diameter spinal afferents (12) whereas pain fibers are typically much smaller in diameter (59) and should not be directly recruited. The pain sensation was most likely explained by the activation of local pain fibers in the skin (60). In our experience, this sensation could be minimized or prevented by cleaning the skin and ensuring a homogenous contact of the stimulation electrodes with the skin surface. Overall, our data showed a good safety profile of tSCS.

A limitation of our study is the potential inter-rater variability of the MAS. This has been readily described in literature (61–63). Still the MAS remains one of the main primary outcome measures for rating of spasticity (63, 64). To reduce the impact of inter-rater variability in this study, we appointed the same blinded rater for all measurements in a given participant. Biomechanical assessments using isokinetic dynamometers can in principle offer complementary measurements of spasticity in future studies (65).

Another limitation of our study is the small sample size of 16 participants, which is appropriate for a pilot trial design, yet limits the statistical power and increases the risk of a type II error. The dataset establishes effects sizes and can contribute to prospective group planning. Yet, to confirm the observed effects of tSCS and establish statistical significance, a larger number of patients, ideally distributed across multiple centers, will be required in the future.

Additionally, this study investigated tSCS as a standalone intervention. Transcutaneous spinal cord stimulation paired with functional training has been demonstrated to be an effective method for spasticity reduction and functional recovery in SCI research (30, 50, 66–68) and should be considered in future studies.

As stated in the discussion, finding a suitable sham condition for tSCS is challenging. Despite the precautions taken in this study, the possibility of unblinding and it's resulting expectation bias cannot be entirely excluded.

Another limitation of the study protocol is the lack of follow-up measurements, e. g. after several hours or days after the interventions (23). Additional assessment time-points in future investigations would give additional insights on the sustainability of effects.