The study is an investigator-initiated, phase II, double-blind, placebo-controlled, randomized, parallel group clinical trial with a compassionate crossover design. All study activities were conducted at the Tisch MS Research Center of New York, USA. The study was conducted as an FDA investigational new drug, and is registered with ClinicalTrials.gov, number NCT03355365. The ethics of the study was approved on 11/28/2016 by Western Institutional Review Board (reference number 20,162,572). All patients gave written informed consent. An independent external data and safety monitoring board evaluated all safety data on a yearly basis.

Eligible patients had clinically definite SPMS or PPMS with EDSS disability score between 3.0 and 6.5 with stable disease as determined by less than a 1.0 point change in EDSS and an absence of clinical relapses in the 12 months prior to enrollment, and by lack of gadolinium-enhancing lesions with stable MRI disease burden (number and size of T2 lesions) in the prior six months. Enrollment criteria included disease duration of less than 20 years at time of screening (based on onset of symptoms when symptom onset was clearly defined or based on date of diagnosis if symptom onset was difficult to determine); no change of disease modifying therapy (DMT) less than 12 months prior to beginning the trial; and no change in MS symptom medications, including dalfampridine less than six months prior to study treatment. Subjects with existing medical comorbidities or cancer history that might complicate safety outcomes of the experimental treatment were excluded.

Eligible subjects were stratified by disease subtype (SPMS or PPMS) and baseline EDSS score (3.0–4.0, 4.5–5.5, 6.0, and 6.5). The study aimed to enroll a total of 50 subjects, including 40 SPMS subjects distributed in each of the 4 EDSS blocks (10 subjects per block), and 10 PPMS subjects with four in EDSS block 3.0–4.0 and two each in EDSS blocks 4.5–5.5, 6.0, and 6.5. After ensuring eligibility criteria, subjects in each stratum were block randomized into either the saline or MSC-NP treatment group in order of date of consent based on a pre-determined randomization scheme generated by LG who was not involved in patient enrollment or treatment. Subject allocation and coordination of cell manufacturing was performed by VH, who was unblinded. The single procedure neurologist (SS) who performed all lumbar punctures and intrathecal injection procedures was also unblinded and did not perform any clinical or safety assessments on study participants. All study participants, neurologists assessing clinical outcomes or adverse events, and clinical study coordinators involved in enrollment and data collection were blinded to the treatment assignment.

Cell manufacturing was performed in a cGMP facility at the Tisch MS Research Center of New York. Detailed methods of MSC and MSC-NP manufacturing and release criteria have been described previously and are included in the supplementary material (Supplemental Methods) [9, 13]. In brief, MSCs were isolated from a single bone marrow aspirate from each subject, expanded ex vivo, and cryopreserved at passages 2 and 3. Prior to each treatment, a portion of cryopreserved MSCs were thawed, expanded for 2 more passages, and cultured for 2 weeks in neural maintenance media to generate MSC-NPs. All individual batches of MSC-NPs conformed to release criteria as detailed in the Supplemental Methods. To preserve functionality of the cells, final product autologous MSC-NPs were collected directly from cell culture, washed, and cell count/viability was determined just prior to each injection. Viability was > 80% for all batches of MSC-NPs. A dose of up to 10 million cells was resuspended in preservative-free saline, transferred to an unlabeled sterile tube and immediately transported to the procedure neurologist in a container with a pre-specified label. Placebo samples consisted of an empty tube transported in an identical container with a pre-specified placebo label.

In year one of the study, subjects assigned to the MSC-NP group received six separate IT injections of up to 1 × 107 autologous MSC-NPs spaced two months apart (treatments one through six). Subjects in the placebo group received IT injections of saline following the same schedule. Lumbar puncture and CSF aspiration was performed as previously described [13]. The cell suspension was removed from the transport vial using a 22-gauge needle 10 cc syringe and diluted in 3 ml of preservative-free sterile saline before IT injection. Injection was followed by a 2 ml saline flush. For placebo subjects, the entire procedure was performed with saline only. In two of the subjects, cells were administered intrathecally via an in-dwelling access port of an implanted baclofen pump. All procedures took place posterior to the study subject and behind a screen to ensure that the patient and assisting nurse remained blinded to the treatment. Following each procedure, subjects were placed in the Trendelenburg position for one hour. Prophylactic IV infusion of antibiotics (80 mg of tobramycin and 500 mg of vancomycin) was co-administered, and prophylactic oral acetaminophen was administered to minimize headaches.

The experimental design included a compassionate crossover element for purposes of masking (ensuring all subjects had a bone marrow procedure) and compassionate use of the MSC-NP treatment for subjects receiving placebo. Therefore, in year two of the study, placebo subjects crossed over into the MSC-NP treatment group and received six separate IT injections of autologous MSC-NPs as part of a compassionate crossover design (treatments seven through 12). Similarly, subjects receiving MSC-NPs in year one received saline injections in year two.

The primary outcome was EDSS Plus, defined as improvement in either EDSS, timed 25-ft walk (T25FW), or nine-hole peg test (9HPT) [16]. Improvement was defined by at least one of the following three pre-specified measures: ≥0.5 improvement in EDSS (if EDSS at entry is ≥ 6.0) or ≥ 1.0 improvement in EDSS (if EDSS at entry is ≤ 5.5), ≥ 20% improvement in T25FW, and ≥ 20% improvement in 9HPT in either dominant or non-dominant upper limb. Secondary outcome measures included EDSS, T25FW, 9HPT, six-minute walk test (6MWT), 12-item walking scale (MSWS-12), multiple sclerosis functional composite (MSFC), paced auditory serial addition test (PASAT), urodynamics and MRI imaging. T25FW and 9HPT were analyzed as both a continuous measure (% change after one year) and a binary measure (≥ 20% improvement or ≥ 20% worsening compared to baseline) based on previously defined cutoffs determined to be clinically meaningful [16]. All outcome assessments were performed at baseline, two months after the sixth treatment and two months after the 12th treatment. Additional EDSS Plus assessments (EDSS, T25FW and 9HPT) were performed at mid-year timepoints (treatments four and ten).

To assess bladder function, a history of symptoms was taken and any use of medications affecting bladder function was noted. Subjects underwent urodynamic testing performed in the same laboratory and results interpreted by a single neuro-urologist. Bladder function measurements included post-void residual volume (PVR) and maximum bladder capacity. Bladder function improvement was determined by a decrease in PVR of > 20% after 1 year of treatment compared to baseline.

Brain, cervical and thoracic spine MRI scans with and without gadolinium enhancement were read by a neuro-radiologist. As an exploratory outcome, images were analyzed using the NeuroQuant® volumetric MRI software (CorTechs Labs, Inc.; La Jolla, California, USA) which measures and compares volumes of brain structures to a normative database adjusted for age, gender and intracranial volume. A multi-structure atrophy report was generated for all brain MRI scans as well as a LesionQuant™ Flair assessment of total number and volume of lesions.

Each subject was assigned to an assessing neurologist who performed and oversaw all outcome assessments, and a separate adverse event (AE) neurologist to whom all AEs were reported. All subjects were contacted two, seven, and 30 days following each procedure and asked if they experienced any AEs including headaches. All AEs were reported and graded following the NCI Common Terminology Criteria for Adverse Events (CTCAE version 4.0). If a subject reported a headache, they completed the headache pain scale (0–10 scale). Headaches were categorized as mild (0–3), moderate (4–7) or severe (8–10). The clinical data collection and management was performed using the OpenClinica open-source software, version 3.13.1.

As part of the study protocol, CSF was collected at each IT procedure just prior to cell or saline injection. Cell-free CSF was processed as previously described [17]. For biomarker discovery, two CSF samples representing baseline and post-MSC-NP treatment from 36 subjects were included for a total of 72 samples. For the majority of sample pairs (26/36), CSF obtained at the time of the first MSC-NP treatment was used as a baseline sample, and CSF obtained at the time of the sixth MSC-NP treatment (reflecting five treatments) was used as a post-treatment sample. In the remaining pairs (10/36) the timing of baseline or post-treatment samples was modified to avoid blood contaminated samples. Twenty subjects received MSC-NP treatments in year one, and 16 subjects received MSC-NP treatments in year two.

Proteomic analysis of CSF was performed using Slow Off-rate Modified Aptamers assay (SOMAScan Assay v4.1, SomaLogic, Inc). SOMAmer-protein binding was quantified using DNA-hybridization microarrays and normalized using hybridization controls. Median signal normalization was performed using Adaptive Normalization by Maximum Likelihood. On average, 17.2% of SOMAmers fell below the limit of detection (range 2.3–46.4%), suggesting low signal in CSF samples. Signal was log2 transformed for statistical analysis. Candidate biomarkers were validated in CSF samples from all trial subjects who received MSC-NP treatment. MMP9 and CCL2 were measured in undiluted CSF using human MMP9 magnetic Luminex performance assay (R&D Systems) and Bio-Plex Pro human cytokine panel from Bio-Rad, respectively. Bio-Plex Pro 200 system (Bio-Rad) was used for analyte detection. Responders to MSC-NP treatment were defined by improvement in any of the following outcomes following MSC-NP treatment: EDSS (≥ 0.5 point improvement), timed 25-foot walk (≥ 20% improvement), or 9-hole peg test (≥ 20% improvement). Non-responders were defined by stability or worsening in any of the outcomes.

Sample size of 20 in the treatment group and 20 in the placebo group was calculated to achieve more than 80% power to detect a difference between the group proportions of 35%, using a two-sided Z-test with pooled variance and using a significance level of 0.05. This calculation assumes that 40% of patients in the treatment group and 5% of patients in the placebo group achieve at least one of the EDSS-Plus components. The prevalence of patients with improved outcome was informed by the results of the phase I trial, in which 40% of patients experienced a decrease of 0.5 or more on EDSS [13]. Target enrollment of 50 patients (25 patients in each group) was designed to account for an expected attrition rate of 20%.

The primary outcome, EDSS Plus improvement, and secondary outcomes were analyzed by treatment in the first year. Counts and percentages are displayed for categorical variables (EDSS Plus improvement, T25FW improvement, and T25FW worsening) and median and interquartile range for continuous variables (percent change in EDSS Plus, T25FW, 6MWT, 9HPT in the dominant hand, 9HPT in the non-dominant hand, MSFC, MSWS-12 and PASAT). Univariate hypothesis tests (Mann-Whitney U test for continuous variables, Chi squared tests or Fisher’s exact tests for categorical variables) were performed to compare each variable between placebo and MSC-NP treated patients using R version 4.2.3. The T25FW and 6MWT walking outcomes were then reassessed within low (3.0-5.5) and high (6.0-6.5) EDSS strata.

Differentially expressed proteins were identified from proteomic analysis using a multivariate linear mixed effect model with patient age and EDSS modeled as a continuous covariate and other independent variables (treatment, diagnosis, and gender) as discrete. The model was performed using R version 4.0.3. Pairwise contrasts (post-treatment minus pre-treatment) were extracted with least squares means function and candidate biomarkers selected based on unadjusted p-values (p < 0.01). Differences between biomarker levels before and after treatment were analyzed using Wilcoxon matched-pairs signed rank test, and significance of longitudinal values was determined by Friedman test followed by Dunn’s multiple comparisons test using GraphPad Prism 9. Differences in PVR volumes and grey matter volumes were determined by Mann-Whitney test.