{"paper_id":"276c1cd8-d809-4d5e-9cb9-bac8bb18188c","body_text":"Intrathecal mesenchymal stem cell therapy in progressive multiple sclerosis: cross-compartment immune profiling in the SMART-MS randomized trial | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Intrathecal mesenchymal stem cell therapy in progressive multiple sclerosis: cross-compartment immune profiling in the SMART-MS randomized trial Dimitrios Kleftogiannis, Sonia Gavasso, Live Egeland Eidem, Jonas Bull Haugsøen, and 16 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9629755/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Therapeutic options for progressive multiple sclerosis (MS) remain limited, and the biological mechanisms engaged by intrathecal mesenchymal stem cell (MSC) therapy are incompletely understood. MSCs are proposed to exert immunomodulatory and trophic effects, yet most studies rely on targeted biomarkers and lack systems-level analysis across immune compartments. Here, we applied an integrated multi-omics framework to characterize immune and cerebrospinal fluid (CSF) responses to MSC therapy in patients with progressive MS enrolled in the SMART-MS trial, a randomized, placebo-controlled crossover study of a single intrathecal MSC injection. Although the clinical trial did not demonstrate a clear neuro-regenerative signal as the primary endpoint, exploratory MRI findings and adverse events (e.g., fever, back pain) suggested localized biological responses following intrathecal administration. Longitudinal peripheral blood mass cytometry and matched CSF proteomics were analysed from 18 participants sampled at baseline, 6 months, and 12 months. Immune phenotypes and composite functional program scores were quantified using mixed-effects modelling. To synthesize CSF proteomic changes into biologically interpretable patterns, we summarized protein-level responses into composite CSF functional programs reflecting extracellular matrix and CNS interface remodelling, innate and vascular inflammatory stress, metabolic and cytoskeletal adaptation, and biological reactivity. These program-level scores enabled structured cross-compartment integration with circulating immune programs. MSC exposure did not broadly alter peripheral immune composition or functional programs across circulating lymphocyte and monocyte populations. Instead, the dominant signal emerged within the CSF proteome, where treatment was associated with coordinated extracellular matrix and CNS interface remodelling alongside attenuation of acute-phase inflammatory pathways in many individuals. Elevated inflammatory and metabolic signatures were largely confined to patients with clinical adverse events or spinal MRI reactivity, consistent with amplified biological responsiveness rather than distinct MSC-specific mechanisms. Cross-compartment analyses revealed weak and heterogeneous coupling between circulating immune programs and CSF remodelling, supporting predominantly compartmentalized intrathecal effects. Together, these findings suggest that intrathecal MSC therapy in progressive MS is associated with selective remodelling at the immune–CNS interface rather than broad systemic immunosuppression and demonstrate the value of integrated multi-omics approaches for dissecting treatment-associated biology in neuroinflammatory disease. Multiple sclerosis mesenchymal stem cell therapy mass cytometry immune profiling CSF proteomics bioinformatics single cell analysis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. BACKGROUND Multiple sclerosis (MS) is a chronic immune-mediated disease of the central nervous system (CNS) affecting more than 2.9 million individuals worldwide, with typical onset between 20 and 40 years of age. Clinically, approximately 90% of patients initially present with relapsing–remitting MS (RRMS), whereas around 10% are diagnosed with progressive MS, a disease course characterized by gradual accumulation of disability, impaired mobility, cognitive decline, and ongoing neuroaxonal loss, particularly within the spinal cord [ 1 ]. Despite substantial advances in the treatment of RRMS, there remains a major unmet need for effective therapies in progressive MS. To date, ocrelizumab is the only approved therapy shown to slow disability progression in this population, underscoring the limited therapeutic options available [ 2 ]. Progressive disease is increasingly understood to reflect compartmentalized inflammation within CNS border regions that is poorly captured by systemic immune readouts. Consequently, therapeutic strategies capable of simultaneously modulating immune activity, promoting neuroprotection, and supporting tissue repair are considered particularly relevant for altering long-term outcomes in progressive MS [ 3 – 4 ]. Against this background, mesenchymal stem cells (MSCs) have emerged as a promising therapeutic candidate. MSCs are multipotent stromal cells that can be isolated from adult tissues such as bone marrow or adipose tissue, as well as from perinatal sources, and can be administered in autologous or allogeneic settings without the ethical concerns associated with embryonic stem cells [ 5 , 6 ]. Unlike induced pluripotent stem cell–based approaches, MSC therapies do not require genetic reprogramming and are supported by extensive preclinical evidence demonstrating immunomodulatory, trophic, and tissue-repair–associated properties. Early-phase clinical studies and small randomized trials of intrathecal MSC or MSC-derived neural progenitor therapies in progressive MS have reported acceptable safety profiles and signals suggestive of biological activity. A recent example is the SMART-MS trial [ 7 ], a randomized, double-blind, placebo-controlled phase I/II crossover study evaluating intrathecal administration of autologous bone marrow–derived MSCs in progressive MS (NCT04749667). In this trial, 18 patients with non-active progressive disease received a single intrathecal injection of MSCs or placebo followed by crossover at six months, enabling within-patient comparisons while maintaining blinding. Although the study did not demonstrate clear neuro-regenerative efficacy on its primary endpoint, MSC exposure was associated with transient biological signals, including exploratory MRI signals suggestive of transient tissue effects. Adverse events such as fever, back pain, and spinal MRI abnormalities suggested localized inflammatory or reactive responses following intrathecal administration, highlighting the need to define how intrathecal MSC therapy reshapes immune-CNS interface biology in vivo [ 8 – 14 ]. In parallel, rapid advances in high-dimensional omics technologies now enable detailed interrogation of immune cell composition, activation states, and functional programs in MS, offering new opportunities to dissect treatment-associated biological effects in vivo [ 15 – 17 ]. However, most MSC studies to date have relied on targeted biomarker approaches focusing on a limited number of inflammatory chemokines, complement components, matrix metalloproteinases, or neuroaxonal injury markers, often assessed in isolation within cerebrospinal fluid (CSF) and with minimal integration of comprehensive immune profiling in peripheral blood or cross-compartment analyses [ 18 , 19 ]. As a result, how MSC therapy modulates immune and tissue-associated programs across systemic and intrathecal compartments in progressive MS remains incompletely understood. To address this gap, we performed integrated longitudinal immune profiling in patients with progressive MS enrolled in the SMART-MS trial [ 7 ], combining mass cytometry (CyTOF – Cytometry by Time-of-Flight)–based single-cell analysis of peripheral blood at baseline, 6 months, and 12 months with matched CSF proteomic profiling in 18 participants. By jointly interrogating immune composition, functional programs, and CSF proteomic remodelling, we provide a systems-level view of MSC-associated biological effects. Our findings indicate that intrathecal MSC therapy is associated with selective remodelling of the CSF interface and extracellular matrix rather than broad modulation of circulating lymphocyte and monocyte programs. Inflammatory and metabolic proteomic responses were primarily enriched in individuals experiencing clinical or radiological reactivity, highlighting substantial inter-individual heterogeneity in treatment-associated biology. 2. RESULTS 2.1 Baseline peripheral immune composition in progressive MS To characterise the baseline peripheral immune landscape of patients with progressive MS, we performed high-dimensional profiling using CyTOF with a 41-antibody panel (Supplementary Table 1). An overview of the SMART-MS trial design and the integrative analysis workflow is provided in Fig. 1. Phenotypic characterisation of peripheral blood leukocytes was performed using unsupervised clustering of CD45⁺ and CD66b low/neg cells with the FlowSOM algorithm [20] (n = 2,558,027 cells at baseline from m = 14 patients; Supplementary Fig. 1), yielding 25 distinct populations spanning CD4⁺ and CD8⁺ T cells, B cells, monocytes, natural killer (NK) cells, and dendritic cells (Fig. 2A). In the two-dimensional Uniform Manifold Approximation and Projection (UMAP) representation, these phenotypes formed well-separated and coherent clusters, indicating that the unsupervised approach captured biologically meaningful populations rather than arbitrary data partitions. To validate the phenotypic assignments, we examined average marker expression profiles across all 25 clusters. The heatmap in Fig. 2B revealed clear lineage- and differentiation-specific patterns: CD4⁺ T cell clusters were enriched for CD4 and displayed CD45RA/RO-defined naïve, central memory (CM), effector memory (EM), and effector memory cells re-expressing CD45RA (TEMRA) phenotypes; CD8⁺ T cell clusters expressed CD8a and corresponding differentiation markers; B cell clusters were defined by CD19/CD20 and CD27/CD38-based naïve and memory states; and myeloid and NK-cell clusters were distinguished by CD14/CD16/CD11c/HLA-DR and CD56/CD16 expression, respectively. Together, these expression profiles re-confirmed that the identified clusters correspond to well-recognised immune subsets and established a robust framework for downstream compositional and functional analyses. We next examined lineage-level distributions at the patient level (Fig. 2C). Stacked bar plots demonstrated modest but consistent inter-individual variability in immune lineage proportions. As expected at baseline, the peripheral immune compartment (CD45 + CD66b low/neg ) was dominated by T cells and monocytes, with smaller contributions from B cells, NK cells, and dendritic cells. Considerable heterogeneity in the abundance of specific immune subsets across patients was observed (Fig. 2D), highlighting the inter-patient variability characteristic of progressive MS. Collectively, these analyses provide a comprehensive baseline reference of peripheral immune composition prior to treatment, forming the foundation for subsequent evaluation of MSC-associated effects on immune cell frequencies and functional states. (A) Cross-over design of the SMART-MS trial. Eighteen patients with progressive multiple sclerosis were randomized 1:1 to receive either intrathecal mesenchymal stem cell (MSC) therapy followed by control (Sequence A, MSC_first) or control followed by MSC (Sequence B, Control_first). Peripheral blood samples were collected at baseline (BL), 6 months, and 12 months. Cerebrospinal fluid (CSF) samples were collected at BL and 6 months. Yellow arrows indicate periods of MSC exposure, while blue arrows indicate control periods. All patients underwent bone marrow aspiration prior to MSC or control administration. (B) Overview of the integrative data analysis workflow combining longitudinal peripheral blood immune profiling and CSF proteomics. Peripheral blood samples from all time-points were processed for mass cytometry (CyTOF) using in-house protocols and a titrated and validated antibody panel. Unsupervised clustering of CD45⁺ CD66b low/neg leukocytes identified 25 immune phenotypes, which were subsequently analysed along two complementary dimensions: (i) cellular composition, assessed using centred log-ratio–transformed (CLR) cluster frequencies and mixed-effects models accounting for treatment, period, sequence, and patient pairing; and (ii) functional program analysis, based on composite marker-derived scores summarizing activation, differentiation, and trafficking states. In parallel, CSF samples collected at baseline and 6 months underwent proteomic profiling followed by quality control and normalization. Differential protein abundance analysis and GO–based gene set enrichment analysis (GSEA) were performed, and selected CSF functional programs were derived using known protein sets. Mechanistically related functional programs across blood and CSF compartments were subsequently coupled to assess coordinated immune–CNS dynamics. (A ) UMAP projection of all CD45⁺ CD66b low/neg cells at baseline coloured by the 25 immune phenotypes identified by unsupervised clustering. (B) Heatmap of scaled mean marker expression for each phenotype, with side colours indicating major lineages (CD4⁺ T cells, CD8⁺ T cells, B cells, NK cells, dendritic cells, monocytes). (C) Baseline lineage-level composition per patient, showing the relative abundance of each major lineage (sum of corresponding phenotypes) as stacked bars. Each bar corresponds to an individual patient (sample IDs S1–Sxx), displayed without a specific ordering. (D) Baseline relative abundance of each phenotype. Large points represent the cohort median with standard error of the mean, while smaller points in gray indicate individual patients, illustrating inter-individual heterogeneity. 2.2 Peripheral immune composition remains stable during intrathecal MSC therapy We next investigated whether MSC therapy alters the composition of the peripheral immune compartment over time. To this end, CD45⁺ CD66b low/neg peripheral blood leukocytes from SMART-MS participants were analysed by CyTOF at 6 months (6m) and 12 months (12m) following treatment, according to the cross-over trial design. Multidimensional profiling of immune cells at these time points (n = 2,636,447 cells at 6m from m = 16 patients; and n = 1,828,594 cells at 12m from m = 12 patients; Supplementary Fig. 2) yielded UMAP embeddings that were visually comparable to baseline, with well-defined and coherent immune phenotypes preserved over time (Supplementary Fig. 3). At the lineage level, stacked bar plots summarised the relative abundances of CD4⁺ T cells, CD8⁺ T cells, B cells, monocytes, NK cells, and dendritic cells for each patient at 6 and 12 months, stratified by active treatment at the corresponding visit (Fig. 3A). Overall, the peripheral immune composition remained similar during MSC and control periods at both time points. While inter-individual variability in lineage proportions was evident, no lineage exhibited a consistent expansion or depletion associated with MSC exposure. To explore potential treatment effects at higher resolution, we visualised centred log-ratio (CLR)–transformed frequencies of selected immune phenotypes, including CD4 central memory T cells, CD4 TEMRA cells, memory B cells, and CD16 low NK cells, across baseline, 6-month, and 12-month visits, stratified by treatment status (Fig. 3B). Violin and box plots with overlaid patient trajectories revealed modest within-patient fluctuations over time but did not indicate large or systematic differences between MSC and control periods. For example, CD4 TEMRA cells showed slightly higher CLR values during MSC periods in some patients, whereas others displayed minimal or opposing changes, resulting in an overall weak and inconsistent pattern. Similar limited variability without a clear treatment-specific shift was observed for CD4 central memory (CD4_CM), B memory (B_Mem), NK dim CD16 low , and other immune subsets (Supplementary Fig. 12). To formally quantify these observations, we fitted linear mixed-effects models to CLR-transformed frequencies of all 25 immune phenotypes. Across all immune subsets, estimated MSC treatment effects were small, with most effect sizes lying within ± 0.25 on the CLR scale and confidence intervals including zero (Fig. 3C; Supplementary Fig. 13). No phenotype reached statistical significance after correction for multiple testing. The largest positive treatment estimates were observed for transitional B cells and CD4 TEMRA cells, whereas plasmablast B cells showed the most negative estimate; however, these effects were modest in magnitude and not statistically significant (Supplementary Fig. 4). As a sensitivity analysis, we fitted an alternative mixed-effects model comparing MSC exposure to untreated baseline samples. This analysis yielded consistent results, with small and non-significant effect estimates across immune subsets (Supplementary Fig. 14). Collectively, the concordant findings from both mixed-effects modelling approaches indicate that MSC therapy does not induce major changes in systemic immune composition, consistent with compartmentalized CNS inflammation in progressive MS. Given this compositional stability, we next investigated whether MSC treatment might instead modulate more subtle functional properties within otherwise stable immune compartments. (A) Stacked bar plots showing the relative abundances of major immune lineages (CD4⁺ T cells, CD8⁺ T cells, B cells, monocytes, NK cells, and dendritic cells) for each patient at 6 and 12 months, stratified by active treatment at the corresponding visit. (B) Violin and box plots of centred log-ratio (CLR)–transformed frequencies for selected immune phenotypes (CD4 central memory, CD4 TEMRA, memory B cells, and CD16 low NK cells) across baseline, 6-month, and 12-month time points, with individual patient trajectories overlaid. (C) Forest plot of MSC treatment effect estimates from linear mixed-effects models fitted to CLR-transformed frequencies of all 25 immune phenotypes. Points indicate estimated treatment effects, and horizontal bars represent 95% confidence intervals. 2.3 Blood-based functional programs show limited modulation under MSC treatment To move beyond compositional changes in peripheral blood, we derived composite functional scores that summarise activation, differentiation and trafficking programs using relevant markers available in the CyTOF panel. The marker–score matrix generation in Fig. 4A illustrates how individual markers map onto ten functional programs, including T cell activation, memory/effector differentiation, inflammatory homing (e.g., Th17-like), CNS-homing, T regulatory-like features (Treg), B cell activation and memory, myeloid APC activation, non-classical monocyte features and NK-cell activation. We first characterised the baseline (i.e., before treatment) functional landscape across all 25 immune phenotypes in the SMART-MS cohort. A heatmap of median baseline scores per phenotype revealed distinct lineage-associated patterns (Supplementary Fig. 7A). At baseline, within the T cell compartment, CD4 central memory T cells exhibited the highest activation and memory/effector scores, with most CD4 and CD8 memory subsets scoring above the global mean, whereas TEMRA subsets showed lower values consistent with a more terminal, non-classical memory phenotype. inflammatory homing scores were selectively elevated in CD4 effector-memory and CD8 effector/central memory cells, indicating that Th17-like inflammatory trafficking programmes were restricted to a subset of memory T cells rather than being uniformly distributed across the T cell compartment. Treg-like scores were negative across conventional T cell phenotypes, reflecting their low CD25/CD73/HLA-DR and high CD127 expression relative to the global immune cell population, although CD4 memory-like subsets were relatively enriched compared with other T cells. In contrast, CNS-homing scores were highest in CD4 effector-memory and CD8 central memory subsets, highlighting these populations as the main carriers of MS-relevant trafficking signatures at baseline. Importantly, these composite scores are not intended to define cell identity but rather to quantify shared functional and trafficking programmes that may be distributed across multiple immune lineages. Accordingly, elevated scores in non–T cell populations reflect engagement in organizing or licensing components of these programmes rather than direct execution of T cell effector functions. Focusing on other immune compartments, all B cell subsets showed strong B-activation and B-memory/naïve scores in switched memory and plasmablast-like clusters, with naive B cells scoring lower along the memory axis. Myeloid APC activation scores were highest in classical dendritic cells and, to a lesser extent, in plasmacytoid DCs and classical monocytes. Non-classical monocyte scores selectively highlighted the non-classical Monocyte (Mono_nonClassical) subset, while NK-activation scores were most pronounced in CD16⁺ NK subsets. These patterns were consistent when inspecting patient-level distributions for each score and phenotype (Supplementary Fig. 5). In this context, high trafficking- or activation-related scores in antigen-presenting cells likely reflect their role in establishing chemokine gradients, antigen presentation, and immune coordination, which in turn shape downstream lymphocyte recruitment and activation. In addition, Principal Component Analysis (PCA) of baseline functional scores showed clear separation of T, B, myeloid and NK phenotypes along the first two components, supporting that the composite scores capture expected lineage-specific functional programs (Supplementary Fig. 6). Notably, CD4 effector-memory and CD8 central-memory subsets occupied overlapping regions of the functional score PCA space, indicating convergence in activation and trafficking programs despite distinct lineage identities. This overlap reflects shared engagement in MS-relevant immune programs rather than phenotypic equivalence. We next focused on T cell–specific scores to examine functional heterogeneity between CD4⁺ and CD8⁺ subsets. At baseline, patient-level jitter points highlight substantial inter-individual variability within each T cell subset (Fig. 4B). The CD4 central and effector memory exhibited the highest T-activation score with high patient heterogeneity. Inflammatory homing scores were particularly elevated in CD4 effector-memory cells consistent with an activated, migratory CD4 compartment in progressive MS. CD8 effector memory and CD8 central memory subsets, were also elevated with CD8 effector memory subset presenting the highest patient heterogeneity, while CD8 TEMRA and CD4 TEMPRA cells also showed moderate enrichment. CNS-homing scores mirrored the previous pattern, with higher values in memory CD4 effector memory and CD8 effector/central memory populations than in naïve subsets. In contrast, Treg-like scores were generally low/negative across T cell clusters, reflecting their lower expression of CD25, CD73 and HLA-DR and relatively high CD127 compared with the global immune cell population. Having established that the composite scores recapitulate known functional specialisation of immune subsets at baseline, we next applied the same scoring framework to longitudinal samples to test whether MSC therapy modulates these functional programs over time. The same linear mixed-model framework as for the compositional analyses was used. Figure 4C summarises the estimated MSC versus control effects across all score and phenotype combinations. Effect sizes were generally small and symmetrically distributed around zero, with most β coefficients lying within ± 0.1 score units and only a few tiles reaching nominal significance. No coherent pattern emerged across related subsets or scores, suggesting that intrathecal injection/infusion of MSCs do not induce large-scale remodelling of these composite functional programmes in peripheral blood. Next, given the potential role of cell trafficking in progressive MS, we focused more deeply on the CNS-homing score. In Supplementary Fig. 7B, we visualised the relationship between baseline CNS-homing scores and those measured during MSC exposure for four representative compartments: CD4 effector memory (CD4_EM), CD8 central memory (CD8_CM), plasmablast-like B cells (B_pl) and classical monocytes. For patients randomised to the Control_first sequence, baseline values were compared to scores at 12 months, corresponding to the end of their six-month MSC treatment period. For patients in the MSC_first sequence, baseline values were compared to scores at six months, i.e. after an initial six months of MSC treatment. Across all four subsets and in both sequences, individual patients clustered close to the identity line, with baseline and MSC-period scores showing a tight relationship and largely preserved ranking. These trajectories indicate that CNS-homing programmes behave as relatively stable traits over the one-year observation period, with MSC exposure failing to induce a consistent upward or downward shift in the circulating compartment. This interpretation is consistent with the modest, scattered effect sizes observed in the mixed-model heatmap (Fig. 4C) and with the overall stability of cell-type composition reported in Fig. 3. Taken together, our CyTOF analyses indicate that both the composition and the functional wiring of circulating immune cells remain stable in progressive MS over 12 months, with no major MSC-induced shifts detectable at the population level. These findings indicate that any immunomodulatory effects of MSCs in this setting may act primarily within the CNS compartment or at immune–CNS interfaces, or through molecular pathways and mechanisms not captured by our current cell-surface marker panel used for CyTOF. To explore this possibility, we next analysed CSF proteomic profiles available for our clinical trial participants. (A) Schematic representation of composite functional score construction, illustrating how individual CyTOF markers contribute to ten biologically defined functional programs. (B) Baseline distributions of selected T cell functional scores across CD4⁺ and CD8⁺ immune phenotypes, shown as jittered patient-level points in gray including the cohort median in solid color ± standard error. (C) Heatmap of estimated MSC versus control effects on functional scores across all immune phenotypes derived from linear mixed-effects models. Columns represent functional scores, rows represent the 25 phenotypes, and colours indicate the direction and magnitude of the MSC effect (purple = higher under MSC, orange = lower under MSC, white = no effect). Lineage annotation for each phenotype is shown below, nominal significance before correction for multiple testing is marker with *. 2.4 CSF proteomic remodelling during intrathecal MSC therapy in relation to clinical and MRI findings To investigate whether intrathecal MSC therapy is associated with molecular changes within the CNS compartment, we analysed CSF proteomic profiles obtained from all SMART-MS participants (n = 18) at baseline and at 6 months, corresponding to the end of the first treatment period in the cross-over trial design. Normalized and log-transformed protein abundance data were first visualized using classical multidimensional scaling (MDS) to obtain a global overview of the CSF proteomic landscape (Supplementary Fig. 8A). While individual samples showed modest longitudinal displacement over time, no clear global separation between MSC-treated and control samples was observed. Comparable patterns were seen using alternative dimensionality-reduction approaches, including PCA and UMAP (Supplementary Figs. 8B and 8C), indicating that MSC treatment did not produce a dominant proteome-wide or global shift detectable by unsupervised dimensionality reduction. To formally assess whether specific proteins exhibited differential regulation following MSC treatment, we performed protein-level differential abundance analysis in the high-dimensional proteomic space. Because matched CSF samples were available for all patients at baseline and 6 months, longitudinal changes in protein abundance were assessed using within-patient differences over time. This approach revealed treatment-associated shifts in the CSF proteome that were not apparent from absolute abundance levels alone. The resulting analysis identified a subset of proteins exhibiting consistent MSC-associated changes over the 6-month interval. These effects are summarized in a volcano plot (Fig. 5A), which displays estimated MSC-associated protein changes against their statistical significance across all quantified CSF proteins. While the majority of proteins clustered near zero effect size consistent with limited global remodelling of the CSF proteome, a subset of proteins exhibited reproducible MSC-associated changes, providing a focused signal for downstream functional interpretation (Supplementary Table 2). Proteins significantly upregulated (FDR < 0.05) during MSC exposure formed a coherent group related to extracellular matrix organization and tissue remodelling, including periostin (POSTN), cartilage oligomeric matrix protein (COMP), and fibrillar collagens COL1A1 and COL1A2. In addition, SPARC and multiple fibrillar collagens (COL3A1, COL5A1, COL11A1) and the neuronal guidance and plasticity-associated receptor EPHB2 showed increased abundance at a more permissive threshold (FDR < 0.10), suggesting selective modulation of CNS structural and signalling pathways rather than broad immune activation. Several proteins with large apparent effect sizes but lacking statistical significance were also observed. These included hemoglobin subunits (HBA1, HBB) and carbonic anhydrase 3 (CA3), consistent with blood-derived contamination, as well as keratins and epithelial junctional proteins (KRT1, KRT9, KPRP, DSG1, DSP, JUP, PLEC, DCD, AHNAK), indicative of skin contamination during lumbar puncture. Additional intracellular structural and stress-related proteins (e.g. VIM, TUBB2A, TUBA1A, HSPB1, EEF2) were also detected. The concurrent presence of hemoglobin chains together with multiple keratin-associated proteins strongly supports technical contamination rather than biologically meaningful CSF remodelling for these features. Accordingly, these proteins were interpreted cautiously and were not considered central components of the MSC-associated CSF signal. To assess the biological coherence of MSC-associated proteomic changes in a systematic manner, we performed pathway enrichment analysis using Gene Ontology (GO). When all quantified proteins were included without prior filtering, enrichment was dominated by extracellular and barrier-associated processes such as wound healing, coagulation, epidermal development, and tissue remodelling (Supplementary Fig. 9), reflecting the influence of contaminant-derived proteins. After excluding proteins commonly associated with blood and epithelial contamination due to lumbar puncture, enrichment analysis revealed a more refined and biologically interpretable signature. The top 20 significant pathways (Fig. 5B) reflected organic acid metabolism, nucleotide metabolism, cytoskeletal remodelling and vascular/inflammatory interface. Inspection of ranked gene enrichment profiles (Supplementary Fig. 10) demonstrated coordinated downregulation of inflammatory stress signalling programs, coupled with upregulation of pathways related to cellular adaptation, intracellular transport, nucleotide metabolism, metabolic capacity biosynthetic and structural reorganization within the CNS. Together, these patterns are consistent with inflammatory attenuation and adaptive remodelling of CNS interface biology rather than immune activation, suggesting a possible transition toward a more homeostatic and reparative CSF environment. To provide a clinically anchored view of CSF proteomic changes, we incorporated information on adverse events (AE; pain/fever) and spinal cord MRI findings observed six months after treatment. CSF protein deltas were stratified into four groups: Control, MSC with no AE and no MRI findings (MSC no AE no MRI), MSC with AE and no MRI findings (MSC + AE no MRI), and MSC with AE and MRI findings (MSC + AE + MRI). Differentially enriched extracellular matrix and interface-associated proteins including collagens, POSTN, SPARC, COMP, EPHB2, CFL1, and PARK7 showed progressively larger positive deltas across the MSC spectrum, with minimal changes in MSC no AE no MRI patients and the largest shifts observed in MSC + AE no MRI and MSC + AE + MRI groups (Supplementary Fig. 15). Control samples remained centred near zero change. Although subgroup sizes were small (n = 3 per MSC stratum), precluding formal statistical comparisons, the consistent directionality across multiple independent proteins supports a non-random longitudinal trend. As a contextual comparison, we examined established CSF inflammatory markers (including C3/C4, CRP, SAA1/2, HP, A2M, SERPINs, fibrinogen chains, and ORM1/2) together with proteins commonly associated with contamination or procedural carryover. Six-month changes in these markers were generally modest and showed substantial overlap between control and MSC-treated patients, irrespective of adverse events or MRI findings (Supplementary Figs. 16 and 17). Previous analyses of the SMART-MS proteomics dataset identified several inflammation-related proteins as strong discriminators between treatment periods; however, our longitudinal modelling framework focused on coordinated directional changes across protein sets rather than individual discriminatory features [7]. Within this framework, we did not observe a uniform or cohort-wide increase in classical inflammatory markers, indicating that inflammatory responses were heterogeneous and context-dependent rather than representing a global inflammatory shift. Similarly, contamination-associated proteins (including hemoglobin chains, keratins, vimentin, and junctional proteins) displayed heterogeneous and individually variable longitudinal patterns rather than coordinated changes across patients. Because contamination signals are typically stochastic, the absence of a consistent cohort-level increase suggests that the extracellular matrix remodelling signature is unlikely to be explained solely by generalized sampling artifacts or traumatic lumbar puncture. Localized procedural or tissue responses cannot be ruled out, and the extracellular matrix /interface signature should be interpreted as suggestive rather than definitive evidence of treatment-associated intrathecal remodelling. 2.5 Program-level integration of CSF remodelling with circulating T cell trafficking signatures To synthesise the CSF proteomic changes into biologically interpretable patterns, we summarised protein-level responses into composite CSF functional programs reflecting extracellular matrix and CNS interface remodelling, innate and vascular inflammatory stress, and metabolic/cytoskeletal adaptation. In addition, a CSF reactivity score capturing biological responses (e.g., tissue stress reactions) to intrathecal administration was considered. When examined at the patient level, these CSF program scores revealed distinct patterns across MSC-treated individuals stratified by clinical and radiological reactivity (Fig. 5C), providing a structured view of how MSC therapy differentially impacts the CSF compartment. Across MSC-treated individuals, the CSF extracellular matrix/interface remodelling (ECM/interface) score showed a consistent increase, including in patients without adverse events or MRI activity, suggesting a shared treatment-associated pattern within the CSF compartment. Linear modelling indicated that this extracellular matrix/interface remodelling signal was associated with MSC exposure and was not explained by the CSF reactivity score (p = 0.44), supporting the interpretation that it reflects a treatment-related signature rather than nonspecific biological reactivity. In contrast, innate/vascular inflammatory and metabolic/cytoskeletal programs exhibited greater inter-individual variability. Patients experiencing clinical adverse events and/or post-procedural spinal MRI abnormalities following intrathecal MSC administration tended to show higher scores in these dimensions, whereas MSC-treated patients without clinical or radiological reactivity largely overlapped with controls. Consistent with this observation, metabolic/cytoskeletal adaptation scores were strongly associated with the CSF reactivity score in multivariable models (p = 0.004), while treatment effects were modest, indicating that these changes likely reflect secondary, context-dependent responses rather than direct MSC-driven remodelling. Scatter plot analyses (Supplementary Fig. 18) further supported this dissociation: extracellular matrix/interface remodelling occurred largely independently of vascular inflammatory activation, whereas metabolic/cytoskeletal responses closely tracked innate/vascular stress. Together, these patterns suggest that intrathecal MSC therapy primarily reprograms the CNS interface and extracellular matrix environment, while exaggerated inflammatory and metabolic responses emerge only in a subset of patients experiencing heightened CSF reactivity. Finally, to explore potential links between CSF remodelling and peripheral immune dynamics, we examined associations between functional program scores derived from CSF proteomics and blood-based CyTOF analyses. We focused on circulating programs related to immune trafficking and CSF programs capturing extracellular matrix and interface remodelling, given their putative relevance to immune–CNS interactions. Across multiple immune cell types, CSF interface remodelling score showed a clear separation between control and MSC-treated samples. However, within individual CD4⁺ and CD8⁺ T cell subsets, longitudinal changes in the circulating inflammatory homing program did not exhibit consistent linear coupling, supporting a compartmentalized model of intrathecal immune remodelling (Fig. 5D). Instead, the dominant pattern observed was a treatment-associated shift in CSF interface score, with relatively weak and variable within-group relationships to circulating Th17-like inflammatory homing dynamics. Similar exploratory analyses across additional immune compartments, including terminally differentiated or non-trafficking T cell subsets, B cell subsets, and myeloid-derived blood programs, likewise did not reveal reproducible linear coupling with CSF interface remodelling (Supplementary Fig. 11). While some cell-type–specific trends were observed, these relationships were heterogeneous and did not converge on a cell type-specific circulating program that independently explained CSF interface variation. Taken together, these analyses indicate that intrathecal MSC therapy in progressive MS is associated with consistent remodelling of the CSF interface and extracellular matrix, reflected by clear differences between control and MSC-treated samples. In contrast, longitudinal changes in circulating inflammatory Th17-like trafficking programs showed weak and heterogeneous relationships with CSF remodelling at the individual level. The CSF reactivity score was selectively elevated in MSC-treated individuals experiencing post-procedural pain, fever, or spinal MRI enhancement, whereas saline controls and non-reactive MSC-treated patients exhibited only minimal changes, suggesting that this axis reflects biological responsiveness to intrathecal MSC exposure rather than generic procedure-related artifact. Despite the limited sample size, the stability of the interface remodelling signal across patients, together with the selective association of tissue reactivity with metabolic and inflammatory programs, argues against these CSF changes being driven solely by nonspecific injury, stress, or contamination. Collectively, the data support a model in which MSC therapy induces compartmentalized and patient-specific remodelling at the immune–CNS interface, rather than being explained by a single dominant peripheral immune program. (A) Volcano plot showing estimated MSC treatment effects on longitudinal CSF protein changes (6-month minus baseline) plotted against statistical significance. (B) Dot plot summarizing Gene Ontology enrichment analysis of MSC-associated CSF proteomic changes after removal of proteins associated with blood and epithelial contamination. (C) Boxplots show longitudinal changes in three CSF-derived functional program scores. Patients are stratified into four groups based on treatment and clinical/radiological reactivity. (D) Scatter plots illustrating the relationship between longitudinal changes in blood-derived functional programs (CyTOF) and CSF program scores across T cell subsets, with points representing individual patients and regression lines shown separately for MSC and control periods. 3. MATERIALS AND METHODS 3.1 Patients, study design and sample collection Patients were recruited from the SMART-MS clinical trial [ 7 ], a randomized, double-blind, placebo-controlled phase I/II crossover study evaluating intrathecal administration of autologous bone marrow–derived MSCs in progressive MS (ClinicalTrials.gov NCT04749667). Eligible participants were adults aged 18–55 years with primary or secondary progressive MS according to revised McDonald criteria, Expanded Disability Status Scale (EDSS) scores between 4.0 and 7.0, disease duration of 2–18 years, and no clinical relapses or MRI activity for at least 24 months prior to enrolment. Patients receiving disease-modifying therapies during the trial were excluded to avoid confounding treatment effects. Participants were randomized 1:1 to receive either intrathecal MSCs or placebo (0.9% saline) at baseline, followed by crossover at six months, allowing within-patient comparisons while maintaining blinding. Autologous MSCs were generated from bone marrow aspirates under Good Manufacturing Practice (GMP) conditions and administered as a single intrathecal dose of 1×10⁶ cells/kg body weight (maximum 100×10⁶ cells) via lumbar puncture. To preserve blinding, all participants underwent identical bone marrow aspiration and injection procedures regardless of treatment allocation. The primary endpoint of the clinical trial was change in combined evoked potential latency at six months, while secondary endpoints included safety, MRI outcomes, functional and ophthalmological measures, clinical disability scores, and serum biomarkers. Adverse events and serious adverse events were prospectively recorded throughout the study. For the present multi-omics analyses, peripheral blood samples were collected at baseline, 6 months, and 12 months for mass cytometry profiling, and CSF was obtained immediately prior to each intrathecal injection for proteomic analysis. Longitudinal sampling within the crossover design enabled paired comparisons of MSC-treated and control periods within the same individuals. The trial was approved by the Regional Committee for Medical and Health Research Ethics and conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. All participants provided written informed consent prior to inclusion. 3.2 Mass cytometry antibody panel, sample preparation, and data acquisition For mass cytometry, an in-house 41-marker antibody panel was used for surface phenotyping (Supplementary Table 1). Markers were selected based on prior studies and included lineage-defining proteins, differentiation and activation markers, chemokine receptors, adhesion molecules, and stem/progenitor-associated markers relevant to immune trafficking and activation in multiple sclerosis. Antibodies were obtained either pre-conjugated from Standard BioTools or conjugated in-house using Maxpar metal-labeling kits according to the manufacturer’s instructions [ 24 ]. Antibody titrations were performed on barcoded mixtures of stabilized whole blood and reference cell populations to determine optimal staining concentrations, ensuring adequate signal resolution while minimizing channel spillover. Peripheral whole blood (1 mL) was collected into cryotubes, stabilized with SmartTube Proteomic Stabilizer, and stored at − 80°C. After thawing, red blood cells were lysed according to the manufacturer’s protocol, and white blood cells were counted and aliquoted into 3 × 10⁶-cell samples for downstream processing. Samples were barcoded using the Cell-ID 20-plex Pd Barcoding Kit (Standard BioTools), with longitudinal samples from the same patient included within the same barcode pool to reduce technical variability. A healthy donor reference sample was included in each barcode set to monitor consistency across runs. Following barcoding, pooled samples were stained overnight with a pre-prepared antibody cocktail according to the Maxpar staining protocol and incubated with an iridium DNA intercalator for nuclear labeling. After washing, cells were filtered, resuspended in Cell Acquisition Solution Plus containing EQ Six Element Calibration Beads, and acquired on a CyTOF XT instrument (Standard BioTools) at the Flow Cytometry Core Facility, University of Bergen, Norway. Data were collected at approximately 300–500 events per second with signal stability monitored throughout acquisition. 3.3 Mass cytometry data preprocessing and quality control Raw CyTOF data were pre-processed using a standardized and reproducible pipeline implemented in R language. Single-cell debarcoding was subsequently performed using custom R scripts adapted from the premessa package, applying file-specific intensity thresholds to assign events to individual barcodes. Debarcoded samples were exported as individual FCS files for downstream analysis. Quality control and cleanup were performed following established MaxPar/CyTOF cleaning best practices. Briefly, events were filtered based on DNA content using Iridium intercalator channels (Ir191Di/Ir193Di) to remove debris, doublets, and non-cellular events, incorporating signal intensity and event width parameters. Following DNA-based cleanup, an automated Gaussian mixture model (GMM) was applied to partition events based on CD45 and CD66b expression. The leukocyte population used for downstream analysis was defined as CD45⁺ CD66b low/neg cells, thereby enriching for lymphocytes and mononuclear cells while excluding CD66b high granulocytes and residual non-immune events from whole-blood acquisitions. To stabilize variance and improve comparability across markers, signal intensities for all protein channels were arcsinh-transformed using a cofactor of 5 prior to clustering, dimensionality reduction, and functional score computation. Technical variation related to barcode structure and acquisition batches was addressed using the CyCombine framework [ 25 ], which performs batch harmonization while preserving underlying biological structure. Batch correction was applied at the single-cell level prior to unsupervised clustering and compositional analyses. For visualization purposes only, marker expression values were scaled to a 0–1 range using the 1st and 99th percentile expression values derived from a pooled reference dataset of 12 healthy control samples. The resulting scaling parameters were applied uniformly across all patient samples to ensure consistent visual representation. All statistical/computational analyses were performed on transformed but unscaled data unless otherwise stated. 3.4 Immune phenotyping by unsupervised clustering Immune phenotyping was performed using a two-step, sequential unsupervised clustering strategy, designed to first define broad immune lineages and subsequently resolve finer phenotypic heterogeneity within each lineage [ 26 ]. In the first step, all CD66b low/neg cells from all patients and time points were pooled and clustered using the FlowSOM algorithm [ 20 ], employing a 10×10 self-organising map (100 clusters). All markers included in the CyTOF panel were used as input. The mean marker expression of each cluster was visualised using heatmaps, which were manually inspected to assign clusters to five major immune lineages based on canonical lineage-defining markers, including CD3, CD4, CD8, CD19, CD20, CD56, CD16, CD14, and CD11c. This resulted in the identification of CD4⁺ T cells, CD8⁺ T cells, B cells, natural killer (NK) cells, and a combined myeloid compartment comprising monocytes and dendritic cells. In the second step, each major lineage was re-clustered independently using FlowSOM with identical settings and using all available markers. This lineage-restricted clustering enabled higher-resolution identification of phenotypically distinct subpopulations while preserving the global structure defined in the first step. Marker expression profiles were again summarised using heatmaps, and rule-based phenotypic annotation was performed using marker-specific thresholds derived from quantiles of the cluster-level expression distributions. This approach combines data-driven clustering with biologically informed, reproducible annotation criteria. Within the CD4⁺ and CD8⁺ T cell compartments, subsets were annotated as naïve (CD45RA high , CD45RO low , CD27 high ), central memory (CM; CD45RA low , CD45RO high , CD27 high ), effector memory (EM; CD45RA low , CD45RO high , CD27 low ), and TEMRA (CD45RA high , CD45RO low , CD27 low ). Cells not meeting these criteria were grouped as CD4_unclassified or CD8_unclassified, respectively. B cell populations were annotated into six subsets: progenitor B cells (CD34 high ), naïve B cells (Naïve; CD27 low , CD38 low ), memory B cells (Mem; CD27 high , CD38 low ), activated B cells (Activ; HLA-DR high , CD27 low ), plasmablasts (pl; CD38 high , CD20 low , CD19 high ), and transitional B cells (CD38 high , CD27 low , excluding plasmablasts based on CD20/CD19 expression). NK cells were phenotyped into NKT cells (CD56 high and CD3 high ), CD56 bright CD16 high , CD56 bright CD16 low , CD56 dim CD16 high , and CD56 dim CD16 low subsets. The myeloid compartment was further resolved into plasmacytoid dendritic cells (pDCs; CD123 high , CD11c low ), classical dendritic cells (cDCs; CD11c high , CD14 low ), classical monocytes (CD14 high with CD11c low or CD33 high ), and non-classical monocytes (CD16 high , CD14 low ). This hierarchical strategy resulted in the definition of 25 immune phenotypes, which were visualised using heatmaps displaying mean marker expression per subset. For dimensionality reduction and visual inspection, UMAP embeddings were generated using all panel markers, with up to 1,500 cells per patient and time point subsampled when available. 3.5 Immune composition analysis To formally assess longitudinal changes in peripheral immune cell composition, we applied linear mixed-effects modelling using the lme4 and lmerTest R packages. As the relative abundances of the 25 immune phenotypes constitute compositional data, cell-type frequencies were first transformed using the centred log-ratio (CLR) transformation prior to modelling. To account for the cross-over design of the SMART-MS trial, each immune phenotype was modelled using the following fixed-effects structure: CLR(cell frequency) ∼ Treatment + Period + Sequence + Batch + Age + Sex + (1∣ Patient) where Treatment denotes MSC versus control exposure, Period corresponds to the first or second treatment period, Sequence reflects treatment order (MSC-first vs. control-first), Batch represents the CyTOF barcode identifier, Age was mean-centred within the cohort, and Sex was included as a covariate. A random intercept for patient was included to account for repeated measurements and inter-individual baseline differences. For each immune phenotype, model coefficients and associated p-values were extracted and adjusted for multiple testing using false discovery rate (FDR) correction across all subsets. Effect estimates and significance values were visualised using forest plots and volcano plots to summarise global compositional trends. Model assumptions were evaluated by inspection of standard diagnostic plots, including quantile–quantile (QQ) plots of residuals and fitted-versus-residual scatter plots, to assess normality, homoscedasticity, and potential outliers. Representative diagnostic plots are shown in Supplementary Figs. 13 and 14. As a sensitivity analysis, we additionally fitted an alternative mixed-effects model formulation in which treatment arm and time point (including baseline) were included as fixed effects, adjusted for CyTOF batch, age, and sex, with a random intercept for patient. This complementary approach enabled direct comparison of MSC-exposed samples to untreated baseline observations and provided an independent validation of treatment-associated effects observed in the primary cross-over model. 3.6 Construction of blood-based functional program scores To move beyond changes in immune cell composition and capture functional modulation within stable cellular compartments, we derived composite functional program scores from the CyTOF data. This approach was motivated by the observation that immune activation, differentiation, and trafficking states are typically reflected by coordinated expression of multiple markers rather than by changes in single proteins or cell frequencies alone. Composite scores therefore provide a robust summary of biologically coherent functional programs and enable direct comparison of functional states across immune phenotypes and longitudinal samples. Functional programs were defined a priori based on established immunological knowledge and the available CyTOF antibody panel. Each program comprised a small set of markers reflecting a shared biological process, including T cell activation, memory/effector differentiation, inflammatory homing (Th17-like), CNS-homing, Treg-like features, B cell activation and memory, myeloid antigen-presenting cell (APC) activation, non-classical monocyte features, and NK-cell activation. Program names refer to marker-derived functional axes and do not imply lineage identity. For example, the inflammatory homing program reflects expression of trafficking markers associated with inflammatory migration and does not denote a specific T-helper lineage. The mapping between individual markers and functional programs is shown in Fig. 4 A. For selected programs, markers were included with inverse contribution to capture counter-regulatory relationships (e.g. inverse weighting of CD127 in the Treg-like score). Prior to score calculation, marker expression values were z-score normalized across all cells to place markers on a comparable scale. For each cell, functional program scores were computed as the mean of the z-scored expression values of the markers assigned to that program, with inverse markers multiplied by − 1 prior to averaging. Scores were then aggregated at the cell-type and patient level by computing the median score within each immune phenotype for each individual and time point, enabling longitudinal analysis while preserving cell-type specificity. Functional scores are expressed on a relative scale centred around zero. Positive scores indicate relative enrichment of a given functional program compared with the global immune cell population, whereas negative scores indicate relative depletion. Scores were interpreted within the context of each immune lineage, reflecting relative functional wiring rather than absolute activation states. This composite scoring framework formed the basis for the longitudinal and integrative analyses presented in subsequent sections. 3.7 Longitudinal modelling of blood-based functional programs Longitudinal changes in blood-derived functional program scores were assessed using linear mixed-effects models implemented in the lme4 and lmerTest R packages. Modelling followed the same general framework as the immune composition analysis to ensure consistency across analytical layers. For each functional program and immune phenotype, the following model was fitted: Score ∼ Treatment + Period + Sequence + Batch + Age + Sex + (1∣ Patient) where Treatment denotes MSC versus control exposure, Period corresponds to the first or second treatment period of the cross-over design, Sequence reflects treatment order (MSC-first vs. control-first), Batch represents the CyTOF barcode identifier, Age was mean-centred within the cohort, and Sex was included as a covariate. A random intercept for patient was included to account for repeated measurements and inter-individual baseline differences. Model coefficients, standard errors, and p-values were extracted for the treatment effect for each score–cell type combination. To control for multiple testing across immune phenotypes and functional programs, p-values were adjusted using FDR correction. Effect estimates and corresponding significance values were visualised in a heatmap, with rows representing immune phenotypes and columns representing functional programs. Model assumptions were assessed by inspection of residual distributions and fitted-versus-residual plots, with no major deviations observed. 3.8 CSF proteomics data acquisition, preprocessing, and normalization CSF samples collected at baseline and at 6 months (prior to treatment crossover) were analysed using tandem mass tag–based quantitative proteomics. Sample preparation, liquid chromatography–tandem mass spectrometry (LC-MS/MS) acquisition, and primary data processing/cleansing were performed as previously described [ 7 ]. Briefly, samples were processed in three TMT-16plex batches, each including pooled reference controls for normalization. Peptides were analysed on an Orbitrap Eclipse high-resolution mass spectrometer, and raw spectra were processed using Proteome Discoverer v2.5 (Thermo Fisher Scientific), followed by downstream preprocessing and quality control in Perseus framework [ 27 ]. Protein-level intensity data provided from the proteomics pipeline were imported into R for downstream analyses. The resulting dataset consisted of 1325 quantified proteins that were log-transformed and used for all subsequent analyses in this study. These processed protein intensities were subsequently used for differential abundance modelling, pathway enrichment analysis, and derivation of CSF functional program scores. 3.9 Differential abundance analysis of CSF proteomics Differential abundance analysis of the CSF proteome was performed using a paired, longitudinal delta-based framework. As complete CSF samples were available for all participants at BL and 6m, protein-level changes were quantified for each individual as: ΔProtein = Abundance 6m − Abundance BL This approach enabled direct modelling of within-patient proteomic changes over time while controlling for substantial inter-individual variability in baseline protein abundance. By analysing change scores rather than absolute values, treatment-associated effects could be interpreted as differences in longitudinal trajectories attributable to MSC exposure. For each protein, linear regression models were fitted to the delta values using the lm function in R, with Treatment (MSC vs control), Age, and Sex included as covariates. Model fitting and coefficient extraction were performed using the broom R package to ensure consistent handling of estimates, standard errors, and statistical significance. A linear model was chosen in this context, rather than a mixed-effects model, because each individual contributed a single delta value per protein, eliminating repeated measurements within the outcome variable. As such, there was no hierarchical structure requiring random effects, and the paired nature of the design was fully captured by the within-subject delta formulation. Resulting p-values were adjusted for multiple testing across all quantified proteins using the FDR procedure. Differential abundance results were visualised using volcano plots, displaying effect size estimates (MSC-associated change relative to control) against statistical significance. Proteins passing predefined FDR thresholds (FDR < 0.05 and FDR < 0.1) were prioritised for downstream interpretation. To support targeted inspection of biologically relevant signals, selected protein sets were further visualised using boxplots of delta values stratified by treatment and clinical subgroups. These targeted visualisations enabled patient-level assessment of proteomic changes in relation to treatment exposure, adverse events, and radiological findings. Protein identifiers were mapped to gene symbols and descriptive annotations using the AnnotationDbi and org.Hs.eg.db R packages. Where necessary, UniProt identifiers were converted to HGNC gene symbols to ensure consistent representation across analyses and figures. 3.10 Pathway enrichment analysis of CSF proteomics To identify coordinated biological processes associated with MSC-induced CSF proteomic changes, pathway enrichment analysis was performed using a ranked gene set enrichment approach. Proteins were ranked according to MSC-associated effect estimates derived from linear modelling of longitudinal CSF protein changes, with positive values indicating relative enrichment during MSC exposure and negative values indicating relative depletion. Gene set enrichment analysis (GSEA) was conducted using the gseGO function from the clusterProfiler R package, focusing on Gene Ontology Biological Process (GO:BP) terms. Analyses were performed with parameters minGSSize = 10, maxGSSize = 300, and pvalueCutoff = 0.1 , with p-values adjusted for multiple testing using the FDR procedure. GSEA was selected over over-representation analysis to leverage the full ranked proteomic dataset and detect subtle but coordinated pathway-level changes without imposing arbitrary significance thresholds at the protein level. An initial GSEA including all quantified proteins revealed enrichment of pathways related to tissue remodelling, coagulation, and epithelial or barrier-associated processes, consistent with the presence of blood- and skin-derived proteins introduced during lumbar puncture. To mitigate bias from technical contamination, a predefined set of proteins commonly associated with blood or epithelial contamination (HBA1, HBB, KRT1, KRT9, KPRP, DSG1, DCD, PLEC, DSP, AHNAK, JUP, CA3) was excluded from the ranked list, and GSEA was repeated on the filtered dataset. Significant pathways were visualized using dot plots displaying normalized enrichment scores (NES), FDR-adjusted p-values, and leading-edge gene ratios, reflecting the proportion of pathway members driving enrichment. Selected pathways were further visualized using gseaplot2 (enrichplot R package) to illustrate enrichment profiles and leading-edge contributions along the ranked protein list. For interpretability, enriched GO terms were grouped post hoc into broader functional categories, including vascular and innate inflammatory processes, cytoskeletal organization, and metabolic pathways; this grouping was applied solely for presentation and did not affect statistical inference. 3.11 Integration of CSF proteomics with blood-based functional programs Composite CSF functional program scores were constructed using a framework conceptually analogous to that applied to the CyTOF-derived blood functional programs. This parallel design enables direct comparison of coordinated biological states inferred from bulk CSF proteomics with cell-type–resolved functional programs measured in peripheral blood, despite differences in measurement modality and biological scale. Protein sets were defined a priori based on established biological knowledge and informed by differential abundance and pathway enrichment analyses, with the aim of capturing mechanistically relevant dimensions of CSF biology. Three primary CSF functional programs were defined: CSF Extracellular matrix / Interface Remodelling , reflecting extracellular matrix organization and CNS interface biology, including fibrillar collagens (COL1A1, COL3A1, COL5A1, COL11A1, COL6A6), extracellular matrix regulators (POSTN, SPARC, COMP), and matrix scaffolds and adhesion molecules (FN1, TNC, VCAN, LAMA2, LAMC1). CSF Innate / Vascular Stress , capturing innate immune and vascular inflammatory responses, comprising complement components (C3, C4A, C4B, C5, CFH, CFB), acute-phase and innate inflammatory proteins (CRP, SAA1, SAA2, HP, ORM1, ORM2, SERPINA1, SERPINA3, A2M, CP), and coagulation-related factors (FGA, FGB, FGG). CSF Metabolic / Cytoskeletal Adaptation , reflecting intracellular structural and metabolic responses, including actin dynamics (CFL1, PFN1, ACTB, ACTG1), microtubule organization and transport (TUBA1A, TUBB2A, TUBB), stress and translational regulation (HSPB1, EEF2), and structural integration (VIM). To account for nonspecific signals potentially arising from tissue injury or blood contamination, rather than MSC-specific biological effects, an additional composite CSF reactivity score was constructed. This score was based on proteins commonly associated with sampling-related artifacts or cellular stress, including hemoglobin subunits (HBA1, HBB), epithelial and junctional proteins (KRT1, KPRP, DSG1, DSP, JUP, AHNAK, DCD), and cytoskeletal or metabolic stress markers (VIM, CA3). The procedural stress score was used as a negative-control axis to aid interpretation of CSF proteomic changes and to distinguish treatment-associated remodelling from nonspecific procedural effects. For each CSF functional program, protein-level longitudinal changes (6-month minus baseline delta values) were z-score normalized across patients and aggregated by computing the mean score per program for each individual. Scores therefore reflect relative enrichment or depletion of coordinated biological programs rather than absolute protein abundance. To evaluate the contribution of nonspecific effects versus MSC exposure, linear models were fitted with each CSF program score as the outcome and CSF reactivity score and treatment (MSC vs. control) as covariates. Finally, CSF functional program scores were integrated with blood-based CyTOF functional programs by matching patient-level longitudinal changes across compartments. Associations were assessed using correlation and linear regression analyses, with a primary focus on mechanistically compatible programs, particularly CSF interface remodelling and circulating T cell trafficking programs (e.g., Th17-like and CNS-homing). This integrative approach enabled evaluation of whether MSC-associated CSF remodelling aligned with specific adaptive immune trafficking states rather than reflecting global immune activation. 4. CONCLUSIONS In this study, we implemented a multi-layer omics integration framework to investigate the biological effects of intrathecal MSC therapy in progressive MS. By combining longitudinal mass cytometry profiling of peripheral blood with matched CSF proteomics, we characterized immune composition, functional immune programs, and protein-level remodelling across systemic and intrathecal compartments within the SMART-MS trial [ 7 ]. This integrative design enabled a systems-level view of MSC-associated changes and revealed mechanistic patterns that would not be apparent from single-modality analyses alone. A central observation of our study is that intrathecal MSC therapy did not broadly reshape peripheral immune composition. These findings are consistent with emerging models of progressive MS in which inflammatory processes become compartmentalized within CNS border regions and are therefore not reflected in peripheral immune composition [ 3 ]. Instead, the most consistent signal emerged within the CSF proteome, where MSC exposure was associated with coordinated remodelling of extracellular matrix and CNS interface programs. These findings support a model in which MSC therapy primarily acts within compartmentalized inflammatory niches at the immune-CNS interface. While peripheral functional programs showed stability at the compositional level, our blood and CSF remodelling results suggest that MSC therapy may modulate how immune cells interact with CNS tissue rather than altering global immune architecture [ 21 , 22 ]. The biological interpretation of the CSF proteomic signatures points toward structural and trophic reprogramming of the CNS interface. Proteins related to extracellular matrix organization, cell–matrix interaction, and guidance signalling increased during MSC exposure, consistent with enhanced remodelling of the extracellular environment. Such changes may influence barrier properties, immune cell migration, or local trophic signalling rather than reflecting classical inflammatory activation. Importantly, pathway enrichment analyses indicated attenuation of acute-phase and innate inflammatory programs in many individuals. A distinctive aspect of our analysis was the integration of clinical context, including adverse events and spinal reactive MRI findings. While extracellular matrix/interface remodelling was consistently observed across MSC-treated individuals, several inflammatory and metabolic proteomic changes were preferentially enriched in patients who developed clinical or radiological reactivity. This pattern suggests that a subset of CSF proteomic signals, particularly metabolic/cytoskeletal and vascular stress responses, may reflect secondary biological reactivity to intrathecal MSC exposure rather than core mechanisms of MSC-mediated interface remodelling. Rather than representing distinct protein signatures, clinically reactive patients showed amplified changes within shared extracellular matrix, inflammatory, and metabolic pathways, suggesting increased biological responsiveness rather than fundamentally different mechanisms. These observations highlight the importance of interpreting MSC-associated proteomic changes within clinical context and caution against attributing all CSF alterations solely to direct therapeutic mechanisms. To further disentangle treatment effects from nonspecific tissue responses, we modelled a CSF reactivity score derived from proteins associated with tissue stress, blood leakage, epithelial carryover, and cytoskeletal release. The reactivity score correlated with metabolic/cytoskeletal changes but showed no clear association with the interface remodelling signal. By contrast, interface scores shifted consistently during MSC exposure, indicating that structural remodelling may represent a partially distinct biological response. Together, these findings are not readily explained by procedural effects alone for the observed CSF remodelling and instead support a model in which structural interface changes represent a primary MSC-associated effect, whereas metabolic and cytoskeletal responses reflect context-dependent biological reactivity influenced by post-treatment tissue stress and individual sensitivity thresholds. Integration of CSF and blood-derived functional programs provided additional context for MSC-associated immune dynamics across compartments. Exploratory analyses suggested that changes in circulating inflammatory trafficking programs (e.g., Th17-like) occasionally paralleled shifts in CSF interface remodelling within memory T cell subsets; however, these relationships were heterogeneous and did not demonstrate consistent or statistically robust linear coupling at the individual level. Instead, the dominant pattern was a treatment-associated increase in CSF extracellular matrix/interface remodelling largely independent of specific peripheral immune programs. These findings support a model in which intrathecal MSC therapy primarily induces local remodelling of the immune–CNS interface, while a subset of inflammatory and metabolic CSF changes appear to track more closely with adverse event–associated biology than with consistent MSC-driven immune modulation. While adaptive trafficking signatures may reflect broader immune context, our data suggest that MSC-associated CSF remodelling represents a compartmental effect linked to immune–CNS communication rather than generalized immunosuppression [ 21 , 23 ]. From a clinical perspective, program-level CSF readouts may provide a framework for distinguishing adaptive biological remodelling from excessive inflammatory reactivity following intrathecal therapies. Although exploratory, our findings raise the possibility that extracellular matrix/interface signatures represent desired treatment responses, while elevated metabolic and vascular stress signals may identify patients with heightened sensitivity to intrathecal interventions. These observations further suggest that certain CSF proteomic signatures may serve as indicators of biological reactivity or treatment-associated adverse responses rather than biomarkers of therapeutic efficacy alone. Future clinical trials incorporating longitudinal multi-omics profiling may therefore benefit from integrating functional program scores as biomarkers of therapeutic engagement or over-reactivity. Similar to other omics studies, several limitations should be considered. The SMART-MS cohort was relatively small (n = 18), limiting statistical power for subgroup analyses and precluding formal inference regarding adverse events or MRI-defined reactivity. Longitudinal follow-up was restricted to 12 months and may therefore not capture longer-term remodelling or neurodegenerative trajectories. Although bulk CSF proteomics provides valuable insight into extracellular and secreted protein dynamics, it lacks cellular resolution. In parallel, the CyTOF analysis was constrained by a predefined antibody panel focused on “canonical” immune populations. Consequently, functional scores were derived from the available markers and may not fully capture alternative activation states or rare immune subsets. The composite functional programs should therefore be interpreted as structured summaries of measurable signalling axes rather than exhaustive representations of immune function and CSF processes. Future studies incorporating broader single-cell technologies, such as scRNA-seq, CITE-seq, or spatial profiling, may further refine the cellular origin and context of the observed signatures [ 17 ]. Replication in independent cohorts will be essential, although assembling comparable intrathecal MSC datasets in progressive MS remains challenging. Despite these limitations, our study provides several advances. To our knowledge, this represents the first integrated analysis combining high-dimensional single-cell immune profiling with CSF proteomics in MSC-treated progressive MS patients. The program-level analytical framework enabled cross-compartment comparisons and highlighted dissociable biological axes of interface remodelling, inflammatory stress, and metabolic adaptation. Moreover, explicit modelling of CSF reactivity provides a practical strategy to address a long-standing challenge in CSF proteomics studies of intrathecal therapies. In summary, our results support a model in which intrathecal MSC therapy primarily reshapes the immune–CNS interface and extracellular environment within the intrathecal compartment, with peripheral immune programs remaining largely stable. In most patients, this remodelling occurs alongside reduced innate inflammatory signalling, whereas a subset of individuals exhibits amplified inflammatory and metabolic responses associated with clinical or radiological reactivity. By integrating multi-omics data across compartments, this work provides a systems-level framework for understanding MSC therapy in progressive MS and establishes a generalizable analytical framework for future translational studies. Abbreviations multiple sclerosis (MS) mesenchymal stem cell (MSC) cerebrospinal fluid (CSF) magnetic resonance imaging (MRI) central nervous system (CNS) relapsing–remitting MS (RRMS) cytometry by time-of-flight (CyTOF) tandem mass spectrometry (LC-MS/MS) uniform manifold approximation and projection (UMAP) central memory (CM) effector memory (EM) effector memory cells re-expressing CD45RA (TEMRA) centred log-ratio (CLR) antigen-presenting cell (APC) inflammatory homing-like (Th17-like) principal Component analysis (PCA) T regulatory-like (Treg) dendritic cells (DCs) natural killer (NK) multidimensional scaling (MDS) false discovery rate (FDR) quantile–quantile (QQ) cartilage oligomeric matrix protein (COMP) carbonic anhydrase 3 (CA3) gene ontology (GO) adverse events (AE; pain/fever) MSC with no AE and no MRI findings (MSC no AE no MRI) MSC with AE and no MRI findings (MSC + AE no MRI) MSC with AE and MRI findings (MSC + AE + MRI) extracellular matrix/interface remodelling (ECM/interface) expanded disability status scale (EDSS) good manufacturing practice (GMP) gene set enrichment analysis (GSEA) Declarations 6.1 Ethics approval and consent to participate The SMART-MS study received approval from the Regional Committee for Medical and Health Research Ethics (reference no. 159326) and the Norwegian Medicines Agency. The trial was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice principles. Written informed consent was obtained from all participants before inclusion in the study. Trial conduct was overseen by external monitors from Western Norway Health Trust Research & Development, and safety was reviewed regularly by an independent Data Monitoring Committee. The study was prospectively registered at ClinicalTrials.gov (NCT04749667) and in the European Union Clinical Trials Register (EudraCT no. 2020-002373-95). Consent for publication The manuscript does not contain any individual person’s identifiable data. 6.3 Availability of datasets and codes All analysis code required to reproduce the CyTOF and CSF proteomics analyses is publicly available at https://github.com/dkleftogi/ms-immune-signatures, including pipelines for data preprocessing, batch harmonization, immune phenotyping, functional program scoring, and statistical modelling. Processed and pseudonymised CyTOF phenotypic data and CSF proteomics datasets are deposited in the associated Zenodo repository, as referenced in the GitHub documentation. Shared datasets contain no directly identifiable information and include only study variables necessary to reproduce the analyses. Raw data and additional metadata can be made available upon reasonable request, subject to institutional ethical approvals, data protection regulations, and governance policies of the SMART-MS study. Competing interests T.H. has received speaker honoraria and/or participated in clinical multiple sclerosis trials sponsored by Biogen, Sanofi, Merck, Roche, Amgen, and Novartis. Ø.T. has participated in advisory boards and received speaker honoraria from Biogen, Merck, Novartis, Teva, Roche, Sanofi, and Bristol Myers Squibb, and has participated in clinical trials sponsored by Merck, Novartis, Roche, and Sanofi. All other authors declare no competing interests. Funding This study was funded by KLINBEFORSK (The Norwegian Clinical Research Program), Helse Vest, the Norwegian Red Cross, and the Norwegian MS Society. The funders had no role in the design, conduct, analysis, or reporting of the trial. Authors contributions D.K. conceived the study, developed the computational framework, performed the bioinformatics and statistical analyses, and drafted the manuscript. S.G. was responsible for CyTOF-based immune profiling and contributed to data interpretation and manuscript preparation. L.E.E. and J.B.H. contributed to data analysis and critically reviewed the manuscript. T.H., L.S., and K.W. were responsible for patient recruitment and clinical data acquisition and critically reviewed the manuscript. M.R. and H.S. were responsible for mesenchymal stem cell production and quality control and critically reviewed the manuscript. H.B., L.E.E. and F.S.B. contributed to the CSF proteomics data analysis and critically reviewed the manuscript. N.A.-S., S.M.-A., M.Y., C.E.S., Ø.T., K.M., and L.B. contributed to study design and critically reviewed the manuscript. T.K. and C.E.K. contributed to study conception, led the clinical trial and sample collection, contributed to data interpretation, and critically reviewed the manuscript.All authors read and approved the final manuscript. Acknowledgements The authors gratefully acknowledge the Flow Cytometry Core Facility at the University of Bergen for technical support in CyTOF data acquisition. We sincerely thank all patients who participated in the SMART-MS trial, and their families, for their invaluable contribution to this research. We also thank the laboratory and clinical personnel involved in sample handling, processing, and trial coordination for their dedicated support throughout the study. References Wallin MT, Culpepper WJ, Nichols E, Bhutta ZA, Gebrehiwot TT, Hay SI, et al. Global, regional, and national burden of multiple sclerosis 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Neurol Elsevier. 2019;18:269–85. https://doi.org/10.1016/S1474-4422(18)30443-5 . Lin M, Zhang J, Zhang Y, Luo J, Shi S. Ocrelizumab for multiple sclerosis. Cochrane Database Syst Rev. 2022;5:CD013247. https://doi.org/10.1002/14651858.CD013247.pub2 . Lassmann H, van Horssen J, Mahad D. Progressive multiple sclerosis: pathology and pathogenesis. Nat Rev Neurol Nat Publishing Group. 2012;8:647–56. https://doi.org/10.1038/nrneurol.2012.168 . Giovannoni G, Popescu V, Wuerfel J, Hellwig K, Iacobaeus E, Jensen MB, et al. Smouldering multiple sclerosis: the real MS. Ther Adv Neurol Disord. 2022;15:17562864211066751. https://doi.org/10.1177/17562864211066751 . Gavasso S, Kråkenes T, Olsen H, Evjenth EC, Ytterdal M, Haugsøen JB, et al. The Therapeutic Mechanisms of Mesenchymal Stem Cells in MS—A Review Focusing on Neuroprotective Properties. Int J Mol Sci. 2024;25:1365. https://doi.org/10.3390/ijms25031365 . Smith JA, Nicaise AM, Ionescu R-B, Hamel R, Peruzzotti-Jametti L, Pluchino S. Stem Cell Therapies for Progressive Multiple Sclerosis. Front Cell Dev Biol [Internet] Front. 2021. https://doi.org/10.3389/fcell.2021.696434 . [cited 2026 Feb 10];9. Kvistad CE, Kråkenes T, Holmøy T, Eidem LE, Steffensen L, Wesnes K, et al. Intrathecal Mesenchymal Stem Cells in Progressive Multiple Sclerosis. Neurol Wolters Kluwer. 2026;106:e214915. https://doi.org/10.1212/WNL.0000000000214915 . Connick P, Kolappan M, Crawley C, Webber DJ, Patani R, Michell AW, et al. Autologous mesenchymal stem cells for the treatment of secondary progressive multiple sclerosis: an open-label phase 2a proof-of-concept study. Lancet Neurol. 2012;11:150–6. https://doi.org/10.1016/S1474-4422(11)70305-2 . Bonab MM, Sahraian MA, Aghsaie A, Karvigh SA, Hosseinian SM, Nikbin B, et al. Autologous mesenchymal stem cell therapy in progressive multiple sclerosis: an open label study. Curr Stem Cell Res Ther. 2012;7:407–14. https://doi.org/10.2174/157488812804484648 . Petrou P, Kassis I, Ginzberg A, Halimi M, Yaghmour N, Abramsky O, et al. Long-Term Clinical and Immunological Effects of Repeated Mesenchymal Stem Cell Injections in Patients With Progressive Forms of Multiple Sclerosis. Front Neurol [Internet] Front. 2021. https://doi.org/10.3389/fneur.2021.639315 . [cited 2026 Feb 10];12. Harris VK, Yan QJ, Vyshkina T, Sahabi S, Liu X, Sadiq SA. Clinical and pathological effects of intrathecal injection of mesenchymal stem cell-derived neural progenitors in an experimental model of multiple sclerosis. J Neurol Sci. 2012;313:167–77. https://doi.org/10.1016/j.jns.2011.08.036 . Harris VK, Stark J, Williams A, Roche M, Malin M, Kumar A, et al. Efficacy of intrathecal mesenchymal stem cell-neural progenitor therapy in progressive MS: results from a phase II, randomized, placebo-controlled clinical trial. Stem Cell Res Ther. 2024;15:151. https://doi.org/10.1186/s13287-024-03765-6 . Genchi A, Brambilla E, Sangalli F, Radaelli M, Bacigaluppi M, Furlan R, et al. Neural stem cell transplantation in patients with progressive multiple sclerosis: an open-label, phase 1 study. Nat Med. 2023;29:75–85. https://doi.org/10.1038/s41591-022-02097-3 . Barmada A, Sharan J, Band N, Prodromos C. Serious Adverse Events Have Not Been Reported with Spinal Intrathecal Injection of Mesenchymal Stem Cells: A Systematic Review. Curr Stem Cell Res Ther. 2023;18:829–33. https://doi.org/10.2174/1574888X17666220817125324 . Böttcher C, Fernández-Zapata C, Schlickeiser S, Kunkel D, Schulz AR, Mei HE, et al. Multi-parameter immune profiling of peripheral blood mononuclear cells by multiplexed single-cell mass cytometry in patients with early multiple sclerosis. Sci Rep Nat Publishing Group. 2019;9:19471. https://doi.org/10.1038/s41598-019-55852-x . Zhang T, Warden AR, Li Y, Ding X. Progress and applications of mass cytometry in sketching immune landscapes. Clin Translational Med. 2020;10:e206. https://doi.org/10.1002/ctm2.206 . Single-cell analysis. of cerebrospinal fluid reveals common features of neuroinflammation: Cell Reports Medicine [Internet]. [cited 2026 Feb 11]. https://www.cell.com/cell-reports-medicine/fulltext/S2666-3791(24)00463-4 . Accessed 11 Feb 2026. Targeted proteomics of cerebrospinal fluid. in treatment naïve multiple sclerosis patients identifies immune biomarkers of clinical phenotypes | Scientific Reports [Internet]. [cited 2026 Feb 10]. https://www.nature.com/articles/s41598-024-67769-1 . Accessed 10 Feb 2026. Jagid N, Kizilbash C, Harris V, Sadiq S. CSF Biomarkers of Intrathecal Cell Therapy in Progressive Multiple Sclerosis Reflect Changes in Microglia and Astrocyte Polarization (P8-6.009). Neurol Wolters Kluwer. 2024;102:5500. https://doi.org/10.1212/WNL.0000000000205860 . Van Gassen S, Callebaut B, Van Helden MJ, Lambrecht BN, Demeester P, Dhaene T, et al. FlowSOM: Using self-organizing maps for visualization and interpretation of cytometry data. Cytometry A. 2015;87:636–45. https://doi.org/10.1002/cyto.a.22625 . Machado-Santos J, Saji E, Tröscher AR, Paunovic M, Liblau R, Gabriely G, et al. The compartmentalized inflammatory response in the multiple sclerosis brain is composed of tissue-resident CD8 + T lymphocytes and B cells. Brain. 2018;141:2066–82. https://doi.org/10.1093/brain/awy151 . Apolloni S, Tortoriello S, Milani M, Rossi S. Extracellular Matrix Remodelling in Motor Neuron Diseases. Int J Mol Sci. 2025;26:11376. https://doi.org/10.3390/ijms262311376 . Hu H, Li H, Li R, Liu P, Liu H. Re-establishing immune tolerance in multiple sclerosis: focusing on novel mechanisms of mesenchymal stem cell regulation of Th17/Treg balance. J Transl Med. 2024;22:663. https://doi.org/10.1186/s12967-024-05450-x . Gavasso S, Haugsøen JB, Kleftogiannis D, Gerking Y, Anandan S, Herdlevær I et al. High-dimensional Immune Profiling Following Autologous Hematopoietic Stem Cell Transplantation in Relapsing-Remitting Multiple Sclerosis [Internet]. bioRxiv; 2025 [cited 2026 Feb 26]. p. 2025.07.01.662494. https://doi.org/10.1101/2025.07.01.662494 Pedersen CB, Dam SH, Barnkob MB, Leipold MD, Purroy N, Rassenti LZ, et al. cyCombine allows for robust integration of single-cell cytometry datasets within and across technologies. Nat Commun Nat Publishing Group. 2022;13:1698. https://doi.org/10.1038/s41467-022-29383-5 . Global. and single-cell proteomics view of the co-evolution between neural progenitors and breast cancer cells in a co-culture model - eBioMedicine [Internet]. [cited 2026 Feb 11]. https://www.thelancet.com/journals/ebiom/article/PIIS2352-3964(24)00361-X/fulltext . Accessed 11 Feb 2026. Tyanova S, Temu T, Sinitcyn P, Carlson A, Hein MY, Geiger T, et al. The Perseus computational platform for comprehensive analysis of (prote)omics data. Nat Methods Nat Publishing Group. 2016;13:731–40. https://doi.org/10.1038/nmeth.3901 . Additional Declarations Competing interest reported. TH has received speaker honoraria and/or participated in clinical multiple sclerosis trials sponsored by Biogen, Sanofi, Merck, Roche, Amgen, and Novartis. ØT has participated in advisory boards and received speaker honoraria from Biogen, Merck, Novartis, Teva, Roche, Sanofi, and Bristol Myers Squibb, and has participated in clinical trials sponsored by Merck, Novartis, Roche, and Sanofi. All other authors declare no competing interests. Supplementary Files supplementarymaterial.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-9629755\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":true,\"archivedVersions\":[],\"articleType\":\"Research Article\",\"associatedPublications\":[],\"authors\":[{\"id\":638234230,\"identity\":\"406685a2-8d1e-4382-8a1b-789a1e49459c\",\"order_by\":0,\"name\":\"Dimitrios Kleftogiannis\",\"email\":\"data:image/png;base64,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\",\"orcid\":\"\",\"institution\":\"University of Bergen\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Dimitrios\",\"middleName\":\"\",\"lastName\":\"Kleftogiannis\",\"suffix\":\"\"},{\"id\":638234231,\"identity\":\"f927dd1d-0be7-485a-a165-6fd8868a2fb6\",\"order_by\":1,\"name\":\"Sonia Gavasso\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Haukeland University Hospital\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Sonia\",\"middleName\":\"\",\"lastName\":\"Gavasso\",\"suffix\":\"\"},{\"id\":638234234,\"identity\":\"0803bf77-47a6-44de-8af8-b0bbd043d4f6\",\"order_by\":2,\"name\":\"Live Egeland Eidem\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Stavanger University Hospital\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Live\",\"middleName\":\"Egeland\",\"lastName\":\"Eidem\",\"suffix\":\"\"},{\"id\":638234236,\"identity\":\"d24012a5-5583-45c8-9020-73f3e3335449\",\"order_by\":3,\"name\":\"Jonas Bull Haugsøen\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Haukeland University Hospital\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Jonas\",\"middleName\":\"Bull\",\"lastName\":\"Haugsøen\",\"suffix\":\"\"},{\"id\":638234237,\"identity\":\"925df7d5-7c6b-40e8-aa51-bf1b50216480\",\"order_by\":4,\"name\":\"Trygve Holmøy\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Akershus University Hospital\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Trygve\",\"middleName\":\"\",\"lastName\":\"Holmøy\",\"suffix\":\"\"},{\"id\":638234239,\"identity\":\"1f24e6d3-6654-46bd-a523-517c54189034\",\"order_by\":5,\"name\":\"Linn Steffensen\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University Hospital of North Norway\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Linn\",\"middleName\":\"\",\"lastName\":\"Steffensen\",\"suffix\":\"\"},{\"id\":638234240,\"identity\":\"8cad7dc1-7379-4766-b3af-6425c5f3c874\",\"order_by\":6,\"name\":\"Kristin Wesnes\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"St Olav's University Hospital\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Kristin\",\"middleName\":\"\",\"lastName\":\"Wesnes\",\"suffix\":\"\"},{\"id\":638234241,\"identity\":\"e6639176-6f19-411e-8aee-5729b8c8a85d\",\"order_by\":7,\"name\":\"Markus Rojewski\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Institut für Klinische Transfusionsmedizin und Immungenetik\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Markus\",\"middleName\":\"\",\"lastName\":\"Rojewski\",\"suffix\":\"\"},{\"id\":638234242,\"identity\":\"5bcc2547-1e29-46c3-a311-c1e4ae83c9e0\",\"order_by\":8,\"name\":\"Harald Barsnes\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University of Bergen\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Harald\",\"middleName\":\"\",\"lastName\":\"Barsnes\",\"suffix\":\"\"},{\"id\":638234243,\"identity\":\"37b2dac4-48c8-42eb-b7a3-e184eb83e1e8\",\"order_by\":9,\"name\":\"Niyaz Al-Sharabi\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University of Bergen\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Niyaz\",\"middleName\":\"\",\"lastName\":\"Al-Sharabi\",\"suffix\":\"\"},{\"id\":638234244,\"identity\":\"89a5f278-be11-43b9-9053-21a9899d7f5b\",\"order_by\":10,\"name\":\"Samih Mohamed-Ahmed\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University of Bergen\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Samih\",\"middleName\":\"\",\"lastName\":\"Mohamed-Ahmed\",\"suffix\":\"\"},{\"id\":638234245,\"identity\":\"de141a07-d018-4aa3-9ecb-6d17c2adfd20\",\"order_by\":11,\"name\":\"Marie Ytterdal\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Haukeland University Hospital\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Marie\",\"middleName\":\"\",\"lastName\":\"Ytterdal\",\"suffix\":\"\"},{\"id\":638234246,\"identity\":\"a864af01-db32-43ff-914d-dbf7dcf46daa\",\"order_by\":12,\"name\":\"Casper Eugen Sandvik\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Haukeland University Hospital\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Casper\",\"middleName\":\"Eugen\",\"lastName\":\"Sandvik\",\"suffix\":\"\"},{\"id\":638234247,\"identity\":\"51bd2099-e9f1-4935-8e7b-bedca1fc67db\",\"order_by\":13,\"name\":\"Øivind Torkildsen\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University of Bergen\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Øivind\",\"middleName\":\"\",\"lastName\":\"Torkildsen\",\"suffix\":\"\"},{\"id\":638234250,\"identity\":\"c5b7bcb5-f3a1-4498-a804-06b4226463b2\",\"order_by\":14,\"name\":\"Hubert Schrezenmeier\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Institut für Klinische Transfusionsmedizin und Immungenetik\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Hubert\",\"middleName\":\"\",\"lastName\":\"Schrezenmeier\",\"suffix\":\"\"},{\"id\":638234251,\"identity\":\"addf3610-cfc5-496f-83fc-432d8688476e\",\"order_by\":15,\"name\":\"Kamal Mustafa\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University of Bergen\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Kamal\",\"middleName\":\"\",\"lastName\":\"Mustafa\",\"suffix\":\"\"},{\"id\":638234252,\"identity\":\"bcd63186-0974-43bd-9bec-b4f5a8cb3c22\",\"order_by\":16,\"name\":\"Frode Steingrimsen Berven\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University of Bergen\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Frode\",\"middleName\":\"Steingrimsen\",\"lastName\":\"Berven\",\"suffix\":\"\"},{\"id\":638234253,\"identity\":\"36fb6a69-0420-4cca-9e4f-5f8fe3979047\",\"order_by\":17,\"name\":\"Lars Bø\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University of Bergen\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Lars\",\"middleName\":\"\",\"lastName\":\"Bø\",\"suffix\":\"\"},{\"id\":638234255,\"identity\":\"232535c0-b385-442d-94d7-19dc0c2717ad\",\"order_by\":18,\"name\":\"Torbjørn Kråkenes\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Haukeland University Hospital\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Torbjørn\",\"middleName\":\"\",\"lastName\":\"Kråkenes\",\"suffix\":\"\"},{\"id\":638234256,\"identity\":\"2f2998a5-cae8-491a-b9b3-7755c5d9ebbe\",\"order_by\":19,\"name\":\"Christopher Elnan Kvistad\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Haukeland University Hospital\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Christopher\",\"middleName\":\"Elnan\",\"lastName\":\"Kvistad\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2026-05-06 11:24:35\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-9629755/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-9629755/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":109103521,\"identity\":\"01204a80-d506-4f37-becc-0f74cb66390c\",\"added_by\":\"auto\",\"created_at\":\"2026-05-12 14:41:16\",\"extension\":\"png\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":1229102,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eSMART-MS clinical trial design and integrative blood–CSF analysis framework\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(A) \\u003c/strong\\u003eCross-over design of the SMART-MS trial. Eighteen patients with progressive multiple sclerosis were randomized 1:1 to receive either intrathecal mesenchymal stem cell (MSC) therapy followed by control (Sequence A, MSC_first) or control followed by MSC (Sequence B, Control_first). Peripheral blood samples were collected at baseline (BL), 6 months, and 12 months. Cerebrospinal fluid (CSF) samples were collected at BL and 6 months. Yellow arrows indicate periods of MSC exposure, while blue arrows indicate control periods. All patients underwent bone marrow aspiration prior to MSC or control administration.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(B) \\u003c/strong\\u003eOverview of the integrative data analysis workflow combining longitudinal peripheral blood immune profiling and CSF proteomics. Peripheral blood samples from all time-points were processed for mass cytometry (CyTOF) using in-house protocols and a titrated and validated antibody panel. Unsupervised clustering of CD45⁺ CD66b\\u003csup\\u003elow/neg \\u003c/sup\\u003eleukocytes identified 25 immune phenotypes, which were subsequently analysed along two complementary dimensions: (i) cellular composition, assessed using centred log-ratio–transformed (CLR) cluster frequencies and mixed-effects models accounting for treatment, period, sequence, and patient pairing; and (ii) functional program analysis, based on composite marker-derived scores summarizing activation, differentiation, and trafficking states. In parallel, CSF samples collected at baseline and 6 months underwent proteomic profiling followed by quality control and normalization. Differential protein abundance analysis and GO–based gene set enrichment analysis (GSEA) were performed, and selected CSF functional programs were derived using known protein sets. Mechanistically related functional programs across blood and CSF compartments were subsequently coupled to assess coordinated immune–CNS dynamics.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9629755/v1/2b4eedd1782f540503ee93c0.png\"},{\"id\":109103411,\"identity\":\"ccd364c6-1087-4c76-8103-663834be111a\",\"added_by\":\"auto\",\"created_at\":\"2026-05-12 14:40:53\",\"extension\":\"png\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":585164,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eBaseline peripheral immune landscape in progressive MS\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(A\\u003c/strong\\u003e) UMAP projection of all CD45⁺ CD66b\\u003csup\\u003elow/neg\\u003c/sup\\u003e cells at baseline coloured by the 25 immune phenotypes identified by unsupervised clustering.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(B)\\u003c/strong\\u003e Heatmap of scaled mean marker expression for each phenotype, with side colours indicating major lineages (CD4⁺ T cells, CD8⁺ T cells, B cells, NK cells, dendritic cells, monocytes).\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(C)\\u003c/strong\\u003e Baseline lineage-level composition per patient, showing the relative abundance of each major lineage (sum of corresponding phenotypes) as stacked bars. Each bar corresponds to an individual patient (sample IDs S1–Sxx), displayed without a specific ordering.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(D)\\u003c/strong\\u003e Baseline relative abundance of each phenotype. Large points represent the cohort median with standard error of the mean, while smaller points in gray indicate individual patients, illustrating inter-individual heterogeneity.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9629755/v1/4e6c4b48e9e4af00912fd9da.png\"},{\"id\":109103317,\"identity\":\"4b350ac1-c9ea-46a9-8635-0e2af905e41a\",\"added_by\":\"auto\",\"created_at\":\"2026-05-12 14:40:23\",\"extension\":\"png\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":618684,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eLongitudinal peripheral immune composition during MSC and control periods.\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(A)\\u003c/strong\\u003e Stacked bar plots showing the relative abundances of major immune lineages (CD4⁺ T cells, CD8⁺ T cells, B cells, monocytes, NK cells, and dendritic cells) for each patient at 6 and 12 months, stratified by active treatment at the corresponding visit.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(B)\\u003c/strong\\u003e Violin and box plots of centred log-ratio (CLR)–transformed frequencies for selected immune phenotypes (CD4 central memory, CD4 TEMRA, memory B cells, and CD16\\u003csup\\u003elow \\u003c/sup\\u003eNK cells) across baseline, 6-month, and 12-month time points, with individual patient trajectories overlaid.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(C)\\u003c/strong\\u003e Forest plot of MSC treatment effect estimates from linear mixed-effects models fitted to CLR-transformed frequencies of all 25 immune phenotypes. Points indicate estimated treatment effects, and horizontal bars represent 95% confidence intervals.\\u0026nbsp;\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9629755/v1/96cbc1bffb2ef2be50f08885.png\"},{\"id\":109103926,\"identity\":\"1c8895fe-1693-4d17-b01a-8100579fc625\",\"added_by\":\"auto\",\"created_at\":\"2026-05-12 14:43:24\",\"extension\":\"png\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":475923,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eBlood-based functional immune programs and their longitudinal behaviour under MSC treatment\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(A)\\u003c/strong\\u003e Schematic representation of composite functional score construction, illustrating how individual CyTOF markers contribute to ten biologically defined functional programs.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(B)\\u003c/strong\\u003e Baseline distributions of selected T cell functional scores across CD4⁺ and CD8⁺ immune phenotypes, shown as jittered patient-level points in gray including the cohort median in solid color ± standard error.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(C)\\u003c/strong\\u003e Heatmap of estimated MSC versus control effects on functional scores across all immune phenotypes derived from linear mixed-effects models. Columns represent functional scores, rows represent the 25 phenotypes, and colours indicate the direction and magnitude of the MSC effect (purple = higher under MSC, orange = lower under MSC, white = no effect). Lineage annotation for each phenotype is shown below, nominal significance before correction for multiple testing is marker with *.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9629755/v1/1af7f07a163e8881d9a5bfa2.png\"},{\"id\":109103302,\"identity\":\"96899efd-b5bf-4111-8c57-7a3c5552dcb6\",\"added_by\":\"auto\",\"created_at\":\"2026-05-12 14:40:16\",\"extension\":\"png\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":525240,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e\\u003cstrong\\u003eCSF proteomic remodelling under MSC treatment and integration with circulating immune functional programs.\\u003c/strong\\u003e\\u003cbr\\u003e\\n \\u003cstrong\\u003e(A)\\u003c/strong\\u003e Volcano plot showing estimated MSC treatment effects on longitudinal CSF protein changes (6-month minus baseline) plotted against statistical significance.\\u003cbr\\u003e\\n \\u003cstrong\\u003e(B)\\u003c/strong\\u003e Dot plot summarizing Gene Ontology enrichment analysis of MSC-associated CSF proteomic changes after removal of proteins associated with blood and epithelial contamination.\\u003cbr\\u003e\\n \\u003cstrong\\u003e(C) \\u003c/strong\\u003eBoxplots show longitudinal changes in three CSF-derived functional program scores. Patients are stratified into four groups based on treatment and clinical/radiological reactivity.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(D)\\u003c/strong\\u003e Scatter plots illustrating the relationship between longitudinal changes in blood-derived functional programs (CyTOF) and CSF program scores across T cell subsets, with points representing individual patients and regression lines shown separately for MSC and control periods.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage5.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9629755/v1/edd7a98c9f3b275c341b843e.png\"},{\"id\":109104306,\"identity\":\"2a4f0b72-9932-445f-bd61-127947577806\",\"added_by\":\"auto\",\"created_at\":\"2026-05-12 14:45:53\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":3699753,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9629755/v1/931ac698-a8ae-4023-a846-e5cb965f6a2b.pdf\"},{\"id\":109103145,\"identity\":\"bc9d96a8-9957-4051-8fbf-de93da900962\",\"added_by\":\"auto\",\"created_at\":\"2026-05-12 14:38:56\",\"extension\":\"docx\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":5374254,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"supplementarymaterial.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9629755/v1/022d4d7de1aa7981c799eec4.docx\"}],\"financialInterests\":\"Competing interest reported. TH has received speaker honoraria and/or participated in clinical multiple sclerosis trials sponsored by Biogen, Sanofi, Merck, Roche, Amgen, and Novartis. ØT has participated in advisory boards and received speaker honoraria from Biogen, Merck, Novartis, Teva, Roche, Sanofi, and Bristol Myers Squibb, and has participated in clinical trials sponsored by Merck, Novartis, Roche, and Sanofi. All other authors declare no competing interests.\",\"formattedTitle\":\"\\u003cp\\u003eIntrathecal mesenchymal stem cell therapy in progressive multiple sclerosis: cross-compartment immune profiling in the SMART-MS randomized trial\\u003c/p\\u003e\",\"fulltext\":[{\"header\":\"1. BACKGROUND\",\"content\":\"\\u003cp\\u003eMultiple sclerosis (MS) is a chronic immune-mediated disease of the central nervous system (CNS) affecting more than 2.9\\u0026nbsp;million individuals worldwide, with typical onset between 20 and 40 years of age. Clinically, approximately 90% of patients initially present with relapsing\\u0026ndash;remitting MS (RRMS), whereas around 10% are diagnosed with progressive MS, a disease course characterized by gradual accumulation of disability, impaired mobility, cognitive decline, and ongoing neuroaxonal loss, particularly within the spinal cord [\\u003cspan citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e1\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eDespite substantial advances in the treatment of RRMS, there remains a major unmet need for effective therapies in progressive MS. To date, ocrelizumab is the only approved therapy shown to slow disability progression in this population, underscoring the limited therapeutic options available [\\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2\\u003c/span\\u003e]. Progressive disease is increasingly understood to reflect compartmentalized inflammation within CNS border regions that is poorly captured by systemic immune readouts. Consequently, therapeutic strategies capable of simultaneously modulating immune activity, promoting neuroprotection, and supporting tissue repair are considered particularly relevant for altering long-term outcomes in progressive MS [\\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e3\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR4\\\" class=\\\"CitationRef\\\"\\u003e4\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eAgainst this background, mesenchymal stem cells (MSCs) have emerged as a promising therapeutic candidate. MSCs are multipotent stromal cells that can be isolated from adult tissues such as bone marrow or adipose tissue, as well as from perinatal sources, and can be administered in autologous or allogeneic settings without the ethical concerns associated with embryonic stem cells [\\u003cspan citationid=\\\"CR5\\\" class=\\\"CitationRef\\\"\\u003e5\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e6\\u003c/span\\u003e]. Unlike induced pluripotent stem cell\\u0026ndash;based approaches, MSC therapies do not require genetic reprogramming and are supported by extensive preclinical evidence demonstrating immunomodulatory, trophic, and tissue-repair\\u0026ndash;associated properties. Early-phase clinical studies and small randomized trials of intrathecal MSC or MSC-derived neural progenitor therapies in progressive MS have reported acceptable safety profiles and signals suggestive of biological activity.\\u003c/p\\u003e \\u003cp\\u003eA recent example is the SMART-MS trial [\\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e], a randomized, double-blind, placebo-controlled phase I/II crossover study evaluating intrathecal administration of autologous bone marrow\\u0026ndash;derived MSCs in progressive MS (NCT04749667). In this trial, 18 patients with non-active progressive disease received a single intrathecal injection of MSCs or placebo followed by crossover at six months, enabling within-patient comparisons while maintaining blinding. Although the study did not demonstrate clear neuro-regenerative efficacy on its primary endpoint, MSC exposure was associated with transient biological signals, including exploratory MRI signals suggestive of transient tissue effects. Adverse events such as fever, back pain, and spinal MRI abnormalities suggested localized inflammatory or reactive responses following intrathecal administration, highlighting the need to define how intrathecal MSC therapy reshapes immune-CNS interface biology in vivo [\\u003cspan additionalcitationids=\\\"CR9 CR10 CR11 CR12 CR13\\\" citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e8\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e14\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eIn parallel, rapid advances in high-dimensional omics technologies now enable detailed interrogation of immune cell composition, activation states, and functional programs in MS, offering new opportunities to dissect treatment-associated biological effects in vivo [\\u003cspan additionalcitationids=\\\"CR16\\\" citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR17\\\" class=\\\"CitationRef\\\"\\u003e17\\u003c/span\\u003e]. However, most MSC studies to date have relied on targeted biomarker approaches focusing on a limited number of inflammatory chemokines, complement components, matrix metalloproteinases, or neuroaxonal injury markers, often assessed in isolation within cerebrospinal fluid (CSF) and with minimal integration of comprehensive immune profiling in peripheral blood or cross-compartment analyses [\\u003cspan citationid=\\\"CR18\\\" class=\\\"CitationRef\\\"\\u003e18\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR19\\\" class=\\\"CitationRef\\\"\\u003e19\\u003c/span\\u003e]. As a result, how MSC therapy modulates immune and tissue-associated programs across systemic and intrathecal compartments in progressive MS remains incompletely understood.\\u003c/p\\u003e \\u003cp\\u003eTo address this gap, we performed integrated longitudinal immune profiling in patients with progressive MS enrolled in the SMART-MS trial [\\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e], combining mass cytometry (CyTOF \\u0026ndash; Cytometry by Time-of-Flight)\\u0026ndash;based single-cell analysis of peripheral blood at baseline, 6 months, and 12 months with matched CSF proteomic profiling in 18 participants. By jointly interrogating immune composition, functional programs, and CSF proteomic remodelling, we provide a systems-level view of MSC-associated biological effects. Our findings indicate that intrathecal MSC therapy is associated with selective remodelling of the CSF interface and extracellular matrix rather than broad modulation of circulating lymphocyte and monocyte programs. Inflammatory and metabolic proteomic responses were primarily enriched in individuals experiencing clinical or radiological reactivity, highlighting substantial inter-individual heterogeneity in treatment-associated biology.\\u003c/p\\u003e\"},{\"header\":\"2. RESULTS\",\"content\":\"\\u003cdiv id=\\\"Sec3\\\"\\u003e\\n \\u003ch2\\u003e2.1 Baseline peripheral immune composition in progressive MS\\u003c/h2\\u003e\\n \\u003cp\\u003eTo characterise the baseline peripheral immune landscape of patients with progressive MS, we performed high-dimensional profiling using CyTOF with a 41-antibody panel (Supplementary Table\\u0026nbsp;1). An overview of the SMART-MS trial design and the integrative analysis workflow is provided in Fig.\\u0026nbsp;1. Phenotypic characterisation of peripheral blood leukocytes was performed using unsupervised clustering of CD45⁺ and CD66b\\u003csup\\u003elow/neg\\u003c/sup\\u003e cells with the FlowSOM algorithm [20] (n = 2,558,027 cells at baseline from m = 14 patients; Supplementary Fig.\\u0026nbsp;1), yielding 25 distinct populations spanning CD4⁺ and CD8⁺ T cells, B cells, monocytes, natural killer (NK) cells, and dendritic cells (Fig.\\u0026nbsp;2A).\\u003c/p\\u003e\\n \\u003cp\\u003eIn the two-dimensional Uniform Manifold Approximation and Projection (UMAP) representation, these phenotypes formed well-separated and coherent clusters, indicating that the unsupervised approach captured biologically meaningful populations rather than arbitrary data partitions. To validate the phenotypic assignments, we examined average marker expression profiles across all 25 clusters. The heatmap in Fig. 2B revealed clear lineage- and differentiation-specific patterns: CD4⁺ T cell clusters were enriched for CD4 and displayed CD45RA/RO-defined naïve, central memory (CM), effector memory (EM), and effector memory cells re-expressing CD45RA (TEMRA) phenotypes; CD8⁺ T cell clusters expressed CD8a and corresponding differentiation markers; B cell clusters were defined by CD19/CD20 and CD27/CD38-based naïve and memory states; and myeloid and NK-cell clusters were distinguished by CD14/CD16/CD11c/HLA-DR and CD56/CD16 expression, respectively. Together, these expression profiles re-confirmed that the identified clusters correspond to well-recognised immune subsets and established a robust framework for downstream compositional and functional analyses.\\u003c/p\\u003e\\n \\u003cp\\u003eWe next examined lineage-level distributions at the patient level (Fig. 2C). Stacked bar plots demonstrated modest but consistent inter-individual variability in immune lineage proportions. As expected at baseline, the peripheral immune compartment (CD45\\u003csup\\u003e+\\u003c/sup\\u003eCD66b\\u003csup\\u003elow/neg\\u003c/sup\\u003e) was dominated by T cells and monocytes, with smaller contributions from B cells, NK cells, and dendritic cells. Considerable heterogeneity in the abundance of specific immune subsets across patients was observed (Fig. 2D), highlighting the inter-patient variability characteristic of progressive MS. Collectively, these analyses provide a comprehensive baseline reference of peripheral immune composition prior to treatment, forming the foundation for subsequent evaluation of MSC-associated effects on immune cell frequencies and functional states.\\u003c/p\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003e(A)\\u003c/strong\\u003e Cross-over design of the SMART-MS trial. Eighteen patients with progressive multiple sclerosis were randomized 1:1 to receive either intrathecal mesenchymal stem cell (MSC) therapy followed by control (Sequence A, MSC_first) or control followed by MSC (Sequence B, Control_first). Peripheral blood samples were collected at baseline (BL), 6 months, and 12 months. Cerebrospinal fluid (CSF) samples were collected at BL and 6 months. Yellow arrows indicate periods of MSC exposure, while blue arrows indicate control periods. All patients underwent bone marrow aspiration prior to MSC or control administration.\\u003cbr\\u003e \\u003cstrong\\u003e(B)\\u003c/strong\\u003e Overview of the integrative data analysis workflow combining longitudinal peripheral blood immune profiling and CSF proteomics. Peripheral blood samples from all time-points were processed for mass cytometry (CyTOF) using in-house protocols and a titrated and validated antibody panel. Unsupervised clustering of CD45⁺ CD66b\\u003csup\\u003elow/neg\\u003c/sup\\u003e leukocytes identified 25 immune phenotypes, which were subsequently analysed along two complementary dimensions: (i) cellular composition, assessed using centred log-ratio–transformed (CLR) cluster frequencies and mixed-effects models accounting for treatment, period, sequence, and patient pairing; and (ii) functional program analysis, based on composite marker-derived scores summarizing activation, differentiation, and trafficking states. In parallel, CSF samples collected at baseline and 6 months underwent proteomic profiling followed by quality control and normalization. Differential protein abundance analysis and GO–based gene set enrichment analysis (GSEA) were performed, and selected CSF functional programs were derived using known protein sets. Mechanistically related functional programs across blood and CSF compartments were subsequently coupled to assess coordinated immune–CNS dynamics.\\u003cbr\\u003e\\u003cstrong\\u003e(A\\u003c/strong\\u003e) UMAP projection of all CD45⁺ CD66b\\u003csup\\u003elow/neg\\u003c/sup\\u003e cells at baseline coloured by the 25 immune phenotypes identified by unsupervised clustering.\\u003cbr\\u003e \\u003cstrong\\u003e(B)\\u003c/strong\\u003e Heatmap of scaled mean marker expression for each phenotype, with side colours indicating major lineages (CD4⁺ T cells, CD8⁺ T cells, B cells, NK cells, dendritic cells, monocytes).\\u003cbr\\u003e \\u003cstrong\\u003e(C)\\u003c/strong\\u003e Baseline lineage-level composition per patient, showing the relative abundance of each major lineage (sum of corresponding phenotypes) as stacked bars. Each bar corresponds to an individual patient (sample IDs S1–Sxx), displayed without a specific ordering.\\u003cbr\\u003e \\u003cstrong\\u003e(D)\\u003c/strong\\u003e Baseline relative abundance of each phenotype. Large points represent the cohort median with standard error of the mean, while smaller points in gray indicate individual patients, illustrating inter-individual heterogeneity.\\u003cbr\\u003e\\u003c/p\\u003e\\n\\u003c/div\\u003e\\n\\u003cdiv id=\\\"Sec4\\\"\\u003e\\n \\u003ch2\\u003e2.2 Peripheral immune composition remains stable during intrathecal MSC therapy\\u003c/h2\\u003e\\n \\u003cp\\u003eWe next investigated whether MSC therapy alters the composition of the peripheral immune compartment over time. To this end, CD45⁺ CD66b\\u003csup\\u003elow/neg\\u003c/sup\\u003e peripheral blood leukocytes from SMART-MS participants were analysed by CyTOF at 6 months (6m) and 12 months (12m) following treatment, according to the cross-over trial design. Multidimensional profiling of immune cells at these time points (n = 2,636,447 cells at 6m from m = 16 patients; and n = 1,828,594 cells at 12m from m = 12 patients; Supplementary Fig. 2) yielded UMAP embeddings that were visually comparable to baseline, with well-defined and coherent immune phenotypes preserved over time (Supplementary Fig. 3).\\u003c/p\\u003e\\n \\u003cp\\u003eAt the lineage level, stacked bar plots summarised the relative abundances of CD4⁺ T cells, CD8⁺ T cells, B cells, monocytes, NK cells, and dendritic cells for each patient at 6 and 12 months, stratified by active treatment at the corresponding visit (Fig. 3A). Overall, the peripheral immune composition remained similar during MSC and control periods at both time points. While inter-individual variability in lineage proportions was evident, no lineage exhibited a consistent expansion or depletion associated with MSC exposure.\\u003c/p\\u003e\\n \\u003cp\\u003eTo explore potential treatment effects at higher resolution, we visualised centred log-ratio (CLR)–transformed frequencies of selected immune phenotypes, including CD4 central memory T cells, CD4 TEMRA cells, memory B cells, and CD16\\u003csup\\u003elow\\u003c/sup\\u003e NK cells, across baseline, 6-month, and 12-month visits, stratified by treatment status (Fig. 3B). Violin and box plots with overlaid patient trajectories revealed modest within-patient fluctuations over time but did not indicate large or systematic differences between MSC and control periods. For example, CD4 TEMRA cells showed slightly higher CLR values during MSC periods in some patients, whereas others displayed minimal or opposing changes, resulting in an overall weak and inconsistent pattern. Similar limited variability without a clear treatment-specific shift was observed for CD4 central memory (CD4_CM), B memory (B_Mem), NK dim CD16\\u003csup\\u003elow\\u003c/sup\\u003e, and other immune subsets (Supplementary Fig.\\u0026nbsp;12).\\u003c/p\\u003e\\n \\u003cp\\u003eTo formally quantify these observations, we fitted linear mixed-effects models to CLR-transformed frequencies of all 25 immune phenotypes. Across all immune subsets, estimated MSC treatment effects were small, with most effect sizes lying within ± 0.25 on the CLR scale and confidence intervals including zero (Fig. 3C; Supplementary Fig.\\u0026nbsp;13). No phenotype reached statistical significance after correction for multiple testing. The largest positive treatment estimates were observed for transitional B cells and CD4 TEMRA cells, whereas plasmablast B cells showed the most negative estimate; however, these effects were modest in magnitude and not statistically significant (Supplementary Fig.\\u0026nbsp;4).\\u003c/p\\u003e\\n \\u003cp\\u003eAs a sensitivity analysis, we fitted an alternative mixed-effects model comparing MSC exposure to untreated baseline samples. This analysis yielded consistent results, with small and non-significant effect estimates across immune subsets (Supplementary Fig. 14). Collectively, the concordant findings from both mixed-effects modelling approaches indicate that MSC therapy does not induce major changes in systemic immune composition, consistent with compartmentalized CNS inflammation in progressive MS. Given this compositional stability, we next investigated whether MSC treatment might instead modulate more subtle functional properties within otherwise stable immune compartments.\\u003c/p\\u003e\\n \\u003cp\\u003e\\u003cstrong\\u003e(A)\\u003c/strong\\u003e Stacked bar plots showing the relative abundances of major immune lineages (CD4⁺ T cells, CD8⁺ T cells, B cells, monocytes, NK cells, and dendritic cells) for each patient at 6 and 12 months, stratified by active treatment at the corresponding visit.\\u003cbr\\u003e \\u003cstrong\\u003e(B)\\u003c/strong\\u003e Violin and box plots of centred log-ratio (CLR)–transformed frequencies for selected immune phenotypes (CD4 central memory, CD4 TEMRA, memory B cells, and CD16\\u003csup\\u003elow\\u003c/sup\\u003e NK cells) across baseline, 6-month, and 12-month time points, with individual patient trajectories overlaid.\\u003cbr\\u003e \\u003cstrong\\u003e(C)\\u003c/strong\\u003e Forest plot of MSC treatment effect estimates from linear mixed-effects models fitted to CLR-transformed frequencies of all 25 immune phenotypes. Points indicate estimated treatment effects, and horizontal bars represent 95% confidence intervals.\\u003cbr\\u003e\\u003c/p\\u003e\\u003cbr\\u003e\\n\\u003c/div\\u003e\\n\\u003cdiv id=\\\"Sec5\\\"\\u003e\\n \\u003ch2\\u003e2.3 Blood-based functional programs show limited modulation under MSC treatment\\u003c/h2\\u003e\\n \\u003cp\\u003eTo move beyond compositional changes in peripheral blood, we derived composite functional scores that summarise activation, differentiation and trafficking programs using relevant markers available in the CyTOF panel. The marker–score matrix generation in Fig. 4A illustrates how individual markers map onto ten functional programs, including T cell activation, memory/effector differentiation, inflammatory homing (e.g., Th17-like), CNS-homing, T regulatory-like features (Treg), B cell activation and memory, myeloid APC activation, non-classical monocyte features and NK-cell activation.\\u003c/p\\u003e\\n \\u003cp\\u003eWe first characterised the baseline (i.e., before treatment) functional landscape across all 25 immune phenotypes in the SMART-MS cohort. A heatmap of median baseline scores per phenotype revealed distinct lineage-associated patterns (Supplementary Fig.\\u0026nbsp;7A). At baseline, within the T cell compartment, CD4 central memory T cells exhibited the highest activation and memory/effector scores, with most CD4 and CD8 memory subsets scoring above the global mean, whereas TEMRA subsets showed lower values consistent with a more terminal, non-classical memory phenotype. inflammatory homing scores were selectively elevated in CD4 effector-memory and CD8 effector/central memory cells, indicating that Th17-like inflammatory trafficking programmes were restricted to a subset of memory T cells rather than being uniformly distributed across the T cell compartment. Treg-like scores were negative across conventional T cell phenotypes, reflecting their low CD25/CD73/HLA-DR and high CD127 expression relative to the global immune cell population, although CD4 memory-like subsets were relatively enriched compared with other T cells. In contrast, CNS-homing scores were highest in CD4 effector-memory and CD8 central memory subsets, highlighting these populations as the main carriers of MS-relevant trafficking signatures at baseline. Importantly, these composite scores are not intended to define cell identity but rather to quantify shared functional and trafficking programmes that may be distributed across multiple immune lineages. Accordingly, elevated scores in non–T cell populations reflect engagement in organizing or licensing components of these programmes rather than direct execution of T cell effector functions.\\u003c/p\\u003e\\n \\u003cp\\u003eFocusing on other immune compartments, all B cell subsets showed strong B-activation and B-memory/naïve scores in switched memory and plasmablast-like clusters, with naive B cells scoring lower along the memory axis. Myeloid APC activation scores were highest in classical dendritic cells and, to a lesser extent, in plasmacytoid DCs and classical monocytes. Non-classical monocyte scores selectively highlighted the non-classical Monocyte (Mono_nonClassical) subset, while NK-activation scores were most pronounced in CD16⁺ NK subsets. These patterns were consistent when inspecting patient-level distributions for each score and phenotype (Supplementary Fig.\\u0026nbsp;5). In this context, high trafficking- or activation-related scores in antigen-presenting cells likely reflect their role in establishing chemokine gradients, antigen presentation, and immune coordination, which in turn shape downstream lymphocyte recruitment and activation. In addition, Principal Component Analysis (PCA) of baseline functional scores showed clear separation of T, B, myeloid and NK phenotypes along the first two components, supporting that the composite scores capture expected lineage-specific functional programs (Supplementary Fig.\\u0026nbsp;6). Notably, CD4 effector-memory and CD8 central-memory subsets occupied overlapping regions of the functional score PCA space, indicating convergence in activation and trafficking programs despite distinct lineage identities. This overlap reflects shared engagement in MS-relevant immune programs rather than phenotypic equivalence.\\u003c/p\\u003e\\n \\u003cp\\u003eWe next focused on T cell–specific scores to examine functional heterogeneity between CD4⁺ and CD8⁺ subsets. At baseline, patient-level jitter points highlight substantial inter-individual variability within each T cell subset (Fig. 4B). The CD4 central and effector memory exhibited the highest T-activation score with high patient heterogeneity. Inflammatory homing scores were particularly elevated in CD4 effector-memory cells consistent with an activated, migratory CD4 compartment in progressive MS. CD8 effector memory and CD8 central memory subsets, were also elevated with CD8 effector memory subset presenting the highest patient heterogeneity, while CD8 TEMRA and CD4 TEMPRA cells also showed moderate enrichment. CNS-homing scores mirrored the previous pattern, with higher values in memory CD4 effector memory and CD8 effector/central memory populations than in naïve subsets. In contrast, Treg-like scores were generally low/negative across T cell clusters, reflecting their lower expression of CD25, CD73 and HLA-DR and relatively high CD127 compared with the global immune cell population.\\u003c/p\\u003e\\n \\u003cp\\u003eHaving established that the composite scores recapitulate known functional specialisation of immune subsets at baseline, we next applied the same scoring framework to longitudinal samples to test whether MSC therapy modulates these functional programs over time. The same linear mixed-model framework as for the compositional analyses was used. Figure 4C summarises the estimated MSC versus control effects across all score and phenotype combinations. Effect sizes were generally small and symmetrically distributed around zero, with most β coefficients lying within ± 0.1 score units and only a few tiles reaching nominal significance. No coherent pattern emerged across related subsets or scores, suggesting that intrathecal injection/infusion of MSCs do not induce large-scale remodelling of these composite functional programmes in peripheral blood.\\u003c/p\\u003e\\n \\u003cp\\u003eNext, given the potential role of cell trafficking in progressive MS, we focused more deeply on the CNS-homing score. In Supplementary Fig. 7B, we visualised the relationship between baseline CNS-homing scores and those measured during MSC exposure for four representative compartments: CD4 effector memory (CD4_EM), CD8 central memory (CD8_CM), plasmablast-like B cells (B_pl) and classical monocytes. For patients randomised to the Control_first sequence, baseline values were compared to scores at 12 months, corresponding to the end of their six-month MSC treatment period. For patients in the MSC_first sequence, baseline values were compared to scores at six months, i.e. after an initial six months of MSC treatment. Across all four subsets and in both sequences, individual patients clustered close to the identity line, with baseline and MSC-period scores showing a tight relationship and largely preserved ranking. These trajectories indicate that CNS-homing programmes behave as relatively stable traits over the one-year observation period, with MSC exposure failing to induce a consistent upward or downward shift in the circulating compartment. This interpretation is consistent with the modest, scattered effect sizes observed in the mixed-model heatmap (Fig. 4C) and with the overall stability of cell-type composition reported in Fig. 3.\\u003c/p\\u003e\\n \\u003cp\\u003eTaken together, our CyTOF analyses indicate that both the composition and the functional wiring of circulating immune cells remain stable in progressive MS over 12 months, with no major MSC-induced shifts detectable at the population level. These findings indicate that any immunomodulatory effects of MSCs in this setting may act primarily within the CNS compartment or at immune–CNS interfaces, or through molecular pathways and mechanisms not captured by our current cell-surface marker panel used for CyTOF. To explore this possibility, we next analysed CSF proteomic profiles available for our clinical trial participants.\\u003c/p\\u003e\\u003cstrong\\u003e(A)\\u003c/strong\\u003e Schematic representation of composite functional score construction, illustrating how individual CyTOF markers contribute to ten biologically defined functional programs.\\u003cbr\\u003e \\u003cstrong\\u003e(B)\\u003c/strong\\u003e Baseline distributions of selected T cell functional scores across CD4⁺ and CD8⁺ immune phenotypes, shown as jittered patient-level points in gray including the cohort median in solid color ± standard error.\\u003cbr\\u003e \\u003cstrong\\u003e(C)\\u003c/strong\\u003e Heatmap of estimated MSC versus control effects on functional scores across all immune phenotypes derived from linear mixed-effects models. Columns represent functional scores, rows represent the 25 phenotypes, and colours indicate the direction and magnitude of the MSC effect (purple = higher under MSC, orange = lower under MSC, white = no effect). Lineage annotation for each phenotype is shown below, nominal significance before correction for multiple testing is marker with *.\\u003cbr\\u003e\\n\\u003c/div\\u003e\\n\\u003cdiv id=\\\"Sec6\\\"\\u003e\\n \\u003ch2\\u003e2.4 CSF proteomic remodelling during intrathecal MSC therapy in relation to clinical and MRI findings\\u003c/h2\\u003e\\n \\u003cp\\u003eTo investigate whether intrathecal MSC therapy is associated with molecular changes within the CNS compartment, we analysed CSF proteomic profiles obtained from all SMART-MS participants (n = 18) at baseline and at 6 months, corresponding to the end of the first treatment period in the cross-over trial design. Normalized and log-transformed protein abundance data were first visualized using classical multidimensional scaling (MDS) to obtain a global overview of the CSF proteomic landscape (Supplementary Fig.\\u0026nbsp;8A). While individual samples showed modest longitudinal displacement over time, no clear global separation between MSC-treated and control samples was observed. Comparable patterns were seen using alternative dimensionality-reduction approaches, including PCA and UMAP (Supplementary Figs.\\u0026nbsp;8B and 8C), indicating that MSC treatment did not produce a dominant proteome-wide or global shift detectable by unsupervised dimensionality reduction.\\u003c/p\\u003e\\n \\u003cp\\u003eTo formally assess whether specific proteins exhibited differential regulation following MSC treatment, we performed protein-level differential abundance analysis in the high-dimensional proteomic space. Because matched CSF samples were available for all patients at baseline and 6 months, longitudinal changes in protein abundance were assessed using within-patient differences over time. This approach revealed treatment-associated shifts in the CSF proteome that were not apparent from absolute abundance levels alone. The resulting analysis identified a subset of proteins exhibiting consistent MSC-associated changes over the 6-month interval. These effects are summarized in a volcano plot (Fig. 5A), which displays estimated MSC-associated protein changes against their statistical significance across all quantified CSF proteins.\\u003c/p\\u003e\\n \\u003cp\\u003eWhile the majority of proteins clustered near zero effect size consistent with limited global remodelling of the CSF proteome, a subset of proteins exhibited reproducible MSC-associated changes, providing a focused signal for downstream functional interpretation (Supplementary Table\\u0026nbsp;2). Proteins significantly upregulated (FDR \\u0026lt; 0.05) during MSC exposure formed a coherent group related to extracellular matrix organization and tissue remodelling, including periostin (POSTN), cartilage oligomeric matrix protein (COMP), and fibrillar collagens COL1A1 and COL1A2. In addition, SPARC and multiple fibrillar collagens (COL3A1, COL5A1, COL11A1) and the neuronal guidance and plasticity-associated receptor EPHB2 showed increased abundance at a more permissive threshold (FDR \\u0026lt; 0.10), suggesting selective modulation of CNS structural and signalling pathways rather than broad immune activation.\\u003c/p\\u003e\\n \\u003cp\\u003eSeveral proteins with large apparent effect sizes but lacking statistical significance were also observed. These included hemoglobin subunits (HBA1, HBB) and carbonic anhydrase 3 (CA3), consistent with blood-derived contamination, as well as keratins and epithelial junctional proteins (KRT1, KRT9, KPRP, DSG1, DSP, JUP, PLEC, DCD, AHNAK), indicative of skin contamination during lumbar puncture. Additional intracellular structural and stress-related proteins (e.g. VIM, TUBB2A, TUBA1A, HSPB1, EEF2) were also detected. The concurrent presence of hemoglobin chains together with multiple keratin-associated proteins strongly supports technical contamination rather than biologically meaningful CSF remodelling for these features. Accordingly, these proteins were interpreted cautiously and were not considered central components of the MSC-associated CSF signal.\\u003c/p\\u003e\\n \\u003cp\\u003eTo assess the biological coherence of MSC-associated proteomic changes in a systematic manner, we performed pathway enrichment analysis using Gene Ontology (GO). When all quantified proteins were included without prior filtering, enrichment was dominated by extracellular and barrier-associated processes such as wound healing, coagulation, epidermal development, and tissue remodelling (Supplementary Fig. 9), reflecting the influence of contaminant-derived proteins. After excluding proteins commonly associated with blood and epithelial contamination due to lumbar puncture, enrichment analysis revealed a more refined and biologically interpretable signature. The top 20 significant pathways (Fig. 5B) reflected organic acid metabolism, nucleotide metabolism, cytoskeletal remodelling and vascular/inflammatory interface. Inspection of ranked gene enrichment profiles (Supplementary Fig.\\u0026nbsp;10) demonstrated coordinated downregulation of inflammatory stress signalling programs, coupled with upregulation of pathways related to cellular adaptation, intracellular transport, nucleotide metabolism, metabolic capacity biosynthetic and structural reorganization within the CNS. Together, these patterns are consistent with inflammatory attenuation and adaptive remodelling of CNS interface biology rather than immune activation, suggesting a possible transition toward a more homeostatic and reparative CSF environment.\\u003c/p\\u003e\\n \\u003cp\\u003eTo provide a clinically anchored view of CSF proteomic changes, we incorporated information on adverse events (AE; pain/fever) and spinal cord MRI findings observed six months after treatment. CSF protein deltas were stratified into four groups: Control, MSC with no AE and no MRI findings (MSC no AE no MRI), MSC with AE and no MRI findings (MSC + AE no MRI), and MSC with AE and MRI findings (MSC + AE + MRI).\\u003c/p\\u003e\\n \\u003cp\\u003eDifferentially enriched extracellular matrix and interface-associated proteins including collagens, POSTN, SPARC, COMP, EPHB2, CFL1, and PARK7 showed progressively larger positive deltas across the MSC spectrum, with minimal changes in MSC no AE no MRI patients and the largest shifts observed in MSC + AE no MRI and MSC + AE + MRI groups (Supplementary Fig.\\u0026nbsp;15). Control samples remained centred near zero change. Although subgroup sizes were small (n = 3 per MSC stratum), precluding formal statistical comparisons, the consistent directionality across multiple independent proteins supports a non-random longitudinal trend.\\u003c/p\\u003e\\n \\u003cp\\u003eAs a contextual comparison, we examined established CSF inflammatory markers (including C3/C4, CRP, SAA1/2, HP, A2M, SERPINs, fibrinogen chains, and ORM1/2) together with proteins commonly associated with contamination or procedural carryover. Six-month changes in these markers were generally modest and showed substantial overlap between control and MSC-treated patients, irrespective of adverse events or MRI findings (Supplementary Figs.\\u0026nbsp;16 and 17). Previous analyses of the SMART-MS proteomics dataset identified several inflammation-related proteins as strong discriminators between treatment periods; however, our longitudinal modelling framework focused on coordinated directional changes across protein sets rather than individual discriminatory features [7]. Within this framework, we did not observe a uniform or cohort-wide increase in classical inflammatory markers, indicating that inflammatory responses were heterogeneous and context-dependent rather than representing a global inflammatory shift.\\u003c/p\\u003e\\n \\u003cp\\u003eSimilarly, contamination-associated proteins (including hemoglobin chains, keratins, vimentin, and junctional proteins) displayed heterogeneous and individually variable longitudinal patterns rather than coordinated changes across patients. Because contamination signals are typically stochastic, the absence of a consistent cohort-level increase suggests that the extracellular matrix remodelling signature is unlikely to be explained solely by generalized sampling artifacts or traumatic lumbar puncture. Localized procedural or tissue responses cannot be ruled out, and the extracellular matrix /interface signature should be interpreted as suggestive rather than definitive evidence of treatment-associated intrathecal remodelling.\\u003c/p\\u003e\\n\\u003c/div\\u003e\\n\\u003cp\\u003e2.5 Program-level integration of CSF remodelling with circulating T cell trafficking signatures\\u003c/p\\u003e\\n\\u003cp\\u003eTo synthesise the CSF proteomic changes into biologically interpretable patterns, we summarised protein-level responses into composite CSF functional programs reflecting extracellular matrix and CNS interface remodelling, innate and vascular inflammatory stress, and metabolic/cytoskeletal adaptation. In addition, a CSF reactivity score capturing biological responses (e.g., tissue stress reactions) to intrathecal administration was considered. When examined at the patient level, these CSF program scores revealed distinct patterns across MSC-treated individuals stratified by clinical and radiological reactivity (Fig. 5C), providing a structured view of how MSC therapy differentially impacts the CSF compartment.\\u003c/p\\u003e\\n\\u003cp\\u003eAcross MSC-treated individuals, the CSF extracellular matrix/interface remodelling (ECM/interface) score showed a consistent increase, including in patients without adverse events or MRI activity, suggesting a shared treatment-associated pattern within the CSF compartment. Linear modelling indicated that this extracellular matrix/interface remodelling signal was associated with MSC exposure and was not explained by the CSF reactivity score (p = 0.44), supporting the interpretation that it reflects a treatment-related signature rather than nonspecific biological reactivity.\\u003c/p\\u003e\\n\\u003cp\\u003eIn contrast, innate/vascular inflammatory and metabolic/cytoskeletal programs exhibited greater inter-individual variability. Patients experiencing clinical adverse events and/or post-procedural spinal MRI abnormalities following intrathecal MSC administration tended to show higher scores in these dimensions, whereas MSC-treated patients without clinical or radiological reactivity largely overlapped with controls. Consistent with this observation, metabolic/cytoskeletal adaptation scores were strongly associated with the CSF reactivity score in multivariable models (p = 0.004), while treatment effects were modest, indicating that these changes likely reflect secondary, context-dependent responses rather than direct MSC-driven remodelling.\\u003c/p\\u003e\\n\\u003cp\\u003eScatter plot analyses (Supplementary Fig.\\u0026nbsp;18) further supported this dissociation: extracellular matrix/interface remodelling occurred largely independently of vascular inflammatory activation, whereas metabolic/cytoskeletal responses closely tracked innate/vascular stress. Together, these patterns suggest that intrathecal MSC therapy primarily reprograms the CNS interface and extracellular matrix environment, while exaggerated inflammatory and metabolic responses emerge only in a subset of patients experiencing heightened CSF reactivity.\\u003c/p\\u003e\\n\\u003cp\\u003eFinally, to explore potential links between CSF remodelling and peripheral immune dynamics, we examined associations between functional program scores derived from CSF proteomics and blood-based CyTOF analyses. We focused on circulating programs related to immune trafficking and CSF programs capturing extracellular matrix and interface remodelling, given their putative relevance to immune–CNS interactions. Across multiple immune cell types, CSF interface remodelling score showed a clear separation between control and MSC-treated samples. However, within individual CD4⁺ and CD8⁺ T cell subsets, longitudinal changes in the circulating inflammatory homing program did not exhibit consistent linear coupling, supporting a compartmentalized model of intrathecal immune remodelling (Fig. 5D). Instead, the dominant pattern observed was a treatment-associated shift in CSF interface score, with relatively weak and variable within-group relationships to circulating Th17-like inflammatory homing dynamics. Similar exploratory analyses across additional immune compartments, including terminally differentiated or non-trafficking T cell subsets, B cell subsets, and myeloid-derived blood programs, likewise did not reveal reproducible linear coupling with CSF interface remodelling (Supplementary Fig.\\u0026nbsp;11). While some cell-type–specific trends were observed, these relationships were heterogeneous and did not converge on a cell type-specific circulating program that independently explained CSF interface variation.\\u003c/p\\u003e\\n\\u003cp\\u003eTaken together, these analyses indicate that intrathecal MSC therapy in progressive MS is associated with consistent remodelling of the CSF interface and extracellular matrix, reflected by clear differences between control and MSC-treated samples. In contrast, longitudinal changes in circulating inflammatory Th17-like trafficking programs showed weak and heterogeneous relationships with CSF remodelling at the individual level. The CSF reactivity score was selectively elevated in MSC-treated individuals experiencing post-procedural pain, fever, or spinal MRI enhancement, whereas saline controls and non-reactive MSC-treated patients exhibited only minimal changes, suggesting that this axis reflects biological responsiveness to intrathecal MSC exposure rather than generic procedure-related artifact. Despite the limited sample size, the stability of the interface remodelling signal across patients, together with the selective association of tissue reactivity with metabolic and inflammatory programs, argues against these CSF changes being driven solely by nonspecific injury, stress, or contamination. Collectively, the data support a model in which MSC therapy induces compartmentalized and patient-specific remodelling at the immune–CNS interface, rather than being explained by a single dominant peripheral immune program.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e(A)\\u003c/strong\\u003e Volcano plot showing estimated MSC treatment effects on longitudinal CSF protein changes (6-month minus baseline) plotted against statistical significance.\\u003cbr\\u003e\\u003cstrong\\u003e(B)\\u003c/strong\\u003e Dot plot summarizing Gene Ontology enrichment analysis of MSC-associated CSF proteomic changes after removal of proteins associated with blood and epithelial contamination.\\u003cbr\\u003e\\u003cstrong\\u003e(C)\\u003c/strong\\u003e Boxplots show longitudinal changes in three CSF-derived functional program scores. Patients are stratified into four groups based on treatment and clinical/radiological reactivity.\\u003cbr\\u003e\\u003cstrong\\u003e(D)\\u003c/strong\\u003e Scatter plots illustrating the relationship between longitudinal changes in blood-derived functional programs (CyTOF) and CSF program scores across T cell subsets, with points representing individual patients and regression lines shown separately for MSC and control periods.\\u003cbr\\u003e\\u003c/p\\u003e\"},{\"header\":\"3. MATERIALS AND METHODS\",\"content\":\"\\u003cdiv id=\\\"Sec9\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.1 Patients, study design and sample collection\\u003c/h2\\u003e \\u003cp\\u003ePatients were recruited from the SMART-MS clinical trial [\\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e], a randomized, double-blind, placebo-controlled phase I/II crossover study evaluating intrathecal administration of autologous bone marrow\\u0026ndash;derived MSCs in progressive MS (ClinicalTrials.gov NCT04749667). Eligible participants were adults aged 18\\u0026ndash;55 years with primary or secondary progressive MS according to revised McDonald criteria, Expanded Disability Status Scale (EDSS) scores between 4.0 and 7.0, disease duration of 2\\u0026ndash;18 years, and no clinical relapses or MRI activity for at least 24 months prior to enrolment. Patients receiving disease-modifying therapies during the trial were excluded to avoid confounding treatment effects. Participants were randomized 1:1 to receive either intrathecal MSCs or placebo (0.9% saline) at baseline, followed by crossover at six months, allowing within-patient comparisons while maintaining blinding. Autologous MSCs were generated from bone marrow aspirates under Good Manufacturing Practice (GMP) conditions and administered as a single intrathecal dose of 1\\u0026times;10⁶ cells/kg body weight (maximum 100\\u0026times;10⁶ cells) via lumbar puncture. To preserve blinding, all participants underwent identical bone marrow aspiration and injection procedures regardless of treatment allocation. The primary endpoint of the clinical trial was change in combined evoked potential latency at six months, while secondary endpoints included safety, MRI outcomes, functional and ophthalmological measures, clinical disability scores, and serum biomarkers. Adverse events and serious adverse events were prospectively recorded throughout the study. For the present multi-omics analyses, peripheral blood samples were collected at baseline, 6 months, and 12 months for mass cytometry profiling, and CSF was obtained immediately prior to each intrathecal injection for proteomic analysis. Longitudinal sampling within the crossover design enabled paired comparisons of MSC-treated and control periods within the same individuals. The trial was approved by the Regional Committee for Medical and Health Research Ethics and conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. All participants provided written informed consent prior to inclusion.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec10\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.2 Mass cytometry antibody panel, sample preparation, and data acquisition\\u003c/h2\\u003e \\u003cp\\u003eFor mass cytometry, an in-house 41-marker antibody panel was used for surface phenotyping (Supplementary Table\\u0026nbsp;1). Markers were selected based on prior studies and included lineage-defining proteins, differentiation and activation markers, chemokine receptors, adhesion molecules, and stem/progenitor-associated markers relevant to immune trafficking and activation in multiple sclerosis. Antibodies were obtained either pre-conjugated from Standard BioTools or conjugated in-house using Maxpar metal-labeling kits according to the manufacturer\\u0026rsquo;s instructions [\\u003cspan citationid=\\\"CR24\\\" class=\\\"CitationRef\\\"\\u003e24\\u003c/span\\u003e]. Antibody titrations were performed on barcoded mixtures of stabilized whole blood and reference cell populations to determine optimal staining concentrations, ensuring adequate signal resolution while minimizing channel spillover. Peripheral whole blood (1 mL) was collected into cryotubes, stabilized with SmartTube Proteomic Stabilizer, and stored at \\u0026minus;\\u0026thinsp;80\\u0026deg;C. After thawing, red blood cells were lysed according to the manufacturer\\u0026rsquo;s protocol, and white blood cells were counted and aliquoted into 3 \\u0026times; 10⁶-cell samples for downstream processing. Samples were barcoded using the Cell-ID 20-plex Pd Barcoding Kit (Standard BioTools), with longitudinal samples from the same patient included within the same barcode pool to reduce technical variability. A healthy donor reference sample was included in each barcode set to monitor consistency across runs. Following barcoding, pooled samples were stained overnight with a pre-prepared antibody cocktail according to the Maxpar staining protocol and incubated with an iridium DNA intercalator for nuclear labeling. After washing, cells were filtered, resuspended in Cell Acquisition Solution Plus containing EQ Six Element Calibration Beads, and acquired on a CyTOF XT instrument (Standard BioTools) at the Flow Cytometry Core Facility, University of Bergen, Norway. Data were collected at approximately 300\\u0026ndash;500 events per second with signal stability monitored throughout acquisition.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec11\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.3 Mass cytometry data preprocessing and quality control\\u003c/h2\\u003e \\u003cp\\u003eRaw CyTOF data were pre-processed using a standardized and reproducible pipeline implemented in R language. Single-cell debarcoding was subsequently performed using custom R scripts adapted from the premessa package, applying file-specific intensity thresholds to assign events to individual barcodes. Debarcoded samples were exported as individual FCS files for downstream analysis. Quality control and cleanup were performed following established MaxPar/CyTOF cleaning best practices. Briefly, events were filtered based on DNA content using Iridium intercalator channels (Ir191Di/Ir193Di) to remove debris, doublets, and non-cellular events, incorporating signal intensity and event width parameters. Following DNA-based cleanup, an automated Gaussian mixture model (GMM) was applied to partition events based on CD45 and CD66b expression. The leukocyte population used for downstream analysis was defined as CD45⁺ CD66b\\u003csup\\u003elow/neg\\u003c/sup\\u003e cells, thereby enriching for lymphocytes and mononuclear cells while excluding CD66b\\u003csup\\u003ehigh\\u003c/sup\\u003e granulocytes and residual non-immune events from whole-blood acquisitions. To stabilize variance and improve comparability across markers, signal intensities for all protein channels were arcsinh-transformed using a cofactor of 5 prior to clustering, dimensionality reduction, and functional score computation. Technical variation related to barcode structure and acquisition batches was addressed using the CyCombine framework [\\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e25\\u003c/span\\u003e], which performs batch harmonization while preserving underlying biological structure. Batch correction was applied at the single-cell level prior to unsupervised clustering and compositional analyses. For visualization purposes only, marker expression values were scaled to a 0\\u0026ndash;1 range using the 1st and 99th percentile expression values derived from a pooled reference dataset of 12 healthy control samples. The resulting scaling parameters were applied uniformly across all patient samples to ensure consistent visual representation. All statistical/computational analyses were performed on transformed but unscaled data unless otherwise stated.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec12\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.4 Immune phenotyping by unsupervised clustering\\u003c/h2\\u003e \\u003cp\\u003eImmune phenotyping was performed using a two-step, sequential unsupervised clustering strategy, designed to first define broad immune lineages and subsequently resolve finer phenotypic heterogeneity within each lineage [\\u003cspan citationid=\\\"CR26\\\" class=\\\"CitationRef\\\"\\u003e26\\u003c/span\\u003e]. In the first step, all CD66b\\u003csup\\u003elow/neg\\u003c/sup\\u003e cells from all patients and time points were pooled and clustered using the FlowSOM algorithm [\\u003cspan citationid=\\\"CR20\\\" class=\\\"CitationRef\\\"\\u003e20\\u003c/span\\u003e], employing a 10\\u0026times;10 self-organising map (100 clusters). All markers included in the CyTOF panel were used as input. The mean marker expression of each cluster was visualised using heatmaps, which were manually inspected to assign clusters to five major immune lineages based on canonical lineage-defining markers, including CD3, CD4, CD8, CD19, CD20, CD56, CD16, CD14, and CD11c. This resulted in the identification of CD4⁺ T cells, CD8⁺ T cells, B cells, natural killer (NK) cells, and a combined myeloid compartment comprising monocytes and dendritic cells. In the second step, each major lineage was re-clustered independently using FlowSOM with identical settings and using all available markers. This lineage-restricted clustering enabled higher-resolution identification of phenotypically distinct subpopulations while preserving the global structure defined in the first step. Marker expression profiles were again summarised using heatmaps, and rule-based phenotypic annotation was performed using marker-specific thresholds derived from quantiles of the cluster-level expression distributions. This approach combines data-driven clustering with biologically informed, reproducible annotation criteria. Within the CD4⁺ and CD8⁺ T cell compartments, subsets were annotated as na\\u0026iuml;ve (CD45RA\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD45RO\\u003csup\\u003elow\\u003c/sup\\u003e, CD27\\u003csup\\u003ehigh\\u003c/sup\\u003e), central memory (CM; CD45RA\\u003csup\\u003elow\\u003c/sup\\u003e, CD45RO\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD27\\u003csup\\u003ehigh\\u003c/sup\\u003e), effector memory (EM; CD45RA\\u003csup\\u003elow\\u003c/sup\\u003e, CD45RO\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD27\\u003csup\\u003elow\\u003c/sup\\u003e), and TEMRA (CD45RA\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD45RO\\u003csup\\u003elow\\u003c/sup\\u003e, CD27\\u003csup\\u003elow\\u003c/sup\\u003e). Cells not meeting these criteria were grouped as CD4_unclassified or CD8_unclassified, respectively. B cell populations were annotated into six subsets: progenitor B cells (CD34\\u003csup\\u003ehigh\\u003c/sup\\u003e), na\\u0026iuml;ve B cells (Na\\u0026iuml;ve; CD27\\u003csup\\u003elow\\u003c/sup\\u003e, CD38\\u003csup\\u003elow\\u003c/sup\\u003e), memory B cells (Mem; CD27\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD38\\u003csup\\u003elow\\u003c/sup\\u003e), activated B cells (Activ; HLA-DR\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD27\\u003csup\\u003elow\\u003c/sup\\u003e), plasmablasts (pl; CD38\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD20\\u003csup\\u003elow\\u003c/sup\\u003e, CD19\\u003csup\\u003ehigh\\u003c/sup\\u003e), and transitional B cells (CD38\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD27\\u003csup\\u003elow\\u003c/sup\\u003e, excluding plasmablasts based on CD20/CD19 expression). NK cells were phenotyped into NKT cells (CD56\\u003csup\\u003ehigh\\u003c/sup\\u003e and CD3\\u003csup\\u003ehigh\\u003c/sup\\u003e), CD56 bright CD16\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD56 bright CD16\\u003csup\\u003elow\\u003c/sup\\u003e, CD56 dim CD16\\u003csup\\u003ehigh\\u003c/sup\\u003e, and CD56 dim CD16\\u003csup\\u003elow\\u003c/sup\\u003e subsets. The myeloid compartment was further resolved into plasmacytoid dendritic cells (pDCs; CD123\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD11c\\u003csup\\u003elow\\u003c/sup\\u003e), classical dendritic cells (cDCs; CD11c\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD14\\u003csup\\u003elow\\u003c/sup\\u003e), classical monocytes (CD14\\u003csup\\u003ehigh\\u003c/sup\\u003e with CD11c\\u003csup\\u003elow\\u003c/sup\\u003e or CD33\\u003csup\\u003ehigh\\u003c/sup\\u003e), and non-classical monocytes (CD16\\u003csup\\u003ehigh\\u003c/sup\\u003e, CD14\\u003csup\\u003elow\\u003c/sup\\u003e). This hierarchical strategy resulted in the definition of 25 immune phenotypes, which were visualised using heatmaps displaying mean marker expression per subset. For dimensionality reduction and visual inspection, UMAP embeddings were generated using all panel markers, with up to 1,500 cells per patient and time point subsampled when available.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec13\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.5 Immune composition analysis\\u003c/h2\\u003e \\u003cp\\u003eTo formally assess longitudinal changes in peripheral immune cell composition, we applied linear mixed-effects modelling using the lme4 and lmerTest R packages. As the relative abundances of the 25 immune phenotypes constitute compositional data, cell-type frequencies were first transformed using the centred log-ratio (CLR) transformation prior to modelling. To account for the cross-over design of the SMART-MS trial, each immune phenotype was modelled using the following fixed-effects structure:\\u003c/p\\u003e \\u003cp\\u003e \\u003cem\\u003eCLR(cell frequency) \\u0026sim; Treatment\\u0026thinsp;+\\u0026thinsp;Period\\u0026thinsp;+\\u0026thinsp;Sequence\\u0026thinsp;+\\u0026thinsp;Batch\\u0026thinsp;+\\u0026thinsp;Age\\u0026thinsp;+\\u0026thinsp;Sex + (1∣ Patient)\\u003c/em\\u003e \\u003c/p\\u003e \\u003cp\\u003ewhere Treatment denotes MSC versus control exposure, Period corresponds to the first or second treatment period, Sequence reflects treatment order (MSC-first vs. control-first), Batch represents the CyTOF barcode identifier, Age was mean-centred within the cohort, and Sex was included as a covariate. A random intercept for patient was included to account for repeated measurements and inter-individual baseline differences. For each immune phenotype, model coefficients and associated p-values were extracted and adjusted for multiple testing using false discovery rate (FDR) correction across all subsets. Effect estimates and significance values were visualised using forest plots and volcano plots to summarise global compositional trends. Model assumptions were evaluated by inspection of standard diagnostic plots, including quantile\\u0026ndash;quantile (QQ) plots of residuals and fitted-versus-residual scatter plots, to assess normality, homoscedasticity, and potential outliers. Representative diagnostic plots are shown in Supplementary Figs.\\u0026nbsp;13 and 14. As a sensitivity analysis, we additionally fitted an alternative mixed-effects model formulation in which treatment arm and time point (including baseline) were included as fixed effects, adjusted for CyTOF batch, age, and sex, with a random intercept for patient. This complementary approach enabled direct comparison of MSC-exposed samples to untreated baseline observations and provided an independent validation of treatment-associated effects observed in the primary cross-over model.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec14\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.6 Construction of blood-based functional program scores\\u003c/h2\\u003e \\u003cp\\u003eTo move beyond changes in immune cell composition and capture functional modulation within stable cellular compartments, we derived composite functional program scores from the CyTOF data. This approach was motivated by the observation that immune activation, differentiation, and trafficking states are typically reflected by coordinated expression of multiple markers rather than by changes in single proteins or cell frequencies alone. Composite scores therefore provide a robust summary of biologically coherent functional programs and enable direct comparison of functional states across immune phenotypes and longitudinal samples. Functional programs were defined a priori based on established immunological knowledge and the available CyTOF antibody panel. Each program comprised a small set of markers reflecting a shared biological process, including T cell activation, memory/effector differentiation, inflammatory homing (Th17-like), CNS-homing, Treg-like features, B cell activation and memory, myeloid antigen-presenting cell (APC) activation, non-classical monocyte features, and NK-cell activation. Program names refer to marker-derived functional axes and do not imply lineage identity. For example, the inflammatory homing program reflects expression of trafficking markers associated with inflammatory migration and does not denote a specific T-helper lineage.\\u003c/p\\u003e \\u003cp\\u003eThe mapping between individual markers and functional programs is shown in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003eA. For selected programs, markers were included with inverse contribution to capture counter-regulatory relationships (e.g. inverse weighting of CD127 in the Treg-like score). Prior to score calculation, marker expression values were z-score normalized across all cells to place markers on a comparable scale. For each cell, functional program scores were computed as the mean of the z-scored expression values of the markers assigned to that program, with inverse markers multiplied by \\u0026minus;\\u0026thinsp;1 prior to averaging. Scores were then aggregated at the cell-type and patient level by computing the median score within each immune phenotype for each individual and time point, enabling longitudinal analysis while preserving cell-type specificity. Functional scores are expressed on a relative scale centred around zero. Positive scores indicate relative enrichment of a given functional program compared with the global immune cell population, whereas negative scores indicate relative depletion. Scores were interpreted within the context of each immune lineage, reflecting relative functional wiring rather than absolute activation states. This composite scoring framework formed the basis for the longitudinal and integrative analyses presented in subsequent sections.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec15\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.7 Longitudinal modelling of blood-based functional programs\\u003c/h2\\u003e \\u003cp\\u003eLongitudinal changes in blood-derived functional program scores were assessed using linear mixed-effects models implemented in the lme4 and lmerTest R packages. Modelling followed the same general framework as the immune composition analysis to ensure consistency across analytical layers. For each functional program and immune phenotype, the following model was fitted:\\u003c/p\\u003e \\u003cp\\u003e \\u003cem\\u003eScore \\u0026sim; Treatment\\u0026thinsp;+\\u0026thinsp;Period\\u0026thinsp;+\\u0026thinsp;Sequence\\u0026thinsp;+\\u0026thinsp;Batch\\u0026thinsp;+\\u0026thinsp;Age\\u0026thinsp;+\\u0026thinsp;Sex + (1∣ Patient)\\u003c/em\\u003e \\u003c/p\\u003e \\u003cp\\u003ewhere Treatment denotes MSC versus control exposure, Period corresponds to the first or second treatment period of the cross-over design, Sequence reflects treatment order (MSC-first vs. control-first), Batch represents the CyTOF barcode identifier, Age was mean-centred within the cohort, and Sex was included as a covariate. A random intercept for patient was included to account for repeated measurements and inter-individual baseline differences. Model coefficients, standard errors, and p-values were extracted for the treatment effect for each score\\u0026ndash;cell type combination. To control for multiple testing across immune phenotypes and functional programs, p-values were adjusted using FDR correction. Effect estimates and corresponding significance values were visualised in a heatmap, with rows representing immune phenotypes and columns representing functional programs. Model assumptions were assessed by inspection of residual distributions and fitted-versus-residual plots, with no major deviations observed.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec16\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.8 CSF proteomics data acquisition, preprocessing, and normalization\\u003c/h2\\u003e \\u003cp\\u003eCSF samples collected at baseline and at 6 months (prior to treatment crossover) were analysed using tandem mass tag\\u0026ndash;based quantitative proteomics. Sample preparation, liquid chromatography\\u0026ndash;tandem mass spectrometry (LC-MS/MS) acquisition, and primary data processing/cleansing were performed as previously described [\\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e]. Briefly, samples were processed in three TMT-16plex batches, each including pooled reference controls for normalization. Peptides were analysed on an Orbitrap Eclipse high-resolution mass spectrometer, and raw spectra were processed using Proteome Discoverer v2.5 (Thermo Fisher Scientific), followed by downstream preprocessing and quality control in Perseus framework [\\u003cspan citationid=\\\"CR27\\\" class=\\\"CitationRef\\\"\\u003e27\\u003c/span\\u003e]. Protein-level intensity data provided from the proteomics pipeline were imported into R for downstream analyses. The resulting dataset consisted of 1325 quantified proteins that were log-transformed and used for all subsequent analyses in this study. These processed protein intensities were subsequently used for differential abundance modelling, pathway enrichment analysis, and derivation of CSF functional program scores.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec17\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.9 Differential abundance analysis of CSF proteomics\\u003c/h2\\u003e \\u003cp\\u003eDifferential abundance analysis of the CSF proteome was performed using a paired, longitudinal delta-based framework. As complete CSF samples were available for all participants at BL and 6m, protein-level changes were quantified for each individual as:\\u003c/p\\u003e \\u003cp\\u003e \\u003cem\\u003eΔProtein\\u0026thinsp;=\\u0026thinsp;Abundance\\u003c/em\\u003e \\u003csub\\u003e \\u003cem\\u003e6m\\u003c/em\\u003e \\u003c/sub\\u003e\\u0026thinsp;\\u003cem\\u003e\\u0026minus;\\u0026thinsp;Abundance\\u003c/em\\u003e\\u003csub\\u003e\\u003cem\\u003eBL\\u003c/em\\u003e\\u003c/sub\\u003e\\u003c/p\\u003e \\u003cp\\u003eThis approach enabled direct modelling of within-patient proteomic changes over time while controlling for substantial inter-individual variability in baseline protein abundance. By analysing change scores rather than absolute values, treatment-associated effects could be interpreted as differences in longitudinal trajectories attributable to MSC exposure. For each protein, linear regression models were fitted to the delta values using the lm function in R, with Treatment (MSC vs control), Age, and Sex included as covariates. Model fitting and coefficient extraction were performed using the broom R package to ensure consistent handling of estimates, standard errors, and statistical significance. A linear model was chosen in this context, rather than a mixed-effects model, because each individual contributed a single delta value per protein, eliminating repeated measurements within the outcome variable. As such, there was no hierarchical structure requiring random effects, and the paired nature of the design was fully captured by the within-subject delta formulation. Resulting p-values were adjusted for multiple testing across all quantified proteins using the FDR procedure. Differential abundance results were visualised using volcano plots, displaying effect size estimates (MSC-associated change relative to control) against statistical significance. Proteins passing predefined FDR thresholds (FDR\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.05 and FDR\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.1) were prioritised for downstream interpretation. To support targeted inspection of biologically relevant signals, selected protein sets were further visualised using boxplots of delta values stratified by treatment and clinical subgroups. These targeted visualisations enabled patient-level assessment of proteomic changes in relation to treatment exposure, adverse events, and radiological findings. Protein identifiers were mapped to gene symbols and descriptive annotations using the AnnotationDbi and org.Hs.eg.db R packages. Where necessary, UniProt identifiers were converted to HGNC gene symbols to ensure consistent representation across analyses and figures.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec18\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.10 Pathway enrichment analysis of CSF proteomics\\u003c/h2\\u003e \\u003cp\\u003eTo identify coordinated biological processes associated with MSC-induced CSF proteomic changes, pathway enrichment analysis was performed using a ranked gene set enrichment approach. Proteins were ranked according to MSC-associated effect estimates derived from linear modelling of longitudinal CSF protein changes, with positive values indicating relative enrichment during MSC exposure and negative values indicating relative depletion. Gene set enrichment analysis (GSEA) was conducted using the gseGO function from the clusterProfiler R package, focusing on Gene Ontology Biological Process (GO:BP) terms. Analyses were performed with parameters \\u003cem\\u003eminGSSize\\u0026thinsp;=\\u0026thinsp;10, maxGSSize\\u0026thinsp;=\\u0026thinsp;300, and pvalueCutoff\\u0026thinsp;=\\u0026thinsp;0.1\\u003c/em\\u003e, with p-values adjusted for multiple testing using the FDR procedure. GSEA was selected over over-representation analysis to leverage the full ranked proteomic dataset and detect subtle but coordinated pathway-level changes without imposing arbitrary significance thresholds at the protein level. An initial GSEA including all quantified proteins revealed enrichment of pathways related to tissue remodelling, coagulation, and epithelial or barrier-associated processes, consistent with the presence of blood- and skin-derived proteins introduced during lumbar puncture. To mitigate bias from technical contamination, a predefined set of proteins commonly associated with blood or epithelial contamination (HBA1, HBB, KRT1, KRT9, KPRP, DSG1, DCD, PLEC, DSP, AHNAK, JUP, CA3) was excluded from the ranked list, and GSEA was repeated on the filtered dataset. Significant pathways were visualized using dot plots displaying normalized enrichment scores (NES), FDR-adjusted p-values, and leading-edge gene ratios, reflecting the proportion of pathway members driving enrichment. Selected pathways were further visualized using gseaplot2 (enrichplot R package) to illustrate enrichment profiles and leading-edge contributions along the ranked protein list. For interpretability, enriched GO terms were grouped post hoc into broader functional categories, including vascular and innate inflammatory processes, cytoskeletal organization, and metabolic pathways; this grouping was applied solely for presentation and did not affect statistical inference.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec19\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.11 Integration of CSF proteomics with blood-based functional programs\\u003c/h2\\u003e \\u003cp\\u003eComposite CSF functional program scores were constructed using a framework conceptually analogous to that applied to the CyTOF-derived blood functional programs. This parallel design enables direct comparison of coordinated biological states inferred from bulk CSF proteomics with cell-type\\u0026ndash;resolved functional programs measured in peripheral blood, despite differences in measurement modality and biological scale. Protein sets were defined a priori based on established biological knowledge and informed by differential abundance and pathway enrichment analyses, with the aim of capturing mechanistically relevant dimensions of CSF biology. Three primary CSF functional programs were defined:\\u003c/p\\u003e \\u003cp\\u003e \\u003cul\\u003e \\u003cli\\u003e \\u003cp\\u003e \\u003cb\\u003eCSF Extracellular matrix / Interface Remodelling\\u003c/b\\u003e, reflecting extracellular matrix organization and CNS interface biology, including fibrillar collagens (COL1A1, COL3A1, COL5A1, COL11A1, COL6A6), extracellular matrix regulators (POSTN, SPARC, COMP), and matrix scaffolds and adhesion molecules (FN1, TNC, VCAN, LAMA2, LAMC1).\\u003c/p\\u003e \\u003c/li\\u003e \\u003cli\\u003e \\u003cp\\u003e \\u003cb\\u003eCSF Innate / Vascular Stress\\u003c/b\\u003e, capturing innate immune and vascular inflammatory responses, comprising complement components (C3, C4A, C4B, C5, CFH, CFB), acute-phase and innate inflammatory proteins (CRP, SAA1, SAA2, HP, ORM1, ORM2, SERPINA1, SERPINA3, A2M, CP), and coagulation-related factors (FGA, FGB, FGG).\\u003c/p\\u003e \\u003c/li\\u003e \\u003cli\\u003e \\u003cp\\u003e \\u003cb\\u003eCSF Metabolic / Cytoskeletal Adaptation\\u003c/b\\u003e, reflecting intracellular structural and metabolic responses, including actin dynamics (CFL1, PFN1, ACTB, ACTG1), microtubule organization and transport (TUBA1A, TUBB2A, TUBB), stress and translational regulation (HSPB1, EEF2), and structural integration (VIM).\\u003c/p\\u003e \\u003c/li\\u003e \\u003c/ul\\u003e \\u003c/p\\u003e \\u003cp\\u003eTo account for nonspecific signals potentially arising from tissue injury or blood contamination, rather than MSC-specific biological effects, an additional composite CSF reactivity score was constructed. This score was based on proteins commonly associated with sampling-related artifacts or cellular stress, including hemoglobin subunits (HBA1, HBB), epithelial and junctional proteins (KRT1, KPRP, DSG1, DSP, JUP, AHNAK, DCD), and cytoskeletal or metabolic stress markers (VIM, CA3). The procedural stress score was used as a negative-control axis to aid interpretation of CSF proteomic changes and to distinguish treatment-associated remodelling from nonspecific procedural effects. For each CSF functional program, protein-level longitudinal changes (6-month minus baseline delta values) were z-score normalized across patients and aggregated by computing the mean score per program for each individual. Scores therefore reflect relative enrichment or depletion of coordinated biological programs rather than absolute protein abundance. To evaluate the contribution of nonspecific effects versus MSC exposure, linear models were fitted with each CSF program score as the outcome and CSF reactivity score and treatment (MSC vs. control) as covariates. Finally, CSF functional program scores were integrated with blood-based CyTOF functional programs by matching patient-level longitudinal changes across compartments. Associations were assessed using correlation and linear regression analyses, with a primary focus on mechanistically compatible programs, particularly CSF interface remodelling and circulating T cell trafficking programs (e.g., Th17-like and CNS-homing). This integrative approach enabled evaluation of whether MSC-associated CSF remodelling aligned with specific adaptive immune trafficking states rather than reflecting global immune activation.\\u003c/p\\u003e \\u003c/div\\u003e\"},{\"header\":\"4. CONCLUSIONS\",\"content\":\"\\u003cp\\u003eIn this study, we implemented a multi-layer omics integration framework to investigate the biological effects of intrathecal MSC therapy in progressive MS. By combining longitudinal mass cytometry profiling of peripheral blood with matched CSF proteomics, we characterized immune composition, functional immune programs, and protein-level remodelling across systemic and intrathecal compartments within the SMART-MS trial [\\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e]. This integrative design enabled a systems-level view of MSC-associated changes and revealed mechanistic patterns that would not be apparent from single-modality analyses alone.\\u003c/p\\u003e \\u003cp\\u003eA central observation of our study is that intrathecal MSC therapy did not broadly reshape peripheral immune composition. These findings are consistent with emerging models of progressive MS in which inflammatory processes become compartmentalized within CNS border regions and are therefore not reflected in peripheral immune composition [\\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e3\\u003c/span\\u003e]. Instead, the most consistent signal emerged within the CSF proteome, where MSC exposure was associated with coordinated remodelling of extracellular matrix and CNS interface programs. These findings support a model in which MSC therapy primarily acts within compartmentalized inflammatory niches at the immune-CNS interface. While peripheral functional programs showed stability at the compositional level, our blood and CSF remodelling results suggest that MSC therapy may modulate how immune cells interact with CNS tissue rather than altering global immune architecture [\\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e21\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR22\\\" class=\\\"CitationRef\\\"\\u003e22\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eThe biological interpretation of the CSF proteomic signatures points toward structural and trophic reprogramming of the CNS interface. Proteins related to extracellular matrix organization, cell\\u0026ndash;matrix interaction, and guidance signalling increased during MSC exposure, consistent with enhanced remodelling of the extracellular environment. Such changes may influence barrier properties, immune cell migration, or local trophic signalling rather than reflecting classical inflammatory activation. Importantly, pathway enrichment analyses indicated attenuation of acute-phase and innate inflammatory programs in many individuals.\\u003c/p\\u003e \\u003cp\\u003eA distinctive aspect of our analysis was the integration of clinical context, including adverse events and spinal reactive MRI findings. While extracellular matrix/interface remodelling was consistently observed across MSC-treated individuals, several inflammatory and metabolic proteomic changes were preferentially enriched in patients who developed clinical or radiological reactivity. This pattern suggests that a subset of CSF proteomic signals, particularly metabolic/cytoskeletal and vascular stress responses, may reflect secondary biological reactivity to intrathecal MSC exposure rather than core mechanisms of MSC-mediated interface remodelling. Rather than representing distinct protein signatures, clinically reactive patients showed amplified changes within shared extracellular matrix, inflammatory, and metabolic pathways, suggesting increased biological responsiveness rather than fundamentally different mechanisms. These observations highlight the importance of interpreting MSC-associated proteomic changes within clinical context and caution against attributing all CSF alterations solely to direct therapeutic mechanisms.\\u003c/p\\u003e \\u003cp\\u003eTo further disentangle treatment effects from nonspecific tissue responses, we modelled a CSF reactivity score derived from proteins associated with tissue stress, blood leakage, epithelial carryover, and cytoskeletal release. The reactivity score correlated with metabolic/cytoskeletal changes but showed no clear association with the interface remodelling signal. By contrast, interface scores shifted consistently during MSC exposure, indicating that structural remodelling may represent a partially distinct biological response. Together, these findings are not readily explained by procedural effects alone for the observed CSF remodelling and instead support a model in which structural interface changes represent a primary MSC-associated effect, whereas metabolic and cytoskeletal responses reflect context-dependent biological reactivity influenced by post-treatment tissue stress and individual sensitivity thresholds.\\u003c/p\\u003e \\u003cp\\u003eIntegration of CSF and blood-derived functional programs provided additional context for MSC-associated immune dynamics across compartments. Exploratory analyses suggested that changes in circulating inflammatory trafficking programs (e.g., Th17-like) occasionally paralleled shifts in CSF interface remodelling within memory T cell subsets; however, these relationships were heterogeneous and did not demonstrate consistent or statistically robust linear coupling at the individual level. Instead, the dominant pattern was a treatment-associated increase in CSF extracellular matrix/interface remodelling largely independent of specific peripheral immune programs. These findings support a model in which intrathecal MSC therapy primarily induces local remodelling of the immune\\u0026ndash;CNS interface, while a subset of inflammatory and metabolic CSF changes appear to track more closely with adverse event\\u0026ndash;associated biology than with consistent MSC-driven immune modulation. While adaptive trafficking signatures may reflect broader immune context, our data suggest that MSC-associated CSF remodelling represents a compartmental effect linked to immune\\u0026ndash;CNS communication rather than generalized immunosuppression [\\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e21\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR23\\\" class=\\\"CitationRef\\\"\\u003e23\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eFrom a clinical perspective, program-level CSF readouts may provide a framework for distinguishing adaptive biological remodelling from excessive inflammatory reactivity following intrathecal therapies. Although exploratory, our findings raise the possibility that extracellular matrix/interface signatures represent desired treatment responses, while elevated metabolic and vascular stress signals may identify patients with heightened sensitivity to intrathecal interventions. These observations further suggest that certain CSF proteomic signatures may serve as indicators of biological reactivity or treatment-associated adverse responses rather than biomarkers of therapeutic efficacy alone. Future clinical trials incorporating longitudinal multi-omics profiling may therefore benefit from integrating functional program scores as biomarkers of therapeutic engagement or over-reactivity.\\u003c/p\\u003e \\u003cp\\u003eSimilar to other omics studies, several limitations should be considered. The SMART-MS cohort was relatively small (n\\u0026thinsp;=\\u0026thinsp;18), limiting statistical power for subgroup analyses and precluding formal inference regarding adverse events or MRI-defined reactivity. Longitudinal follow-up was restricted to 12 months and may therefore not capture longer-term remodelling or neurodegenerative trajectories. Although bulk CSF proteomics provides valuable insight into extracellular and secreted protein dynamics, it lacks cellular resolution. In parallel, the CyTOF analysis was constrained by a predefined antibody panel focused on \\u0026ldquo;canonical\\u0026rdquo; immune populations. Consequently, functional scores were derived from the available markers and may not fully capture alternative activation states or rare immune subsets. The composite functional programs should therefore be interpreted as structured summaries of measurable signalling axes rather than exhaustive representations of immune function and CSF processes. Future studies incorporating broader single-cell technologies, such as scRNA-seq, CITE-seq, or spatial profiling, may further refine the cellular origin and context of the observed signatures [\\u003cspan citationid=\\\"CR17\\\" class=\\\"CitationRef\\\"\\u003e17\\u003c/span\\u003e]. Replication in independent cohorts will be essential, although assembling comparable intrathecal MSC datasets in progressive MS remains challenging.\\u003c/p\\u003e \\u003cp\\u003eDespite these limitations, our study provides several advances. To our knowledge, this represents the first integrated analysis combining high-dimensional single-cell immune profiling with CSF proteomics in MSC-treated progressive MS patients. The program-level analytical framework enabled cross-compartment comparisons and highlighted dissociable biological axes of interface remodelling, inflammatory stress, and metabolic adaptation. Moreover, explicit modelling of CSF reactivity provides a practical strategy to address a long-standing challenge in CSF proteomics studies of intrathecal therapies.\\u003c/p\\u003e \\u003cp\\u003eIn summary, our results support a model in which intrathecal MSC therapy primarily reshapes the immune\\u0026ndash;CNS interface and extracellular environment within the intrathecal compartment, with peripheral immune programs remaining largely stable. In most patients, this remodelling occurs alongside reduced innate inflammatory signalling, whereas a subset of individuals exhibits amplified inflammatory and metabolic responses associated with clinical or radiological reactivity. By integrating multi-omics data across compartments, this work provides a systems-level framework for understanding MSC therapy in progressive MS and establishes a generalizable analytical framework for future translational studies.\\u003c/p\\u003e\"},{\"header\":\"Abbreviations\",\"content\":\"\\u003cp\\u003emultiple sclerosis \\u003cstrong\\u003e(MS)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003emesenchymal stem cell \\u003cstrong\\u003e(MSC)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003ecerebrospinal fluid \\u003cstrong\\u003e(CSF)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003emagnetic resonance imaging \\u003cstrong\\u003e(MRI)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003ecentral nervous system \\u003cstrong\\u003e(CNS)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003erelapsing\\u0026ndash;remitting MS \\u003cstrong\\u003e(RRMS)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003ecytometry by time-of-flight \\u003cstrong\\u003e(CyTOF)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003etandem mass spectrometry \\u003cstrong\\u003e(LC-MS/MS)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003euniform manifold approximation and projection \\u003cstrong\\u003e(UMAP)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003ecentral memory \\u003cstrong\\u003e(CM)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eeffector memory \\u003cstrong\\u003e(EM)\\u0026nbsp;\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eeffector memory cells re-expressing CD45RA \\u003cstrong\\u003e(TEMRA)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003ecentred log-ratio (CLR)\\u003c/p\\u003e\\n\\u003cp\\u003eantigen-presenting cell \\u003cstrong\\u003e(APC)\\u0026nbsp;\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003einflammatory homing-like \\u003cstrong\\u003e(Th17-like)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eprincipal Component analysis \\u003cstrong\\u003e(PCA)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eT regulatory-like \\u003cstrong\\u003e(Treg)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003edendritic cells \\u003cstrong\\u003e(DCs)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003enatural killer \\u003cstrong\\u003e(NK)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003emultidimensional scaling \\u003cstrong\\u003e(MDS)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003efalse discovery rate \\u003cstrong\\u003e(FDR)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003equantile\\u0026ndash;quantile \\u003cstrong\\u003e(QQ)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003ecartilage oligomeric matrix protein \\u003cstrong\\u003e(COMP)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003ecarbonic anhydrase 3 \\u003cstrong\\u003e(CA3)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003egene ontology \\u003cstrong\\u003e(GO)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eadverse events \\u003cstrong\\u003e(AE; pain/fever)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eMSC with no AE and no MRI findings \\u003cstrong\\u003e(MSC no AE no MRI)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eMSC with AE and no MRI findings \\u003cstrong\\u003e(MSC + AE no MRI)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eMSC with AE and MRI findings \\u003cstrong\\u003e(MSC + AE + MRI)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eextracellular matrix/interface remodelling \\u003cstrong\\u003e(ECM/interface)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eexpanded disability status scale \\u003cstrong\\u003e(EDSS)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003egood manufacturing practice \\u003cstrong\\u003e(GMP)\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003egene set enrichment analysis \\u003cstrong\\u003e(GSEA)\\u003c/strong\\u003e\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e\\u003cstrong\\u003e6.1 Ethics approval and consent to participate\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe SMART-MS study received approval from the Regional Committee for Medical and Health Research Ethics (reference no. 159326) and the Norwegian Medicines Agency. The trial was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice principles. Written informed consent was obtained from all participants before inclusion in the study. Trial conduct was overseen by external monitors from Western Norway Health Trust Research \\u0026amp; Development, and safety was reviewed regularly by an independent Data Monitoring Committee. The study was prospectively registered at ClinicalTrials.gov (NCT04749667) and in the European Union Clinical Trials Register (EudraCT no. 2020-002373-95).\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eConsent for publication\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe manuscript does not contain any individual person\\u0026rsquo;s identifiable data.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e6.3 Availability of datasets and codes\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eAll analysis code required to reproduce the CyTOF and CSF proteomics analyses is publicly available at https://github.com/dkleftogi/ms-immune-signatures, including pipelines for data preprocessing, batch harmonization, immune phenotyping, functional program scoring, and statistical modelling. Processed and pseudonymised CyTOF phenotypic data and CSF proteomics datasets are deposited in the associated Zenodo repository, as referenced in the GitHub documentation. Shared datasets contain no directly identifiable information and include only study variables necessary to reproduce the analyses. Raw data and additional metadata can be made available upon reasonable request, subject to institutional ethical approvals, data protection regulations, and governance policies of the SMART-MS study.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eCompeting interests \\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eT.H. has received speaker honoraria and/or participated in clinical multiple sclerosis trials sponsored by Biogen, Sanofi, Merck, Roche, Amgen, and Novartis. \\u0026Oslash;.T. has participated in advisory boards and received speaker honoraria from Biogen, Merck, Novartis, Teva, Roche, Sanofi, and Bristol Myers Squibb, and has participated in clinical trials sponsored by Merck, Novartis, Roche, and Sanofi. All other authors declare no competing interests.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eFunding \\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThis study was funded by KLINBEFORSK (The Norwegian Clinical Research Program), Helse Vest, the Norwegian Red Cross, and the Norwegian MS Society. The funders had no role in the design, conduct, analysis, or reporting of the trial.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAuthors contributions\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eD.K. conceived the study, developed the computational framework, performed the bioinformatics and statistical analyses, and drafted the manuscript. S.G. was responsible for CyTOF-based immune profiling and contributed to data interpretation and manuscript preparation.\\u003c/p\\u003e\\n\\u003cp\\u003eL.E.E. and J.B.H. contributed to data analysis and critically reviewed the manuscript. T.H., L.S., and K.W. were responsible for patient recruitment and clinical data acquisition and critically reviewed the manuscript. M.R. and H.S. were responsible for mesenchymal stem cell production and quality control and critically reviewed the manuscript. H.B., L.E.E. and F.S.B. contributed to the CSF proteomics data analysis and critically reviewed the manuscript.\\u003c/p\\u003e\\n\\u003cp\\u003eN.A.-S., S.M.-A., M.Y., C.E.S., \\u0026Oslash;.T., K.M., and L.B. contributed to study design and critically reviewed the manuscript. T.K. and C.E.K. contributed to study conception, led the clinical trial and sample collection, contributed to data interpretation, and critically reviewed the manuscript.All authors read and approved the final manuscript.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAcknowledgements\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe authors gratefully acknowledge the Flow Cytometry Core Facility at the University of Bergen for technical support in CyTOF data acquisition. We sincerely thank all patients who participated in the SMART-MS trial, and their families, for their invaluable contribution to this research. We also thank the laboratory and clinical personnel involved in sample handling, processing, and trial coordination for their dedicated support throughout the study.\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\u003cli\\u003e\\u003cspan\\u003eWallin MT, Culpepper WJ, Nichols E, Bhutta ZA, Gebrehiwot TT, Hay SI, et al. Global, regional, and national burden of multiple sclerosis 1990\\u0026ndash;2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Neurol Elsevier. 2019;18:269\\u0026ndash;85. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/S1474-4422(18)30443-5\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/S1474-4422(18)30443-5\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eLin M, Zhang J, Zhang Y, Luo J, Shi S. Ocrelizumab for multiple sclerosis. Cochrane Database Syst Rev. 2022;5:CD013247. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1002/14651858.CD013247.pub2\\u003c/span\\u003e\\u003cspan address=\\\"10.1002/14651858.CD013247.pub2\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eLassmann H, van Horssen J, Mahad D. Progressive multiple sclerosis: pathology and pathogenesis. Nat Rev Neurol Nat Publishing Group. 2012;8:647\\u0026ndash;56. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/nrneurol.2012.168\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/nrneurol.2012.168\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eGiovannoni G, Popescu V, Wuerfel J, Hellwig K, Iacobaeus E, Jensen MB, et al. Smouldering multiple sclerosis: the real MS. Ther Adv Neurol Disord. 2022;15:17562864211066751. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1177/17562864211066751\\u003c/span\\u003e\\u003cspan address=\\\"10.1177/17562864211066751\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eGavasso S, Kr\\u0026aring;kenes T, Olsen H, Evjenth EC, Ytterdal M, Haugs\\u0026oslash;en JB, et al. The Therapeutic Mechanisms of Mesenchymal Stem Cells in MS\\u0026mdash;A Review Focusing on Neuroprotective Properties. Int J Mol Sci. 2024;25:1365. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.3390/ijms25031365\\u003c/span\\u003e\\u003cspan address=\\\"10.3390/ijms25031365\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSmith JA, Nicaise AM, Ionescu R-B, Hamel R, Peruzzotti-Jametti L, Pluchino S. Stem Cell Therapies for Progressive Multiple Sclerosis. Front Cell Dev Biol [Internet] Front. 2021. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.3389/fcell.2021.696434\\u003c/span\\u003e\\u003cspan address=\\\"10.3389/fcell.2021.696434\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e. [cited 2026 Feb 10];9.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eKvistad CE, Kr\\u0026aring;kenes T, Holm\\u0026oslash;y T, Eidem LE, Steffensen L, Wesnes K, et al. Intrathecal Mesenchymal Stem Cells in Progressive Multiple Sclerosis. Neurol Wolters Kluwer. 2026;106:e214915. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1212/WNL.0000000000214915\\u003c/span\\u003e\\u003cspan address=\\\"10.1212/WNL.0000000000214915\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eConnick P, Kolappan M, Crawley C, Webber DJ, Patani R, Michell AW, et al. Autologous mesenchymal stem cells for the treatment of secondary progressive multiple sclerosis: an open-label phase 2a proof-of-concept study. Lancet Neurol. 2012;11:150\\u0026ndash;6. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/S1474-4422(11)70305-2\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/S1474-4422(11)70305-2\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBonab MM, Sahraian MA, Aghsaie A, Karvigh SA, Hosseinian SM, Nikbin B, et al. Autologous mesenchymal stem cell therapy in progressive multiple sclerosis: an open label study. Curr Stem Cell Res Ther. 2012;7:407\\u0026ndash;14. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.2174/157488812804484648\\u003c/span\\u003e\\u003cspan address=\\\"10.2174/157488812804484648\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003ePetrou P, Kassis I, Ginzberg A, Halimi M, Yaghmour N, Abramsky O, et al. Long-Term Clinical and Immunological Effects of Repeated Mesenchymal Stem Cell Injections in Patients With Progressive Forms of Multiple Sclerosis. Front Neurol [Internet] Front. 2021. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.3389/fneur.2021.639315\\u003c/span\\u003e\\u003cspan address=\\\"10.3389/fneur.2021.639315\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e. [cited 2026 Feb 10];12.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eHarris VK, Yan QJ, Vyshkina T, Sahabi S, Liu X, Sadiq SA. Clinical and pathological effects of intrathecal injection of mesenchymal stem cell-derived neural progenitors in an experimental model of multiple sclerosis. J Neurol Sci. 2012;313:167\\u0026ndash;77. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.jns.2011.08.036\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.jns.2011.08.036\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eHarris VK, Stark J, Williams A, Roche M, Malin M, Kumar A, et al. Efficacy of intrathecal mesenchymal stem cell-neural progenitor therapy in progressive MS: results from a phase II, randomized, placebo-controlled clinical trial. Stem Cell Res Ther. 2024;15:151. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/s13287-024-03765-6\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/s13287-024-03765-6\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eGenchi A, Brambilla E, Sangalli F, Radaelli M, Bacigaluppi M, Furlan R, et al. Neural stem cell transplantation in patients with progressive multiple sclerosis: an open-label, phase 1 study. Nat Med. 2023;29:75\\u0026ndash;85. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41591-022-02097-3\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41591-022-02097-3\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBarmada A, Sharan J, Band N, Prodromos C. Serious Adverse Events Have Not Been Reported with Spinal Intrathecal Injection of Mesenchymal Stem Cells: A Systematic Review. Curr Stem Cell Res Ther. 2023;18:829\\u0026ndash;33. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.2174/1574888X17666220817125324\\u003c/span\\u003e\\u003cspan address=\\\"10.2174/1574888X17666220817125324\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eB\\u0026ouml;ttcher C, Fern\\u0026aacute;ndez-Zapata C, Schlickeiser S, Kunkel D, Schulz AR, Mei HE, et al. Multi-parameter immune profiling of peripheral blood mononuclear cells by multiplexed single-cell mass cytometry in patients with early multiple sclerosis. Sci Rep Nat Publishing Group. 2019;9:19471. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41598-019-55852-x\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41598-019-55852-x\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eZhang T, Warden AR, Li Y, Ding X. Progress and applications of mass cytometry in sketching immune landscapes. Clin Translational Med. 2020;10:e206. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1002/ctm2.206\\u003c/span\\u003e\\u003cspan address=\\\"10.1002/ctm2.206\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSingle-cell analysis. of cerebrospinal fluid reveals common features of neuroinflammation: Cell Reports Medicine [Internet]. [cited 2026 Feb 11]. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://www.cell.com/cell-reports-medicine/fulltext/S2666-3791(24)00463-4\\u003c/span\\u003e\\u003cspan address=\\\"https://www.cell.com/cell-reports-medicine/fulltext/S2666-3791(24)00463-4\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e. Accessed 11 Feb 2026.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eTargeted proteomics of cerebrospinal fluid. in treatment na\\u0026iuml;ve multiple sclerosis patients identifies immune biomarkers of clinical phenotypes | Scientific Reports [Internet]. [cited 2026 Feb 10]. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://www.nature.com/articles/s41598-024-67769-1\\u003c/span\\u003e\\u003cspan address=\\\"https://www.nature.com/articles/s41598-024-67769-1\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e. Accessed 10 Feb 2026.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eJagid N, Kizilbash C, Harris V, Sadiq S. CSF Biomarkers of Intrathecal Cell Therapy in Progressive Multiple Sclerosis Reflect Changes in Microglia and Astrocyte Polarization (P8-6.009). Neurol Wolters Kluwer. 2024;102:5500. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1212/WNL.0000000000205860\\u003c/span\\u003e\\u003cspan address=\\\"10.1212/WNL.0000000000205860\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eVan Gassen S, Callebaut B, Van Helden MJ, Lambrecht BN, Demeester P, Dhaene T, et al. FlowSOM: Using self-organizing maps for visualization and interpretation of cytometry data. Cytometry A. 2015;87:636\\u0026ndash;45. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1002/cyto.a.22625\\u003c/span\\u003e\\u003cspan address=\\\"10.1002/cyto.a.22625\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eMachado-Santos J, Saji E, Tr\\u0026ouml;scher AR, Paunovic M, Liblau R, Gabriely G, et al. The compartmentalized inflammatory response in the multiple sclerosis brain is composed of tissue-resident CD8\\u0026thinsp;+\\u0026thinsp;T lymphocytes and B cells. Brain. 2018;141:2066\\u0026ndash;82. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1093/brain/awy151\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/brain/awy151\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eApolloni S, Tortoriello S, Milani M, Rossi S. Extracellular Matrix Remodelling in Motor Neuron Diseases. Int J Mol Sci. 2025;26:11376. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.3390/ijms262311376\\u003c/span\\u003e\\u003cspan address=\\\"10.3390/ijms262311376\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eHu H, Li H, Li R, Liu P, Liu H. Re-establishing immune tolerance in multiple sclerosis: focusing on novel mechanisms of mesenchymal stem cell regulation of Th17/Treg balance. J Transl Med. 2024;22:663. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1186/s12967-024-05450-x\\u003c/span\\u003e\\u003cspan address=\\\"10.1186/s12967-024-05450-x\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eGavasso S, Haugs\\u0026oslash;en JB, Kleftogiannis D, Gerking Y, Anandan S, Herdlev\\u0026aelig;r I et al. High-dimensional Immune Profiling Following Autologous Hematopoietic Stem Cell Transplantation in Relapsing-Remitting Multiple Sclerosis [Internet]. bioRxiv; 2025 [cited 2026 Feb 26]. p. 2025.07.01.662494. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1101/2025.07.01.662494\\u003c/span\\u003e\\u003cspan address=\\\"10.1101/2025.07.01.662494\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003ePedersen CB, Dam SH, Barnkob MB, Leipold MD, Purroy N, Rassenti LZ, et al. cyCombine allows for robust integration of single-cell cytometry datasets within and across technologies. Nat Commun Nat Publishing Group. 2022;13:1698. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/s41467-022-29383-5\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/s41467-022-29383-5\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eGlobal. and single-cell proteomics view of the co-evolution between neural progenitors and breast cancer cells in a co-culture model - eBioMedicine [Internet]. [cited 2026 Feb 11]. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://www.thelancet.com/journals/ebiom/article/PIIS2352-3964(24)00361-X/fulltext\\u003c/span\\u003e\\u003cspan address=\\\"https://www.thelancet.com/journals/ebiom/article/PIIS2352-3964(24)00361-X/fulltext\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e. Accessed 11 Feb 2026.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eTyanova S, Temu T, Sinitcyn P, Carlson A, Hein MY, Geiger T, et al. The Perseus computational platform for comprehensive analysis of (prote)omics data. Nat Methods Nat Publishing Group. 2016;13:731\\u0026ndash;40. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1038/nmeth.3901\\u003c/span\\u003e\\u003cspan address=\\\"10.1038/nmeth.3901\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":true,\"hideJournal\":true,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true},\"keywords\":\"Multiple sclerosis, mesenchymal stem cell therapy, mass cytometry, immune profiling, CSF proteomics, bioinformatics, single cell analysis\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-9629755/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-9629755/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eTherapeutic options for progressive multiple sclerosis (MS) remain limited, and the biological mechanisms engaged by intrathecal mesenchymal stem cell (MSC) therapy are incompletely understood. MSCs are proposed to exert immunomodulatory and trophic effects, yet most studies rely on targeted biomarkers and lack systems-level analysis across immune compartments. Here, we applied an integrated multi-omics framework to characterize immune and cerebrospinal fluid (CSF) responses to MSC therapy in patients with progressive MS enrolled in the SMART-MS trial, a randomized, placebo-controlled crossover study of a single intrathecal MSC injection. Although the clinical trial did not demonstrate a clear neuro-regenerative signal as the primary endpoint, exploratory MRI findings and adverse events (e.g., fever, back pain) suggested localized biological responses following intrathecal administration.\\u003c/p\\u003e\\n\\u003cp\\u003eLongitudinal peripheral blood mass cytometry and matched CSF proteomics were analysed from 18 participants sampled at baseline, 6 months, and 12 months. Immune phenotypes and composite functional program scores were quantified using mixed-effects modelling. To synthesize CSF proteomic changes into biologically interpretable patterns, we summarized protein-level responses into composite CSF functional programs reflecting extracellular matrix and CNS interface remodelling, innate and vascular inflammatory stress, metabolic and cytoskeletal adaptation, and biological reactivity. These program-level scores enabled structured cross-compartment integration with circulating immune programs.\\u003c/p\\u003e\\n\\u003cp\\u003eMSC exposure did not broadly alter peripheral immune composition or functional programs across circulating lymphocyte and monocyte populations. Instead, the dominant signal emerged within the CSF proteome, where treatment was associated with coordinated extracellular matrix and CNS interface remodelling alongside attenuation of acute-phase inflammatory pathways in many individuals. Elevated inflammatory and metabolic signatures were largely confined to patients with clinical adverse events or spinal MRI reactivity, consistent with amplified biological responsiveness rather than distinct MSC-specific mechanisms. Cross-compartment analyses revealed weak and heterogeneous coupling between circulating immune programs and CSF remodelling, supporting predominantly compartmentalized intrathecal effects.\\u003c/p\\u003e\\n\\u003cp\\u003eTogether, these findings suggest that intrathecal MSC therapy in progressive MS is associated with selective remodelling at the immune–CNS interface rather than broad systemic immunosuppression and demonstrate the value of integrated multi-omics approaches for dissecting treatment-associated biology in neuroinflammatory disease.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Intrathecal mesenchymal stem cell therapy in progressive multiple sclerosis: cross-compartment immune profiling in the SMART-MS randomized trial\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2026-05-12 14:38:11\",\"doi\":\"10.21203/rs.3.rs-9629755/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"7f74eb3c-2f13-4c5e-92fd-16dcb57f5310\",\"owner\":[],\"postedDate\":\"May 12th, 2026\",\"published\":true,\"recentEditorialEvents\":[{\"type\":\"decision\",\"content\":\"Rejected\",\"date\":\"2026-05-11T12:00:22+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorAssigned\",\"content\":\"\",\"date\":\"2026-05-08T15:10:05+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"checksComplete\",\"content\":\"\",\"date\":\"2026-05-08T03:50:21+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"submitted\",\"content\":\"Journal of Neuroinflammation\",\"date\":\"2026-05-06T11:20:23+00:00\",\"index\":\"\",\"fulltext\":\"\"}],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"posted\",\"subjectAreas\":[],\"tags\":[],\"updatedAt\":\"2026-05-12T14:42:39+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2026-05-12 14:38:11\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-9629755\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-9629755\",\"identity\":\"rs-9629755\",\"version\":[\"v1\"]},\"buildId\":\"XKTyCvWXoU3ODBz1xrDgd\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}