This study was based on two patient cohorts: a main study cohort and a validation cohort ( Figure 1 ). In the main study, MS patients in relapse were recruited at 4 study centers in Germany (Berlin, Greifswald, Jena and Rostock). The validation cohort was enrolled at 3 study centers in Germany (Bochum, Hannover and Regensburg) and comprised patients with MS or NMOSD who also had a relapse. The MS patients had been diagnosed with either CIS/RRMS or SPMS with superimposed relapses according to the revised McDonald criteria (9, 49). The patients with NMOSD were diagnosed according to the criteria of the International Panel for NMO Diagnosis (50). The NMOSD patients were included in this study alongside the MS patients to investigate whether relapse therapy induces similar effects in both diseases. Only patients over the age of 18 were included in the prospective studies, and patients with substantial cognitive deficits that would hamper their study participation were excluded. Sociodemographic, clinical and medication data of the patients (e.g., age, sex, disease duration and relapse presentation) were gathered through questionnaires to be completed by the treating physicians. However, different sets of parameters were recorded for the two cohorts.

Written informed consent was obtained from all patients in advance. The studies were performed in compliance with the Good Clinical Practice guidelines, the Declaration of Helsinki and the European General Data Protection Regulation. Ethical approvals were obtained from the ethics committees of the University of Rostock (approval numbers A 2019-0047 and A 2020-0048) and the Ruhr University Bochum (approval number AZ 15-5395).

The patients were treated and monitored following the guidelines and recommendations of the German Society of Neurology (32). Routine medical care was therefore provided to all patients. The patients either received GCs or underwent apheresis to support relapse recovery. High-dose methylprednisolone was administered as initiation treatment with up to 1000 mg daily for up to 5 days (main study cohort). Therapeutic apheresis (i.e., plasma exchange or immunoadsorption) was performed on several consecutive days, typically as escalation treatment for steroid-refractory relapses after one or two previous GC treatment courses (both cohorts).

To classify clinical responders and non-responders to relapse therapy, the patients were examined at baseline (i.e., prior to initiation or escalation treatment) and at a later date, usually within 4 months. The degree of neurological disability was rated using the Expanded Disability Status Scale (EDSS) (51). In the main study, the treating physicians were asked to judge whether or not an overall symptom improvement could be observed. In the validation cohort, the clinical outcome of therapeutic apheresis was evaluated slightly differently: In patients with optic neuritis, the neurological examination included a high-contrast visual acuity testing, and treatment response was defined as an improvement in the EDSS score of ≥1.0 or in visual acuity of ≥20%.

In the main study, up to ~30 mL peripheral blood was collected immediately before and after relapse therapy in 8 mL BD Vacutainer Cell Preparation Tubes containing the anticoagulant sodium citrate and a blood separation medium consisting of a polyester gel and a Ficoll-Hypaque solution. The blood collection tubes were centrifuged at the respective study center and then sent to the laboratory in Rostock within 24 hours. There, plasma and peripheral blood mononuclear cells (PBMC) were separated according to the manufacturer’s instructions. The plasma samples were stored at -80°C until further use.

The blood samples of the validation cohort were taken in a similar manner before and after therapeutic apheresis. However, different blood collection tubes were used. Serum tubes were centrifuged shortly after blood withdrawal. The sera were then transferred into separate tubes and stored at -80°C. Tubes containing ethylenediaminetetraacetic acid (EDTA) were used to extract PBMC by density gradient centrifugation using Lymphoprep (Carl Roth). The PBMC samples were cryopreserved in CTL-Cryo ABC medium (Immunospot) in Bochum and shipped in liquid nitrogen to Rostock for the analyses of the present study.

The concentrations of the Ig isotypes IgG, IgM and IgA were determined in the plasma samples using a Beckman Coulter AU480 analyzer. Antibody reactivities against selected linear epitopes of proteins from human and EBV were measured in plasma and serum by indirect peptide-based enzyme-linked immunosorbent assays (ELISAs).

The peptide sequences of interest (n=12) were previously described as (auto)antigenic in the literature (12-15, 22, 52, 53) and are listed in Supplementary Table 1 . They were synthesized by the company Miltenyi Biotec with N-terminal acetylation and C-terminal amidation as well as a lysine linker near the N-terminus for the conjugation of biotin to the peptide sequence.

For the ELISAs with plasma samples, 1 µg/mL streptavidin (Merck) was coated on 96-well Immuno Plates (Thermo Scientific) overnight at 4°C. The plates were washed 3 times with phosphate-buffered saline (PBS)/1% Tween 20 and blocked with PBS/1% bovine serum albumin (BSA) buffer at room temperature (RT) for 1 h. Afterwards, 2 µg/mL biotinylated peptide was added and incubated at RT for 1 h. Then, the plates were washed again 3 times. To assess unspecific binding of plasma components, an uncoated control well without streptavidin and peptide was reserved for each sample. Diluted plasma samples (1:200 in blocking buffer) were added and incubated at RT for 2 h on the plates. After washing, a goat anti-human IgG (H+L) horseradish peroxidase-conjugated antibody (Invitrogen, stock concentration: 0.8 mg/mL) was diluted 1:10000, added and incubated for 1 h at RT. After washing 3 times with PBS/1% Tween 20, 100 μL of TMB substrate (SERVA Electrophoresis) was added to each well. After reacting at RT in the dark for 5 min, 100 μL stop solution (1 M H₂SO₄) was added per well. The optical densities (ODs) were measured at 450 nm using an Agilent BioTek Epoch Microplate Spectrophotometer. The average of the measured OD values per sample (doublets or triplets) was normalized by subtracting the mean OD value of blank wells and the OD value of the respective uncoated control. For the statistical analysis, negative values in the data were truncated to zero.

For the serum samples of the validation cohort, the ELISAs were performed with the peptide coating kit from Takara Bio. In 96-well reaction plates, 4 µg/mL of peptides in reaction buffer were coated with coupling reagent for 2 h at 37 °C. The coated wells were blocked for 1 h. The remaining steps were as described above, with the exception of using distilled water as the wash buffer.

For a basic immunophenotyping by flow cytometry, PBMC samples of the validation cohort were thawed at 37°C in a water bath and washed with CTL Anti-Aggregate Wash Supplement (ImmunoSpot). The cells were then washed with PBS and stained with 0.3 µg BD Horizon Fixable Viability Stain 700 per 1×10⁶-1×10⁷ cells in PBS at RT in the dark for live/dead discrimination. After washing, the cells were resuspended in PBS containing 2 mM EDTA and 0.5% BSA, and Fc receptors were blocked with human Fc receptor Blocking Reagent (Miltenyi Biotec) according to the manufacturer’s instructions. Fluorophore-conjugated mouse anti-human antibodies against cell surface markers (CD19-PE/Cy7, CD3-APC-H7 and CD14-BV605, all from BD) were premixed in Brilliant Stain Buffer (BD) at recommended concentrations. The cells were stained at RT in the dark for 30 min. After washing the cells, data were recorded using a BD FACSAria IIIu and the BD FACSDiva software (version 8.0.2) and further processed using FlowJo (version 10.7.1) from BD. The data were automatically cleaned for technical signal anomalies using the flowAI plugin (54). Only live single cells were considered for determining the relative proportions of B cells, T cells and monocytes in PBMC.

Selected fresh PBMC samples (main study cohort) and frozen-thawed PBMC samples (validation cohort) were first washed with autoMACS Running Buffer (Miltenyi Biotec). Afterwards, B cells were positively selected from the PBMC using CD19 MicroBeads and LS Separation Columns (Miltenyi Biotec). CD4⁺ T cells were obtained from the negative fraction of the B-cell separation by removing non-T cells using the Pan T Cell Isolation Kit (Miltenyi Biotec) and isolating CD4⁺ T cells using CD4 MicroBeads (Miltenyi Biotec). The purity of the isolated cells was assessed by staining a small subset of the cells with CD3-PerCP and CD4-FITC or CD20-PE antibodies (Miltenyi Biotec). The flow cytometry analysis was carried out using a FACSCalibur instrument (BD) and Flowing Software version 2.5.1 (Turku Bioscience). The remaining cells were lysed with QIAzol Lysis Reagent (Qiagen) and stored at -20°C until further use.

The RNA from CD19⁺ B cells and CD4⁺ T cells was isolated using the miRNeasy Mini Kit and the RNase-Free DNase Set (Qiagen). RNA concentrations were measured using a NanoDrop ND-1000 Spectrophotometer (Thermo Fisher Scientific), and RNA integrity was assessed with an Agilent 2100 Bioanalyzer using RNA 6000 Pico kits (Agilent Technologies). We used 100 ng of total RNA from each sample to generate amplified, fragmented and biotinylated single-stranded sense strand DNA using the GeneChip WT PLUS Reagent Kit (Applied Biosystems). Poly-A RNA controls were used to monitor the target labeling process, and hybridization controls were spiked into the hybridization cocktail to evaluate sample hybridization efficiency. The hybridization on high-resolution Clariom D arrays for human (Applied Biosystems) was carried out for 16 h at 45°C in a GeneChip Hybridization Oven 645 (Affymetrix) following the manufacturer’s instructions. After washing and staining in a GeneChip Fluidics Station 450 (Affymetrix), the microarrays were scanned using a GeneChip Scanner 3000 7G (Affymetrix).

The Affymetrix GeneChip Command Console version 4.0 was used to extract the signal intensities for the >6.7 million 25mer oligonucleotide probes per array. The Transcriptome Analysis Console (TAC) version 4.0.3 (Applied Biosystems) was then used for processing the data with the signal space transformation robust multi-array average algorithm for background reduction, intensity normalization, probe set summarization and log₂ transformation. The further analysis of the transcriptome data was restricted to gene-level probe sets with annotated NCBI Gene identifier. We also excluded genes that were not expressed by eliminating probe sets with a log₂ signal intensity of <4 in all samples from both the B-cell and T-cell datasets.

Seven genes that are preferentially expressed by CD4⁺ cytotoxic T lymphocytes (CD4-CTLs) (55) were selected to evaluate whether they are differentially expressed in responders vs. non-responders to apheresis. Eligible baseline T-cell samples of MS patients from the validation cohort were used for this analysis. The RNA from the enriched CD4⁺ T cells was isolated as described in the previous section. Afterwards, reverse transcription was performed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Relative quantification of target cDNA levels by real-time polymerase chain reaction (PCR) was performed in triplicate for 45 cycles in a 7900HT Fast Real-Time PCR System (Applied Biosystems) using the TaqMan Universal PCR Master Mix (Applied Biosystems) and inventoried TaqMan assays with FAM dye-labeled minor groove binder probes (Thermo Fisher Scientific) ( Supplementary Table 2 ). GAPDH was measured as reference gene. Raw C_t values were computed automatically using the SDS 2.3 and RQ Manager 1.2 software (Applied Biosystems). We then calculated the mean C_t value of each triplicate, normalized the data to the expression of the reference gene (ΔC_t method) and transformed the data to the linear scale using the equation 2^−ΔCt×1000.

The data were analyzed in R version 4.1.2. The sociodemographic, clinical and medication data of the patients were summarized by descriptive statistics. For this purpose, counts and percentages were calculated for categorical variables, and means, standard deviations (SDs), medians and ranges were calculated for numerical variables. This was done using the valid (i.e., non-missing) data of all patients as well as distinct subgroups in each cohort. For comparing groups of patients, e.g., responders and non-responders to apheresis, we used two-sample two-tailed Welch t-tests and Mann-Whitney U tests for metric data and Fisher exact tests and chi-squared tests for categorical data. The Shapiro-Wilk test was used to determine whether the data was normally distributed. The significance level was set at α=0.05.

The data from the peptide-based ELISAs, flow cytometry, transcriptome profiling and real-time PCR quantification were analyzed basically in the same manner. Means, SDs and standard errors were generally used to describe the data. The baseline and follow-up data were compared using paired t-tests, thereby controlling for individual-level variability by focusing on the differences within each subject. The baseline data of responders and non-responders to relapse therapy were compared with Welch t-tests, not assuming equal variances. The interaction “response × time” was tested for significance using repeated measures analysis of variance (ANOVA) using the rstatix R package. We further calculated t-test effect sizes (Cohen’s d) and partial eta-squared (η_p ²) estimates for the interaction terms. It should be noted, however, that the data were normally distributed for some but not all peptides, cell types and genes. We have therefore checked the key results with the corresponding non-parametric tests (Wilcoxon signed-rank test and Mann-Whitney U tests). Visualizations of the data included bar charts and boxplots.

In the analysis of the transcriptome data, mean differences are presented as log₂ fold changes (FC), as the data are in log₂ scale, and the false discovery rate (FDR) approach was applied to adjust the p-values for multiple testing (56). The criteria to identify differentially expressed genes (DEGs) were a log₂FC of >2.0 or <-2.0 and a raw p-value of <0.05. The overlap of DEGs was visualized in Venn diagrams. For the genes that were significantly upregulated or downregulated in response to therapeutic apheresis in CD19⁺ B cells or CD4⁺ T cells, we computed their enrichment in Reactome pathways (57). This was done by calculating Fisher’s exact test p-values and odds ratios (ORs) and ranking the 10 most significant pathways using the enrichplot R package. Another gene set enrichment analysis was performed with the fgsea R package to test whether MS patients who did not adequately respond to therapeutic apheresis show an elevated expression of the CD4-CTL gene signature (55) in CD4⁺ T cells at baseline as compared to clinical responders.

The main study cohort comprised 33 patients with MS having a relapse ( Table 1 ). For relapse treatment, the patients received either high-dose methylprednisolone (n=27) or therapeutic apheresis (n=3 with plasma exchange and n=3 with immunoadsorption), the latter usually in the escalation treatment phase. The mean age of the patients was 40.6 ± 12.7 years, and the ratio of women to men was approximately 2:1. The patients’ neurological impairment was at a median EDSS score of 3.0. The DMT of the patients was very heterogeneous (n=13 without DMT, n=20 with DMT): There was no single DMT used by 4 or more patients (n=3 each for dimethyl fumarate, glatiramer acetate and teriflunomide). When comparing patients receiving GCs and patients receiving apheresis, there was no statistically significant difference, except for relapse therapy (phase) according to which the patients were categorized. There were a total of 25 responders (R) and 8 non-responders (NR) to the relapse treatment. Among the R, however, only 5 patients showed an improvement in the EDSS score (n=11 unchanged, n=9 with missing value). There was no significant difference in the comparisons of R vs. NR to GCs or apheresis. It is nevertheless worth noting that the NR to apheresis (n=2) had a relatively long disease duration of 12 or more years, while the R to apheresis (n=4) were diagnosed with MS on average 4.5 years ago (p=0.240).

The validation cohort included 30 patients (n=21 with MS, n=9 with NMOSD) who received therapeutic apheresis to manage a relapse. The patients with NMOSD were significantly older and had a significantly higher median EDSS score than the patients with MS ( Table 2 ). The NMOSD group also had a higher proportion of women (88.9% vs. 52.4%) and more NR to apheresis (55.6% vs. 23.8%) than the MS group, though not significantly. There were a total of 20 R and 10 NR. However, there was no significant difference in the comparison of R vs. NR among the parameters recorded, including time to apheresis since relapse onset.

The follow-up blood samples were taken on average 4.5 ± 3.3 days (main study cohort) and 9.0 ± 2.3 days (validation cohort) after baseline. The plasma levels of Ig isotypes remained unchanged after GC therapy (mean decrease <5%), while they fell sharply after apheresis (mean decrease >60%) ( Supplementary Table 3 ). In fact, reduced concentrations of IgA, IgG and IgM were seen in all MS patients who underwent apheresis (n=6), with the decrease in IgG levels being highly significant (from 4.21 to 0.82 g/L or -81% on average, p=0.002).

The plasma samples from the main study cohort were used to measure antibody reactivities against 12 peptides from potential protein antigens ( Supplementary Table 1 ). In the following, the peptides are consistently named with the UniProt database (58) protein identification code, which has at most 5 alphanumeric characters, and the amino acid positions. The highest OD values were detected for EBNA1 (391-410), followed by CRYAB (3-17) and HECAM (370-389), both of which have overlapping epitopes with the EBNA1 (391-410) sequence (“RRPFF” and “SPPR”, respectively). After GC therapy, the IgG reactivities against all peptides did not change much (|d|<0.5 and p>0.05, also in non-parametric tests). In contrast, after therapeutic apheresis, a large decrease (d<-0.8) in antibody reactivity was observed for 8 peptides and a medium decrease (d<-0.5) for 3 peptides (d=-0.48 for MOG (196-215)). Despite the small sample size (n=6 patients), the significance level was reached for 5 of the 12 peptides ( Figure 2 ; Supplementary Table 3 ).

Additional ELISA tests were performed with 9 of the 12 peptides using the sera of the validation cohort. In the subgroup of patients with MS, all IgG reactivities were reduced after apheresis (n=6 peptides with p<0.05, also in signed-rank tests) ( Figure 2 ; Supplementary Table 4 ). The treatment effects were weaker in NMOSD patients, who generally had lower reactivities at baseline than MS patients (although p>0.1). However, the reactivities against EBNA1 (391-410) were also significantly reduced at follow-up in the NMOSD group. EBNA1 (391-410) was thus the only peptide in all 3 apheresis groups (MS main study, MS validation and NMOSD validation) showing a significant decrease in IgG reactivities (in both t-tests and signed-rank tests).

In the main study, the plasma levels of IgA, IgG and IgM at baseline and their change during therapy in MS patients who received apheresis were not significantly related to overall symptom improvement. The IgG reactivities against the 12 peptides and their dynamics were also not significantly associated with the clinical response to apheresis ( Supplementary Table 3 ). Significance was only reached in cases receiving GC therapy for IgG reactivities against PERT (181-195), which were lower in NR than in R at baseline ( Figure 2 ). When analyzing the MS patients of the validation cohort, NR showed higher anti-EBNA1 (391-410) reactivities before apheresis than R (d=1.38, p<0.05 in t-test and U test) ( Supplementary Table 4 ). Higher reactivities against EBNA1 (391-410) in NR vs. R were also noted in the other 2 apheresis groups (MS main study: d=0.61, NMOSD validation: d=1.37), but these differences were not statistically significant, possibly due to smaller sample sizes ( Figure 2 ). Otherwise, no other peptide reached the significance level for R vs. NR at baseline and with regard to the interaction term “response × time”.

The proportions of basic PBMC subpopulations were determined for 29 patients (n=58 samples) of the validation cohort. The data revealed that therapeutic apheresis was accompanied by moderate shifts in the relative frequencies of immune cell subsets: In NMOSD patients, we observed a significant increase in the percentage of CD19⁺ B cells and a significant decrease in the percentage of CD14⁺ monocytes ( Figure 3A ). Over all patients (n=20 with MS and n=9 with NMOSD), the decrease in the proportion of monocytes (from 20.8% to 15.6% on average) and the increase in the proportion of CD3⁺ T cells (from 56.5% to 63.6% on average) were significant. However, the percentages of these cell types at baseline and their change from baseline to follow-up were not significantly associated with the clinical response to apheresis therapy (p>0.05).

The purity of the isolated CD19⁺ B cells (n=6 samples) and CD4⁺ T cells (n=8 samples) for the transcriptome profiling (main study) was at least 88.9% ( Figure 3B ). The enrichment of CD4⁺ T cells for the validation by real-time PCR (n=19 samples) was lower, but still averaged 92.1%. The B-cell and T-cell samples were therefore considered eligible for the gene expression analyses.

We profiled the transcriptome of CD19⁺ B cells from 3 MS patients (n=6 samples) and of CD4⁺ T cells from 4 MS patients (n=8 samples) receiving apheresis as relapse therapy. The RNA integrity numbers (59) of the isolated RNA averaged 8.3 ± 0.4 (B-cell samples) and 8.4 ± 0.3 (T-cell samples). The Clariom D arrays passed all quality control measures, including the expected increases in signal values of the labeling controls and hybridization controls. The data processing yielded expression values for a total of 135750 gene-level probe sets (i.e., transcript clusters). We restricted our analysis to 24322 transcripts with NCBI Gene identifier, and we excluded transcripts that were not expressed in the B-cell samples (n=6353) or T-cell samples (n=4222) (maximum signal <4). The raw and processed data have been deposited in the Gene Expression Omnibus (GEO) database under accession number GSE272973.

By comparing the gene expression at baseline and follow-up, moderate transcriptome changes were detected in response to therapeutic apheresis. For CD19⁺ B cells, 42 DEGs were filtered (10 genes as upregulated and 32 genes as downregulated) ( Supplementary Table 5 ). For CD4⁺ T cells, 69 DEGs were filtered (47 genes as upregulated and 22 genes as downregulated) ( Supplementary Table 6 ). Six gene transcripts were differentially expressed with |log₂FC|>2.0 and p<0.05 in both CD19⁺ B cells and CD4⁺ T cells ( Figure 4A ). However, no gene reached an FDR-adjusted p-value <0.05.

The pathway analysis revealed significant enrichments of the DEGs in interferon (IFN) signaling, with ORs between 6.2 and 13.4 for B cells and CD4⁺ T cells ( Figure 4B ). This was in part attributable to the increased expression of IRF2 and OAS1 in both cell types ( Supplementary Tables 5 , 6 ). The gene products of the DEGs were also overrepresented in costimulation by the CD28 family (OR=12.4 for B cells) and immunoregulatory interactions between lymphoid and non-lymphoid cells (OR=9.5 for CD4⁺ T cells). Furthermore, genes involved in programmed cell death (e.g., CASP9 and DYNLL1) and hematopoietic cell differentiation (e.g., GRAP2 and SH2B3) were enriched in the DEGs from the B-cell dataset (ORs up to 15.6 and 23.1, respectively). Chemokine receptors (e.g., CCR5 and CX3CR1) and rhodopsin-like receptors (e.g., LPAR6 and PTGER2) were enriched in the DEGs from the T-cell dataset (ORs up to 16.0).

A valid comparison of the transcriptome data according to clinical outcome of apheresis treatment was not feasible due to the limited number of cases and quasi-complete separation by sex (1 NR vs. 2 R for B cells and 2 NR vs. 2 R for T cells, one male patient per dataset). After FDR-based correction for multiple testing, there was no gene with p<0.05 for R vs. NR at baseline or for the interaction “response × time”. We nevertheless inspected the T-cell dataset for non-significant expression differences of autosomal genes (n=19265) between R and NR. This revealed that some genes were considerably higher expressed in NR than in R before relapse therapy ( Figure 5A ). Among the top 10 of these genes (log₂ mean difference >3.20, which means >9-fold higher expression in NR vs. R) were 5 genes that have been described as CD4-CTL-related genes (55): FGFBP2, GNLY, GZMH, NKG7 and SLCO4C1 (log₂ mean difference >2.18 for CD4-CTLs vs. central memory T cells). A gene set enrichment analysis confirmed that genes that were found to be more abundantly expressed in CD4-CTLs (n=517) (55) were typically expressed at higher levels in NR than in R at baseline (p=2.8×10^-36 and normalized enrichment score =2.80) ( Figure 5B ).

For the verification of a potential gene expression signature that may predict clinical response to therapeutic apheresis, 7 genes were selected. These genes were more strongly expressed in the T-cell transcriptome data in NR vs. R prior to apheresis, and an elevated expression of these genes is characteristic for CD4-CTLs (55). The real-time PCR analysis was conducted with baseline T-cell samples from 19 MS patients of the validation cohort who underwent apheresis to manage an acute relapse. In the comparison of R (n=14) and NR (n=5), none of the 7 genes (CD244, EOMES, FGFBP2, GNLY, GZMB, NKG7 and PLEK) reached the significance level of α=0.05 in parametric or non-parametric tests. However, for 6 of the 7 CD4-CTL-related genes (all but CD244), the mean mRNA level was at least nominally higher in NR than in R in the real-time PCR data ( Figure 5C ). Particularly high expression levels were measured for NKG7.