The study was composed of the following three phases: (i) peptide autoantigen phage library construction, (ii) screening and extraction of peptide autoantigens, and (iii) validation of the best candidate autoantigens. The last step required testing of serum IgG from independent cohorts of HD, patients with MS and other neurological diseases (OND) for binding with immobilized candidate antigens by ELISA ( Figure 1 ). Finally, the reactivity of IgG to the selected peptides was evaluated using cerebrospinal fluid (CSF) from patients with MS.

Blood samples from the initial cohorts of 11 healthy donors (HD) and 16 MS patients ( Supplementary Table S1 ) were collected in 2020 at the Research Center of Neurology (Moscow) and were used in antigen screening experiments. A total of 9 serum samples were subjected to PhIP-Seq. Later, titers of IgGs that bind identified MS-associated peptides were defined in all 27 samples of donor serum by ELISA. The antigenic peptides extracted from the peptide library were validated by ELISA of serum IgG samples from independent cohorts of 22 HD and 28 MS ( Supplementary Table S2 ), blood samples were collected in 2023 at the Research Center of Neurology (Moscow). The serum of OND patients [14 with amyotrophic lateral sclerosis (ALS) and 5 with neuromyelitis optica spectrum disorder (NMOSD)] served as negative controls in the validation experiments ( Supplementary Table S2 ). An additional cohort of 19 patients with MS and 13 patients with ALS was included in 2024 for analysis of CSF and paired serum samples ( Supplementary Table S3 ). CSF was collected from donors who underwent lumbar puncture during a routine diagnostic examination to confirm neurological condition. Blood and CSF samples were collected and then centrifuged at 4°C (blood: 1500 g, 10 min; CSF: 1000 g, 10 min). Supernatants were immediately frozen, stored at –80°C and shipped on dry ice for further analysis. MS was diagnosed according to the McDonald criteria. Relapsing-remitting active MS was diagnosed based on two or more relapses with incomplete recovery during one year, and high MRI activity (two or more newly developed gadolinium-enhanced (GD+) brain lesions during one year). Relapsing-remitting stable MS was diagnosed if the EDSS score was below 5.0 with no exacerbations, no increase in the EDSS score, and the absence of cerebral enhancing lesion on MRI in the preceding two years. Inclusion criteria for AQP4 seropositive NMOSD were based on the 2015 International Panel on NMOSD Diagnosis which requires AQP4 antibody seropositivity tested by cell-based assay and inflammatory attack(s) of the optic nerve, spinal cord, or brain, including the brainstem and area postrema attacks (30). Patients were diagnosed with clinically definite ALS according to the El Escorial criteria which requires the evidence of lower motor neuron degeneration by clinical and electrophysiological examination, evidence of upper motor neuron degeneration by clinical examination, and progressive spread of symptoms or signs within a region or to other regions, as determined by history or examination (31).

Unbiased design would give enormous number of random peptides; therefore, the complexity of the peptide library was artificially limited. To identify novel IgG autoantigens in MS, we selected human targets among the proteins associated with antibody response during various autoimmune disorders [AAgAtlas project - http://biokb.ncpsb.org/aagatlas (32, 33)], but not previously described as established autoantigens during MS. We also included several CNS-associated proteins involved in neuronal survival, differentiation, or growth (34–36), that had not been previously linked to MS progression. A total of 11,973 unique peptides from 302 (6424 peptides) and 642 (5536 peptides) human proteins associated with various autoimmune pathologies and cellular protein degradation pathways, respectively, were included ( Supplementary Table S4 ). Each protein sequence was split into 44 amino acid peptides with 14 amino acid overlaps. The resulting peptides were computationally reverse translated into 132 nt coding sequences, which were subsequently optimized for prokaryotic expression. For PCR amplification, 17 nt (CCCAGCCGGCCATGGCC) and 20 nt (GCTAGCAGTGGTGGAGGCGG) adapters were added to the 5′ and 3′ ends of each sequence. All formed internal restriction sites NcoI/NheI were changed to alternative codons encoding the same amino acids. The corresponding oligonucleotides were chemically synthesized (Twist Bioscience). To achieve the uniform distribution of peptide sequences and efficient DNA amplification, the final autoantigen library was amplified using emulsion PCR (37). The amplified library was cloned into the fd bacteriophage vector to produce peptide in-frame fusions with the bacteriophage coat protein p3 (fADL-1e-based vector AddgeneID139441) (27, 38) and transformed into E. coli TG-1 strain. Resulting autoantigen library had 94% coverage of unique sequences, as determined by NGS at a mean mapped read depth of 32x. Unique peptide sequences in the designed library were detected with a relatively uniform distribution ( Supplementary Figure S1 ). Phage virions were obtained using the established protocol (38, 39). Briefly, recombinant bacteriophages were purified by double PEG precipitation. The titer of the phage library was 5·10¹³ PFU/ml. The concentration of the bacteriophage particles was determined by ELISA. Purified phage particles were captured on the immunosorbent plate (Nunc) with immobilized anti-M13 antibodies (Sigma-Aldrich) and detected with horseradish peroxidase (HRP) conjugated anti-M13 (GE Healthcare) and anti-FLAG antibodies (Sigma-Aldrich). The resulting phage library was aliquoted and stored at -80°C until further use.

IgGs were isolated from serum by double-repeated 50% ammonium sulfate precipitation, followed by affinity chromatography on protein G-Sepharose (Amersham Biosciences, UK) and size-exclusion chromatography on Superdex 200 (GE Healthcare, UK). The standard protein G PhIP-Seq protocol has been described previously (40). Briefly, 1 mL of phage library (5·10⁹ PFU) was incubated with purified human serum IgG (2 µg) for 18 h at 4°C with rotation. Next day, 10 µL of protein G-coated magnetic beads (Invitrogen) were mixed with each sample and incubated for 4 h at 4°C with rotation. The beads were washed with PhIP-Seq wash buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% NP-40) three times, and the bound phages were lysed at 95 °C for 10 min in water. The initial phage library without immunoprecipitation and mock immunoprecipitates (beads used on the phage library without serum pre-incubation) were used as controls. The final immunoprecipitated phage lysates were used for the preparation of the NGS library. To achieve the uniform distribution of enriched peptide sequences and efficient DNA amplification, we utilized emulsion PCR for DNA amplification as described previously (37) with primers flanking the designed peptide inserts. Illumina sequencing sample indexes and adapters were added within subsequent PCR.

Paired-end libraries were prepared with NEBNext Ultra II DNA Library Prep Kit (New England Biolabs, USA) in accordance with the manufacturer’s recommendations. The libraries were indexed with NEBNext Multiplex Oligos kit for Illumina (96 Index Primers, New England Biolabs, USA). Size distribution of the libraries and their quality were assessed using a high-sensitivity DNA chip (Agilent Technologies, USA). Subsequently, the libraries were quantified by High Sensitivity Quant-iT DNA Assay Kit (Thermo Scientific, USA). DNA sequencing was performed on the MiSeq platform (Illumina, USA) according to the manufacturer’s recommendations using the MiSeq Reagent Kit v2 (300 cycles) and a 10% PhiX spike-in control.

Linked 5′ and 3′ adapters in the NGS reads were identified and trimmed with cutadapt (41). Only reads with both adapters and no insertions or deletions in both paired-end reads were used for further analysis. Trimmed reads were aligned to the oligo library with bowtie (42), allowing for a maximum of three mismatches. Read counts were extracted using phip-stat (40) and normalized across all samples via the size-factors method (43). For each peptide, we computed a z-score using a statistical procedure proposed by Mina et al. (44) and implemented in phippery (https://github.com/matsengrp/phippery). Three mock immunoprecipitations were used for binning. Each bin was composed of minimum of 300 peptides that had similar counts in the mock immunoprecipitations. When analyzing each immunoprecipitation, the peptides that fell within the middle 90% according to the number of counts (a subset assumed not to contain hits) were used to estimate the mean and standard deviation of the peptide counts after immunoprecipitation within each bin. These estimates were then used to calculate the z-score for all peptides from that bin. Standardized enrichments were computed with phippery, and peptides with z-score ≥ 3 and standardized enrichment ≥ 2 were selected. Peptides were considered reactive if the counts exceeded 400 in at least two patients in addition to fulfilling the z-score and enrichment criteria.

Peptide-TRX-coding plasmids were generated by PCR amplification of the coding sequence from the enriched libraries with a set of specific primers ( Supplementary Table S5 ) and transferred into the pET-TRX-6xHis vector. Protein expression of the TRX-fused peptides was performed in the E. coli cell strain BL21 DE3. The overnight culture was inoculated at a 1:100 (v/v) ratio into 2xYT media supplemented with ampicillin (100 μg/mL) and 0.1% glucose. Cell culture was grown at 37°C to an OD600 of 0.4–0.6 and then supplemented with 0.5 mM of the IPTG (isopropylthio-β-galactoside). The expression culture was grown for 4 h at 30 °C with intensive aeration. Collected pellets were lysed with 0.2 mg/mL of lysozyme in PBS supplemented with cOmplete protease inhibitor cocktail (Roche, Germany). Lysates were incubated on ice with 10 μg/mL DNase until the viscosity disappeared. The resulting solution was centrifuged at 20,000 g and 4°C for 20 minutes, filtered through a 0.22 μm membrane and purified utilizing Ni-NTA resin according to the manufacturer’s instructions (Qiagen, USA). 6xHis-tagged peptides fused with TRX were then purified on Superdex 75 size-exclusion column. Recombinant intracellular LMP1 fragments (188-263 a.a.; 257-330 a.a.; 325-386 a.a.) were obtained in bacterial cells, while full-length LMP1 was produced in mammalian cells (19).

The antigens, diluted with 100 mM carbonate buffer (pH 9.0) to a concentration of 5 μg/mL, were absorbed on 96-well plates (MaxiSorb) overnight at 4°C and blocked with 2% nonfat dry milk in carbonate buffer. Starting from the 200 µg/mL dilution, purified serum antibodies were serially diluted at a 1:2 or 1:5 ratio in PBS with 0.5% nonfat dry milk and 0.05% Tween 20, applied to a coated well, and incubated for 1 h at 37°C with rotation. CSF samples were diluted at a 1:1 ratio in the same buffer. For isolated antigen-specific IgG, a concentration of 40 ng/mL was applied. Serum and isolated antigen-specific IgG samples were assayed in triplicate, while CSF samples were assayed in duplicate. As all recombinant antigens contained 6xHis tag, titration of anti-His antibodies (Sigma-Aldrich, USA) was used for calibration. After three washes with PBS containing 0.1% Tween 20, bound antibodies were detected with HRP-conjugated anti-human IgG Fc-specific antibodies (1:5,000) (ThermoFisher Scientific, USA). After 1 h at 37°C with rotation, plates were washed 5 times, and TMB was added for 15 min. The reaction was stopped with 10% H₃PO₄. The OD450 values were measured on the Varioskan Lux microplate reader (Thermo Scientific, USA).

Surface plasmon resonance (SPR) measurements were performed using Biacore T200 (GE Healthcare, USA). Molecular interactions were registered as sensograms representing the records of biosensor signals [in resonance units (RU)] as time function. Using the Biacore amine coupling kit, SPTAN1_601-644 fused with TRX and a reference TRX-carrier without peptide were immobilized in the flow cells 3 and 2 of the CM5 sensor chip, respectively, following the standard protocol. The entire coupling procedure resulted in 800 RU of immobilized proteins. Flow cell 1 with blank immobilization was used as an additional negative control. To evaluate the affinity of purified serum IgG, each patient’s sample was used in concentration 1 mg/mL in HBS-EP buffer with the addition of HSA to 5 mg/mL to eliminate nonspecific binding. Each sample was injected over 500 s, and then dissociation curve in a running buffer was measured over 2000 s. All SPR experiments were performed in triplicate. Assays were carried out at 20°C with a flow rate of 15 µL/min. The association (k_a) and dissociation rate constants (k_d) were calculated from the sensograms using two-state kinetic models in Prism.

To isolate antigen-specific autoantibodies, a preliminarily step involving the depletion of non-specific antibodies was carried out. For this purpose, 15 mg of purified IgG from serum of MS donors were incubated with 5 µL of NHS Mag Sepharose (GE Healthcare, USA) conjugated with TRX-carrier protein in TBS buffer (150 mM NaCl, 50 mM Tris-HCl, pH 7.5) at room temperature with rotation for 1 hour. TRX-specific and non-specifically bound antibodies were then removed from the solution together with NHS-TRX magnetic beads. Next, flow-through from depletion step was incubated with 5 µL of NHS Mag Sepharose conjugated with one of the recombinant peptide antigens fused with TRX-carrier: KRT1_301-344-TRX, SPTAN1_601-644-TRX, or GPI_31-74-TRX. The antibodies were incubated with appropriate antigen-TRX magnetic beads for 1 h at room temperature with rotation. Next, peptide-coupled magnetic Sepharose beads were washed twice in TBS, followed by washing with 50 mM sodium acetate buffer, pH 5.0 to remove non-specific antibodies. Antigen-specific antibodies were then eluted using 100 mM glycine-HCl buffer, pH 2.0, and the pH of the eluate was immediately adjusted to 7.0 using 1 M Tris-HCl buffer to prevent conformational changes of the eluted antibodies. The isolated antigen-specific antibodies were further analyzed using ELISA and polyacrylamide gel electrophoresis (PAGE).

The search for motifs in the analyzed peptides was conducted with the MEME (Multiple Em for Motif Elicitation) tool used in the Classic Discovery mode (http://meme-suite.org/tools/meme). The motifs with E-value < 0.05 were considered significant according to the MEME algorithm guidelines.

The CombiROC method was used to determine the diagnostic accuracy of optimal combinations of serum autoreactive IgG levels (45). This enabled determining the best combinations of biomarkers by a combined analysis of receiver operating characteristic (ROC) curves, calculating the sensitivity (SE) and specificity (SP) of all possible marker combinations (http://CombiROC.eu).

CSF neurofilament light chain (NfL) levels were determined by ELISA using commercially available monoclonal antibodies against human NfL (NF79 and HRP-conjugated NF71, HyTest, Russia) and recombinant NfL standard (RPE038Hu03, Cloud-Clone Corp., Wuhan, China). Measurements with a blank sample and NfL standard were used for signal calibration. Samples were run simultaneously in duplicates. CSF samples were standardly diluted at 1:4 ratio.

Data was analyzed using GraphPad Prism, version 9. The level of significance for the statistical analyses was set at p < 0.05. The antibody index for each sample was calculated as follows: (sample ELISA OD₄₅₀ value – mean blank value)/(mean HD ELISA OD₄₅₀ value – mean blank value). The statistical significance of differences in categorical data (e.g., seroprevalence) was assessed using Fisher’s exact test, as appropriate for relatively small sample sizes. In ELISA assays, samples were defined as seropositive if signals exceeded the mean signal for the HD group by three standard deviations. Several measurements within the HD group were identified as outliers by Grubbs’ test and were subsequently excluded from the cut-off calculation. Then, contingency tables containing the number of hits above this cut-off were analyzed to calculate Fisher’s exact p-values. The Mann–Whitney U test was used to evaluate the statistical significance of differences in ELISA signals between independent samples. Correlations between continuous variables were tested with the Pearson correlation coefficient.

The study was approved by the Local Ethic Committee of the Research Center of Neurology, Moscow, Russia. All patients and HD provided informed consent before participation.

The phage library containing overlapping unique 44-amino acid peptides covering human proteins associated with autoimmune disorders was incubated with serums from patients with two types of MS (stable MS, SMS; and active MS, AMS) or HD ( Supplementary Table S1 ). Characteristics of patients in the MS and HD cohorts are summarized in Supplementary Tables S1 , S2 . After immunoprecipitation with protein G, the enriched peptides were identified via NGS analysis. The number of enriched peptides did not significantly differ between HD and both MS cohorts (mean ± standard deviation of enriched peptides in HD: 181 ± 30; SMS: 164 ± 82; AMS: 143 ± 28) ( Figure 2A ; Supplementary Figure S2 ). To analyze whether different individuals or even the same donor produce antibodies recognizing distinct epitopes of the same protein, the extracted peptides were juxtaposed against the ORF enrichment matrix. The largest number of individual peptides seroactive for all tested MS patients was detected in ninein ( Figure 2B ). Several peptides from mannosyl-oligosaccharide glucosidase (MOGS) were found in the serum of most MS patients but not in HD cohort. While antibodies reactive for most proteins recognized single 44-mer peptides, reactivity to ninein and MOGS among MS patients was detected for several peptides from different parts of the proteins. This indicates that these autoantibodies may bind many epitopes from one polypeptide, suggesting a polyclonal, antigen-driven response during neuroinflammation ( Figure 2B ; Supplementary Figure S4 ).

The peptides ranking was done according to their enrichment in each sample. Significant enrichment of 8 potential autoantigens was found in MS patients but not in HD. A set of peptides (SPTAN1_601-644, DPP4_481-524, GPI_31-74, HAVCR1_31-74, POU4F1_91-134) that had a z-score > 3 in at least three MS patients, but had a z-score < 3 in all healthy donors was identified, and these peptides were on the top of the hit list (UniProt abbreviations of gene names are used). Three peptides (PRX_451-494, PTK6_301-344, KRT1_301-344) with lower enrichment in the MS group but non-immunoreactive (z-score < 0) in healthy donors were also included. The peptide INSR_301-344, which was enriched in both MS and HD groups, was added to the list too. The antigenic peptides selected for further analysis are shown in Table 1 . Corresponding proteins have not yet been linked with MS.

The reactivity of MS and HD serum antibodies to the identified autoreactive peptides was verified by ELISA. Purified total IgG from serums of 16 MS patients and 11 HD were used for ELISA with identified peptides fused to TRX ( Table 1 ). Elevated binding with serum IgG was observed for seven of the nine peptides in MS group, except insulin receptor (INSR_301-344) and transcription factor 1 (POU4F1_91-134), but none in HD samples ( Figures 3A, C ). Although there was a marked increase in mean binding for the majority of the analyzed peptides in MS samples, statistically significant difference in the number of seropositive samples between MS and HD groups was detected only in case of SPTAN1_601-644. Individuals with significantly elevated titers of antibodies against SPTAN1_601-644 were present in both SMS and AMS groups. Antibodies against KRT1_301-344, DPP4_481-524, PRX_451-494, HAVCR1_31-74, PTK6_301-344, and GPI_31-74 were predominantly observed in AMS patients. No difference was found between HD and MS patients for the binding of serum antibodies with well-characterized myelin antigens, such as MBP and MOG ( Figures 3B, C ).

Assuming the association of EBV with MS (12, 20), we studied whether MS patients had elevated titers of serum antibodies against the EBV antigen LMP1, for which cross-reactive anti-LMP1 antibodies associated with MS were earlier reported by our group (19, 23). MS patients had elevated titers of serum antibodies recognizing both the full-length LMP1 and its C-terminal fragments LMP1_188-263 and LMP1_257-330 ( Figures 3B, C ). A statistically significant difference in the number of seropositive patients was detected between the HD and MS groups in the case of the full-length LMP1. Although the number of seropositive patients with IgG against LMP1 fragments did not differ significantly between HD and MS groups, most of the patients reactive to these fragments were also seropositive for the full-length LMP1. Increased titers of anti-LMP1 antibodies were also observed in some individuals in the HD group. However, there was no significant correlation between the binding of LMP1/LMP1-derived peptides and PhIP-Seq-selected antigenic peptides with total IgG from MS donors ( Supplementary Figure S4A ). Likewise, no association between the level of LMP1-binding antibodies and EDSS scores was observed ( Supplementary Figure S4B ).

In all samples, the binding of IgG to TRX-SPTAN1_601-644 was detected ( Supplementary Figure S5 ). Stronger binding was observed for serum samples from MS group and, especially, for AMS group (compared with HD) ( Figure 4A ). Although AMS patients had the highest average titers of IgG against TRX-SPTAN1_601-644, statistically significant difference in maximum resonance units (RUs) signals was observed only when HD group was compared to combined MS group (AMS plus SMS data). This result demonstrates that the size of the cohorts matters. To evaluate the antibody avidity against TRX-SPTAN1_601-644, the dissociation kinetics (off-rate constants) of the antigen-antibody complexes was assessed, as this kinetics does not depend on the concentration of antigen-specific antibodies (46). All analyzed serum IgG from both HD and MS cohorts had high antibody avidity against SPTAN1_601-644 with the off-rates ranging from 2·10^-3 to 6·10^-3 s^-1. It should be emphasized that ELISA analysis of IgG binding to TRX-SPTAN1_601-644 among the same donors demonstrated similar to SPR elevated reactivity for MS patients ( Figure 4B ). Significant positive correlation between the titers of the autoreactive anti-SPTAN1_601-644 IgG detected by SPR and ELISA in the serum samples from HD and MS patients was observed (according to Spearman test r = 0.6430; p= 0.0054) ( Figure 4C ).

Next, we wondered whether autoantibodies from peripheral blood of MS donors, which specifically recognize KRT1_301-344, SPTAN1_601-644, or GPI_31-74 peptides, are polyreactive and can cross-react with other peptides. Thus, autoreactive IgG were incubated with magnetic beads coated by the corresponding autoantigens fused to TRX ( Figure 5A ). IgG fractions both from the donors with increased titers of IgG to the particular analyzed antigen and from those without detectable reactivity to the same antigen were studied. To decrease the non-specific binding of TRX with IgG, antibodies specific to the TRX were depleted from the total IgG pool of MS patients’ and HD sera. The enrichment of the IgG pools with antibodies against the PhIP-Seq-identified peptides (SPTAN1_601-644, KRT1_301-344 or GPI_31-74) was carried out using the so-called positive and negative samples from donors reactive and non-reactive respective to the antigenic peptides. As expected, the positive samples enriched with IgG specific to PhIP-Seq-identified peptides had high affinity to the antigen that was initially chosen for enrichment ( Figures 5B, C ). Concentrations of IgG fractions from positive samples eluted from TRX-peptide—coated beads were relatively high (5-30 µg/mL). In contrast, the IgG preparations obtained in parallel from the “negative” samples contained low levels (< 1 µg/mL) of IgG and demonstrated weak or no binding to all antigens identified in this study ( Figures 5B, C ). At the same time, IgG from HD serum samples and human IVIG (additional negative control representing pooled IgG from 1000 HD) didn’t cross-react with the tested antigens. The antigen-positive antibodies from SMS donors (0/2) showed no cross-reactivity to the analyzed antigens, while IgG from AMS donors (6/8) cross-reacted with at least two analyzed antigens ( Figure 5B ). A detailed comparison of antigen-enriched IgG samples revealed significant cross-reactivity of antigen-specific IgG binding SPTAN1_601-644 and GPI_31-74 ( Supplementary Figure S6 ).

Amino acid sequences of PhIP-Seq-identified peptides showed significant disparity, suggesting a limited level of conservation; only short common motifs for the peptide pairs KRT1 and SPTAN1, GPI and SPTAN1, DPP4 and SPTAN1, KRT1 and GPI were identified ( Supplementary Figure S7 ).

To determine the suitability of the identified antigens as potential markers for MS monitoring, an additional validation was conducted. New independent HD and patients with MS, amyotrophic lateral sclerosis (ALS), and neuromyelitis optica spectrum disorders (NMOSD) were included (for clinical data, refer to Supplementary Table S2 ) to assess the performance and specificity of the best candidate autoantigens across these diverse neurological conditions. In the new cohort, the number of MS patients with elevated titers of autoantibodies against the SPTAN1_601-644 peptide was significantly higher than in the HD group ( Figure 6A ). The reproducible and statistically significant difference in the number of SPTAN1 seropositive patients in both the initial and validation cohorts provides strong evidence for the potential utility of anti-SPTAN1 IgG as a diagnostic marker for MS. Despite the high heterogeneity of MS, no significant association was observed between IgG serum binding to antigenic peptides, EDSS score or disease duration (Figures 6B, C; Supplementary Figures S8 – S10 ).

To further assess the diagnostic accuracy of the identified antigens, ROC analysis was also performed ( Figure 6D , Supplementary Table S6 ). We evaluated whether combinations of several biomarkers selected by combinatorial analysis (45) would result in high predictive power. PTK6_301-344 and PRX_451-494 combination demonstrated the highest sensitivity (SE) and AUC value for two-peptide panel ( Figure 6D ; Supplementary Table S6 ). The antibodies to viral protein LMP1_257-330 fragment can be used as potential biomarker for MS as well ( Figure 6D ). Employing a triple and four-biomarker panel for simultaneous binding of an autoantigens and viral LMP1 allowed improving the predictive values of ROC curves ( Figure 6D ). The simultaneous use of four biomarkers (SPTAN1_601-644, PRX_451-494, PTK6_301-344 and LMP1_257-330) showed the highest prediction value for stratifying subjects with MS (n = 28) from healthy controls (n = 22) providing balanced sensitivity of 79% and specificity of 86% (AUC = 0.818) ( Figure 6D ). To estimate the overall biomarker performance, a permutation test was carried out by resampling the observed data ( Supplementary Figure S11 ). Obtained results imply that analyzed marker combinations provide robust sample classification.

To confirm the presence of autoreactive antibodies targeting the identified peptides in CNS during MS, we examined paired CSF and serum samples from patients with SMS, AMS, and ALS (for clinical data, refer to Supplementary Table S3 ). Firstly, CSF NfL, a marker of axonal injury, was measured. Notably, in line with previous studies, CSF NfL levels correlated with disease activity (47, 48) and were significantly higher in AMS compared to SMS ( Supplementary Figure S12A ). Simultaneously, patients with ALS exhibited elevated CSF NfL levels when compared to both SMS and AMS supporting previous research findings (49). Increased titers of CSF autoreactive antibodies targeting the most reliable discovered autoantigen SPTAN1_601-644 were observed in two out of nine analyzed patients with AMS, while none of the ten patients with SMS or thirteen patients with ALS exhibited similar increased titer ( Supplementary Figure S13A ). Moreover, in patients with MS, the antibody titers of anti-SPTAN1_601-644, anti-PRX_451-494, and anti-LMP1_257-330 IgG in CSF positively correlated with the levels of serum IgG which target the same antigens ( Figure 7 ). In contrast, CSF from ALS patients showed no immunoreactivity to any of the six tested antigens ( Supplementary Figure S13 ), suggesting that the response is specific to MS. Patients with AMS had higher IgG concentrations in CSF than patients with SMS and ALS ( Supplementary Figure S12B ). Both patients with elevated titers of anti-SPTAN1_601-644 IgG in CSF ( Supplementary Figure S13 ) also demonstrated increased total IgG levels in the CSF ( Supplementary Figure S12 ). Given that ALS patients exhibited significantly elevated NfL levels compared to MS patients, we assume that the autoantibodies targeting the identified autoantigens observed in MS group are unlikely a result of non-specific axonal damage and neuroinflammation.