An autoantibody signature predictive for multiple sclerosis

Abstract

Although B cells are implicated in multiple sclerosis (MS) pathophysiology, a predictive or diagnostic autoantibody remains elusive. In this study, the Department of Defense Serum Repository (DoDSR), a cohort of over 10 million individuals, was used to generate whole-proteome autoantibody profiles of hundreds of patients with MS (PwMS) years before and subsequently after MS onset. This analysis defines a unique cluster in approximately 10% of PwMS who share an autoantibody signature against a common motif that has similarity with many human pathogens. These patients exhibit antibody reactivity years before developing MS symptoms and have higher levels of serum neurofilament light (sNfL) compared to other PwMS. Furthermore, this profile is preserved over time, providing molecular evidence for an immunologically active preclinical period years before clinical onset. This autoantibody reactivity was validated in samples from a separate incident MS cohort in both cerebrospinal fluid and serum, where it is highly specific for patients eventually diagnosed with MS. This signature is a starting point for further immunological characterization of this MS patient subset and may be clinically useful as an antigen-specific biomarker for high-risk patients with clinically or radiologically isolated neuroinflammatory syndromes.

Extended Data Fig. 1 |

Heatmap of logRPK values from whole proteome. PhIP-Seq data on MS patient sera from DoDSR cohort. Top peptides shown with peptide enrichment. blowout to demonstrate similarity between pre and post onset sample enrichments over time.

Extended Data Fig. 2 |

We investigated the similarity between different timepoints from the from the DoDSR cohort (n = 500). (Top Left) Unfiltered normalized read counts (rpK) were log transformed, and a correlation matrix was computed using the Euclidean distance measure. The similarity between any two samples was defined by their n-dimensional Euclidean distance, where each peptide in the library was treated as a dimension. The rows and columns of this matrix are ordered such that the early timepoints are followed by the late timepoints. The central diagonal stripe in the heatmap represents a distance of zero where the distance was computed between the same sample. However, the diagonal stripes represent distances calculated between different timepoints of the same sample. (Top Right) Box plots of the Euclidean distance between timepoints, with a trend towards increasing distance at longer intervals. However, patterns of self-reactivity remain stable for more than 15 years. Notches approximate 95% confidence intervals. (Bottom) A boxplot demonstrates the significant difference (two-sided t-test) in Euclidean distance between two timepoints of the same sample versus the distance between two unrelated samples. Notches approximate 95% confidence intervals.

Extended Data Fig. 3 |

A subset of the library capturing relevant motifs from previously identified putative targets of molecular mimicry such as GlialCAM and anoctamin-2. (Top) MS patient samples are ordered so as to preserve the IC clusters. Although individual samples can be seen to enrich peptides of interest, they did not meet the predefined cutoffs for this analysis. (Bottom) HC samples demonstrating background levels of enrichment.

Extended Data Fig. 4 |

Time point specific NfL data from DoD cohort. (1^st time point: HC n = 236, MS_no-IC n = 204, MS_IC n = 13. 2^nd time point: HC n = 234, MS_no-IC n = 191, MS_IC n = 21. Data represented as geometric mean and standard factors for each box and whisker.).

Extended Data Fig. 5 |

Additional OND CSF controls (n = 20), lacking enrichment of the IC motif.

Extended Data Fig. 6 |

Correlation of sum of MFI from all six peptides of interest in Table 1 from Luminex assay with their sum of respective normalized read counts from PhIP-Seq data in both CSF and serum samples from ORIGINS cohort. Shaded area represents 95% confidence intervals.

Extended Data Fig. 7 |

Luminex assay for antibodies against listed pathogen peptides in ORIGINS cohort a) MS patient CSF (n = 104), b) MS patient serum (n = 104), c) OND patient CSF(n = 42) and d) OND patient serum (n = 22). e) Same patient populations respresnted as sum of MFI combining signal across pathogen peptides with highlighted patients having normalized sumMFI values > 3 in CSF.

Fig. 1 |. Overview of MS biomarker study.

a, Schematic of DoDSR cohort and collection. b, Age and time to symptom onset for MS cases (data are presented as median values; box edges are 1st and 3rd quartiles; and whiskers represent 1.5× interquartile range; n = 250 for each group). c,d, Molecular biomarker assays performed on DoDSR cohort of longitudinal sera. NCBI, National Center for Biotechnology Information.

Fig. 2 |. Profiling DoDSR MS cohort.

a, UMAP of autoantigen enrichments in PhIP-Seq screen of DoDSR sera, showing distinct immunogenic clusters (IC1 and IC2). sNfL levels across DoDSR cohort sera with respect to time before onset (b) and grouped by timepoint for cases and controls (c) (data represented as geometric means and standard factors; P values shown above; evaluated via t-test comparing log-transformed NfL values; first timepoint: n = 236 for HC and n = 217 for MS; second timepoint: n = 234 for HC and n = 212 for MS). Sum of normalized reads (d) and individual fold enrichments (e) of top 192 peptides, with IC patientsʼ pre-onset and post-onset samples highlighted on left grouped by hierarchical clustering. f, Blowout of top 26 most enriched peptides in 27 IC patients grouped by gene and respective enrichments in their non-MS matched controls.

Fig. 3 |. IC peptide and cohort analysis.

a, Protein alignment of top 45 peptides within human proteome library that contain the IC regular expression, in order of patient prevalence that exceeded cutoffs in either pre-onset or post-onset samples. Bottom is alignment of regular expression to selected pathogens that infect humans (via PROSITE scan). b, Normalized reads assigned to regex-containing peptides in PwMS within either IC cluster as compared to those without and non-MS controls (data represented as means and s.d.; one-way ANOVA with Tukey’s multiple comparisons adjustment; P values shown above; n = 490 for HC, n = 433 for MSno-IC and n = 44 for MSIC). c, Serum levels of NfL in patients with MS within IC clusters as compared to those without and controls (data represented as geometric means and standard factors; P values shown above; evaluated via t-test with Bonferroni correction comparing log-transformed NfL values; n = 470 for HC, n = 395 for MSno-IC and n = 34 for MSIC). d, Serum NfL levels modeled over time for each group with respect to time to onset.

Fig. 4 |. PhIP-Seq characterization of ORIGINS validation cohort in CSF and sera.

Sum of normalized reads for previously identified top 26 peptides (a) and PhIP-Seq fold enrichments of ORIGINS patient sera and CSF for peptides mapping to same regions of ICs from the DoDSR cohort, relative to controls (b). c, Blowup of IC signature patients with all OND controls with CSF and serum for each patient grouped by column. d, UMAP of PhIP-Seq enrichments for all ORIGINS patients, shaded by sum RPK levels from a. Arrows indicate where IC clusters lie on UMAP.

Fig. 5 |. Luminex validation assay of selected IC peptides.

a, Schematic for barcoded immunofluorescence assay against top peptides from IC motif. Normalized median fluorescence intensity for each patient, matched by colored line, across all ORIGINS patientsʼ CSF (b) and sera (c) (n = 104) as well as OND controls (n = 22 serum and n = 42 CSF) (d,e). f, Same patient population shown with sum of MFI across peptides shown against all four groups, with those exceeding cutoff (normalized logMFI > 3) in CSF highlighted and matched by patient.