Sera from a total of 491 study participants were included in this study. Of these, 91 were people with MS, 32 were people with NMO, 281 were people with PTLD, and 87 were healthy controls. Data and sera from people with MS came from the Johns Hopkins Multiple Sclerosis Center, with recruitment and eligibility criteria described extensively elsewhere (24). Data and sera from people with NMO came from the Johns Hopkins NMO Clinic. Individuals were included in this study if they had a diagnosis of NMO as defined by the 2006 Wingerchuk criteria (25), and all but one individual was positive for aquaporin-4 antibodies. Data and sera from people with PTLD and a subset of the healthy controls (n = 22) came from the Studies of Lyme Disease Immunology and Clinical Events (SLICE) at the Johns Hopkins Lyme Disease Research Center, as previously described, with the exception that people with PTLD included in the current study were not required to have 6 months' illness duration (26). Data and sera from the remaining healthy controls (n = 65) came from an ongoing observational study of healthy donors at the Johns Hopkins Division of Rheumatology. Healthy volunteers who are not pregnant and who do not have a history of cancer, autoimmune disease, or active tuberculosis/HIV/hepatitis infections were eligible for the study. None of the healthy controls had a known history of Lyme disease. All participants signed written informed consent approved by the Johns Hopkins Institutional Review Board.

Demographic variables measured from all study participants include age, sex, and race. Clinical variables measured from people with MS include duration of illness and disease severity measured by the MS severity score (MSSS), as previously described (27, 28). People with MS were grouped according to their MS type, RRMS (n = 41), SPMS (n = 31), or PPMS (n = 16). Three people with MS did not have a documented MS classification, so were exclude from subtype analysis. Anti-aquaporin 4 antibody status was available for individuals with NMO, but no additional clinical information was available. Clinical variables measured from people with PTLD at the time of the study visit included duration of prior antibiotic treatment, duration of illness from onset of initial Lyme disease, two-tier Lyme disease serologic status (29), neurologic Lyme disease status, defined by medical records confirming Bell's Palsy, neuropathy, meningitis, or encephalitis with positive two-tier serology, and symptom severity. Symptom severity associated with PTLD, including the severity of neurologic symptoms, was measured using PLQS as previously described (26). The researchers were blinded to these clinical variables when conducting the experiments to minimize bias.

Single cell sequencing data was retrieved from Gene Expression Omnibus (GEO) database accession no. GSE67835 (30), which contains sequence data from 466 cells (oligodendrocytes (n = 38), astrocytes (n = 62), microglia/macrophages (n = 16), oligodendrocytes precursor cells (n = 18), hybrid cells (n = 46), microglia/macrophage (n = 16), neurons (n = 131), and endothelial cells (n = 20). Cells were isolated from human cortical tissue from eight adults and individual cells were classified into the categories above as described (31). Processing and visualization of the data was carried out in the R statistical language as previously described (32, 33). We retrieved selected brain region expression data for PADI2 from two public data sets. The first was data from five donors with 26 brain regions included in the 2010 Allen Institute for Brain Science, Allen Human Brain Atlas (34). Microarray data from three PADI2 specific probes were normalized across all brains as previously described (https://help.brain-map.org/display/humanbrain/Documentation). The original search can be reproduced at http://human.brain-map.org/microarray/search/show?exact_match=false&search_term=PADI2&search_type=gene&page_num=0. The second dataset was from The Genotype-Tissue Expression (GTEx) Project. The Genotype-Tissue Expression (GTEx) Project was supported by the Common Fund of the Office of the Director of the National Institutes of Health, and by National Cancer Institute (NCI), National Human Genome Research Institute (NHDRI), National Heart, Lung, and Blood Institute (NHLBI), National Institute on Drug Abuse (NIDA), National Institute of Mental Health (NIMH), and National Institute of Neurological Disorders and Stroke (NINDS). The data used for the analyses described in this manuscript were obtained from the GTEx Portal (https://gtexportal.org/home/) on 09/10/2020 using Ensembl Gene ID ENSG00000117115.12. The 17,382 samples included in this dataset are from 54 tissue regions from 948 donors. Details for the exact number of samples for each brain region can be found here: https://gtexportal.org/home/tissueSummaryPage. Data were normalized as previously described (https://gtexportal.org/home/documentationPage#staticTextAnalysisMethods).

Anti-PAD2 antibodies were measured in the serum of study participants using an in-house generated anti-PAD2 enzyme-linked immunosorbent assay (ELISA), as previously described (19). Briefly, recombinant human PAD2 protein containing N-terminal 6 × histidine and T7 tags was bacterially expressed and purified. The histidine tag was removed by thrombin digestion and 100 ng PAD2 protein was coated into each well of a high-binding polystyrene 96-well EIA plate (Costar) overnight. A dilution of 1:250 patient or healthy control sera was used for the primary antibody and a 1:7,500 dilution of goat anti-human IgG HRP was used for the secondary antibody. An 8-point standard curve was present on each plate comprised of dilutions of a known anti-PAD2 positive serum and was used to calculate anti-PAD2 Arbitrary Units (AU) for each sample. Blank wells coated in PBS alone were used to determine the background binding of each serum and these values were subtracted from those obtained from PAD2-coated wells. The cutoff for positivity was set at 4.5 AU as previously defined (6).

GraphPad Prism version 8.3.0 and SAS statistical software (version 9.4, SAS Institute, Cary, NC) were used for all statistical analyses and graphs. A p-value ≤ 0.05 was considered statistically significant throughout. Brown–Forsythe ANOVA with a Dunnett's T3 correction for multiple hypotheses analyses were performed to compare PADI2 expression in oligodendrocytes to other cell types. Fisher's exact tests were used to determine if the proportions of anti-PAD2 antibody positive people differed between groups, and median anti-PAD2 antibody levels between each group were compared using analysis of variance (ANOVA, Kruskal–Wallis), corrected for multiple comparisons. Clinical and demographic variables were compared according to anti-PAD2 antibody positivity for the MS and PTLD groups using Student's t-tests or Mann–Whitney tests for normally and non-normally distributed variables, Chi-squared tests or Fisher's exact tests for categorical variables, as appropriate. The median anti-PAD2 antibody level in each MS subtype was compared using a one-way ANOVA (Kruskal–Wallis) corrected for multiple comparisons. Correlations between anti-PAD2 antibody levels and MS disease severity or PTLD symptom severity were examined using the Kendall's Tau correlation coefficient.

We quantified the expression levels of PADI2 within specific cells and regions of the CNS by in silico analysis of publicly available transcriptomic data (30, 31, 34). This analysis revealed that PADI2 transcripts are detectable in multiple cell types in the CNS (Figure 4A). PADI2 expression was highest in oligodendrocytes [mean reads per kilobase of transcript per million (RPKM) ± standard deviation; 6.4 ± 2.2], as expected. We compared the expression of PADI2 in all other cell types to oligodendrocytes by Brown–Forsythe ANOVA with a Dunnett's T3 correction for multiple hypotheses. Astrocytes expressed PADI2 at similarly high levels as oligodendrocytes (5.5 ± 2.6 RPKM, p = 0.38), followed by microglia/macrophages (4.5 ± 3.5 RPKM, p = 0.29), which demonstrated bimodal expression of PADI2, with one subpopulation of cells expressing high levels of PADI2 and other low levels. Oligodendrocytes precursor cells (3.2 ± 3.3 RPKM, p = 0.006), neurons (0.9 ± 1.8 RPKM, p < 0.0001) and endothelial cells (1.3 ± 2.4 RPKM, p < 0.0001) were the lowest expressors of PADI2. Analysis of the GTEx database indicated that PADI2 was expressed in multiple regions of the CNS with the highest expression observed in the spinal cord and substantia nigra (Figure 4B). Analysis of gene expression data from the Allen Human Brain Atlas (34) showed similar results, with PADI2 transcripts detectable in multiple regions of the human brain, including the white matter, basal ganglia, midbrain, thalamus, and pons (Figure 4C). Together, the data reveal widespread expression of PADI2 within multiple CNS regions and cell types.

Given the widespread expression of PADI2 throughout the brain and spinal cord, and the finding that PAD2 is a known autoantigen in RA (19), we reasoned that PAD2 may become a target of the immune responses in individuals with CNS symptoms in whom an autoreactive process has been implicated. To address this hypothesis, we performed an ELISA to determine the prevalence of anti-PAD2 antibodies in the sera of individuals with diseases known or suspected to have an autoimmune process affecting the CNS: MS (n = 91), NMO (n = 32), and PTLD (n = 281). Sera from a group of healthy donors was included as a control population (n = 87). Demographic variables for each group including age, sex, and race are shown in Table 1. The average age of the healthy controls was lower than in the disease groups, but age was similar among the MS, NMO, and PTLD groups. In addition, the PTLD group contained more individuals who identified as male and white. Analysis of anti-PAD2 antibodies revealed an increased proportion of anti-PAD2 positivity in the MS (19.8%, 95% CI: 12.9%−29.1%; p = 0.007) and PTLD groups (13.9%, 95% CI: 10.3%−18.4%; p = 0.057) relative to the healthy controls (5.7%, 95% CI: 2.5%−12.8%; Table 1), with significantly higher median antibody levels in people with MS compared to healthy controls [median (interquartile range) of 1.96 (1.17–4.14) vs. 1.34 (0.78–1.98) anti-PAD2 AU; p_adj = < 0.001; Figure 2]. Patients with NMO had significantly lower levels of anti-PAD2 antibodies [0.21 (0.15–0.46) anti-PAD2 AU] than patients with MS, PTLD, or healthy controls (p_adj < 0.0001 for all; Figure 2). Together, these data revealed that anti-PAD2 antibodies are present in a subset of people with MS and PTLD.

The finding of anti-PAD2 antibodies in a subset of people with MS coupled to the previously reported association of anti-PAD2 antibodies with less severe disease in RA, led us to explore whether anti-PAD2 antibodies were associated with disease severity in MS (Table 2). There were no significant differences in demographic variables, disease duration, or current treatment between anti-PAD2 positive and negative people with MS. However, there was a modest and inverse correlation between anti-PAD2 antibody levels and disease severity, as measured by MSSS in a univariate analysis (τ = −0.145, p = 0.02) of all patients with MS, but this trend was not maintained in a multivariable model adjusting for age, sex, and treatment (Figure 3A). In addition, when assessed by MS subtype, anti-PAD2 antibody levels were significantly higher in people with RRMS and SPMS, compared to those with PPMS (Figure 3B). While 22.0% of people with RRMS (9/41) and 22.6% of people with SPMS (7/31) were anti-PAD2 positive, these antibodies were only detected in one individual with PPMS (1/16; 6.3%).

Given the clinical heterogeneity in PTLD and the finding that anti-PAD2 antibodies associated with less severe disease in MS, we sought to define whether these autoantibodies associated with neurologic symptom severity in PTLD. When grouped by anti-PAD2 antibody status, anti-PAD2 positive people with PTLD were significantly older, but did not differ by race, sex, two-tier serologic positivity for Lyme disease, duration of illness, duration of antibiotic treatment since the onset of their Lyme disease, or diagnosis of neurologic Lyme disease during the acute infection (Table 3). In a univariate analysis, there was a modest and inverse correlation between anti-PAD2 antibody levels and neurocognitive score (p = 0.027), with difficulty focusing or concentrating, memory changes, fatigue, and difficulty finding words contributing most strongly to the effect, but this trend was not maintained in a multivariable model adjusting for age and duration of antibiotic treatment (Figure 4A). Analysis of other systemic symptoms assessed by the PLQS revealed that difficulty sleeping and changes in the frequency or urgency of urination also negatively correlated with anti-PAD2 antibodies in PTLD (Figure 4B). Given that age positively correlated with anti-PAD2 antibodies, we performed a case-control study with 30 individuals with PTLD who had the lowest scores on the PLQS (<5) and 30 individuals with the highest neurological scores (≥17) matched for age, sex, and disease duration. Consistent with the previous observations, anti-PAD2 antibody levels were significantly higher in patients with a lower burden of neurocognitive symptoms (p = 0.036).