Phenotypic and Ig Repertoire Analyses Indicate a Common Origin of IgD⁻CD27⁻ Double Negative B Cells in Healthy Individuals and Multiple Sclerosis Patients

Multiple sclerosis (MS) is known as an inflammatory autoimmune demyelinating disease of the CNS. Although T cells have been regarded as the principal effectors, an important role of B cells in MS pathogenesis is now widely accepted. B cells are found at sites of tissue injury in the CNS. They are also found in the cerebrospinal fluid, white matter lesions, gray matter, and in the meninges, where they form lymphoid-like tissue aggregates (1) that associate with proximal tissue damage (2). The B cells that populate distinct compartments of the CNS are clonally related (3) and link to populations in the periphery (4). Furthermore, they are responsible for the production of the oligoclonal Ig bands in the cerebrospinal fluid that remain a hallmark of the disease (5, 6). The strongest evidence supporting their role comes from B cell–depleting therapy, which demonstrates remarkable efficacy in relapsing-remitting MS (RRMS) and even primary progressive MS (PPMS) (7–9). These collective findings notwithstanding, the mechanisms underpinning the pathogenic contribution of B cells require further understanding.

Abnormalities in specific subsets of the B cell lineage have been increasingly identified in autoimmune diseases, including MS (10). One such B cell subset is the age-associated IgD⁻CD27⁻ double negative (DN) B cell. We recently reported an increased proportion of young MS patients <60 y old (20% of subjects) with increased peripheral frequencies of DN B cells compared with age-matched healthy controls (HC; 3% of subjects) (11). The proportion of DN B cells was also increased in MS cerebrospinal fluid. Their potential to induce (proinflammatory) T cell responses was indicated by expression of MHC class II, CD80, and CD86 (11). Further, DN B cells produced proinflammatory and cytotoxic cytokines following ex vivo stimulation. These results, together with the finding of clonal relations between Ig class-switched DN B cells in the peripheral blood of MS patients and intrathecal Ig repertoires (4), point toward the possible involvement of DN B cells in MS pathogenesis. However, the origin and developmental pathway of DN B cells in MS patients remain unknown.

DN B cells are elevated in aged individuals (11, 12), in rotavirus (13) and HIV infection (14), and in several autoimmune diseases, such as systemic lupus erythematosus (SLE) (15, 16) and rheumatoid arthritis (RA) (17, 18). In SLE, their frequency was associated with more severe disease status and increased titers of disease-specific autoantibodies (15, 19), indicating that they may contribute to autoimmune pathology. DN B cells show similarities with the recently described CD21^lowCD11c⁺T-bet⁺ age-associated B cells in aged and autoimmune mice and autoimmune individuals (20–22). Further, DN B cells constitute a heterogeneous population of IgG⁺, IgA⁺, and IgM⁺ isotypes (11, 15, 17, 23). They resemble IgD⁻CD27⁺ class-switched memory (CSM) B cells in their shortened telomeres (12), their expression of somatically mutated IgH V region genes (15), and their inability to extrude rhodamine or similar dyes (15) because of the lack of the transmembrane protein ATP-binding-cassette-B1 (ABCB1) transporter expression (12). The absence of ABCB1 expression was previously indicated as a characteristic of CSM B cells compared with CD27⁻ B cells (24). There furthermore appears to be a clonal relationship between DN and CD27⁺ memory B cells in HC (23). In addition, DN B cells demonstrated a decreased IgH mutation frequency compared with CSM B cells (17, 23, 25, 26).

In this study, we further investigated the origin and selection characteristics of DN B cells in MS patients and HC. First, we determined the expression of several Ig isotype and developmental markers on peripheral blood DN, naive, and CSM B cells of MS patients and HC using flow cytometry. Next, we examined the H and L chain Ig repertoire of both DN and IgD⁻CD27⁺ memory B cells of MS patients and HC using high-throughput adaptive immune receptor repertoire (AIRR) sequencing. This analysis focused on clonality, V gene usage, mutational profiles, and CDR3 physicochemical properties.

The study was approved by both the Human Research Protection Program at Yale School of Medicine and Hasselt University Commissions of Medical Ethics. Written informed consent was obtained from all participants in accordance with the Declaration of Helsinki. MS patients were recruited at the Rehabilitation and MS-Center (Pelt, Belgium) or Hospital Ramón y Cajal (Madrid, Spain) and were diagnosed according to the McDonald criteria (27). HC were recruited at Hasselt University (Hasselt, Belgium). Samples were cryopreserved at the University Biobank Limburg.

For flow cytometry, peripheral blood was collected from 63 RRMS patients, 20 secondary progressive MS (SPMS) patients, 13 PPMS patients, and 48 HC (Table I). All PPMS patients and 54 RRMS patients were treatment naive, whereas nine RRMS patients were previously being treated with first-line MS therapy. Of the SPMS patients, 17 were untreated and three were treated with IFN-β within 6 mo prior to sampling. For AIRR sequencing, peripheral blood was collected from five untreated RRMS patients and four HC (Table II), representing the earliest subject enrollment of the study. Of the MS patients, four were treatment naive and one was untreated for at least 3 mo after a short IFN-β treatment regimen.

All RRMS patients were younger than 60 y to exclude findings related to aging. Individuals older than 60 y were included in the PPMS and SPMS groups as the progressive forms of MS are rare in younger individuals. HC reported no history of autoimmune disease or malignancies and were matched to MS patients as closely as possible with regard to age and gender. Relevant demographic and clinical data (Tables I, II) were collected from all study subjects.

PBMC were isolated from whole blood by Ficoll density gradient centrifugation (Lympholyte; Cedarlane Laboratories, SanBio B.V., Uden, The Netherlands). Cryopreserved PBMC were used for flow cytometry. After thawing, PBMC were immediately recovered in 20% FBS in RPMI 1640 (Lonza, Basel, Switzerland) and 0.5 mg/ml DNAse. For AIRR sequencing, fresh samples were used except for MS469, MS495, and HC209. B cells were purified from the PBMC using negative magnetic selection (STEMCELL Technologies SARL, Grenoble, France) according to the manufacturer’s instructions. Purity of the isolated B cells was routinely ≥99.5% as determined by flow cytometry on a FACSAria II flow cytometer (BD Biosciences, Erembodegem, Belgium), following the staining procedure prior to sorting described below.

DN B cells were analyzed using anti-human CD19 Brilliant Violet 421 (clone HIB19), IgM PerCP-Cy5.5 (clone MHM-88), IgD PE-Cy7 (clone IA6-2), CD95 PE-Dazzle594 (clone DX2), CD5 BV605 (clone L17F12), CD38 BV711 (clone HIT2), CD20 allophycocyanin-Cy7 (clone 2H7) (all from BioLegend, London, U.K.), CD27 allophycocyanin (clone M-T271), IgG FITC (clone G18-145), CD10 BV786 (clone H110a) (all from BD Biosciences), and IgA PE (clone IS118E10) (Miltenyi Biotec, Leiden, The Netherlands). Viable cells were selected using the Fixable Viability Dye eFluor 506 (eBioscience, San Diego, CA). Fluorescence minus one controls were used for gating. The gating strategy is depicted in Supplemental Fig. 1A. All flow cytometry was performed on a LSRFortessa flow cytometer (BD Biosciences) and analysis was executed using FACSDiva (BD Biosciences) or FlowJo software (FlowJo, Ashland, OR). The cutoff to identify donors with expanded DN B cells was the mean percentage of DN B cells from healthy donors <60 y plus two times the SD (cutoff: 7%).

For the isolation of DN and IgD⁻CD27⁺ memory B cells, enriched B cells were stained with anti-human CD27 allophycocyanin (clone M-T271), IgD PE-CF594 (clone IA6-2) (both from BD Biosciences), and CD19 PE-Cy7 (clone HIB19; BioLegend) during 30 min at 4°C. At least 2 × 10⁴ DN or IgD⁻CD27⁺ memory B cells were then sorted on a FACSAria II flow cytometer (BD Biosciences), according to the gating strategy shown in Supplemental Fig. 1C. Sorted cells were pelleted or suspended in RNAlater Stabilization Solution (Thermo Fisher Scientific, Erembodegem, Belgium) before storage at −80°C.

Total RNA was extracted from the sorted cells with RNeasy kits (QIAGEN, Germantown, MD) according to the manufacturer’s protocol. Amplicons for sequencing on the Illumina MiSEQ platform were synthesized using reagents and a protocol provided by the manufacturer (New England Biolabs, Ipswich, MA). Briefly, RNA was reverse transcribed into cDNA using a biotinylated oligo dT primer. An adaptor sequence, containing a universal priming site and a 17-nt unique molecular identifier, was added to the 3′ end of all cDNA. Following purification using streptavidin-coated magnetic beads, PCR was performed to enrich for Ig sequences using a pool of primers targeting the IGHA, IGHD, IGHE, IGHG, IGHM, IGKC, and IGLC regions. This Ig-specific primer pool contained tailed sequences with a priming site for a secondary PCR step. The second primer was specific to the adaptor sequence added during the reverse transcription step and contained a sample index for downstream pooling of samples prior to sequencing. Following purification of PCR products using AMPure XP beads, the secondary PCR was performed to add the full-length Illumina P5 Adaptor sequence to the end of each Ig amplicon. Final products were purified using AMPure XP beads, then quantified with a TapeStation (Agilent Genomics, Santa Clara, CA).

Raw read quality control and assembly was performed as we have previously described with pRESTO version 0.5.2.999-2017.01.18 (28). For each sequence, germline segments were identified with IMGT/HighV-QUEST (http://imgt.org) using the February 2, 2017 version of the site (29). Subsequent analyses on the annotated sequences were performed using the immcantation tool suite (http://www.immcantation.org) version 1.8.0, which contains Change-O (version 0.3.9.999-2018.01.15) (30), alakazam (version 0.2.10), and shazam (version 0.1.9). Nonfunctional sequences, as defined by IMGT, as well as sequences for which the C region primer and the V or J gene calls were inconsistent (suggesting a chimera) were excluded from further analysis. Sequences with >20 Ns distributed in >15 intervals of Ns in the V region were removed. Duplicated sequences were collapsed. For each subject, sequences were assigned into clones using the DefineClones imgt function of Change-O by single-linkage clustering excluding sequences with non-ACGT characters in the junction (--maxmiss 0) and requiring the same junction region length, common IGHV and IGHJ gene annotations, and a maximum length normalized nucleotide Hamming distance of 0.19 (--model ham --dist 0.19) between their junction regions (31). For each clone, a germline sequence was reconstructed from the alignment data masking the N/P nucleotides and D segments of the junction. Mutation frequencies for the V region were calculated as the number of mismatches from the germline sequence. Sequencing data were deposited in the Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra) under BioProject accession PRJNA429427. Processed AIRR sequencing data from naive B cell sorts of HC were obtained from a previously published study (32) to be used as a reference control.

For flow cytometry, statistical analyses were performed using GraphPad Prism version 8. Multiple groups were compared by Kruskal–Wallis with Tukey post hoc testing. B cell subsets were compared within one study group using the Friedman test with Dunn multiple comparisons correction and between two study groups using Mann–Whitney U test. Fisher exact test was used for differences in proportions. For AIRR sequencing, statistical analyses were performed using the two-sided t test, using the paired version when comparing HC DN B cells versus HC IgD⁻CD27⁺ memory B cells or MS DN B cells versus MS IgD⁻CD27⁺ memory B cells. The Benjamini–Hochberg false discovery rate (FDR) was used to correct for multiple hypothesis testing. To assess the significance of the D frame usage between populations, a sign test was used.

First, frequencies of IgD⁺CD27⁻ naive B cells, IgD⁻CD27⁺ CSM B cells, and IgD⁻CD27⁻ DN B cells were measured in the peripheral blood of HC (n = 48), RRMS patients (n = 63), SPMS patients (n = 20), and PPMS patients (n = 13) (Table I). The frequency of total B cells within the PBMC population was significantly increased in all MS patient groups when compared with HC (p = 0.004 RRMS; p = 0.002 SPMS; p = 0.04 PPMS; Fig. 1A). Frequencies of naive and CSM B cells were similar between HC and MS patient groups (Fig. 1B, 1C). Interestingly, the frequency of DN B cells was significantly elevated in the peripheral blood of RRMS patients (6.1 ± 3.5%) when compared with HC (4.3 ± 1.9%; p = 0.004), all younger than 60 y (Fig. 1D). Moreover, the proportion of RRMS patients (18/63; 28.6%) and PPMS patients (4/11; 36.4%) younger than 60 y who presented with increased peripheral blood frequencies of DN B cells (>7% of CD19⁺ B cells, Fig. 1E) was significantly elevated compared with age-matched HC (4/42; 9.5%; p = 0.027 and p = 0.048, respectively; Fig. 1F). In total, 26/88 (29.5%) MS patients younger than 60 y showed an elevated frequency of DN B cells in the peripheral blood.

To evaluate the expression of well-described B cell developmental markers, we measured the distribution of Ig isotypes and the developmental markers CD5, CD10, CD38, CD20, and CD95 on naive, CSM, and DN B cells (Fig. 2A). In all B cell subsets, >93% of the cells were CD20⁺. As there was no statistically significant difference in the distribution of Ig isotypes in DN, CSM, or naive B cells between HC, RRMS, SPMS, and PPMS patients (Supplemental Fig. 1B), B cell subsets were compared between HC and total MS patients. On average, the largest proportion of DN B cells was IgG⁺ in both HC and MS patients (46.4 ± 14.3 and 44.6 ± 14.2%, respectively), followed by an equal proportion of IgM⁺ cells (20 ± 16.1 and 21.5 ± 15.9%, respectively) and IgA⁺ cells (20.3 ± 9.8 and 19.8 ± 8.9%, respectively) (Fig. 2B). Nevertheless, a high interdonor variability was observed in the frequency of isotype-specific DN and CSM B cells. DN B cells resembled CSM B cells in the frequency of IgM⁺ and IgG⁺ cells, although the frequency of IgA⁺ cells was significantly lower in the DN versus CSM B cell population, both for HC and MS patients (p < 0.0001).

In addition, DN B cells resembled CSM B cells in the expression of CD5, CD95, CD10, and CD38 (Fig. 2C), although quantitative analyses indicated several differences. The frequency of CD5⁺, CD38⁺, and CD95⁺ cells in the DN B cell population was in between that of CSM and naive B cells (Fig. 2D). In this regard, naive B cells demonstrated the highest frequency of CD5⁺ and CD38⁺ cells whereas CSM B cells presented with the highest frequency of CD95⁺ cells. In general, CD38⁺ and CD95⁺ cells were less frequent in all included B cell subsets from MS patients when compared with HC (Fig. 2D). No difference was observed in the frequency of CD10⁺ cells between DN, CSM, and naive B cells.

Thus, DN B cells resemble CSM B cells in Ig and developmental marker expression.

Differences in the expression of the developmental markers CD5, CD10, CD38, and CD95 were observed between IgM⁺, IgG⁺, and IgA⁺ DN B cells (Fig. 3A). IgM⁺ DN B cells differed from both IgG⁺ and IgA⁺ DN B cells in their significantly increased frequency of CD5⁺, CD10⁺, or CD38⁺ cells (Fig. 3B). In this regard, IgM⁺ DN B cells resembled naive B cells (Fig. 2D). IgM⁺ and IgG⁺ DN B cells showed a similar frequency of CD95⁺ cells, which was significantly lower than that of IgA⁺ DN B cells (p < 0.0001, Fig. 3B). The frequency of CD95⁺ cells within the IgA⁺ DN B cell population (Fig. 3B) was comparable to that of the CSM B cell population (Fig. 2D). IgG⁺ DN B cells further differed from IgA⁺ DN B cells in the frequency of CD38⁺ cells, which was significantly lower for IgG⁺ DN B cells (p = 0.02 HC and p = 0.0002 MS).

When comparing DN B cells of HC and MS patients, HC presented with a higher frequency of CD5⁺ (p = 0.03), CD10⁺ (p = 0.04), and CD95⁺ (p = 0.002) IgM⁺ DN B cells (Fig. 3B). Furthermore, higher frequencies of CD38⁺ and CD95⁺ IgG⁺ DN B cells (p = 0.0002 and p = 0.02, respectively) and IgA⁺ DN B cells (p = 0.03 and p = 0.0002, respectively) were observed in HC compared with MS patients (Fig. 3B).

Together, these data demonstrate that IgM⁺ DN B cells have the closest resemblance to naive B cells, whereas IgA⁺ DN B cells are most similar to CSM B cells, considering the expression of developmental surface markers.

To further look into the developmental pathways of DN B cells, we performed AIRR sequencing of DN B cells and compared their Ig repertoires to those of IgD⁻CD27⁺ memory (CSM) B cells. B cell subpopulations of IgD⁻CD27⁻ DN and IgD⁻CD27⁺ memory B cells were sorted from PBMC of HC (n = 4) and RRMS patients (n = 5) (Supplemental Fig. 1C, Table II). DN B cells were again elevated in the MS group relative to the HC group, although not statistically significant (Supplemental Fig. 1D), because of the limited sample size. We performed deep sequencing of the Ig repertoire of sorted DN and IgD⁻CD27⁺ memory B cells in MS patients (n = 5) and HC (n = 4). Specimens from two MS patients and one HC were excluded from this analysis because of low RNA quality, which can adversely affect accurate representation of the repertoire diversity. Therefore, sequencing of the VH and VL from sorted B cell subsets from MS patients (n = 3) and HC (n = 3) produced a total of 10,427,119 raw reads. After quality control and processing, a high-fidelity data set was generated that comprised 75,098 unique error-corrected H chain sequences and 212,846 unique error-corrected IgL and IgK sequences (Table III). As expected, the number of sequences from the DN population was lower than that of the CSM (Table III); however, in several instances, the DN/CSM sequence ratio was lower than that measured by flow cytometry (Supplemental Fig. 1D). This difference may be a consequence of several variables that affect sequencing output, such as RNA quality and nonlinear sequence recovery. Focusing on unique sequences equalizes the contribution of cells expressing different levels of mRNA but does not differentiate between BCR sequences from different B cells expressing identical receptors. It represents a sampling of the total cell population that can be used to estimate Ig repertoire features, as described below.

The DN VH repertoire was composed of a median of 21.7% (range, 8.7–64.1%) IgM, 43.6% (9.0–67.8%) IgG, and 23.9% (6.5–33.9%) IgA sequences, whereas the IgD⁻CD27⁺ memory VH repertoire consisted of 15.5% (8.1–43.3%) IgM, 31.3% (22.4–39.1%) IgG, and 52.1% (31.1–56.9%) IgA sequences (Fig. 4). A significantly higher proportion of IgA sequences was present in the IgD⁻CD27⁺ memory compared with the DN B cell repertoire of MS patients (p = 0.005 and FDR = 0.067) and HC (p = 0.008 and FDR = 0.067). For the L chain, IgK made up 53.3% (range, 42.4–62.1%) of the DN and 57.4% (44.1–70.6%) of the IgD⁻CD27⁺ memory B cell repertoire, followed by IgL in 46.7% (37.9–57.6%) of the DN and 42.6% (29.4–55.9%) of the IgD⁻CD27⁺ repertoire. No significant differences were observed between HC and MS patients in terms of isotype usage. Comparing DN and IgD⁻CD27⁺ memory B cells, the only difference was an increased proportion of IgA⁺ sequences in the IgD⁻CD27⁺ memory B cell population.

To test whether the DN and IgD⁻CD27⁺ B cell repertoires of MS patients and HC were characterized by conspicuous clonal expansions, sequences were clustered into clonal groups, with each member of the group representing one unique variant (including variants that differed only by isotype). In the IgD⁻CD27⁺ memory B cell population, expanded clones were identified in all MS patients and HC, but even the most abundant clones did not represent >1% of the repertoire (Fig. 5A). DN B cells, however, showed a small number of expanded clones in one out of three MS patients and one out of three HC. The largest identified clone size was 2% of the repertoire. Repertoire diversity of IgD⁻CD27⁺ and DN B cells was further explored by comparison of the evenness of both B cell populations. Evenness quantifies the extent to which an abundance distribution deviates from a uniform distribution. Diversity was higher in the IgD⁻CD27⁺ B cell samples when compared with DN B cells in both HC and MS patients, although this was not statistically significant (Fig. 5B).

Overall, repertoire diversity was not significantly different between the IgD⁻CD27⁺ and DN B cell populations. Nevertheless, in some donors belonging to both the HC and MS groups, DN B cells displayed more expansion and less repertoire diversity, suggesting that DN B cell oligoclonality could be donor specific.

We next examined whether DN B cells are related to IgD⁻CD27⁺ memory B cells by analysis of clonal overlap between both populations. Focusing on the H chain locus, overlap of IgD⁻CD27⁺ memory B cell clones with DN B cell clones ranged between 0.2 and 2.2% in HC and between 0.3 and 1.3% in MS patients (Fig. 6A). Interestingly, all isotypes of DN B cell sequences showed a connection to the IgD⁻CD27⁺ memory B cell compartment, although IgD⁻CD27⁺ memory B cell clones more frequently contained IgA⁺ DN B cell sequences than IgM⁺ DN B cell sequences (Fig. 6B, Supplemental Fig. 2). Thus, a small fraction of clones connected the DN and IgD⁻CD27⁺ memory B cell populations, although overall, these populations were clonally distinct.

Biased usage of IGHV, IGHD, or IGHJ genes has been demonstrated in the B cells present in some autoimmune diseases, including those present in the CNS of MS patients (3, 33–36). Therefore, the V(D)J family and gene usage was compared between DN and IgD⁻CD27⁺ memory B cells of HC and MS patients. Although the relative rankings of gene usage frequencies in both subsets were similar to naive B cells from a set of HC (Fig. 7), significant differences were apparent between them. IGHV2 family (p = 1.460 × 10⁻⁴ and FDR = 0.011) sequences were decreased in DN B cells (Fig. 7A), whereas IGLV3 (p = 2.810 × 10⁻⁴ and FDR = 0.028) was increased (Fig. 7B) when compared with IgD⁻CD27⁺ memory B cells of MS patients. Similar results were observed in DN B cells of HC, although not statistically significant. Frequencies of IGHV6 and IGHV7 family usage did not exceed 1.5% (data not shown). No significant differences were present in IGHD and IGHJ family usage between DN and IgD⁻CD27⁺ memory B cells of HC and MS patients (data not shown). When considering V(D)J gene usage (Fig. 7C), DN B cells of MS patients showed decreased usage of IGHV2-5 (p = 0.001 and FDR = 0.163) and increased usage of IGHV4-34 (p = 0.001 and FDR = 0.163) genes when compared with IgD⁻CD27⁺ memory B cells. No significant differences in V(D)J family or gene usage were found between HC and MS patients.

We next focused on IgM⁺ and IgG⁺ DN B cells separately. In MS patients, IgM⁺ DN B cells showed significantly higher usage of IGHV1 compared with IgM⁺ IgD⁻CD27⁺ memory B cells (p = 7.503 × 10⁻⁵ and FDR = 0.033, Supplemental Fig. 3A). Interestingly, IGHV4-34 gene usage was nonsignificantly elevated in IgM⁺ DN B cells compared with IgG⁺ DN B cells of HC and also in two out of three MS patients (Supplemental Fig. 3B). Thus, DN and IgD⁻CD27⁺ memory B cells, particularly IgM⁺, showed differences in V(D)J family and gene usage.

Somatic hypermutation (SHM) is associated with affinity maturation during the germinal center reaction in which B cells are exposed to Ag and T cell help and undergo multiple rounds of proliferation. Our data demonstrated SHM in both DN and IgD⁻CD27⁺ memory B cell sequences (Fig. 8). However, mean mutation loads of DN B cells were significantly lower than those of IgD⁻CD27⁺ memory B cells, both in the IgM⁺ (2 and 5%, respectively; p = 0.008 and FDR = 0.056) and IgG⁺ (4 and 7%, respectively; p = 0.009 and FDR = 0.056) isotype in HC (Fig. 8A). In MS patients, the same SHM pattern was observed in IgG⁺ sequences (p = 0.041 and FDR = 0.165). The higher SHM in IgD⁻CD27⁺ memory B cell sequences was consistent in all tested subjects except one (MS537, which had a very low number of sequences, Fig. 8B). There was no difference in IGHV mutation frequency of DN or IgD⁻CD27⁺ memory B cells between HC and MS patients.

Ag-selected Ig sequences usually have a higher frequency of mutations that lead to the replacement of amino acids in the CDR and a low frequency of such mutations in the framework region that confer structural stability (37, 38). As expected, both DN and IgD⁻CD27⁺ memory B cells showed negative selection in the framework region and neutral to positive selection in the CDR, according to BASELINe analysis of mutation patterns (Fig. 8C) (39).

The mutational load within clonal populations was also examined. Although we found that IgD⁻CD27⁺ memory B cells were in general more mutated than DN B cells, this was not the case within clones that included members from both subsets (Supplemental Fig. 4). The average mutation frequency was similar when comparing sequences in the same clone spanning the DN and IgD⁻CD27⁺ compartments in both HC and MS patients. Clones exclusive of the DN B cells showed the lowest average clonal mutation frequency (Supplemental Fig. 4).

Collectively, these results indicate that DN B cells are Ag-experienced cells but differ from IgD⁻CD27⁺ memory B cells when considering mutation load, except for clonally related DN and IgD⁻CD27⁺ memory B cells.

The hypervariable region CDR3 makes major contributions to B cell repertoire diversity as it is generated by random selection and recombination of V, D, and J gene segments in the H chain. Consequently, the physicochemical properties of CDR3 can reflect the function of the repertoire. Differences in CDR3 H chain properties are found between different B cell subsets with smaller, more hydrophilic, and more basic CDR3 for memory B cells compared with naive B cells (40). Longer CDR length, hydrophobicity, and positive charge of Abs have been associated with autoimmunity (40–43). In this study, differences were observed in polarity (p = 0.013 and FDR = 0.05), aromatic (p = 0.027 and FDR = 0.109), basic (p = 0.048 and FDR = 0.194), and acidic (p = 0.039 and FDR = 0.101) amino acid residue content between DN and IgD⁻CD27⁺ memory B cells of MS patients (Fig. 9A). DN B cells showed longer CDR3 (p = 0.032 and FDR = 0.128) in HC than IgD⁻CD27⁺ memory B cells (Fig. 9B). As DN B cell sequences showed a nonsignificant higher usage of the longer IGHJ6 family genes (data not shown), CDR3 length was analyzed in DN B cell sequences using the different IGHJ families. However, CDR3 regions of DN B cells were longer than that of IgD⁻CD27⁺ memory B cells, irrespective of the contiguous IGHJ family used (Fig. 9C).

We next considered CDR3 properties for different isotypes (Supplemental Fig. 3C). For HC, IgM⁺ B cells showed longer CDR3 in DN B cells compared with IgD⁻CD27⁺ memory B cells (p = 0.017 and FDR = 0.185). In MS, IgG⁺ B cells showed longer CDR3 in DN B cells compared with IgD⁻CD27⁺ memory B cells (p = 0.04 and FDR = 0.185). Further, the acidic and aromatic residue content was significantly higher for IgG⁺ DN B cells compared with IgG⁺ IgD⁻CD27⁺ memory B cells of MS patients (p = 0.005 and FDR = 0.114; p = 0.01 and FDR = 0.16, respectively). For HC, mean side-chain bulkiness was significantly higher for IgM⁺ DN B cells compared with IgD⁻CD27⁺ memory B cells (p = 0.004 and FDR = 0.048). When comparing IgM⁺ and IgG⁺ DN B cells, IgM⁺ DN B cells displayed a higher CDR3 charge compared with IgG⁺ DN B cells of HC (p = 0.003 and FDR = 0.081) (Supplemental Fig. 3C). No differences were observed in DN B cells between HC and MS patients for any of the CDR3 physicochemical properties. Thus, CDR3 sequences of DN B cells differ in length, side-chain bulkiness, and acidic and aromatic residue content from those of IgD⁻CD27⁺ memory B cells. IgM⁺ DN B cells further differ from IgG⁺ DN B cells in CDR3 charge.

As part of the H chain CDR3, the D segment is positioned in the center of the Ag binding site. D segments can be used in three different reading frames because of the addition and deletion of nontemplated nucleotides at the junctions between the recombining gene segments. Hydrophobic reading frames were previously reported to favor Ab self-reactivity and are normally counter selected in B cells from HC (44). For each D allele germline sequence in IMGT, we identified the D reading frames with the highest and lowest grand average of hydropathy (GRAVY) or hydropathy index to examine whether these were used differently between DN and IgD⁻CD27⁺ memory B cells in HC and MS patients. When analyzing the distribution of the frame usage difference between HC and MS for all the alleles, MS samples showed an increased usage of the frame with lower GRAVY index for DN B cells (p = 0.001, sign test) but not IgD⁻CD27⁺ memory B cells (Fig. 10A). D frame usage associated with hydrophobicity was similar for DN and IgD⁻CD27⁺ memory B cells in both HC and MS patients (Fig. 10B). Thus, the DN B cell population tended to use D frames with a lower GRAVY index in MS patients compared with HC.

In this study, we report the absence of major differences in both the developmental marker expression and the Ig repertoires of DN B cells in HC and MS patients. This leads us to suggest that these cells have a similar origin in both cohorts. Elevated frequencies of DN B cells could be triggered because of the aging process that is common in the immune system of the elderly (45) and occurs prematurely in a proportion of MS patients (46). Aging of the immune system is accompanied by a low-grade chronic inflammation, termed “inflammaging,” that is characterized by an increase in proinflammatory cytokines (IL-6, IL-15, IL-8), an increase in other inflammatory mediators (such as coagulation factors), and subclinical infections with common viruses (47). This chronic inflammatory environment combined with genetic predisposition could result in the expansion of DN B cells in aged individuals and a proportion of MS patients.

Flow cytometric analysis confirmed the finding of abnormal elevations in the frequency of DN B cells in the peripheral blood of MS patients. The proportion of MS patients (29.5%) and HC (9.5%) younger than 60 y with an elevated frequency of DN B cells was higher compared with our previous study (20% MS, 3% HC) (11), which could be explained by the higher number of included individuals in the current study.

The isotype distribution of DN B cells from HC and MS patients was similar in our flow cytometry and AIRR sequencing analyses. IgG⁺ cells made up the largest proportion of DN B cells, which is in agreement with our previous results (11) and with studies in aged healthy individuals (12, 48) and RA patients (17). The finding of a significantly lower frequency of IgA⁺ cells and sequences in DN versus CSM B cells is also in line with findings in healthy individuals (26) and RA patients (17). It has previously been suggested that IgA⁺CD27⁻ and IgA⁺CD27⁺ memory B cells are generated via separate response pathways because of molecular differences between these cell populations (49). Thereby, IgA⁺CD27⁻ B cells demonstrated a germinal center–independent origin and were suggested to be a blood counterpart of IgA⁺ cells from the gut lamina propria. Another report, however, indicated a high level of clonal relationships between the IgA⁺CD27⁻ and IgA⁺CD27⁺ B cell pool, with only few clonal relationships between germinal center–independent IgD⁺CD27⁺ B cells and IgA⁺CD27⁻ B cells (50).

DN B cells are mature Ag-experienced B cells with an expression profile of developmental markers more similar to CSM than naive B cells. Their mature state was demonstrated by the low expression of CD38 and the immature markers CD5 and CD10. Previously, the majority of DN B cells in SLE patients were also shown to be CD5⁻ and CD10⁻ (19). The absence of CD38 is further characteristic of CD21^lowCD11c⁺T-bet⁺ age-associated B cells (21, 51) and DN B cells in SLE (16). The most prominent difference between DN and CSM B cells, however, was the lower mutation load of DN B cell Ig repertoires in HC and MS patients. This is in agreement with previous reports in HC (23, 25, 26, 48), SLE (16, 19), and RA patients (17). Overall, DN and CSM B cell populations were clonally distinct, with differences in V(D)J family and gene usage, which could indicate differential activation pathways or differences in the Ags driving the response. Furthermore, the intermediate frequency of CD5⁺, CD38⁺, and CD95⁺ cells in the DN B cell population between that of the naive and CSM B cell populations could point to an intermediate developmental state. Our results are in line with a previous report of increased CD95 transcripts in CD27⁺ versus CD27⁻ memory B cells in a gene expression profile analysis (52). Nevertheless, a small fraction of clonally related cells with similar mutation loads was demonstrated in DN and IgD⁻CD27⁺ memory B cells. Therefore, part of DN B cells could be exhausted memory B cells that have lost CD27 expression because of chronic Ag stimulation, as previously suggested (48, 53). Taken together, our results indicate that the majority of DN B cells show earlier maturation features compared with CSM B cells and are formed via differential activation pathways, whereas a small part of DN B cells are clonally related to CSM B cells and could be exhausted memory B cells.

IgG⁺ and IgA⁺ DN B cells displayed several characteristics of memory B cells, including the absence of CD5 and CD10 surface expression, a class-switched state, and SHM. Given that the Ig mutation frequency was lower in class-switched CD27⁻ versus CD27⁺ B cells and that isotype switching occurs before SHM during the germinal center response (54), we propose that class-switched DN B cells are involved in (primary) T cell–dependent immune responses that prematurely exit the germinal center response. In line with this, a similar number of cell divisions and Ig mutation level was reported for DN and germinal center B cells (49). In addition, the lower mutation load of DN B cells could be due to counterselection mechanisms. AIRR sequencing provided some suggestive data for autoreactivity of a subset of the DN B cell population. IGHV4-34 gene segments were increased in DN B cells of MS patients when compared with IgD⁻CD27⁺ memory B cells. Previous reports demonstrated the expression of 9G4 (VH4-34), which encodes intrinsically autoreactive B cells and Abs, by DN B cells in SLE (15) and by B cells and plasma cells in MS brain and cerebrospinal fluid (33–35, 55, 56). The longer CDR3 length of DN B cells compared with IgD⁻CD27⁺ memory B cells could further point to autoreactivity of a proportion of DN B cells (40, 41, 43). Contradictory to these postulations is the tendency of DN B cells and IgD⁻CD27⁺ memory B cells of MS patients to select for D frames with a lower hydrophobicity when compared with those of HC. Hydrophobic D reading frames have previously been described to favor Ab self-reactivity, although low patient numbers were used (44).

IgG⁺ DN B cells could further be discriminated from IgA⁺ DN B cells because of their lower frequency of CD38⁺ and CD95⁺ cells and, thus, their decreased activation status. Although IgA⁺ DN B cell sequences showed the highest clonal overlap with IgD⁻CD27⁺ memory B cells, overall, these populations were clonally distinct. It thus remains unclear whether IgA⁺ DN B cells represent germinal center–independent or –dependent cells.

IgM⁺ DN B cells showed similarity to naive B cells in their frequency of CD5⁺, CD10⁺, and CD38⁺ cells. Although there was no significant difference in Ig mutation load of IgM⁺ versus IgG⁺ DN B cells, the lower frequency trend we observed in IgM⁺ was consistent with previous findings (23). The increased IGHV4-34 usage in IgM⁺ versus IgG⁺ DN B cells was not significant, but the trend we observed could be explained by the negative selection that naive VH4-34⁺ B cells tend to undergo, which results in their underrepresentation in the memory compartment (57). Further, naive B cells were previously reported to express more negatively charged H-CDR3s than memory B cells (32). The increased CDR3 charge of IgM⁺ versus IgG⁺ DN B cells of HC could therefore point to an alternative developmental pathway. IgM⁺ DN B cells could display characteristics of IgD⁺CD27⁺ non–class-switched memory B cells, which have been termed “innate-like” B cells as they develop independent of germinal centers and can respond to T cell–independent Ags (58). This is in line with the described repertoire similarity of IgD⁻IgM⁺CD27⁻ or IgD⁻IgM⁺CD27⁺ memory B cells and IgD⁺CD27⁺ B cells (23). Nevertheless, a proportion of the IgM⁺ DN B cells showed clonal overlap with IgD⁻CD27⁺ memory B cells. These results lead us to suggest that part of the IgM⁺ DN B cells could be precursors of CD27⁺ memory B cells in T cell–dependent immune responses. They could be involved in the extrafollicular B cell response that produces an initial burst of low-affinity Abs early after B cell activation. This response can occur with or without T cell help and does not induce SHM (59).

A strength of this study is the combination of protein expression of developmental surface markers and Ig repertoire analysis to study the origin and selection characteristics of DN B cells in HC and MS patients. Results obtained in the flow cytometry analysis were further explored in the AIRR sequencing. Although our Ig repertoire dataset, which included 287,944 unique sequences, was large, the number of included HC and MS patients was limited. Inclusion of a larger cohort of MS patients, including subjects that represent the different clinical subtypes, is necessary to study the selection and differentiation mechanisms of DN B cells in relation to disease subtype and severity, age, and response to treatment.

DN B cells of MS patients resemble DN B cells described in SLE and RA patients in several characteristics: surface expression of IgM, IgG, CD10, and CD38 (15, 60); Ig mutation load (15, 16); and expression of VH4-34–encoded Abs (16). In RA, the distribution of IgM⁺, IgG⁺, and IgA⁺ cells within the DN B cell population differed from our results (17, 18), although IgG, again, was the predominant isotype (17). Positive correlations between DN B cell frequency and disease activity or autoantibody titers have been demonstrated in SLE (15), but not in RA (18), and remain to be determined in MS. In RA, DN B cells correlated with a clinical response to IL-6R inhibition therapy, namely tocilizumab (17). It remains unknown whether DN B cells resemble the T-bet–driven CD21^low B cells. CD21^low B cells in mice are activated by TLR7/9 stimulation in combination with IFN-γ and IL-21, which can be amplified by BCR ligation (61, 62). Therefore, it was proposed that Ags with a capacity to engage TLRs, including viral particles containing nucleic acids or self-antigens associated with RNA or DNA, are able to drive T-bet expression in B cells in which the BCR could play a role by amplifying T-bet expression through delivery of nucleic acids containing Ags (61, 62). Recently, DN B cell expansions in SLE were assigned to a subset of CXCR5⁻CD21⁻CD11c⁺T-bet⁺ cells (16). These “DN2” cells differentiated into autoantibody-producing plasma cells driven by TLR7, which led to their characterization as extrafollicular B cells responding to innate stimuli. As our previous study indicated, ∼20% of DN B cells could be retraced to the CD21^lowCD11c⁺ B cell gate (11); we expect that the majority of DN B cells in MS patients are not DN2 cells. This could be explained by the difference in pathological mechanisms in SLE and MS that center around Ab-mediated and cell-mediated immune mechanisms, respectively. Moreover, a subset of DN B cells with cytoplasmic FOXO1, a transcription factor involved in B cell development, was found to be increased in SLE patients and positively correlated with disease activity (60). Whether this DN B cell subset corresponds with DN2 cells is not clear. Involvement of the T-bet–driven pathway in the abnormal elevations of DN B cells in MS remains to be determined.

In conclusion, developmental surface marker expression and Ig repertoire characteristics suggest that DN B cells share a common developmental pathway in HC and MS patients that leads to expansion of these cells because of inflammaging. We postulate that DN B cells consist of B cells with different origins dependent on the isotype. IgG⁺ DN B cells are more related to CSM B cells, whereas IgM⁺ DN B cells are more similar to naive and non–class-switched IgD⁺CD27⁺ memory B cells. Further study is warranted to examine the downstream pathways that lead to expansion of DN B cells in a proportion of MS patients and whether DN B cells play a role in MS pathology.

We thank Igna Rutten, Kim Ulenaers (Hasselt University and University Biobank Limburg, Hasselt), and Anita Knevels (Revalidation and MS-Center, Pelt) for patient recruitment and sample collection. The authors also thank Drs. Song Chen and Eileen Dimalanta (New England Biolabs, MA) for providing technical guidance during the B cell sequencing library production.

K.C.O. has received speaking fees provided by New England Biolabs. The other authors have no financial conflicts of interest.