The Plasma Cell Infiltrate Populating the Muscle Tissue of Patients with Inclusion Body Myositis Features Distinct B Cell Receptor Repertoire Properties

Sporadic inclusion body myositis (IBM) is the most common idiopathic inflammatory myopathy (IIM) in patients >50 y of age (1). IBM leads to progressive muscle atrophy, and patients experience asymmetric muscle weakness, predominantly involving finger flexors and quadriceps. Muscle histopathology usually shows inflammatory infiltrates with rimmed vacuoles (2). IBM is refractory to conventional immunotherapy in contrast to dermatomyositis (DM) and polymyositis (PM). The disease pathophysiology is not clearly understood but is thought to consist of a complex network of inflammatory, autoimmune, and degenerative mechanisms.

Immune cell infiltrates are invariably found in the muscle tissue of patients with IBM, DM, or PM. Those populating IBM muscle tissue include a large fraction of CD8⁺ T cells and fully differentiated CD138⁺CD20⁻ Ab-secreting plasma cells with scarce CD20⁺ naive or memory B cells (3–5). The elevated myonuclei reactivity of IBM serum (6) and identification of autoantibodies targeting cytosolic 5′-nucleotidase 1A (cN1A) suggest the existence of a selected population of autoantibody-secreting plasma cells in IBM that could be relevant to disease pathology (7–10). Additionally, clonally related plasma cells in the absence of visible germinal centers were identified via Sanger sequencing of laser-microdissected IBM tissue (11). This was an unexpected finding given that clonally expanded B cells observed in the affected tissue compartments of autoimmune diseases such as the myasthenia gravis (MG) thymus, rheumatoid arthritis joints, and Sjögren’s syndrome (SS) glands are characteristically associated with conspicuous, tissue-localized germinal centers with B cell follicles that contribute to the local B cell response (12–14). Aside from the known predominance of plasma cells in IBM and PM tissue, differences in the repertoire of B cells in the muscle tissue of IBM from IIM tissue have yet to be identified. Moreover, differences between the plasma cells found in IBM relative to Ag-experienced B cells in the circulation have also not yet been characterized. These differences may be important for understanding their role in the disease pathology.

Accordingly, we sought to compare the BCR repertoire in IBM muscle tissue to the repertoire found in DM and PM muscle tissue, and also to circulating peripheral subsets of Ab-secreting cells (ASCs) from vaccinated individuals and to CD27⁺ memory B cells from healthy donors. We applied adaptive immune receptor repertoire sequencing, as this approach provides large BCR repertoire datasets and, importantly, the data are not influenced by PCR-based biases that occur in Sanger-based sequencing approaches, including those used in previous studies of IBM infiltrates (11, 15). We found that the plasma cells found in IBM tissue reflect the hallmarks of Ag-driven selection and differentiation, including isotype class switching, accumulation of somatic mutations, and large clonal expansions. The BCR repertoire properties of the IBM infiltrate differed from those found in circulating ASC populations and in DM and PM infiltrates, and included unswitched IgM-expressing B cells with elevated somatic mutation frequency and unique CDR3 physiochemical properties. These findings collectively suggest that the plasma cells present in IBM muscle may be selected by a distinct set of Ags.

Deidentified muscle specimens and peripheral blood samples were collected under an exempt protocol approved by the Human Research Protection Program at Yale School of Medicine and the University of Connecticut School of Medicine. Additional tissue from muscle biopsy samples collected for histopathological diagnosis from three PM, three DM, and four IBM patients were snap-frozen and stored at −80°C immediately after collection. These biopsies were performed for clinical indications, independent of the current study. The diagnoses of PM, DM, and IBM were based on clinical and histopathological findings, including immunohistochemical stains and electron microscopy. However, given the deidentified nature of the samples, detailed clinical information on the muscle biopsy samples were not available.

The IIM biopsy tissue was prepared for sequencing by first isolating RNA from frozen sections using the Qiagen RNA mini kit according to the manufacturer’s instructions. RNA libraries were made using reagents supplied by New England Biolabs (provided by Eileen Dimalanta and Chen Song, Ipswich, MA) as part of the NEBNext immune sequencing kit, using approaches we have previously described (16, 17).

The healthy donor BCR repertoire data originated from previous studies performed by our group. Specifically, the repertoires of B cell subsets isolated using CD27⁺ flow cytometry–based sorting of the peripheral blood of six healthy donors (HDs), four of whom were used in a previous study on MG (18) and two of whom were used in a previous study on multiple sclerosis (19). Moreover, reference repertoires and gene expression for ASC subsets were obtained from three studies: single-cell repertoires from a previous study on MG (17), bulk repertoires from a previous study on influenza vaccination (20), and a third single-cell transcriptome generated for another study (M. Wang, R. Ruoyi Jiang, S. Mohanty, H. Meng, A.C. Shaw, and S.H. Kleinstein, submitted for publication) from a subject 7 d after influenza vaccination (used solely to visualize the ASC-specific expression of gene markers later examined in the analysis of microarray transcriptomes). This collection of sequence data are referred to as the ASC group throughout this manuscript.

For the single-cell library preparation, a human pan-B cell immunomagnetic negative selection kit (STEMCELL Technologies) was used to isolate B cells from frozen PBMCs after thawing per the manufacturer’s instructions. B cells were loaded directly into the Chromium Controller (10x Genomics) and prepared using the Chromium single-cell 5′ reagent kit (10x Genomics) for version 3 chemistry per the manufacturer’s protocol.

Standard preprocessing for V(D)J repertoires was carried out, as previously described (16, 17) using pRESTO v0.5.4 (http://presto.readthedocs.io). Briefly, reads with a Phred-like score <20, without a C region (maximum error, 0.2) or template switch site (maximum error, 0.5), were removed and excised from reads. A unique molecular identifier (UMI) annotation was added to each read corresponding to the 17-nt barcode preceding the template switch site. Reads with the same UMI were multiple aligned using MUSCLE v3.8.31 (21). The following steps were then applied to correct for sequencing errors in the UMI region. A threshold was identified for clustering UMIs by constructing a histogram from all pairwise Hamming distances by sampling 10,000 UMI sequences and finding the minima below a normalized Hamming distance of 0.25 (expected Hamming distance between two random 4-nt sequences). Sequences were then clustered into groups based on their UMIs using this threshold by the CD-HIT-EST algorithm (22). A histogram for all pairwise distances within each group was then computed and summed together to identify a minimum threshold for clustering below a normalized Hamming distance of 0.25. A default threshold of 0.8 was used when the optimal threshold was <0.8 given clustering algorithm limitations. Reads with the same UMI/read cluster annotation were assigned to the majority sample to remove multiplex index misassignments. Reads with the same final UMI/read cluster annotation were collapsed into a single consensus sequence and discarded when the error was >0.1 or the majority isotype assignment was <0.6. Mate pairs were assembled first by overlap with a minimum allowed overlap length of 8 bp, a maximum error rate of 0.3, and a p value threshold of 1 × 10⁻⁵. Those failing this initial assignment were assembled by alignment against the March 30, 2017 version of IMGT V-gene germlines with a minimum identity of 0.5 and a E-value threshold of 1 × 10⁻⁵ (23). A second isotype assignment was made by alignment of the last 100 bp of the 3′ end of the of the consensus sequences against possible C region primer sequences derived from IMGT C region reference sequences (maximum error, 0.4). Duplicate sequences were discarded. Sequences that were represented by only a single sequence in a single UMI/read cluster were discarded.

V(D)J germline genes were then assigned with IgBLAST v1.7.0 (24) using the March 30, 2017 version of the IMGT gene database. Nonfunctional sequences were then removed from further analysis and assigned into clonal groups using Change-O v0.3.4 (25). First, sequences were partitioned based on common IGHV gene annotations, IGHJ gene annotations, and junction region lengths. IGHV and IGHJ gene annotations were determined by the union of ambiguous gene assignments with each junction length partition having at least one overlapping gene annotation. Sequences were then clustered based on the within-group junction Hamming distance by single linkage using a threshold determined by finding the minimum from the fit of a bimodal distance-to-nearest histogram obtained using a kernel density estimate (26, 27). Full-length germline sequences were reconstructed for each clonal cluster (VH) or sequence (VL) with D segment and N/P regions masked (replaced with Ns); any ambiguous gene assignments within clonal groups were resolved by majority rule. Sequences shorter than 200 bp or with >10% “N” ambiguous nucleotides were removed. A sliding window of 10 nt with a maximum of 5 nt mismatches was used to remove possible chimera sequences. Mutations and selection strength of framework regions (FWRs) and CDRs were quantified using the implementation of BASELINe in SHazaM v0.1.8 in R v3.4.2 (R Core Team, n.d.). To remove the confounding effect of VH gene usage on the ratio of replacement and silent mutations, and given that the somatic hypermutation (SHM) frequency was elevated in the IgM-switched B cells in IBM in comparison with all others, our examinations were restricted to IgG-switched V(D)J sequences where specified.

Alakazam v0.2.7 in R was used for analysis of CDR3 physiochemical properties and diversity analysis (28). Diversity analysis was performed using methods that account for low clone numbers and variability in sampling depth. The generalized Hill index with uniform downsampling to the minimum recovered sequence count from any sample was computed. A total of 2000 replicates of bootstrapping from the inferred clonal abundance derived using a one-parameter Chao estimator was performed. Briefly, clones were determined by grouping V(D)J sequences utilizing identical IGHV and IGHJ genes with equivalent length junctions (CDR3 sequences) and clustered into clones based on a Hamming distance threshold. Of note, automated thresholds identified for cloning did not differ significantly between myositis samples and HD CD27⁺ samples for all pairwise comparisons (Supplemental Fig. 1). To further control for different numbers of B cells in tissue sections, abundance calculations were performed by downsampling to an equivalent number of unique V(D)J sequences based on the count in the sample with the fewest sequences. Consequently, biases in our analysis from differences in sequencing depth were not expected to impact our calculation of diversity for the switched compartment.

We also analyzed the observed patterns of SHM found in B cells and compared them to patterns expected under a setting without selection. We first assessed the overall frequency of hot-spot (WA/TA, WRC/GYW) versus cold-spot (SYC/GRS) to confirm that patterns of SHM observed in IIM samples are consistent with activation-induced cytidine deaminase (AID) and SHM targeting and not sequencing error. That is, we assessed the overall frequency of hot-spot (WA/TA, WRC/GYW) versus cold-spot (SYC/GRS) in samples with >100 VDJ sequences to confirm that the patterns of SHM observed in IIM samples are consistent with AID targeting using the createMutabilityMatrix function in SHazaM v0.1.8 (29, 30).

We then quantified selection in terms of differences in the expected versus observed replacement/silent mutation ratio given models for SHM using BASELINe (28). BASELINe uses an empirically derived model of SHM to quantify the expected number of replacement or silent mutations and uses a binomial model to test whether the observed numbers of these types of mutations differ—removing the confounding effect of VH gene usage on the ratio of replacement and silent mutations. Furthermore, to examine whether differences in the selection of the IgG BCR repertoire reflect selection for the enrichment of certain receptor specificities in IBM, we examined the usage of different VH and JH genes or CDR3 sequences with specific biochemical properties.

For the single-cell library, samples were sequenced on the NovaSeq 6000 with 100 × 100 or 150 × 150 paired-end reads for gene expression and BCR libraries, respectively. The cellranger mkfastq function from 10x Genomics v2.2.0 was used to convert base calls to fastq sequences and perform demultiplexing; sequences were then aligned to the GCRhg38 genome supplied by 10x Genomics. Sparse count matrices, barcode assignments, and feature calls were made using the cellranger count function, and Seurat v.2.3.4 was used for visualization and cluster assignment. Cells with <50 genes detected or high mitochondrial content >0.2 of all transcripts were discarded. Log-normalized count values (31) were used to assign clusters and compute t-distributed stochastic neighbor embedding from highly variable genes identified using FindVariableGenes. The first 10 dimensions were used for principal-component analysis (PCA) based on the location of the inflection point from PCElbowPlot. FindClusters was then used (default threshold of 0.8) to identify clusters (shared nearest neighbor clustering after PCA). A cluster was annotated as being associated with ASCs based on the visualization of a cluster with the highest expression of CD38, BLIMP1, and CD138. V(D)J sequences were reconstructed using the cellranger vdj function from fastq reads. V(D)J germline segments were reassigned using IgBLAST v.1.7.034 with the June 21, 2018 version of the IMGT gene database. Cells with multiple IGH V(D)J sequences were assigned to the most abundant IGH V(D)J sequence based on UMI count, and nonfunctional sequences were excluded; V(D)J sequences were then assigned into clonal groups using a default threshold of 0.1 identified in the same manner as for bulk repertoires. IGH sequences were then further grouped based on whether the cells shared any common combinations of IGKV with IGKJ or IGLV with IGLJ gene annotations for paired L chains. Germline sequences were reconstructed for each clonal cluster with masking of D segment and N and P regions as described previously. V(D)J reads associated with cells belonging to the ASC cluster were extracted for further analysis.

Three processed microarray gene datasets (GSE3112, GSE39454, GSE48280) examining IBM and control tissue biopsies (3, 32, 33), available through National Center for Biotechnology Information GEO, were downloaded and annotated with the appropriate package for each platform using GEOquery (34). Of note, the gene set associated with GSE3112 was not normalized; a log₂ transformation of the gene expression matrix was performed to permit comparisons with other gene sets (3). We selected a set of genes of interest relevant to B cell biology and the SHM process. Statistical testing of microarray gene expression was performed in R using the base pairwise.wilcox.test function. Meta-analysis of the p values from each of the three studies for this set of genes was performed using Stouffer’s method from the metap v1.0 R package (30).

All statistical testing of repertoire features was performed in R 3.6.0. For comparison between groups, a t test was used. A p value of <0.05 was considered to be statistically significant.

Sequencing data are available at the GEO database (https://www.ncbi.nlm.nih.gov/geo/) under study accession number GSE227124, as well as other publicly available datasets (Supplemental Table I).

We generated BCR repertoires, using high-throughput unbiased adaptive immune receptor repertoire sequencing, from the muscle biopsy samples of four patients with IBM, three patients with DM, and three patients with PM. All patients had confirmed clinical diagnoses, which included histology and electron microscopy (Supplemental Fig. 2, Table I). Muscle biopsies from patients without an inflammatory myopathy diagnosis were not used as controls owing to the sparsity of recoverable BCR sequences from these samples (8). It has been demonstrated that the B cells in the IBM infiltrate are predominantly comprised of differentiated plasma cells (4). Accordingly, we reasoned that Ag-experienced B cells would provide the best comparison with this mature infiltrate. To this end, reference BCR repertoires included CD27⁺ memory B cell subsets (HD CD27⁺) from healthy donors (herein referred to as B_mem), and ASC subsets from patients with an autoantibody-mediated autoimmune disease (MG) and vaccinated healthy donors (see Materials and Methods). All repertoires included the full-length H chain V(D)J regions and the sequence-identified isotypes, that is, unswitched (IgM) and the switched isotypes (IgG and IgA). The myositis-associated muscle biopsies repertoires included 4899, 5284, and 4714 unique sequences from the DM, PM and IBM specimens, respectively (Table II). The repertoires from the B_mem and ASC cohorts included 37,007 and 463,896 unique sequences, respectively.

To quantify the presence of clonal expansions in IBM tissue compared with DM, PM, and to the HD CD27⁺ B_mem and ASC populations, we first visualized the distribution of clone sizes from repertoire sequencing of the IgG isotype, given the frequent involvement of this isotype in human autoimmune disease (Fig. 1A, 1B). Comparisons of diversity and abundance involved analysis of only bulk repertoire sequences. As expected, IgG-expressing B cell clones in the B_mem population were observed to have a higher Simpson’s diversity (low clonality) relative to those from the ASC population, where a low Simpson’s diversity (greater clonality) was observed (p = 0.006), given the focused Ag response in this group. B cell clones in IBM and DM were also observed to have lower Simpson’s diversity (greater clonality) relative to HD CD27⁺ B_mem (p = 0.039 and 0.019, respectively). No significant difference between HD CD27⁺ B_mem and PM was observed (p = 0.22). Thus, the infiltrating switched (IgG) B cell clones in IBM as well as in DM are clonally expanded relative to circulating HD CD27⁺ B_mem populations, demonstrating expansion similar to ASCs. One sample (IBM4) from the IBM patient cohort demonstrated considerable clonal restriction, exceeding that measured in all of the other samples. An illustrative example of an expanded B cell clonal family present in an IBM muscle infiltrate is shown in Fig. 1C and 1D. These clonal variants (produced during the evolution of an immune response) are members of the same clone and contain both shared and unique mutations, which arise by affinity maturation processes. The extensive clonality of the IBM infiltrate repertoire suggests that Ag-specific proliferation of selected clones has occurred.

We next investigated isotype usage and the frequency of somatic mutations in the variable regions. Both isotype class switching and mutation analyses can approximate Ag exposure and experience of the BCR repertoire. The isotype usage in both the HD CD27⁺ B_mem and ASC populations included IgG, IgM, and IgA (Fig. 2A). The HD CD27⁺ B_mem population included a large fraction of IgM-expressing cells. A consistently higher frequency of IgG isotype usage in IIM samples (60–68% average per status) versus HD CD27⁺ samples (22%) was observed. HD CD27⁺ B_mem populations were observed to have a higher fraction of IgM (p = 0.005 for IBM versus HD CD27⁺, p = 0.003 for DM versus HD CD27⁺, p = 0.003 for PM versus HD CD27⁺, p = 0.028 for ASC versus HD CD27⁺) and lower fraction of IgG (p = 0.001 for DM versus HD CD27⁺, p = 0.026 for IBM versus HD CD27⁺). These findings demonstrate that B cells in IIM tissue are largely switched to IgG expression, a feature distinct from circulating HD CD27⁺ B_mem cells.

The distribution of somatic mutation frequencies (Fig. 2B) showed that although IBM infiltrates include very few IgM-expressing B cells, it was interesting to find those that are present had a higher mutation frequency than DM and PM infiltrates and also higher than the control ASC and B_mem repertoires (p = 0.012 for IBM versus HD CD27⁺, p = 0.011 for IBM versus DM, p = 0.024 for IBM versus PM) and ASC populations (p = 0.028). The average mutation frequency of IBM IgG-expressing B cells was more elevated than circulating B_mem IgGs (p = 0.022) but was similar to that of DM, PM, and ASC populations (p = 0.279 for IBM versus DM, p = 0.198 for IBM versus PM, p = 0.815 for IBM versus ASC). Differences in the mutation frequency of the IgA-expressing cells were found only between the B_mem and ASC repertoires. The isotype class switch to IgG and IgA as well as the elevated somatic mutation frequency in IgG and IgM found in the IBM infiltrate suggest that these B cells are Ag experienced.

Given that the IBM infiltrate included an elevated mutation frequency, we next sought to determine whether mutations in the IBM-derived BCR sequences were introduced by the AID-governed SHM mechanism (Fig. 3). The relative frequency of mutations at hot-spot (WA/TW and WRC/GYW) and cold-spot (SYC/GRS) motifs in the IIM and control BCR repertoires was assessed to quantify SHM targeting preferences (29, 30). No significant differences were observed for all pairwise comparisons between the IIM and control repertoires (Fig. 3A), suggesting similar AID and error-prone repair pathways operating across IIM and control BCR repertoires.

With this possible clonal enrichment ruled out, we then quantified Ag-driven selection in terms of differences in the expected versus observed replacement/silent (R/S) mutation ratio using BASELINe (28), which is an empirically derived model that provides measures of selection pressure. The FWRs are critical to preserving the BCR structure and thus they show evidence of negative selection (silent mutations are more frequently observed), whereas positive selection (replacement mutations are more frequently observed) is found in the CDRs, as they include Ag contact residues. The IgG-switched BCR sequences from ASCs displayed increased selection pressure relative to B_mem (p = 0.023 for CDR and p = 0.048 for FWR) (Fig. 3B). The IgG-switched sequences from PM and IBM samples, but not DM samples, displayed decreased selection pressure compared with ASCs for CDR (p ≤ 0.001 for ASC versus PM, p ≤ 0.001 for ASC versus IBM) and for FWR (p ≤ 0.001 for ASC versus PM, p ≤ 0.001 for ASC versus IBM) regions (Fig. 3B). Moreover, IBM had the lowest BASELINe selection pressure compared with DM for CDR (p ≤ 0.001 for IBM versus DM, p = 0.013 for IBM versus PM) and FWR (p ≤ 0.001 for IBM versus DM, p = 0.095 [not significant] for IBM versus PM) regions. These findings suggest that there are unique differences in the Ag-driven selection pressure that gave rise to the resident B cell infiltrate in IBM and PM compared with the other IIM subsets and circulating ACSs and B_mem populations.

To examine whether differences in the selection of the BCR repertoire reflect selection for the enrichment of certain receptor specificities, we next examined the usage of different VH and JH genes across IIM and control BCR repertoires. DM but not IBM displayed slightly increased usage of VH1 (p = 0.043 versus HD CD27⁺ and p = 0.031 versus ASC) and decreased usage of VH3 (p = 0.039 versus HD CD27⁺) (Fig. 4A). We next assessed the differential expression of four VH genes frequently investigated in the context of autoimmunity: VH1-18, VH1-69, VH3-23, and VH4-34 (Fig. 4B). DM was observed to use more VH1-18 (p = 0.019 versus HD CD27⁺) as well as less VH3–23 (p ≤ 0.001 versus HD CD27⁺). IBM was observed to use less VH1-18 compared with DM (p = 0.025). No other significant differences were observed in JH gene usage (Fig. 4C) between different IIMs, except minor differences in DM and IBM when compared with the HD CD27⁺ B cell repertoire.

Given the importance of the CDR3 region for Ag contact, the CDR3 region of V(D)J sequences from IBM samples were then examined (Fig. 5). Several physiochemical properties were broadly observed to be associated with IBM samples, including increased hydrophobicity. Higher GRAVY hydrophobicity scores and aliphatic index for IBM-associated IgM compared with DM were observed (p = 0.044 and 0.047, respectively). Differences were also observed for IgAs in comparisons between IBM and PM: IBM was associated with a higher aliphatic index (p = 0.009) and lower aromatic index (p = 0.026). Moreover, a higher overall aliphatic index was observed for IBM for IgM (p = 0.028 versus HD CD27⁺, p = 0.009 versus ASC), IgG (p = 0.038 versus HD CD27⁺), and IgA (p ≤ 0.001 versus HD CD27⁺, p = 0.002 versus ASC, and p = 0.005 versus DM). Likewise, a lower aromatic index was observed for IBM for IgM (p = 0.013 versus HD CD27⁺, p = 0.018 versus ASC), IgG (p = 0.029 versus HD CD27⁺), and IgA (p = 0.029 versus HD CD27⁺, p = 0.009 versus ASC). The differences in amino acid properties in the CDR3 region in IBM, especially in the IgM-expressing population, which also included elevated somatic mutation frequency, may suggest differences in selective mechanisms and Ag exposure.

Given that our analysis of the IBM plasma cell infiltrate indicated that this repertoire includes properties consistent with Ag-driven selection and affinity maturation, we next investigated whether these processes occur within the muscle tissue. To this end, we first examined IBM muscle tissue for the expression of germinal center–associated gene expression signatures, including AID, querying publicly available IBM muscle-derived microarray datasets. An upregulation of AID in IBM tissue relative to healthy control was not observed given that there were no differences in the AICDA gene expression and low BCL6 expression (Supplemental Fig. 3, Table III). However, we did observe statistically significant (p < 0.05) increases in expression of plasma cell signatures (SDC1, XBP1, PRDM1), including low PAX5 expression (which is downregulated at the stage of plasma cell expression). Furthermore, transcription factors related to plasma cells (IRF4, IRF8) and cell cycle (increased CDK1 expression but not for MYC) were increased. Moreover, a gene marker for inflammation (TNF) was also upregulated.

Germinal centers are classically associated with Ag-selected B cells undergoing proliferation, affinity maturation, and differentiation. Accordingly, we examined the IBM muscle tissue specimens used in this study for the presence of germinal center structures by histology. As in previous studies (11), no follicular structures classically associated with germinal centers were observed in H&E-stained sections for all myositis patients examined (Supplemental Fig. 2). Thus, although the plasma cells found in IBM tissue reflect the hallmarks of Ag-driven selection and differentiation, the absence of local AID expression and germinal centers suggests that this infiltrate is recruited to the tissue and then proliferated locally in the muscle.

In this study, through the application of high-throughput BCR sequence analysis, we have identified differences in the IBM B cell repertoire compared with both DM and PM, and to circulating reference B cell subsets, including Ag-driven ASC populations. We specifically identified remarkable clonal expansions as well as unique features associated with the phenotype of infiltrating B cell subsets that included prominent IgG and IgA isotype usage, an elevated SHM frequency of all isotypes, and a unique selection pressure profile. These findings highlight the distinct features of the IBM infiltrates in muscle tissue that reinforce the inflammatory nature of IBM (35).

We first demonstrated that clonal expansions in IBM are consistent in size with those associated with other myositis subtypes despite the absence of visible germinal centers. These findings replicate previous studies, notably our own (15) and those from other groups (8) who have also observed the absence of B cell follicles and presence of clonal variants from microdissected tissues consistent with clonal expansions (11, 36). Moreover, these observations are similar to the findings of CD8 T cell oligoclonal expansions noted in prior studies of IBM tissue (37).

We found increased IgG isotype usage in IBM, which is aligned with previous findings (11). Similarly, SHM is involved in affinity maturation, which eventually selects B cells with higher affinity receptors and leads to the development of memory B cells and plasma cells. We noted that a distinguishing feature of IBM B cells was an elevated frequency of IgM SHM, suggesting that the IgM-switched plasma cell infiltrate in IBM is more somatically mutated than healthy peripheral counterparts or those in DM and PM. Importantly, the overall distribution of hot- and cold-spot mutability did not significantly differ between IIM subtypes, which argues against the possibility of sequencing errors contributing to our findings related to SHM. Overall, Ig isotype usage and increased SHM in IgM in IBM reinforce the role of adaptive immunity in IBM.

To quantify the selection for B cells of certain specificities for tissue entry, we used a combination of analysis for global SHM patterns and more specific analysis of V, J gene usage and CDR3 physiochemical properties. Compared to the expected frequency based on the intrinsic hot- and cold-spot biases of SHM, we observed a significant accumulation of silent mutations over replacement mutations in repertoires from IBM tissue by BASELINe analysis. Interestingly, our BASELINe results reflect observations from our own group and others for MG (18), rheumatoid arthritis (38), and SS (39) where similar observations have been made in association with repertoires from patients with autoimmune diseases characterized by ectopic B cell follicles showing an increase in the ratio of silent over replacement mutations in CDR and FWR regions.

We observed increased frequencies of VH1 and VH4 family genes such as VH1-18 and decreased usage of the most common V gene VH3-23 in DM, and usage of the VH1 family (including VH1-18) was lower in IBM compared with DM. The pattern of VH gene usage in DM patients differs to some extent from a previous report (40). However, in the same report, there was the suggestion of a different pattern of VH usage between patients with DM. Whether such differences stem from small population size or from different subtypes of DM is not clear. However, differences in VH gene usage provide additional evidence that the characteristics of the ASCs in IBM are distinct from those typically associated with circulating ASC subsets and DM. We speculated that the distinct pattern of SHM and VDJ gene usage observed for B cells in IBM (and other IIM) tissue may reflect broader differences in the regulation of plasma cell fate in situ. We also speculated that these defects in selection could contribute to the accumulation of autoreactive features in the serum IBM repertoire observed in the literature (7, 8, 15). For instance, we observed that the CDR3s of IBM samples were notably more hydrophobic likely because of increased aliphatic character and reduced aromatic character.

Along with the absence of germinal centers, we noted an absence of AID gene expression on meta-analysis of publicly available gene expression datasets. It is highly unlikely that the plasma cell population expresses AID given that Blimp-1 (PRDM1) is a strong repressor of AID upregulation (30) and populations in PM and IBM tissue are characterized by a strong PRDM1 signature (including IRF4 and IRF8 expression). Nevertheless, visible germinal centers in other autoimmune diseases are usually described in a relatively smaller tissue, and muscle tissue is larger. A sampling-related absence of germinal centers cannot be confidently excluded. Of note, germinal centers are not always present even in DM, and only ∼21% cases of juvenile DM showed follicle-like structures (41, 42). Questions remain whether the plasma cells noted in IBM are trafficked from outside the muscle or developed locally. Interestingly, PAX5 expression in muscle tissue in IBM was low, which may indirectly suggest that the B cell maturation to plasma cells may have occurred locally (43).

However, given the increased plasma cell signature in IBM, some of the differences noted in our study between IBM and DM/PM are related to abundance of plasma cell infiltrates in IBM muscle tissues. High frequencies of IgG plasma cells and autoreactive B cell features have also been described in the salivary gland infiltrates in primary SS (13, 44–46). Interestingly, SS is more frequent in patients with IBM compared with other autoimmune rheumatological diseases, but a clear relationship has not been defined (47–50). However, as mentioned earlier, unlike SS, there is no evidence for ectopic germinal centers in IBM, except in the context of other autoimmune disease (51).

Apart from the abundance of plasma cells in muscle tissue, the presence of cytotoxic T cells (CD8⁺ T cells) is a hallmark of IBM muscle biopsy, and there are clonally expanded CD8⁺CD57⁺ cells in the blood and muscle tissue in patients with IBM. It is possible that these highly differentiated T cells are stimulated by chronic Ag presentation by B cells, but such a link is not well established (35, 52, 53).

Interestingly, there are some reports of plasma cell infiltrates in T lymphocyte–mediated disease such as lichen planus and IgG4-related disease (54, 55). Whereas regulatory T cells and Th cells possibly play a role in IgG4-producing plasma cells in IgG4-related disease, highly differentiated cytotoxic T cells have been implicated in IBM pathogenesis (52, 55). Moreover, whereas lichen planus, primary SS, and IgG4-related diseases are responsive to immunomodulators, IBM does not respond to conventional immunotherapy. Although the phenomenon of linked T cell and plasma cell expansion may be tied by mechanisms of cross presentation and the linked recognition of common Ags responsible for T cell and B cell expansions, it needs further exploration in IBM (4, 52). The presence of public B cell clones, identified through highly similar BCR sequences, can be found among individuals responding to the same Ag (56, 57). Although the identification of public clones among our IBM patient cohort would support the presence of a common IBM Ag, this analysis was not explored given the requirement for exceptionally large BCR sequencing datasets derived from B cells that specifically target an identical Ag.

There are several limitations for this study. First, we did not have detailed clinical information on the patients with IIM, which limited our scope to translate the study findings directly into the clinical context. Second, given the rarity of IIM and tissue biopsies from IIM patients, the overall number of samples used for each disease group was low, at three from DM, three for PM, and four samples for IBM. Future studies should increase the limited sample count used for these experiments to better enhance the ability to detect significant repertoire differences. Third, our scientific rationale for comparing the muscle tissue BCR repertoire to that of the experienced circulating B cell repertoire was based on our view that known Ag-experienced cells offered the most direct comparison with the infiltrate in myositis, which are also known to be mature B cell types (4). Additionally, resident B cells are exceptionally sparse in healthy human muscle tissue, which we opted not to include. We further considered that comparing the muscle infiltrates to total circulating cells (PBMCs) would bias the results due to the high frequency of naive B cell populations in PBMCs. We also did not have access to circulating CD27⁺ B cells from the IIM patients, and thus we opted for highly enriched Ag-experienced B cell populations from other sources. Fourth, due to constraints associated with human tissue access, these sequencing control groups were generated from different sequencing runs and some from experiments with distinct repertoire sequencing protocols (bulk versus single cell). Fifth, one of the patients with IBM was previously treated with methotrexate (an immunomodulator) and may have influenced some of the findings. Finally, we attempted to address potential PCR error and batch effects using proven methods described in Materials and Methods. The possibility that these sources of error may not be eliminated should nevertheless be considered when interpreting our findings as is the case for other next-generation sequencing of adaptive immune receptor repertoires investigations.

These limitations notwithstanding, our results demonstrate that some features of the B cell repertoire associated with IBM muscle tissue are distinct from those of DM and PM, paralleling the large differences observed in the clinical presentation of the disease. The distinct character of the serum autoantibody profile of IBM has recently been demonstrated through a retrospective study that performed an unsupervised hierarchical clustering based on serum Ab specificities and showed that the first predominant division separates IBM from all other myositis subtypes (58). Shared features of IBM with DM and PM are also broadly associated with autoimmunity, and thus we speculate that the differences in the BCR repertoire we observed in this study suggest that the B cell infiltrate in IBM tissue is active, dividing, and could harbor a source of plasma cells that produce autoreactive Abs. However, several questions related to the pathophysiology of IBM remain unanswered, and we expect our study to stimulate further investigation to better understand the possible interplay between the plasma cell infiltration and cytotoxic T cells. Eventually, such work may hold the key to therapeutic development for this debilitating disease.

The authors have no financial conflicts of interest.

We gratefully acknowledge the generosity and support of the Dr. Martin Shubik Inclusion Body Myositis Fund.