Characterizing local antibody responses in the muscle of inclusion body myositis patients

Abstract

Objective:

Sporadic inclusion body myositis (IBM) is the most common adult idiopathic inflammatory myopathy. IBM etiology has been elusive, due to both degenerative and autoimmune disease features found on muscle biopsy, and significant disease heterogeneity. Investigating the role of antibodies in the muscle of IBM patients may improve our understanding of disease pathogenesis.

Methods:

We used an IBM xenograft mouse model in which muscle biopsy tissue from IBM patients (n=98) and controls (n=131; including 54 from other types of myopathy) were implanted into immunodeficient mice (NOD-Rag1^null-IL2rγ^null). We quantified the amount of IgG, IgA, and IgM in xenografted mouse sera using MesoScale Diagnostics (MSD) assay. We detected antibody reactivities targeting autoantigens and infectious agents using Phage ImmunoPrecipitation Sequencing (PhIP-Seq). Finally, we used FR3 AmplifiKation Sequencing (FR3AK-Seq) to sequence the antibody mRNAs from a separate cohort of 146 patient biopsies (14 IBM, 22 healthy controls, 110 other myositis subtypes).

Results:

With the MSD assay we found IgG, IgA, and IgM in a larger percentage of IBM xenografted mice versus controls. Using PhIP-Seq, we found anti-microbial reactivities secreted from IBM muscle are prevalent amongst a healthy control population but autoantigen reactivities in IBM are more unique at the peptide and protein level. Additionally, NT5C1A (IgG/IgA and IgM) and TIF1γ (IgG/A) autoantibodies are secreted from muscle tissues of 4/18 and 10/18 IBM xenograft donors, respectively.

Conclusion:

Our characterization of antibody responses within the muscle of IBM patients reveals that muscle-infiltrating B cells produce both disease-associated autoantibodies and a broad spectrum of antibodies targeting non-self peptides.

GRAPHICAL ABSTRACT

Introduction

Sporadic inclusion body myositis (IBM) is the most common adult idiopathic inflammatory myopathy (IIM) with a prevalence of up to 50 cases per million adults.^1,2 Symptomatically, while slow progression of muscle weakness is common, some patients experience rapid progression of muscle weakness.³ Muscle weakness is generally found with finger flexion, knee extension, and ankle dorsiflexion; the specific muscles affected, combined with variable disease progression, and the presence or absence of dysphagia, all contribute to disease course heterogeneity.¹

The etiology of IBM has been elusive, due to both degenerative and autoimmune disease features found on muscle biopsy. Degenerative findings include mitochondrial dysfunction (cytochrome-oxidase negativity), rimmed vacuoles, and protein aggregates, while autoimmune findings include infiltrating CD8+ T cells, macrophages, upregulation of major histocompatibility complex-I (MHC-I) expression, and an abundance of antibody-secreting plasma cells.^1,4,5 Although immunosuppressive medications have been shown to decrease muscle enzyme levels in the blood, muscle weakness worsens in patients, and inflammatory infiltrates persist. Additionally, the disease course is generally slowly progressive, making it difficult to discern treatment benefits,^6,7 especially since there are no established prognostic biomarkers (one study has suggested that loss of nuclear TDP-43 may be a potential biomarker, though further research is needed).⁸ The clinically utilized diagnostic NT5C1A autoantibody is only found in ~50% of IBM patients and does not correlate robustly with disease severity or prognosis.^1,9,10

On biopsy, there is not only heterogeneity in the amount of degenerative and autoimmune features present but also heterogeneity in the composition of the immune infiltrates. Although the vast majority of patients have a prominent CD8+ T cell infiltration in their muscle,¹ many patients also have large numbers of plasma cells¹¹ along with abundant immunoglobulin transcripts.¹² Studies investigating the role of the adaptive immune response in IBM are needed for developing a better understanding of disease pathogenesis and to identify new diagnostic and prognostic biomarkers.

Here we utilized IBM muscle biopsies xenografted into immunodeficient mice (NOD-Rag1^null-IL2rγ^null) to investigate the antibody-producing compartment of the immune infiltrate.⁸ It has been previously shown that the xenografts retain muscle-infiltrated lymphocytes from patient tissue.⁸ As NOD-Rag1^null-IL2rγ^null mice do not produce their own B cells or secrete antibodies, antibodies produced by human muscle-infiltrated B cells within the mouse can be characterized even months after xenotransplantation.

While several studies have explored autoantibody associations with IBM, antibodies that recognize non-self antigens have yet to be extensively explored. The only reported associations between IBM and environmental exposures have been limited to case reports and small studies of IBM patients with chronic viral diseases including hepatitis C, HIV, and HTLV-1.^13–15 It is thought that inflammatory stimuli may contribute to disease initiation or progression via molecular mimicry or other mechanism. Phage ImmunoPrecipitation Sequencing (PhIP-Seq) ¹⁶ is an efficient method for characterizing the (auto)antibody reactome.

Here we applied PhIP-Seq to identify antibody reactivities in the muscle using the following libraries: VirScan (covering all viruses known to infect humans),^17,18 ToxScan¹⁹ (covering proteins in the UniProt database that have ‘toxin’ or ‘virulence’ listed as a keyword association), and a library consisting of overlapping 90 amino acid peptide tiles spanning the entire human proteome.²⁰ Here we used FR3 AmplifiKation Sequencing (FR3AK-Seq)²¹ for sequencing antibody mRNAs in the muscle of IBM patients. Together, PhIP-Seq and FR3AK-Seq provide an unbiased characterization of the local antibody response in the muscle of IBM patients.

Methods

Xenograft mice and sera

We obtained sera from 229 total xenografted mice – 98 mice with tissue from inclusion body myositis (IBM) patients, 54 mice with tissue from patients with other forms of myopathy, and 77 mice with tissue from non-myopathic patients including healthy muscle (summarized in Table S1). All animal experiments were approved by the Johns Hopkins University Institutional Animal Care and Use Committee (IACUC) (protocol number: MO22M407). NRG mice were anesthetized with 1.5% isoflurane, and 0.1 mg/kg Buprenorphine (ZooPharm) was administered subcutaneously for pre-emptive analgesia. The tibialis anterior and extensor digitorum longus muscles were removed, and a 7 x 3 x 3 mm strip of dissected human muscle was placed in the tibial compartment and ligated with non-absorbable sutures (6–0 Surgipro, Covidien) to the tendons of the peroneus longus muscle. The incision was closed with surgical glue (Histoacryl, Tissue seal) and stainless-steel wound clips (AutoClip System, F.S.T). The analgesic Carprofen (Rimadyl, Patterson Veterinary) was given subcutaneously after the surgery at a dose of 5 mg/kg. To harvest the muscle, NRG host mice were anesthetized with 1.5% isofluorane, the leg was shaved, and the skin overlying the xenograft was opened, and the location of the graft was determined by the position of the non-absorbable sutures. The xenograft was then removed, snap-frozen in 2-methylbutane (Fisher, 03551). Cardiac punctures were performed to collect blood which was then centrifuged and sera was aliquoted and frozen. Following collection, the NRG host mice were euthanized consistent with AVMA guidelines.

OKT3 T Cell Depletion in mice

For OKT3 experiments, xenografted mice were given OKT3 (Fisher, 50561956) at 10 mg/kg. OKT3 was diluted in sterile PBS and intraperitoneal (IP) injections were performed immediately after xenograft surgery and once weekly until xenograft collection. Untreated xenografted mice were given sterile PBS following the same injection regimen.

Idiopathic inflammatory myopathy muscle biopsy cohort

Muscle biopsies were obtained from 146 patients for RNA extraction (14 from IBM, 110 from other IIMs, and 22 from healthy controls). This cohort was separate from the xenograft cohort. All patients provided informed consent. Patients were not selected to be treatment naive. All subjects were enrolled under the following institutional review boards: National Institutes of Health (IRB number 91-AR-0196), the Johns Hopkins Myositis Center (IRB number NA_00007454), the Clinic Hospital (Barcelona; IRB number HCB/2015/0479), and the Vall d'Hebron Hospital (Barcelona; IRB number PR (AG) 68/2008).

Xenograft RNA Extraction

RNA extraction was performed using TRIzol. Libraries were prepared using the NEBNext Poly(A) mRNA Magnetic Isolation Module and Ultra^™ II Directional RNA Library Prep Kit for Illumina (New England BioLabs, ref. #E7490 and #E7760) and sequenced using the Illumina Novaseq 6000 sequencer.

Muscle Biopsy RNA Extraction

As previously described,^22,23 and briefly summarized here, RNA extraction was performed using TRIzol on 146 muscle biopsy samples. Libraries were prepared either using the NeoPrep system with the TruSeq Stranded mRNA Library Prep protocol (Illumina, San Diego, CA) or with the NEBNext Poly(A) mRNA Magnetic Isolation Module and Ultra^™ II Directional RNA Library Prep Kit for Illumina (New England BioLabs, ref. #E7490 and #E7760) and sequenced using the Illumina HiSeq 2500 or 3000, NextSeq 2000 or NovaSeq X plus sequencers.

RNAseq Analysis

Sequencing reads were aligned to the hg38 and mm10 reference genomes using STAR v2.7.2 and segregated into species-specific alignments with Disambiguate v1.0.0.²⁴ Post-alignment quality was assessed using FastQC v0.11.8, and summary metrics were compiled with MultiQC. Counts were normalized using the Trimmed Mean of M-values (TMM) method from edgeR v3.34.1 for visualization. Differential expression analysis was conducted with limma v3.48.3. Graphical analyses were generated in Python using the seaborn library.

MesoScale Diagnostics Assay to measure antibody concentrations

IgG, IgA, and IgM concentrations were measured using the MesoScale Diagnostics (MSD) Isotyping Panel 1 (Human/NHP) kit (MesoScale Diagnostics, K15203D-2) and its accompanying protocol. Briefly, calibrators were prepared as recommended by the kit instructions, and samples were diluted 1:100 in 200 μL. Plates were blocked with 150 μL of Blocker A for 35 min at room temperature, with shaking at 800 rpm. Plates were washed 6 times, with 200 μL of PBS-T (PBS with 0.05% Tween-20) and an automatic plate washer. At wash 3, plates were rotated 180 degrees. 25 μL of diluted sample or calibrator were added to the plates and incubated at room temperature for 2 hrs with shaking at 800rpm. The plates were washed as previously described. 25 μL of detection antibody mixture was added to the plates and incubated at room temperature for 2 hrs with shaking at 800rpm. The plates were washed as previously described. 150 μL of 2X Read Buffer T was added to each well and samples were read on an MSD instrument. MSD values were converted to concentrations using myassays.com and the “four parameter logistic fit (1/y² weighting).” This regression is recommended in the MSD human isotyping kit protocol.

PhIP-Seq

PhIP-Seq, using sera with at least 0.05 ng of any antibody isotype per 100 μL from xenografted mice, which included mice with tissue from 18 IBM donors and 8 control donors, was performed as described previously.¹⁶ Briefly, xenograft serum samples were randomized across two 96-well plates. Serum from matched donors that were available, were included on the same plates that contained the corresponding xenograft samples. Samples with higher antibody concentrations and sufficient volume were normalized to 10ng of total input of the most abundant isotype (10 ng was the maximum amount input), otherwise 100 μL of neat xenograft serum was used in each reaction. If less than 100 μL of serum was available, the maximum volume was used and PBS was added up to 100 μL. A duplicate of each sample plate was prepared so that IgG/IgA capture and IgM capture could be performed separately.

Each sample was mixed with a 90 amino acid peptide-tile library spanning the human proteome and a 56 amino acid peptide-tile library consisting of the following sublibraries: VirScan, and ToxScan. Each library was added at a volume corresponding to an excess of 0.5x10⁵ per peptide, based on library complexity. Bovine serum albumin (BSA) was added to samples and libraries at a final concentration of 0.1%, in 150 μL total volume. For IgA capture, 2 μL of biotinylated goat anti-human IgA (Southern Biotech, Cat# 2050–08) was added to each sample overnight. For IgM capture, 2 μL of biotinylated goat anti-human IgM (Southern Biotech, Cat# 2020–08) was added to each sample overnight. The following day, protein A and protein G beads were used for immunocapture of IgG. All beads were washed and resuspended in 1% BSA in PBS, before being added to samples. 10 μL of protein A, 10 μL of protein G, and 5 μL of Streptavidin Magnetic beads (Pierce Cat# 88817) were added to each sample, on plates with IgA capture. 5 μL of streptavidin magnetic beads were added to each sample on plates with IgM capture. A set of 8 “mock” immunoprecipitations (libraries with no serum input), with appropriate beads, was run on each of the sample plates.

Following bead binding for four hours, beads are now part of a complex in which the beads bound to the antibodies are in turn bound to their phage-displayed peptide ligands. The phages displaying the peptides contain the peptide-encoding DNA. After bead binding, the bead complex is subsequently washed and resuspended in PCR master mix, then thermocycled to amplify the bound peptide-encoding phage DNA. 20 cycles of PCR were performed (PCR1), followed by 20 more cycles of PCR (PCR2) with adapter and sample barcode-containing primers. 5 μL of each PCR2 sample was pooled together and PCR products column was purified, for sequencing on an Illumina NovaSeq, with an SP100 flow cell.

To obtain greater sequencing depth for the human library, pooled PCR2 samples were diluted to 10 ng/μL. An additional 3 cycles of PCR were performed on 20ng total of PCR2 product. These samples were subsequently purified with SparQ beads at 0.6X bead volume to reaction volume, then run on a 3% agarose gel with 1X TAE buffer. The product was gel-extracted and re-amplified with an additional 3 cycles of PCR, then purified with SparQ beads at 1X bead volume to reaction volume. Products were pooled, column-purified to remove PCR reaction buffer, and then sequenced on an Illumina NovaSeq, with an SP100 flow cell.

PhIP-Seq Data Analysis

Sequencing reads were mapped to the libraries using perfect matching. By comparing against the set of mock IPs, a fold change and p-value were calculated for each peptide in each sample using the edgeR Bioconductor package.²⁵ Peptides were considered “hits” for a sample if the following conditions were met: count ≥ 100, p-value ≤ 0.001, and fold-change ≥ 5. For IgM data from the human library, hit criteria were adjusted to fold-change ≥ 6 due to the higher level of variability. Hits fold-change (HFC) values report the fold-changes of hits and is set to 1 for non-hits. All statistical tests for antibody concentrations and PhIP-Seq data were performed in R.

Antibody FR3AK-seq assay

cDNA was generated using the Superscript III First-Strand Synthesis System kit according to the manufacturer’s instructions (ThermoFisher, cat # 18080–051) with isotype-specific reverse transcription primers. 4 μL of RNA was used for each sample, or if the sample volume was lower, nuclease-free water was added to up to 4 μL. 500ng of PBMC RNA was used as a positive control sample. RNA corresponding to 100 cells of the Ramos cell-line was used as a spike-in in all samples. cDNA samples were cleaned up and concentrated using the Zymo Research ZR-96 DNA Clean and Concentrator kit (cat# D4023). 20 cycles of PCR1 were performed using the KAPA2G Fast Multiplex Mix (Roche, Cat# 07961430001), 17 forward primers designed to encompass all human IgHV alleles, and 5 reverse primers specific for each isotype (IgG, IgA, IgG, IgD, and IgM). 20 cycles of PCR2 using a Herculase Fusion DNA Polymerase kit (Agilent, Cat# 600677), was performed for sample barcoding. PCR2 samples were pooled together and cleaned up using the QIAquick PCR Purification Kit (Qiagen, Cat# 28104). Sequencing was performed on an Illumina NextSeq.

Reverse transcription	IgA_RT	CCG CTT TCG CTC CAG GTC ACA C
	IgM_RT	GTG ATG GAG TCG GGA AGG AAG TC
	IgG_RT	CCT GAG TTC CAC GAC ACC GTC
	IgD_RT	CCA TGT ACC AGG TGA CAG TCA CG
	IgE_RT	CAG GTC ACC ATC ACC GGC TCC
Reverse primers: PCRI	IgG_rev	CTG GAG TTC AGA CGT GTG CTC TTC CGA TCC GAC ACC GTC ACC GGT TC
	IgE_rev	CTG GAG TTC AGA CGT GTG CTC TTC CGA TCC CGG CTC CGG GAA GTA G
	IgM_rev	CTG GAG TTC AGA CGT GTG CTC TTC CGA TCG AAG GAA GTC CTG TGC GAG G
	IgD_rev	CTG GAG TTC AGA CGT GTG CTC TTC CGA TCT CAC GGA CGT TGG GTG GTA CC
	IgA_rev	CTG GAG TTC AGA CGT GTG CTC TTC CGA TCC AGG TCA CAC TGA GTG GCT C

1MM hIGHV: 17 primers
1	ACA CGG CCG TGT ATT ACT GT
2	ACA CAG CCA CGT ATT ACT GT
3	ACA CCG CCT TGT ATT ACT GT
4	ACA TGG CCT TGT ATT ACT GT
5	ACA TGG CTG TGT ATT ACT GT
6	ACA CGG CTG TTT ATT ACT GT
7	ACA CGG CCG TGT ATT GCT GT
8	ACA CGG CCT TGT ATC ACT GT
9	ACA CTG CCG TGT GTT ACT GT
10	ACA CAG CCA CGT ATT ATT GT
11	ACA CAG CTG TGT ATT ACT GT
12	ACA CCG GCG TGT ATT ACT GT
13	ACA CAG GCA CAT ATT ACT GT
14	ACT TGG CCG TGT ATT ACT GT
15	GCA CGG CCG CGT ATT ACT GT
16	CAC GGC TGT GTA TTA CTG TG
17	ACA TGG CTG TGT ATT ACT AT

For the 2¹⁰ dilution experiment, a PBMC PCR1 sample was diluted 1:256 in nuclease-free water. 5 μL of the dilution was added to a 20 μL PCR reaction using the KAPA2G Fast Multiple Mix, such that the final dilution of the sample in the PCR mastermix was a 2^10 dilution. The PCR reaction was amplified for 10 cycles (theoretically 2¹⁰-fold amplification). 2 μL of undiluted PCR1 product, and 2 μL of diluted and re-amplified PCR1 product, were used in separate PCR2 reactions, and performed in duplicate. PCR2 was performed (20 cycles). 5 μL of each PCR2 sample was pooled together and cleaned up using the QIAquick PCR kit and sequenced on an Illumina MiSeq.

Antibody FR3AK-seq analysis

CDR3s were identified and quantified using MiXCR software (v 3.0.13).²⁶ The following MiXCR parameters were used: …. --species hs --starting-material rna \ --receptor-type IGH \ --5-end v-primers \ --3-end c-primers \ --adapters no-adapters \ --assemble "-OseparateByC=true" --assemble "-OaddReadsCountOnClustering=true" \.

The Ramos spike-in was used to normalize MiXCR clone counts to generate cellular equivalents for each clone, and cellular equivalent values for each sample were normalized by μg of RNA input into the cDNA reaction. Cellular equivalent calculations and normalization were performed in R. Further analyses to generate figures were performed in R, were performed using only sequences with clone counts >=3.

For the 2^10 dilution experiment, the following MiXCR parameters were used: --species hs \ --starting-material rna \ --receptor-type IGH \--5-end v-primers \ --3-end c-primers \ --adapters no-adapters \ --assemble "-OseparateByV=false"\ --assemble "-OseparateByJ=false"\ --assemble "-OseparateByC=false"\ --assemble "-OaddReadsCountOnClustering=true" \. Only clone counts >= 10 were used in the analysis. Clone counts were then normalized to counts per million, before plotting.

The BCR clustering analysis was performed with GIANA4²⁷ without V gene information.

Results

Human IgG, IgA, and IgM antibodies are detected in sera from immunodeficient mice harboring IBM muscle xenografts

A MesoScale Diagnostics (MSD) assay was used to quantify IgG, IgA, and IgM antibody concentrations from 229 xenograft-harboring immunodeficient mouse sera (98 IBM xenografts and 131 controls). These numbers include some sera from different mice harboring xenografts derived from separate pieces of the same donor muscle biopsy (Fig. S1). Control xenografts consisted of muscle tissue from patients with other myopathies, mitochondrial disorders, neuropathies, and normal muscle (Table S1). All three isotypes were detectable in a larger, significantly different, fraction of IBM xenograft samples versus both myopathy controls and non-myopathy controls (by t-Wilcox test and Fisher’s (all p-values between 10^-5 and 10^-12, Fig. 1). Of the 131 control xenografts, 54 were from other myopathies and 77 were from non-myopathic controls. Measurable antibody concentrations of the three isotypes, from the sera of these mice, suggest that xenotransplanted muscle tissues continue to produce antibodies and that IBM muscle tissues continue to produce more antibodies versus other myopathies. Notably, many control samples have undetectable antibodies, supporting our assumption that detectable antibody is primarily derived from the active production in the muscle, and not passively transferred circulating antibodies.

Figure 1. Concentrations of IgG, IgA, and IgM isotypes of each xenograft serum sample measured by MesoScale Diagnostics (MSD) assay.

A) IgG, B) IgA and C) IgM. Percentages shown on top indicate percentage of xenograft samples with antibody concentrations above detection thresholds. Control group is non-myopathic controls, n=77 xenograft samples from 16 donors. myoCont is myopathic controls, n=54 xenograft samples from 10 donors. IBM group is n=98 xenograft samples from 23 donors. For IgG and IgM, the threshold was the limit of detection and concentrations below the detection limit are represented by 0. For IgA, threshold of log10=1.8 was set based on concentration of most dilute MSD kit calibrator. Comparison of concentrations of each isotype between the following groups of xenograft serum samples: IBM, myopathy controls (myoCont), and non-myopathic controls. Wilcox and Fisher’s tests comparing IBM to each control group for each isotype were statistically significant (all p-values between 10^-5 and 10^-12).

IBM muscle-derived antibodies recognize both highly prevalent pathogen reactivities and unique autoreactivities

Mouse serum samples with at least 0.05 ng of any antibody isotype per 100 μL were selected for PhIP-Seq analysis. 63 IBM xenograft samples from 18 donors and 16 control xenograft samples from 8 donors were characterized using the VirScan, ToxScan, and human proteome libraries. IgG and IgA antibodies were co-immunoprecipitated (Methods). IgM antibodies were immunoprecipitated separately. The majority of IBM xenograft-derived IgG/IgA reactivities from the VirScan and ToxScan libraries were also found to be highly prevalent among sera from a cohort of healthy controls (Vaccine Research Center cohort, VRC)¹⁹ (Fig. 2). Table S2 lists the specific peptides from VirScan and ToxScan that were enriched in the muscle of IBM donors compared to controls, along with high prevalence in the serum of VRC healthy controls. This observation is consistent with pre-existing class-switched B cells having trafficked into inflamed muscle tissue, where they continue to secrete infection-associated immunoglobulin. Unlike the ToxScan and VirScan reactivities, however, the majority of IgG/IgA autoreactivities identified in IBM xenografts were not commonly detected in the VRC cohort (Fig 2). Interestingly, the peptide autoreactivities identified in IBM xenografted mice were typically not detected in the sera of the IBM donors (Table S2).

Figure 2.

Comparison of IBM xenograft cohort prevalence and VRC control cohort prevalence for peptides in A) ToxScan, B) VirScan and C) human PhIP-Seq libraries. Only peptides with reactivity in at least 1 IBM xenograft sample are shown. VRC reactivities are from human serum samples.

Collapsing to the protein level by considering the most reactive peptide from each protein (“promax”), 295 proteins were significantly more reactive in IBM xenografted mice compared to controls (Table S3). Of these 295 proteins, 67% were identified as IgM autoreactivities, with the six most statistically significant proteins found reactive with the IgM isotype. Interestingly, the most significantly IBM-associated IgG/IgA protein-level reactivity is E3 ubiquitin-protein ligase TRIM33 (TRIM33, also known as TIF1γ), which is a diagnostic autoantibody in dermatomyositis also known to be associated with risk of malignancy.^28,29 Ten of the 18 IBM xenografted mouse sera tested were reactive to different peptides within the TIF1γ protein; no single peptide was reactive in more than 2 donors.

NT5C1A autoantibodies are identified in the muscle of IBM patients

Since the NT5C1A autoantibody is found in the serum of ~50% of IBM patients, we investigated whether it was also secreted locally in the muscle (findings summarized in Table 1). Of the two IBM patients who were identified as NT5C1A positive by clinical testing, and screened by PhIP-Seq, IgG/IgA NT5C1A was detected in the donor serum, as well as 3/3 corresponding xenografted mouse sera. Furthermore, these NT5C1A autoantibodies targeted the same peptide, which is known contain an immunodominant epitope of NT5C1A. Although IgM NT5C1A was not detected by PhIP-Seq in donor serum, it was detected in 3/3 mouse xenograft sera.

Table 1. Comparison of NT5C1A autoantibody status by clinical assay, with detection by PhIP-Seq in donor serum and xenograft mouse serum.

Values in parentheses denote NT5C1A peptide annotation in the PhIP-Seq library. Amino acid positions covered by peptide are indicated in parentheses. NT5C1A accession number: Q9BXI3

IBM NT5C1A status	Total Xeno Donors in PhIP Cohort	PhIP Donor Serum Screened	PhIP Donor Serum IgG/IgA Hits	PhIP Xeno Serum IgG/IgA Hits	PhIP Donor Serum IgM	PhIP Xeno Serum IgM
Positive NT5C1A	3: xeno20, 40, 44	2: xeno20, 44	1: xeno20 (141–196)	3/3 in xeno20 (141–196)	0	3/3 in xeno20 (141–196)
Negative NT5C1A	6: xeno5, 8, 9, 17, 45, 26	4: xeno5, 8, 17, 45	1: xeno17 (141–196)	1/4 in xeno17 (141–196)	0	2/4 in xeno17 (141–196) 1/8 in xeno26 (113–168)
Weak Positive NT5C1A	3: xeno36, 46, 42	2: xeno42,46	0	1/4 in xeno46 (197–252)	0	0

Of the IBM patients who were identified as anti-NT5C1A negative by clinical testing, PhIP-Seq of the donor serum identified IgG/IgA NT5C1A reactivity in one IBM patient. One out of four xenografted mouse sera from the same donor had IgG/IgA NT5C1A reactivity, while two of four had IgM NT5C1A reactivity. NT5C1A reactivities identified in both the serum and muscle of this patient recognized the same immunodominant peptide. Interestingly, PhIP-Seq identified IgM NT5C1A reactivity targeting a different peptide in one of eight xenograft sera from a different donor, though this peptide shares a 28 amino acid overlap with the immunodominant peptide. Of the patients who were identified as weakly NT5C1A positive by clinical testing, PhIP-Seq identified IgG/IgA NT5C1A reactivity in 1/4 xenograft sera from an IBM patient, targeting a peptide immediately adjacent to an immunodominant peptide (no amino acid overlap).

Taken together, these results suggest that IgG/IgA and IgM NT5C1A autoantibodies are produced in IBM muscle. Furthermore, class-switched NT5C1A autoantibodies mostly target the same peptides as IgM antibodies, and autoantibodies found in both the muscle and the blood target the same NT5C1A immunodominant peptides. However, additional isotype reactivities to additional NT5C1A peptides may be found in the muscle versus the blood, potentially indicative of a more robust and diverse response inside the muscle tissue.

B cell RNA transcripts are enriched in IBM muscle

We also investigated antibody mRNA sequences within the muscle of 14 IBM patients. Muscle tissues from several patients with other inflammatory myopathies and healthy controls were also analyzed for comparison (Table S1). We performed isotype-specific antibody mRNA sequencing of muscle biopsies from 146 individuals. Total RNA was extracted from muscle, from which cDNA specific for the five heavy-chain antibody isotypes were generated. cDNA was PCR amplified with isotype-specific reverse primers, and a pool of 17 forward primers to amplify all heavy chain V alleles (Fig. S2A). This novel primer set demonstrated minimal PCR bias, as shown by a dilution and re-amplification experiment in Fig. S2B. Using RNA extracted from peripheral blood mononuclear cells, IgG and IgA isotype-specific clones demonstrate minimal overlap with IgM clones, as expected (Fig. S2C).

Using a Ramos cell line as a spike-in, total cellular equivalents were calculated for each patient, representing the absolute number of B cells identified within each biopsy. As shown in Fig. 3A, B cells are abundant in IBM and Jo1 patient muscle, compared to other subtypes of myositis. Furthermore, most IBM patients have high clonal diversity within the muscle (Fig. 3B). Compared with other autoantibody-specific myositis subtypes (Jo1, HMGCR, and NXP2), a smaller fraction of the antibody clones is hyperexpanded in IBM, suggesting that transcripts tend to be more polyclonal in IBM (Fig. 3C). Together, the increased clonal diversity and decreased hyperexpansion in IBM suggests that antibody specificities likely correspond to a diversity of antigens. This diversity at the transcript level is consistent with the diversity of autoantigenic reactivity at the secreted antibody level measured by PhIP-Seq as described above.

Figure 3. Characteristics of B cell clones for each cohort.

Each point represents a different sample. NT indicates control cohort and all other cohorts are myositis subtypes. A) Log10 total cellular equivalents (calculated with respect to Ramos spike-in) of each cohort. B) Diversity, represented by number of unique nucleotide CDR3 sequences, of each cohort. C) Fraction of hyperexpanded clones of each cohort. Frequencies of all clones larger than 0.01 were summed to obtain the fraction hyperexpanded value.

IgG, IgA, and IgM antibody transcript levels are correlated in IBM muscle

IgG, IgA, and IgM transcripts are most abundant in IBM and Jo1 patients (Fig. 4A). Across all samples, IgE transcripts were nearly undetectable in muscle. IgD transcripts were much less abundant compared to IgG, IgA, and IgM transcripts, but were detected at higher levels in samples with overall higher levels of antibody transcripts. In all primary muscle biopsies, IgG, IgA, and IgM transcript levels demonstrate strong positive correlations with one another (Fig. 4B–D). In IBM xenograft mouse serum samples, however, we observed a much weaker, yet statistically significant, correlation between IgG, IgA, and IgM measurements (Fig. S3).

Figure 4. Levels of IgG, IgA, and IgM isotype mRNAs and Jo1 convergent clone.

A) Heat map, grouped by cohort (NT is control cohort and all other cohorts are myositis subtypes), where values correspond to log10 of total cellular equivalents of each isotype (calculated with respect to Ramos spike-in). B-D) Correlation between total cellular equivalents of two isotypes for each sample. Linear regression lines are in blue with 95% confidence interval in gray. IBM samples are red and all other samples (including controls) are in black. Each dot represents a sample. Correlation between IgG and IgM transcript levels (R² = 0.73). Correlation between IgA and IgM transcript levels (R² = 0.74). Correlation between IgG and IgA transcript levels (R² = 0.77). E) Motif and multiple sequence alignment of convergent Jo1 sequences. Each sequence is labelled with Jo1 sample number and isotype.

We also performed RNA sequencing on IBM and non-IBM muscle biopsies and xenografted muscle, as well as on IBM xenografted muscle after mice received treatment with the OKT3 T cell depleting therapy (Fig. S4). Consistent with Fig 3A, IBM biopsies and xenografts exhibited high levels of antibody mRNA. Interestingly, muscle from OKT3 treated mice demonstrated even more increased mRNA expression of all IgG subclasses compared to non-OKT3 treated mice. IgA, IgD and IgM gene expression was similar between the muscle tissue of OKT3 and non-OKT3 treated mice.

CDR3 clustering reveals motif in multiple Jo1 patients

Given the abundance of antibody transcript in the muscle, especially in IBM and Jo1 patients, we performed clustering of CDR3 sequences to determine if motifs were shared between multiple patients. There were no IBM-specific clusters containing sequences from more than two IBM patients (no clusters contained more than 3 sequences total). However, the cluster with the largest number of sequences was identified in Jo1 samples (from four unique donors) and contained sequences from IgG, IgA, and IgM transcripts (Fig. 4E).

Discussion

While antibody sequences of B cells (and especially plasma cells) in the muscle of IBM patients, and other inflammatory myopathy patients, have been previously reported,^{11,12,30–32} this study is the first to investigate the reactome of the antibodies produced within the muscle. We demonstrated that B cells infiltrating IBM muscle secrete IgG, IgA, and IgM antibodies, which is consistent with the expression of those antibodies at the messenger RNA level. In the muscle of IBM patients, IgG/IgA antibody reactivities targeting pathogen peptides, identified via ToxScan and VirScan, are also prevalent in the blood of healthy controls. This finding suggests that many B cells found in inflamed muscle are likely associated with bystander B cell recruitment. Further research is needed to determine whether antibodies targeting environmental epitopes in IBM may also damage muscle, or simply mark the general inflammatory status of the muscle.

We identified 5 peptides and 295 protein reactivities associated with IBM xenografted mouse sera compared with controls. The majority of reactivities at the protein level were IgM, potentially reflective of the lower specificity associated with the pentameric isotype. However, IgG/IgA reactivities (at both the peptide and protein levels) suggest there is likely also a local, muscle-specific inflammatory response leading to the production of class-switched antibodies. A limitation of our PhIP-Seq analyses is a potential increase in sensitivity of antibody detection in our IBM xenografts compared to control xenografts. We analyzed 63 xenografts from 18 IBM donors compared to 18 control xenografts from 8 donors. The average number of xenografts per donor is 3.5 for IBM compared to 2 for controls. While we grouped our analyses at the donor level to minimize bias in the peptides we associated with IBM, the additional xenograft per donor does provide additional samples and thus sensitivity to detect reactivities.

PhIP-Seq was previously used to discover that NT5C1A was a major target of autoantibodies in IBM patients, and one to three immunodominant regions of the protein have been reported.¹⁰ Although IgG antibodies are typically the target of clinical autoantibody testing, IgG, IgA, and IgM NT5C1A autoantibodies have previously been identified in the blood of IBM patients; additionally, the use of all three isotypes was shown to improve diagnostic sensitivity.³³ It is important to note that while sensitivity has been shown to increase using all three isotypes, there will likely be a decrease in specificity, which is also an important consideration for a rare disease like IBM. Further research is needed to optimize the use of isotypes in relation to both sensitivity and specificity. Using PhIP-Seq to profile the sera of IBM muscle xenografted immunodeficient mice, we determined that IgG/IgA and IgM NT5C1A autoantibodies are secreted from B cells within the muscle of IBM patients. In two donors for which IgG/IgA and IgM NT5C1A autoantibodies were detected in the muscle, PhIP-Seq also identified IgG/IgA NT5C1A autoantibodies in the blood. Because PhIP-Seq utilizes peptide libraries, these peptides contain limited conformational structure and lack post-translational modifications. These limitations of PhIP-Seq likely explain why PhIP-Seq did not identify NT5C1A autoantibodies in all clinically positive or weakly positive NT5C1A patients, as clinical assays use whole protein for autoantigen detection. Furthermore, it should be noted that NT5C1A autoantibodies were not identified in all xenografts from the same donor, which is in accordance with the high level of immune heterogeneity among tissue biopsies.

In addition to NT5C1A, we also identified reactivity to TRIM33 (TIF1γ) at the protein level, in the muscle of 55% of IBM donors (i.e., 55% of xenografted mouse sera grouped by donor). TIF1γ is a well-known autoantibody in dermatomyositis, but has not been associated with IBM. Interestingly, we did not identify TIF1γ reactivity in the serum of the same IBM donors. One potential explanation is that the B cells producing these autoantibodies are at a higher concentration in the muscle, and thus the autoantibodies are more detectable in the muscle compared to the blood. Another possibility is that these muscle-derived autoantibodies tend to be antigen-occupied in the blood, and thus not detectable in binding assays. It should be noted that TIF1γ positive patients were not incorporated in the xenograft cohort, only in the RNA sequencing cohort. Further research is needed to understand the origins of the IBM muscle-infiltrate derived TIF1γ antibody, why it is not detectable in the serum, whether it is associated with malignancy, and whether it can contribute to muscle damage in IBM.

A recent study identified IgG reactivity to NADH dehydrogenase 1 α subcomplex 11 (NDUFA11) in about 10% and 3.5% of IBM patients from an exploratory and validation cohort.³⁴ Reactivity to one NDUFA11 peptide was found in IgG/IgA in two IBM xenografts (two different donors) and no control xenografts. These findings are thus consistent with the low prevalence of NDUFA11 antibodies reported in IBM. Additionally, neither of the two patients with NDUFA11 reactivity also had NT5C1A reactivity, similar to Notaricola et. al.’s report that no IBM patient had both NT5C1A and NDUFA11 antibodies.³⁴

We performed antibody mRNA transcript analyses with the FR3AK-seq assay on a large set of muscle biopsies. A dilution and re-amplification experiment confirmed that our FR3AK-seq antibody assay minimally distorts the repertoire during amplification, similar to the previously published T cell receptor version of the assay.²¹ Using this low-cost, high-through sequencing methodology, we sequenced the antibody transcripts contained within 146 biopsy samples. Previous studies identified class-switched clonally related sequences in IBM patients but found fewer IgM sequences compared to IgG and IgA.^11,12,31,32 We detected IgG, IgA, and IgM sequences all in similar proportions, but the difference may be due to previous use of laser capture microdissection versus total mRNA extraction used here. However, in these previous studies IBM patients had more B cells than dermatomyositis (DM) patients¹¹ and more IgM sequences compared to the DM and polymyositis patients (exact criteria of polymyositis not provided)³¹ whereas we found similar cellular equivalents in IBM as well as other subtypes Jo1, Mi2 and TIF1γ.

Previous research has also found evidence of B cell clonal expansion and a polyclonal response in IBM muscle,^12,32 similar to our findings. While polyclonal B cell responses suggests that B cell depletion therapy may be beneficial, several previous studies have noted high levels of plasma cells in the muscle of IBM patients. When considering the efficacy of a B cell depletion therapy, its activity on plasma cells must therefore be considered. While we show that IBM has a polyclonal response, we also show that IBM has less hyperexpansion than some autoantibody subtypes of myositis, but more hyperexpansion than other autoantibody subtypes (Fig. 3C). IBM also has a similar level of clonal diversity compared to autoantibody subtypes we analyzed (Fig. 3B). Our study suggests that B cell diversity is not a unique characteristic of IBM compared to autoantibody subtypes of myositis.

Interestingly, IBM xenografted mice that received OKT3 T cell depletion therapy had increased expression of IgG mRNA compared to untreated IBM xenografted mice. Since CD4 T cells typically support B cell expansion, this observation is surprising. However, a previous study showed that while kidney transplant recipients who received anti-thymocyte globulin (ATG) had effective T cell depletion, subsequent T cell reconstitution favored Tfh cells, which increased IgG B cell numbers compared to non-ATG treated patients.³⁵ While the xenograft model did not have T cell reconstitution, it is possible that in IBM muscle, T cell subset(s) were negatively regulating antibody production prior to depletion. Further research is needed to determine why IgG expression is favored as opposed to IgM and IgA isotypes. The paradoxical increase of IgG in the setting of T cell depletion with OKT3 suggests that antibody-secreting cells may be a driver of perpetuating the autoimmune response in IBM.

While we previously reported a convergent T cell clone among four different IBM patients,²¹ our antibody repertoire sequencing analysis of those same patients did not reveal a convergent B cell clone. However, we did identify a convergent B cell clone in the muscle of four different Jo1 patients involving IgM, IgA, and IgG, which has not been previously described. A previous study reported IgM autoantibody predominance and minimal class-switching in Jo1 patients.³⁶ A previous study identified an expanded Jo1 specific B cell clone,³⁷ but the reported CDR3 motif was different than ours.

While our FR3AK-seq antibody transcript sequencing has beneficial features (low-cost, high-throughput, isotype-specific, and quantitative), it also has limitations. Because our forward PCR primers land at the 3’ end of the framework 3 region of the V gene, just upstream of the CDR3, we do not get full V gene information, including the gene sequence and somatic hypermutations outside the CDR3 region. FR3AK-seq is also unable to generate paired heavy and light chain sequences. Combining FR3AK-seq with longer transcript sequencing assays, and with single-cell mRNA sequencing assays can help to overcome these limitations.

Conclusion

We employed both antibody reactome studies and antibody heavy chain repertoire sequencing to characterize the muscle-infiltrated B cell response in IBM patients. In a xenograft transplant system, IBM muscle biopsies demonstrated increased secretion of IgG, IgA and IgM antibodies versus controls, which included other types of myositis. Secreted antibodies were specific for many environmental antigens, as well as the NT5C1A IBM autoantigen, and the TIF1γ autoantigen for which reactivity was identified only in IBM muscle. This suggests that muscle-infiltrated B cells in IBM produce antibodies frequently observed in the blood, as well as autoantibodies that are distinct from those easily detected in the blood. Using repertoire sequencing, we identified an increased level of antibody transcripts in IBM and Jo1 patients compared to other myositis patients, which also revealed a public Jo1 antibody clone. This study provides new information about muscle-infiltrated antibody responses in IBM, which may help to improve the diagnosis and treatment of this heterogenous autoimmune disease.

Highlights:

• Xenograft mouse model to study myositis muscle associated B cells and antibodies

• Biopsy secreted IgG, IgA, and IgM higher in inclusion body myositis (IBM)

• Reactivities to microbial antigens detected similarly in IBM and control muscle

• Autoantibodies targeting NT5C1A and TIF-1γ secreted preferentially from IBM muscle

• IgG, IgA, IgM mRNA expressed at high and correlated levels in IBM muscle