Single-cell sequencing of immune cells from anticitrullinated peptide antibody positive and negative rheumatoid arthritis

The presence or absence of anti-citrullinated peptide antibodies (ACPA) and associated disparities in patients with rheumatoid arthritis (RA) implies disease heterogeneity with unknown diverse immunopathological mechanisms. Here we profile CD45+ hematopoietic cells from peripheral blood or synovial tissues from both ACPA+ and ACPA- RA patients by single-cell RNA sequencing and identify subsets of immune cells that contribute to the pathogenesis of RA subtypes. We find several synovial immune cell abnormalities, including up-regulation of CCL13, CCL18 and MMP3 in myeloid cell subsets of ACPA- RA compared with ACPA+ RA. Also evident is a lack of HLA-DRB5 expression and lower expression of cytotoxic and exhaustion related genes in the synovial tissues of patients with ACPA- RA. Furthermore, the HLA-DR15 haplotype (DRB1/DRB5) conveys an increased risk of developing active disease in ACPA+ RA in a large cohort of patients with treatment-naive RA. Immunohistochemical staining shows increased infiltration of CCL13 and CCL18-expressing immune cells in synovial tissues of ACPA- RA. Collectively, our data provide evidence of the differential involvement of cellular and molecular pathways involved in the pathogenesis of seropositive and seronegative RA subtypes and reveal the importance of precision therapy based on ACPA status.

C haracterized by joint inflammation and bone destruction, rheumatoid arthritis (RA) is considered a heterogeneous disease, which can be categorized by the presence or absence of anticitrullinated-peptide antibodies (ACPA) 1 . ACPAs are disease-specific biomarkers for RA and are present in approximately two-thirds of these patients 2 . Genetic, clinical, and therapeutic differences exist between ACPA-and ACPA+ RA patients [3][4][5] . However, the differences in immunopathogenesis between these two RA subtypes are not well understood. Citrullinated autoantigens in ACPA+ RA probably activate monocyte/ macrophages directly by forming immune complexes (IC) with ACPA. ICs subsequently trigger a proinflammatory cytokine response in macrophage colony-stimulating factor (M-CSF) differentiated macrophages and induce TNF secretion by blood and synovium-derived macrophages in patients with RA 6,7 . Extracellular citrullinated proteins presented by antigen-presenting cells (APCs) to auto-reactive T and B cells trigger an adaptive immune response 8 . The study of peripheral blood-derived citrullinated peptide-specific B cells in RA has revealed that these cells have a memory phenotype, expressing IgA or IgG, and are present at relatively high frequencies 9 . The synovial T cell repertoire differs between ACPA+ and ACPA-RA 10 . Furthermore, ACPA might contribute to bone loss and the development of joint pathology by facilitating synovium-derived fibroblast migration 11 . As the immunopathogenesis of ACPA-RA is still not clear, an understanding of the mechanisms involved is warranted to enable the management of this disease subtype. The single-cell technologies have helped to identify inflammatory cell states and specific immune cell populations associated with RA 12,13 .
Here we apply single-cell RNA sequencing (scRNA-seq) to characterize the cell composition, proportion, gene expression signature, and developmental trajectories of CD45 + cells in the peripheral blood and synovial membrane of ACPA+ and ACPA-RA patients. We present this single-cell immunological landscape and identify immune cell types in ACPA-RA that might contribute to disease development.

Results
The immunological landscape of immune cells from ACPA+ and ACPA-RA. To investigate the heterogeneity of immune cell phenotype in different subtypes of RA patients, we carried out scRNA-seq on CD45 + peripheral blood mononuclear cells (PBMCs) isolated from healthy controls (HCs) (n = 4), PBMCs and synovial tissue mononuclear cells (STMCs) isolated from ACPA+ (n = 10), and ACPA-(n = 10) RA patients (Fig. 1a). The studied patients were not taking disease-modifying antirheumatic drugs (DMARDS), corticosteroids, or targeted therapies at the time of sampling (some patients chose to use physical therapies, such as heat incubation or acupuncture to alleviate pain). After data preprocessing and quality control, we obtained single-cell transcriptomes of 135,429 immune cells from PBMCs and 71,073 immune cells from STMCs. Using unsupervised graph-based clustering, we identified 25 clusters in PBMCs, which consisted of three major populations, B cells, T/Natural Killer (NK) cells, and myeloid cells (monocytes and dendritic cells), by examining the expression of canonical markers CD3D, CD79A, CD14, FCER1A, FCGR3A, IL3RA, NKG7, PPBP, and MZB1 (Fig. 1b, Supplementary Fig. 1a).
We identified a total of four clusters of B cells in PBMCs from all individuals by comparing the expression of IGHD, CD27, IGHM, IGHG1, IGHG3, IGHA1, and MZB1; these B cell clusters included naïve B cells, memory-unswitched B cells (mem-unsw B), memory-switched B cells (mem-sw B), and plasma B cells ( Supplementary Fig. 1b). We found that the proportion of B cell populations were similar in RA patients in comparison to those in HCs (Fig. 1c). We identified a total of eight distinct subsets of T/ NK cells in PBMCs. Among T cells, naïve T cells expressed CCR7 and SELL, exhausted T cells expressed high amounts of PDCD1, LAG3, and TIGIT, (Interferon) IFN-act T cells expressed IFI6, ISG15, and STAT1, GZMK hi CD4 T cells and GZMK hi CD8 T cells highly expressed GZMK, and cytotoxic T cells expressed cytotoxic genes (NKG7, PRF1, GNLY, and GZMB) (Supplementary Fig. 1b). We observed lower proportions of naïve T and higher proportions of cytotoxic T cells in ACPA+ RA compared with HCs. No significant differences were found in the proportions of T cell subsets between ACPA-and ACPA+ RA (Fig. 1c, Supplementary Fig. 1c). Analysis of myeloid cells in the PBMC on the basis of differential expression of CD14, FCGR3A, FCER1A, IRF8, IL3RA, HLA-DRB5, S100B, and IFN activation markers (ISG15, IFI6, IFI44, and STAT1) revealed no significant differences in Monocyte/Dendritic cells (Mono_DC) populations in ACPA-RA compared with ACPA+ RA and HCs, respectively (Fig. 1c, Supplementary Fig. 1b).
ACPA-RA lacks B cell expression of HLA-DRB5. To characterize the specific signature of immune/inflammatory cells in the two RA subtypes, we examined the cellular-subtype proportions within each cell populations from PBMCs and synovial tissues (ST). Using graph-based clustering, we sub-grouped PBMC-derived B cells into seven cell subtypes and designated each subset according to their "core gene signature." Based on IGHD, CD27, IGHM, IGHG1, IGHA1, GNLY, GPR183, IGHG3, IGHG4, MZB1, and HLA-DRB5 expression, we defined naïve B, Mem-unsw B, Mem-sw B, Plasma B, IGHG4 + Plasma B, HLA-DRB5 + MZB1 -B, and HLA-DRB5 + Plasma B subsets ( Fig. 2a-b), respectively. Next, we compared the proportions of B cell subsets in different RA subtypes. It was noteworthy that HLA-DRB5 + Plasma B cells, expressing HLA-DRB5, IGHM, IGHG1, and MZB1, were significantly decreased; however, IGHG4 + Plasma B cells, distinctly expressing IGHG4 from other plasma B subsets, were significantly increased in the peripheral blood of ACPA-RA patients (Fig. 2c, Supplementary Fig. 3a). We sub-grouped STderived B cells into eight-cell subtypes (Fig. 2d-e). We identified three memory B and five plasma cell subtypes in ST, respectively. Among them, HLA-DRB5 + and HLA-DRB5 + HOPX + memory B cells, which were absent in more than half (four of seven) of STs from ACPA-RA patients, shared common highly expressed genes (GPR183 and HLA-DRB5) and distinct HOPX expression in HLA-DRB5 + HOPX + memory B cells ( Fig. 2d-f, Supplementary  Fig. 3b). CCL + memory B cells, which express a set of chemokine genes that include CCL3, CCL13, and CCL18, were enriched in the synovial tissues ( Fig. 2d-e). Relatively higher proportions of CCL + memory B cells were observed in the synovial membrane of ACPA-RA patients but this difference was not statistically significant from that in ACPA+ RA patients (Fig. 2f).
To further explore the continuum of class-switching states in B cell subsets, we carried out pseudotime ordering of single cells using Monocle 3. Starting from naïve B cells, we reconstructed a trajectory with three branches in PBMCs. We noted that although HLA-DRB5 + Plasma B subsets expressed plasma B-related genes (MZB1), they are enriched for "peptide antigen binding" and "antigen processing and presentation," which resemble an "intermediate" subset (Fig. 3a). In HCs, progression proceeded from naïve B cells to either HLA-DRB5 + Plasma B cells and then to plasma cells or along the trajectory from Mem-unsw B to Mem-sw B (Fig. 3b)          Plasma B cells along with the two branches observed in HCs (Fig. 3b). We next compared the functional and molecular differences of IGHG4 + Plasma B, HLA-DRB5 + Memory B, and CCL + Memory B between ACPA-and ACPA+ RA patients. We found a lower transcriptional profile, with enrichment for "protein processing in endoplasmic reticulum" and "B receptor signaling pathway" but a relative higher transcriptional profile, with enrichment for "osteoclast differentiation" of IGHG4 + Plasma B in ACPA-RA patients compared with ACPA+ RA patients (Fig. 3c). ACPA-RA patients-derived plasma B cells in peripheral blood displayed a higher expression of IGHM and lower expressions of IGHG3 and IGHA1 (Fig. 3d). HLA-DRB5 + memory B cells from ACPA-RA patients were absent for HLA-DRB5 expression and exhibited a lower transcriptional profile, with enrichment for "antigen processing and presentation", "peptide antigen binding", "peptide binding", "antigen binding", and "MHC class II receptor activity" compared with ACPA+ RA patients (Fig. 3c, e). CCL + memory B cells from ACPA-RA patients were additionally transcriptionally distinct, being enriched for "immunoglobulin receptor binding", "antigen binding", and "protein processing in endoplasmic reticulum" but lower for "glycolysis/gluconeogenesis", "phagosome", and "antigen processing and presentation activity" (Fig. 3c). In summary, we demonstrated a lack of an "intermediate" HLA-DRB5 + Plasma B and abnormal IGHG4 + Plasma B in peripheral blood and a lack of HLA-DRB5 expression as well as impaired "antigen processing and presentation activity" on memory B cells in ST of ACPA-RA.  Fig. 5a, b). No obvious differences in the proportions of DC subsets in ST were observed between ACPA-and ACPA+ RA (Fig. 4c). Notably, CCL13, CCL18, and MMP3 were the most significantly upregulated genes, and HLA-DRB5, SPP1, BRI3 expression were absent in ACPA-RA ST-derived DC subsets (Fig. 4d). When we further applied Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway and Gene Ontology (GO) enrichment analyses, we found "phagosome", "glycolysis/  Asterisks indicate the significance. The y axis represents log-scaled normalized counts. P values were calculated by the two-sided Wilcoxon test. *p < 0.05, **p < 0.01, ***p < 0.001.
HLA-DR beta gene alleles are located on chromosome 6; DRB1 is constitutively expressed in all individuals, and the presence of DRB3, 4, and 5 varies with each individual 14 . We performed HLA-DRB1 and DRB5 genotyping on an additional cohort of 209 treatment naïve RA patients to further characterize the contribution of the HLA-DRB5 genotype to the development of different RA subtypes. Although the HLA-DRB5 genotype was not significantly more prevalent in ACPA+ RA (49/141, 34.8%) compared with ACPA-RA (15/48, 31.3%) and HCs (10/34,29.4%) ( Supplementary  Fig. 8a), evidence of disease activity, including DAS28, CRP, and IgM were significantly higher in ACPA+ RA with HLA-DRB5 genotype, whilst this was not observed in ACPA-RA (Supplementary Fig. 8b). These results indicate a close association of HLA-DRB5 genotype with disease activity in ACPA+ RA.
Interaction between macrophage subsets, DCs, and CD4 + T cell subsets. To further investigate the interactions between macrophage subsets, DCs, and CD4 T cells in RA, we utilized CellPhoneDB to better understand the regulatory relationships among these cell clusters. While the interactions of HLA-DRB5 + DC and HLA-DRB5 -DC with S100A8 hi Cytotoxic CD4 T or NK/ CD4 T were similar, HLA-DRB5 -pDC, which were more abundant in ACPA-RA, exhibited stronger interactions with cytotoxic T subsets than HLA-DRB5 + pDC (Fig. 10a). Both HLA-DRB5 -CCL + and HLA-DRB5 hi CD36 + macrophages exhibited stronger interactions with T/NK subsets in ST than CCL + IL18 + macrophages (Fig. 10b).
We further utilized a set of immune-related ligand-receptor (L-R) pairs to predict those involved in these cell-cell interactions. DC subtypes employed CD74-APP/COPA/MIF to interact with Teff cell subtypes (Fig. 10c). HLA-DRB5 -CCL + macrophages,  which were significantly upregulated in the ST of ACPA-RA, was more prone to utilize CCR8-CCL18, CCR1-CCL18, CCR1-CCL13, and CCR2-CCL13 to interact with ST T subsets (Fig. 10d). ST T subsets interact with macrophage subsets via CCL3-CCR1, CCL3-IDE, and CCL3-CCR5 (Fig. 10e). Altogether, our interaction data reveal that abnormal DC, T cell, and macrophage subsets present in RA patients display a stronger interaction with each other via the CD74, CCL13, CCL18, and CCL3 molecules, which could confer signals contributing to the abnormal immune/inflammatory responses in RA.

Discussion
The immune response against citrullinated antigens is a hallmark of ACPA+ patients with RA, although it remains uncertain whether RA can be truly divided into two clinical subtypes with distinct mechanisms of immune-dysregulation based solely on serum ACPA status. To our knowledge, this study is the first to comprehensively address differences in the distribution of CD45 + immune cell subsets by single-cell RNA sequencing in both the peripheral blood and synovial tissue, and systematically compare them between ACPA-and ACPA+ RA.
In ACPA-RA, we observed several specific abnormalities in the synovial membrane (Supplementary Table 3), including significant upregulation of CCL13, CCL18 expression in B, DC, and macrophage subsets, and upregulation of MMP3 expression in DC, macrophage, and T cells subsets, respectively. We also observed a lack of HLA-DRB5 expression on B, DC and macrophage subsets, and decreased cytotoxic T cells and exhausted T cells in the synovial membrane of ACPA-RA. This constellation of abnormally distributed immune/inflammatory cells suggests that the local synovial response is substantially different in ACPA-and ACPA+ RA. We observed significantly upregulated expression of CCL13, CCL18, and MMP3. CCL13 was found aberrantly expressed in the serum, synovial fluid, and synovial tissues from RA patients compared with those with osteoarthropathy, which could enhance fibroblast-like synoviocytes (FLS) proliferation, increase macrophage infiltration, and synovial tissue angiogenesis 15,16 . CCL18 levels in serum and synovial tissues were found to be elevated in RA patients compared with controls, and were associated with RA disease progression 17,18 . MMP3 was found to be elevated in the serum of RA and associated with bone erosion development 19 . We found that CCL + macrophages were more prone to utilize CCR8-CCL18, CCR1-CCL18, CCR1-CCL13, and T CCR2-CL13 to interact with ST T subsets. Notably, CCL18 has been reported to inhibit CCR1 mediated chemotaxis 20 , which might explain the observed significantly lower proportions of NK and exhausted CD4 T cells in ST of ACPA-RA. Lewis et al. have reported clustering of patients into three categories as lympho-myeloid (CD20 B cell aggregate rich), diffuse-myeloid (CD68 rich in the lining or sublining layer but poor in B cells), and pauci-immune (with scant infiltration of immune cells and predominantly of FLS) 21 . Synovial RNA-seq comparing differential gene expression between ACPA+ and ACPA-RA patients showed increased plasma cell genes such as XBP1, ODC1, and EAF2, indicating a pro-lympho-myeloid pathotype in ACPA+ RA 21 . We further compared gene expression in different ST-derived plasma B between ACPA-and ACPA+ RA. Consistent with Lewis' findings, we observed increased ODC1 expression in HLA-DRB5 + memory B and IGLL5 + Plasma B (Supplementary Fig. 9). Moreover, we also found the absence of exhausted T cell markers and decreased cytotoxic T/NK cells in ST of ACPA-RA, further supporting that ACPA-RA patients are more likely clustered into diffuse-myeloid or pauci-immune category. Humby et al. also revealed that diffuse-myeloid or pauci-immune RA were less likely to develop joint damage progression than lympho-myeloid pathotype 22 . However, the pauci-immune RA patients predicted inadequate response to TNF blockade treatment 23 . Our present study demonstrates that although exhibiting similar clinical arthritic symptom, the immune pathogenesis of ACPA-and ACPA+ RA patients is quite different, which contributes to the histological, disease progression, and therapeutic strategy heterogeneity. Our study demonstrated a close association of the HLA-DRB5 genotype with disease activity in ACPA+ RA. Human leukocyte antigen-DR15 is a haplotype that results in the expression of DRB1 and DRB5, most frequently HLA-DRB1*15:01 and HLA-DRB5*01:01. Previous studies demonstrated the association of the HLA-DR15 haplotype with multiple sclerosis, a chronic T cell-mediated autoimmune disease 24,25 . HLA-DR15 status was also reported as an important marker in autoimmune bone marrow disease 26 . The study of the role of HLA-DR15 in multiple sclerosis demonstrated that patients displayed T cell proliferative responses to myelin-associated autoantigens, which was related to the presence and level of HLA-DR15 expression 27 . In addition, a recent study described autoproliferation of peripheral Th1 cells mediated by memory B cells in an HLA-DR15 haplotypedependent manner 28,29 . In this study, we observed the lack of HLA-DRB5 expression and lower antigen-presenting ability in ST memory B cell subsets, as well as macrophage and DC subsets of ACPA-RA. Therefore, we speculate that autoimmune B and T cell subsets are not major players in the pathogenesis of synovium phenotype of ACPA-RA. The presence of HLA-DRB5 genotype in ACPA+ RA might also indicate autoproliferation of Th1 cells to present autoantigen, contributing to disease activity. Furthermore, the absence of an "intermediate" HLA-DRB5 + Plasma B, higher expression of IGHM, and lower expressions of IGHG3 in the peripheral blood of ACPA-RA, indicated the abnormal classswitching of B cell subsets in ACPA-RA.
All samples included were taken at the time of arthroplasty, indicating the potential contribution of OA to clinical symptom in our present study. Although synovitis is a common feature of OA and characterized by immune cells infiltration, a recent review summarized the published studies regarding the presence of inflammatory cells and their cytokines in OA ST 30 , and suggested that the number of infiltrating cells as well as cytokine were significantly lower in OA than in RA ST 30 . Moreover, a previous study performed by Raychaudhuri et al. applied scRNA-seq, mass cytometry, bulk RNA-seq, and flow cytometry to define inflammatory cell states in RA joint synovial tissues 12 . They reported that OA samples were characterized by a high abundance of fibroblast and endothelial cells but not significant in B cells, T cells, DC, or macrophages, as observed in ST between ACPAand ACPA+ RA patients in our study.
Kurowska-Stolarska et al. used CD163, MerTK, and CD206 to define ST macrophages (STM) and found MerTK pos CD206 pos STM were associated with remission maintenance 31 . They also performed scRNA-seq of MerTK pos CD206 pos and MerT-K neg CD206 neg STMs, and identified nine distinct STM clusters. Of which, HLA high CLEC10A high , TREM2 high TIMD4 pos and S100A12 pos macrophages were consistent with CLEC10A + macrophages, TIMD4 + CCL + macrophages, IL1B hi S100A12 hi macrophage identified in our present study, respectively. Intriguingly, they found that, in contrast to proinflammatory MerT-K neg CD206 neg STMs, MerTK pos CD206 pos could induce FLS TGF-β response genes and a MerTK inhibitor increases FLS expression of MMPs1/3/14 and IL-6. With this study, we propose that ACPA-RA ST macrophage subsets might be MerT-K neg CD206 neg STM and ACPA+ RA ST macrophage subsets might be MerTK pos CD206 pos . Should our proposition be validated by future studies, MerTK neg CD206 neg STM might become a promising therapeutic target in the clinic.
In summary, our study provides a transcriptional landscape of immune cell status at single-cell resolution in ACPA-and ACPA + RA. We identified the HLA-DR15 haplotype as a risk factor for developing the active disease in ACPA+ RA. We further demonstrated that significantly higher expression of CCL13, CCL18, and MMP3 in DC and macrophage subsets, together with poor B cells and T cells responses are salient immune features in ST of ACPA-RA. Our findings suggest that ACPA-RA may preferentially utilize different immune mechanisms and pathways than ACPA+ RA. In ACPA-RA ST, inflammatory myeloid cells, including M1 macrophages (IL1B) and MMP3 secreting DC subsets contribute to synovial pathogenesis while in ACPA+ RA ST, lymphoid cells (B cells and T cells) are major contributors. Therefore, they might respond to different therapeutic strategies.

Methods
Patient recruitment and ethics. Peripheral blood and synovial membrane samples were obtained from patients who fulfilled the 2010 ACR/EULAR Rheumatoid Arthritis classification criteria and who underwent arthroplasty at Peking Union Medical College Hospital. Patient basic information as well as C-reactive protein, rheumatoid factor (RF), anti-CCP antibody status, erythrocyte sedimentation rate (ESR), disease duration, and DAS28 score were recorded and presented in Supplementary Table 1 Sample preparation for 10× genomics. Synovial membrane tissues were washed with cold PBS, cut into small pieces, and digested with 3 mg/ml collagenase IV, 1.5 mg/ml dispase II, and 0.1 mg/ml DNase I in PBS containing 10% FBS at 37°C for 5 min. After three rounds of digestion, the cell supernatants were collected, combined, and passed through a 70-μm cell strainer. Mononuclear cells were isolated from peripheral blood and synovial membrane tissues using Ficoll-Paque gradient centrifugation (CL5020, CEDARLANE). CD45 + cells were isolated using magnetic beads (130-045-801, Miltenyi Biotech). Cell quantity and viability were assessed using Trypan Blue staining.
10× genomics single-cell RNA sequencing. Single-cell 3' gene expression libraries for CD45 + immune cells were prepared by strictly following the protocols of the Chromium Single Cell 3'v2 Library kit (10× Genomics). All resulting libraries were sequenced on the Illumina Xten PE-150 platform (ANOROAD and Novogene, Beijing, China).
Single-cell RNA-seq data alignment and quality control. Raw 10× Genomics sequencing data were processed using the CellRanger software v. 2.2.0, and the 10× human transcriptome GRCh38-1.2.0 was used as the reference. Single-cell read counts from all samples were first converted to individual Seurat objects using the Seurat (v3.1.1) analysis package in R (v3.6.1). For each object, we filtered data based on the number of unique molecular identifiers (UMI) and the number of genes detected. We retained genes expressed in at least five cells and cells in which the number of genes detected ranged from 500 to 3500. Each filtered Seurat object was then normalized using Seurat's NormalizaData function. Highly variable genes were detected using the FindVariableFeatures function in Seurat.
Integration of scRNA-seq data from the same tissue. All individual Seurat objects from the same tissue (PBMC or synovial tissue) were integrated using the FindIntegratedAnchors and IntegrateData functions in Seurat. Canonical correlations with highly variably expressed genes were then used for downstream dimension reduction and clustering analysis.
Dimension reduction and major cell type annotation. Peripheral blood mononuclear cells (PBMC) and synovial tissue objects were analyzed separately. The number of unique molecular identifiers, percentage of mitochondrial genes, and cell cycle genes were regressed out, and genes were scaled to unit variance. Principle component analysis (PCA) was performed. Clusters were then identified using Uniform Manifold Approximation and Projection (UMAP). Cell identity was assigned using known markers shown in Fig. 1C and Supplementary Figure 1A, C.
Sub-clustering of B cells, DCs, monocytes, macrophages, and T Cells. B cells, DC, monocytes, and macrophages, and T Cells were extracted from PBMC and synovial tissues for further sub-clustering. After extraction, genes were scaled to unit variance. PCA and clustering were performed as described in Dimension Reduction and Major Cell Type Annotation section. Doublet clusters were removed following the criteria described below: (1) B cell, DC, monocytes and macrophage subclusters in PBMC with CD3 expression (calculated as the mean expression of CD3D, CD3E, and CD3G) > 0.1; (2) B cell, DC, T cell subclusters in synovial with CD14 expression > 0.1.
Detection of differentially expressed genes and pathway analysis. Differential gene expression (DEG) testing was performed using the FindMarkers function in Seurat with Wilcoxon test and p values were adjusted using Bonferroni correction. DEGs were filtered using a minimum log 2 (fold change) of 0.5 and a maximum adjusted p value of 0.05 and were then ranked by average log 2 (fold change) and FDR. Enrichment analysis for the functions of the DEGs was conducted using the clusterProfiler (v3.12.0) package 32 and DAVID 6.8 33,34 . The gene sets were based on Gene Ontology terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways.
Trajectory analysis. The Monocle 3 R package was used to construct single-cell developmental trajectories for B cells [35][36][37][38] . The raw expression matrices were fetched from the associated Seurat object using the GetAssayData function 39,40 . After creating the CellDataSet object in Monocle 3, data were normalized and preprocessed using the preprocess_cds function. Next, the dimensionality of the data was reduced using UMAP with the reduce_dimension function. Trajectory inference analysis was performed using the cluster_cells and learn_graph functions.
Cell-cell interaction analysis. To comprehensively analyze cell-cell interactions between immune cells, we used CellPhoneDB (v2.1.2) 41,42 . We derived potential ligand-receptor interactions based on the expression of a receptor by one cell subpopulation and ligand expression by another. We separately fetched the normalized counts of HCs, ACPA-positive, and ACPA-negative cells and used them as input for the CellPhoneDB algorithm.
Data visualization. All plots were generated using the ggplot2 (v 3.2.1), pheatmap (v 1.0.12), and EnhancedVolcano (v 1.2.0) packages in R 3.6.1. Box plots are defined as follows: the middle line corresponds to the median; the lower and upper hinges correspond to the first and third quartiles, respectively; the upper whisker extends from the hinge to the largest value no further than 1.5× the interquartile range (or the distance between the first and third quartiles) from the hinge; and the lower whisker extends from the hinge to the smallest value at most 1.5× the interquartile range from the hinge. Data beyond the end of the whiskers were designated as "outliers" and are plotted individually.
HLA genotyping. DRB1-DRB5 haplotypes were determined by BGI (Beijing Genomics Institute) using the super high-solution single-molecule sequence-based typing (SS-SBT) method 43 . In brief, genomic DNA was extracted from whole blood samples and sequencing was performed using an ABI3130 (Life Technologies, Carlsbad, CA).
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Sequence data have been deposited in the Genome Sequence Archive in BIG Data Center, Beijing Institute of Genomics (BIG), Chinese Academy of Sciences, under the accession code HRA000155. Source data are provided with this paper.