Progression to rheumatoid arthritis in at-risk individuals is defined by systemic inflammation and by T and B cell dysregulation

Abstract

Rheumatoid arthritis (RA) is preceded by an at-risk stage of disease that can be marked by the presence of anti-citrullinated protein antibodies (ACPA) but absence of clinically-apparent synovitis (clinical RA). Preemptive intervention in at-risk individuals could prevent or delay future tissue damage, however the immunobiology of this stage is unclear. Using integrative multi-omics, we longitudinally profiled at-risk individuals, where one-third of participants developed clinical RA on study. We found evidence of systemic inflammation and signatures of activation in naïve T and B cells of at-risk individuals. During progression to clinical RA, proinflammatory-skewing of atypical B cells and expansion of memory CD4 T cells with signatures of activation and B cell help were present without elevations in circulating ACPA titers. Epigenetic changes in naive CD4 T cells suggested a predisposition to differentiate into effector cells capable of B cell help. These findings characterize pathogenesis of the ACPA+ at-risk stage and support the concept that disease begins much earlier than clinically-apparent synovitis. Additionally, an extensive immune resource of the at-risk stage and progression to clinical RA with interactive tools was developed to enable further investigation.

One Sentence Summary:

Individuals at risk for rheumatoid arthritis exhibit inflammation and immune dysregulation prior to and during transition to active disease.

Introduction

Rheumatoid arthritis (RA) is a destructive, systemic autoimmune disease estimated to affect 0.5–1% of the population. Disease symptoms prompt treatment with disease-modifying anti-rheumatic drugs (DMARDs) in a reactive response-dependent manner. Treatment failures or disease relapse are common despite treatment with best-in-class DMARD therapies (1). Data from various autoimmune diseases suggests that early diagnosis and treatment can reduce disease activity and prevent or delay tissue damage (2, 3).

Anti-citrullinated protein antibodies (ACPA) and rheumatoid factor (RF) are sensitive biomarkers of active disease but can often be detected, on average, 3–5 years prior to the onset of clinically evident synovitis (4–6), with a positive predictive value of 30–60% (7, 8). Individuals that are ACPA+ but otherwise clinically healthy are considered “at-risk” for future clinical RA (at-risk individuals, ARI). Building on successes with preemptive treatment in individuals at-risk for type 1 diabetes (3), a proactive intervention strategy for preventing clinical RA in ACPA+ ARI was proposed. Initial clinical trials that tested this hypothesis have achieved some successes. Rituximab delayed onset of clinical RA in ACPA+ and RF+ ARI, but did not reduce the overall rate of developing clinical RA compared to placebo in the PRAIRI trial (9). Abatacept is the only preemptive therapy that has decreased progression to clinical RA in ACPA+ ARI in the ARIAA and APIPPRA trials (10, 11), suggesting that adaptive immune responses play key roles in development of clinical RA in ACPA+ ARI.

Overall, a detailed immunobiological understanding of the “at-risk” period is lacking, hindering progress toward effective preemptive therapies. Cross-sectional studies have provided clues to immune effector mechanisms active in the at-risk state. These include elevated circulating inflammatory cytokines (12–16), expansion of pro-inflammatory CD4 effector T cell populations including antigen-specific Th17 cells, and lymphocyte activation characterized by lower glycolytic enzyme expression (17–21). Expanded T peripheral helper (Tph) cells and ACPA specificities suggest increased T cell-driven autoreactive B cell responses occur during the at-risk period. Additionally, increased IgA+ plasmablasts and shared IgA+/IgG+ clonal families suggest potential mucosal drivers of disease pathogenesis (14, 18, 22, 23). However, studies attempting to assemble a systematic view of immunobiology at baseline in ARI and longitudinally as they progress toward clinical RA are lacking.

Here, we performed a prospective, longitudinal study of ACPA+ ARI in which one-third of ARI developed clinical RA during the study. Utilizing a multi-omics systems immunology approach, we show that ACPA+ ARI with no evidence of active synovitis already have RA-like inflammation. We identify signatures of activation in naïve lymphocytes and expansion of key effector lymphocyte populations in ARI who develop clinical RA. Our findings suggest that inflammatory disease begins well before development of synovitis. An extensive immune resource of the at-risk stage and progression to clinical RA with interactive tools was developed to enable further investigation and can be explored at https://apps.allenimmunology.org/aifi/insights/ra-progression/.

Results

ARI exhibit signs of systemic inflammation prior to the onset of clinical RA

To define changes that contribute to the development of clinical RA, we studied a prospective cohort of 45 clinically healthy ACPA+ ARI (mean follow-up of 533 days), 11 patients with ‘early’ clinical RA (ERA), and 38 ACPA− healthy controls (HC1; Fig. 1A, fig. S1A, tables S1 and S2). During the study, 16 ARI (36%) progressed to clinical RA (‘converters’) (fig. S1B, table S3). Baseline (initial) ACPA titers were similar between ARI converters, non-converters, and ERA (Fig. 1B), whereas rheumatoid factor (RF) IgM and IgA were elevated in ERA (for RF IgA, FDR ≤ 0.04 compared to CONV and NONC; for RF IgM, FDR ≤ 0.04 compared to CONV, NONC, and HC1) (fig. S1C). To understand the systemic immune state in ARI, we compared the plasma proteome from their baseline visit to controls. We identified 272 proteins that differ between the two (249 (92%) higher in ARI and the remainder lower) (fig. S1D, table S4). We also observed enrichment of inflammatory pathways, including cytokine and chemokine signaling (fig. S1E, table S5). Ethnicity, which differed between controls and ARI, did not overtly contribute to these protein changes (fig. S1F). Baseline protein signatures were similar between converters and non-converters, but distinct from controls (fig. S2A and B, tables S6 to 8). Both converters and non-converters showed minimal differences in protein abundance compared to ERA (fig. S2C, tables S9 and 10), suggesting that some molecular features of RA begin prior to clinical manifestation.

Fig. 1: Active inflammation is observed in ARI prior to disease onset.

(A) Overview of study and multimodal workflow. HC1, controls; ARI, at-risk individuals; ERA, early RA; CONV, converters. (B) First sample (baseline) ACPA (anti-CCP3) measurements from HC1, NONC (non-converters), CONV, and ERA. (C) k-means clustering (k=6) of z-scored normalized protein expression (NPX) values from differential proteins in ARI vs. HC1 and ERA vs. HC1 (FDR<0.1). Rows denote proteins, columns denote baseline samples. ARI are annotated as CONV or NONC. (D) Abundance of select inflammatory plasma proteins elevated in ARI. Dots represent participant samples in each cluster. (E) Absolute concentration of select plasma proteins from participants in Prot-C1 and Prot-C6 clusters as assayed by Meso Scale Discovery or LegendPlex. (F) Number of differentially expressed genes (DEGs; FDR<0.1 and absolute log2 fold change≥0.1) per immune cell type, elevated in ARI (above 0) or HC1 (below 0). Cell types are based on (24). CM, central memory; EM, effector memory; ISG, interferon-stimulated gene. Boxplots show median (centerline) and first and third quartiles (lower and upper bound of the box); whiskers show the 1.5x interquartile range of data. Effect sizes and P values were determined by linear regression models (C), Wald test (F), Kruskal-Wallis test with Dunn’s post-hoc testing (B), or Wilcoxon rank-sum test (D and E). FDR values are indicated for all panels and values below 0.1 were considered significant.

To evaluate heterogeneity between individuals, we clustered all participants (ARI, ERA, HC1) using differential proteins identified in ARI and ERA (tables S4 and S11; see Methods). Most (40/56) ARI and ERA participants, including 13/15 converters and 21/30 non-converters, grouped in three of six clusters (Prot-C4, -C5, and -C6) (Fig. 1C, fig. S2D and E, table S12). Clusters Prot-C6 and Prot-C1 were dominated by ACPA+ individuals and controls, respectively, and exhibited the widest differences in protein abundance. Comparing these two clusters uncovered inflammatory proteins significantly increased in ARI and ERA (FDR < 0.1), including CXCL3 (FDR < 0.01), CXCL5 (FDR < 0.01), and CXCL13 (FDR = 0.05) (Fig. 1D, fig. S2F, table S13). We validated a subset of these protein differences in plasma from participants in clusters Prot-C1 and Prot-C6 (Fig. 1E). Together, these results demonstrate the presence of systemic inflammation in many ACPA+ ARI despite having no evidence of clinical RA.

Immune cells exhibit inflammatory gene programs in clinically healthy ARI

To understand the transcriptional state of immune cells exposed to the inflammatory milieu in ARI, we performed single-cell RNA sequencing (scRNA-seq) on peripheral blood mononuclear cells (PBMCs). Immune cell subsets were defined using the recent Allen Institute for Immunology Immune Cell Atlas (24) and confirmed by manual review (fig. S3A to C; see Methods). We observed increased abundance of CD4 central memory (CM) T cells and CD14 monocytes between ARI and controls (FDR < 0.1) (fig. S3D and E, table S14). In addition, we found a substantial number of transcriptome changes in multiple immune cell subsets (Fig. 1F, data file S1, see (25)) together with enrichment of glycolysis and oxidative phosphorylation pathways in many immune cell types (fig. S4A and B, table S15, data file S2, see (25)). Transcriptional changes were not affected by participant ethnicity (table S16). Transcriptional differences were positively correlated between converters and non-converters, relative to controls, with few differential transcripts directly distinguishing converters from non-converters (fig. S4C and D, data file S3, see (25)). The presence of numerous transcriptional changes in naïve populations from ARI suggested the potential of a primed state that may contribute to disease risk, similar to that described in clinical RA (26). In fact, transcriptional evidence for signaling downstream of inflammatory receptors was increased in CD4 T cells in both converters and non-converters relative to controls (FDR < 0.05) (fig. S4E, table S15, data file S2, see (25)). Taken together, the breadth of plasma proteomic and transcriptional changes in circulating immune cells are indicative of an ongoing inflammatory state in ACPA+ ARI, regardless of future conversion status.

Progression from at-risk to clinical RA is marked by systemic immune changes and emergence of inflammatory monocyte activity upon clinical RA diagnosis

Based on prior studies (4–6, 12–14), we hypothesized that an immune triggering event would precede onset of clinical RA. We focused on longitudinal samples from 13 female converters (Fig. 2A, see Methods). Converters had a range of baseline ACPA concentrations (fig. S5A), but most had no appreciable longitudinal increase in ACPA titers prior to and including the time of diagnosis of clinical RA. Among clinical labs, only platelet counts increased in converters during progression to clinical RA (FDR = 0.1) (fig. S5B, table S1). To evaluate global gene expression differences, we modeled longitudinal transcriptome variability in converters (mean 468 days to diagnosis) relative to ACPA− healthy controls (HC2; mean 355 days on study) (24) (fig. S5C, table S17). By coefficient of variation (CV), 23/27 cell types exhibited more intra-donor transcriptomic variability in converters compared with controls over a similar timeframe (FDR < 0.01) (Fig. 2B, data file S4, see (25)). Despite this temporal variability, the plasma proteome exhibited little change and most cell subsets had few consistent longitudinal transcriptome and abundance changes in converters (Fig. 2C and D, table S18, data file S5, see (25)). In contrast, naïve and CM CD4 T cells showed a large degree of transcriptional reprogramming without a change in abundance (Fig. 2D, fig. S5D to F, table S19 and 20). Thus, progression to clinical RA is marked by peripheral immune variability, with CD4 T cells emerging as the most differentially impacted.

Fig. 2: Longitudinal changes in naïve and CM CD4 T cells dominate progression to clinical RA.

(A) Overview of longitudinal comparison of converters (CONV) from ‘at-risk’ to clinical RA. (B) Number of genes per cell type with higher average intra-donor coefficients of variation (CVs) over time in CONV during progression to clinical RA (orange) or in HC2 (green). cDC, conventional dendritic cell (DC); NK, natural killer cell; pDC, plasmacytoid DC. (C) Comparison of the number of differentially expressed genes (DEGs) (y-axis) with the change in frequency over time (x-axis; centered log-ratio (CLR) transformed) as CONV progress to clinical RA. Bubble size corresponds to the number of DEGs. All cell type abundance changes were above FDR of 0.1. (D) Number of DEGs from longitudinal model (FDR<0.1) per cell type, elevated (above 0) or diminished (below 0) in CONV progressing to clinical RA. CM, central memory; EM, effector memory; ISG, interferon-stimulated gene. (E) Overview of paired comparison in converters at their last ‘at-risk’ pre-symptomatic visit vs. time of their clinical RA diagnosis. (F) Normalized RNA expression of TNF in Core CD16 monocytes and CXCL10 in ISG+ CD16 monocytes. (G) Mean RNA expression of select inflammatory genes amongst monocyte subtypes. IL1B+ CD14 monocytes (red) have highest expression of pro-inflammatory genes. (H) Gene scores calculated by comparing marker genes from FOLR2+ICAM+ RA synovial tissue macrophages (30) among all monocyte cell types. STM, synovial tissue macrophages. (I) Frequency of IL1B+ CD14 monocytes within total CD14 monocytes. Violin plot shape is proportional to the density of data points (dots). Larger width represents a higher density of data points. Effect sizes and P values were determined by linear mixed effect models (C, D), paired Wald test (F), ANOVA (H) or paired Wilcoxon test (I). Nominal P values are indicated for (H, I). FDR values are indicated for remaining panels. Values below 0.1 were considered significant.

In clinical RA, cellular and transcriptional changes precede RA flares (27). To determine if similar changes accompany clinical RA diagnosis, we performed a paired analysis comparing each converter’s PBMC from their last ‘at-risk’ pre-clinical visit (mean 122 days before diagnosis) to their disease diagnosis visit (Fig. 2E). Whereas most cell subsets had minimal transcript changes, TNF (FDR = 0.05) and CXCL10 (FDR = 0.01) increased in core CD16 monocytes and interferon-stimulated gene positive (ISG+) CD16 monocytes, respectively (Fig. 2F), and IGHA1 (FDR = 0.04) and JCHAIN (FDR < 0.01) increased in CD95+ memory B cells (fig. S5G, data file S6, see (25)). The highest expression of TNF and other pro-inflammatory genes was found in IL1B+ CD14 monocytes (28, 29) (Fig. 2G, fig. S5H), which exhibit a gene signature resembling FOLR2+ ICAM+ macrophages from RA synovial tissue (30) (Fig. 2H). TNF did not increase between visits in these cells (fig. S5I), but the number of IL1B+ CD14 monocytes expanded as a proportion of total monocytes at the onset of clinical RA (Fig. 2I, table S21). Together, longitudinal transcriptome changes in CD4 T cells with inflammatory monocyte changes emerging at the time of diagnosis characterize converters as they progress from at-risk to clinical disease.

B cells exhibit pro-inflammatory skewing during progression to clinical RA

Our cross-sectional analysis of ARI relative to controls revealed a large number of transcriptional changes in both naïve and memory B cells (Fig. 1F). Further, the presence of autoantibodies suggested autoreactive memory B cells (MBC) would be present among ARI and the finding of increased circulating inflammatory proteins indicated potential for chronic B cell activation before onset of clinical RA. We first focused on MBC, particularly CD27− effector B cells that have been shown to be expanded in RA synovial tissue (28). We identified two clusters of CD27− effector B cells, Beff-C8 and Beff-C9 (Fig. 3A, fig. S6A, table S22). Compared with Beff-C8, Beff-C9 had higher expression of genes corresponding to ‘age-associated’, double negative (DN) 2, Tbet+, or atypical B cells, including ITGAX, TBX21, and ZEB2 (31) (FDR < 0.01) (Fig. 3B, fig. S6B, table S23). Beff-C9 also had a higher proportion of immunoglobin heavy chain (IgH) class-switched cells (P < 0.01) (Fig. 3C) and higher mean expression of class-switched IgH genes compared with Beff-C8 (FDR < 0.01) (fig. S6C). An MBC cluster transcriptionally related to Beff-C9 in healthy controls remained stable over 2 years (fig. S6D and E). These data suggest that a subset of CD27− atypical or effector B cells with an activated transcriptional profile are found in converters as clinical RA approaches.

Fig. 3: The B cell compartment exhibits a pro-inflammatory skewing during progression to clinical RA.

(A) Uniform Manifold Approximation and Projection (UMAP) plots of MBCs from ARI, HC1 and ERA showing B cell population labels (left) and Leiden clusters (right). (B) DEGs for Beff-C8 compared with Beff-C9 with selected genes labeled (left). Dot size in heatmap (right) indicates the fraction of cells with positive expression for selected effector population marker genes. (C) Specified IgH isotype identity, as frequency within each population, for Beff-C8 and Beff-C9. UND, undetermined. (D) IGHG3 gene expression by core naïve B cells of ARI and HC1. (E) Normalized expression of IgH constant gene germline transcription of IGHMD⁺ naïve B cells that differ between ARI and HC1. (F) CLR-transformed cytometry frequencies of naïve B cell cluster Bnve-S5 as CONV progress to clinical RA. Each participant’s longitudinal series is connected by a gray line, with a group trendline and 95% confidence interval in purple. (G) GSEA enrichment analysis with the top Reactome pathways among naïve B cells of ARI compared with HC1. (H and I) B cells were stimulated ex vivo and analyzed by intracellular flow cytometry. Experimental workflow (H) and RANKL+, IL-6+ and TNF+ cell frequencies within the stimulated naïve B cell populations of ARI and HC2 (I). Boxplots show median (centerline) and first and third quartiles (lower and upper bound of the box); whiskers show the 1.5x interquartile range of data. Violin plot shape is proportional to the density of data points (dots). Larger width represents a higher density of data points. P values were determined by a linear mixed model (B, G), Wald test (E, F), or Wilcoxon rank-sum test (J). FDR values are indicated for all panels. Values below 0.1 were considered significant.

We hypothesized that chronic activation may influence B cell receptor signaling and IgH class-switching in ARI. IGHG3 (FDR = 0.1) and IGHM (FDR = 0.05) gene expression were elevated in naïve B cells among ARI (Fig. 3D, fig. S6F). Naïve B cell identity was confirmed by canonical marker gene expression and unswitched IgH identity (fig. S6G and H), suggesting that expression of IGHG3 is due to sterile or non-productive germline transcription (GLT). Using productive IgH isotype labeling, only IGHG3 GLT by unswitched (IGMD+) naïve cells was significantly increased among ARI (FDR = 0.02) (Fig. 3E). Detection of IGHG3 GLT by unswitched B cells was validated with scVDJ/RNA-seq data from HC2 (n=4; fig. S6I). This suggests a greater likelihood that naïve B cells of ARI will class-switch to the IgG3+ isotype upon activation, an intriguing finding given that patients with RA and other select autoimmune diseases are reported to have higher total IgG1 and IgG3 titers (32). A subset of naïve B cells (Bnve-S5) expanded in converters (Fig. 3F, fig. S6J and K), a feature previously reported to correlate with reduced PAX5 expression and activation in ACPA+ individuals (18). Indeed, PAX5 expression within naïve B cells from ARI was reduced (FDR = 0.04) (fig. S6L) and pathway analysis showed increased activation and antigen presentation activity compared with controls (FDR < 0.01) (Fig. 3G). An increased activation signature and IGHG3 GLT suggest that naïve B cells from ARI may be in a functionally “primed” state.

To determine whether the activated molecular profiles of naïve B cells corresponded to functional differences upon targeted perturbation, we stimulated ARI and control PBMCs and evaluated B cell cytokine expression (Fig. 3H). After stimulation, naïve B cells from ARI had higher frequencies of IL-6+ (FDR = 0.03) and RANKL+ (FDR < 0.01) cells compared with controls (Fig. 3I), suggesting they are primed for proinflammatory cytokine and RANKL secretion. Together these data suggest naïve and memory ARI B cell populations show signatures of activation with evidence of a naïve B cell primed state in ACPA+ ARI.

CD4 memory T cells with a B cell helper signature expand during progression to clinical RA

Changes in class-switched B cells imply concurrent activity in the T cell compartment. In fact, CM CD4 T cells had the highest number of longitudinal transcriptome changes in converters as they progressed to clinical RA (Fig. 2D). This included a signature of T cell activation, with downregulation of CD3G and CD247 (CD3Z) and upregulation of genes related to cytokine and antigen receptor signaling (STAT5B, STAT2, STIM2, AKT3, CD28 and FOSB) (Fig. 4A and B, fig. S7A, table S24; see T cell activation score in Methods). These results suggest ongoing involvement of CD4 T cells during preclinical disease development.

Fig. 4: Effector and memory T cells with pathogenic signatures expand during progression to clinical RA.

(A to C) Longitudinal analysis of CONV who progress to clinical RA. (A) RNA expression differences in central memory (CM) CD4 T cells. Genes associated with T cell activation are noted. (B) T cell RNA activation metric in CM CD4 T cells. See Supplemental Methods for details. Each participant’s longitudinal series is connected by a gray line, with a group trendline and 95% confidence interval in purple. (C) Frequency of CD4mem-C3 cells over time, as in (B). (D) Cells expressing Tfh gene program are distinguished based on the non-negative matrix factorization projection using a pre-computed weight matrix of CD4 T cells from (33) (left). A UMAP density plot of cluster CD4mem-C3 is shown (right). (E) Gene scores calculated by comparing the Tph/Tfh gene signature from Zhang et al. (28) among all CD4mem clusters. Cluster C3 (red) was expanded during progression to clinical RA. (F) Mean RNA expression of select genes across CD4mem clusters. Cluster C3 (red) was expanded during progression to clinical RA. (G) Normalized RNA expression of select genes that promote differentiation to B cell helper and Th17 cells in CD4mem-C3 (red) vs. remaining CD4mem clusters (blue). (H) Differentially expressed genes between CD4mem-C3 (red) and remaining CD4mem clusters (blue). Select genes associated with T cell activation are labeled. Violin plot shape is proportional to the density of data points (dots). Larger width represents a higher density of data points. P values were determined by linear mixed models (A, C), the Kruskal-Wallis test with pairwise Dunn’s posthoc test (E) or the Wilcoxon rank-sum test (G-H). Nominal P values are indicated for (B, E). FDR values are indicated for all other plots. Values below 0.1 were considered significant.

To understand how the activation signature impacts effector identities during clinical RA development, we leveraged a human CD4 T cell reference dataset (33) to analyze memory gene programs in converters. Cluster CD4mem-C3, consisting of 80% CM CD4 T cells (fig. S7B to D), increased in abundance during progression to clinical RA (FDR = 0.03) (Fig. 4C, fig. S7E and F, table S25). This cluster aligned to the T follicular helper cell (Tfh) gene program from the reference dataset (Fig. 4D, fig. S8A and B) and a Tfh/Tph signature from RA synovial tissue (28) (Fig. 4E). Cluster CD4mem-C3 exhibited elevated expression of genes important for T cell differentiation, effector polarization and B cell help (PDCD1, CXCR5, ICOS, MAF, NEAT1, KLRB1) (FDR < 0.01) (Fig. 4F to H, table S26) (34–37). In an independent multi-modal single cell dataset (38), the CD4mem-C3 conversion signature overlapped a PD1+CXCR5+ICOS+KLRB1+ population (cluster 7) at both the transcriptome and surface proteome level (fig. S8C to G). Indeed, this population had the highest surface protein expression of PD1 and ICOS (fig. S8F and G). Cluster CD4mem-C3 also expressed an overlapping activation signature with CM CD4 T cells (Fig. 4A and H, fig. S8H). The RNA CD4mem-C3 frequencies showed a modest correlation with frequencies of an activated Tfh-like cluster (CD4mem-E6: PD1⁺CXCR5^midTIGIT⁺ICOS⁺) by flow cytometry (rho=0.38) (fig. S9A to D, table S27), which may reflect the heterogeneity and sample size limitations of our cohort. These findings suggest that the increased population of CM CD4 T cells in converters approaching clinical RA is driven predominantly by changes in activated T cells with a B cell helper gene signature.

Prior studies indicate that Tfh cells directly support effector or atypical B cell development and activation (39, 40) and, conversely, that effector B cells help drive optimal Tfh cell formation through antigen presentation (41, 42). To gain insight into T-B interactions that support the observed activation and expansion in converters during progression to clinical RA, we inferred ligand-receptor interactions and potential downstream gene regulation in CD4mem-C3 and Beff-C9 (fig. S10A to D, tables S28 and 29). We identified key activation complexes between T cell ligands and B cell receptors, including SEMA4D-CD72, CD40LG-CD40, and ST6GAL1-CD22, and found upregulation of ligand genes in CD4mem-C3 cells (fig. S10A). Several of these T cell ligands showed potential for regulation of B cell maturation and homing genes, such as AICDA, TBX21, and CXCR4 (fig. S10B). Further, Beff-C9 cells highly expressed several ligand genes that engage with critical activating T cell receptors, including CD86-CD28, HLA-DRB1-CD4, and HLA-DQB1-CD4 (fig. S10C), corresponding to potential for Tfh priming through antigen-presentation by activated effector B cells, with putative regulation of genes for T cell function and differentiation, such as EOMES, IKZF2, MAF and AIM2 (fig. S10D). Together, these data suggest that CD4 memory and CD27− effector B cell subsets may be providing bi-directional activating signals that support their mutual activation and expansion in progression to clinical RA.

Epigenetic changes in naïve CD4 T cells are linked to NFAT-calcium activation and B cell helper phenotype bias in ARI

Given the systemic inflammation and longitudinal increase of a CD4 memory T cell population in converters, we hypothesized that naïve CD4 T cells are in a state of heightened activation. We performed Multi-Omics Factor Analysis (MOFA) (43) on data from a subset of participants’ plasma proteomes and from a trimodal single-cell assay (44) on their PBMCs to link the systemic milieu, surface protein phenotype, transcriptional program, and transcription factor (TF) activity (Fig. 5A, table S1, fig. S11A; see supplementary methods). This analysis resulted in grouping the various data into one of multiple factors that describe variation within the combined dataset. In comparing a subset of ARI and matched control participants, Factor 1 explained the highest variation in the transcriptome (Fig. 5B) and best differentiated ARI from controls (FDR < 0.01) (Fig. 5C, fig. S11B, table S30), with transcripts encoding calcium responsive proteins (NFAT5, NFATC3, STIM2; FDR < 0.1) elevated in ARI (Fig. 5D). Moreover, within Factor 1, we detected increased NFAT TF motifs and decreased FOX TF motifs in accessible chromatin regions of ARI (FDR < 0.05) (Fig. 5E, fig. S11C), suggesting naïve CD4 T cells from ARI are in an activated state.

Fig. 5: Naïve CD4 T cells in ARI have a multimodal signature of activation.

(A) PBMC TEA-seq was performed on a subset of ARI and HC2 samples. (B) Percentage of variance in each modality (surface protein, plasma protein, RNA, ATAC) explained by Multi-Omics Factor Analysis (MOFA) factors. (C) Factor 1 scores comparing ARI and HC2. (D) Scaled normalized expression of select genes in Calcium–Calcineurin–NFAT pathway in ARI and HC2. (E) Inferred accessibility for the top 15 transcription factors (TFs) positively or negatively associated with factor 1, ranked by weight. (F) RNA expression differences in core naïve CD4 T cells over time in CONV (orange) who progress to clinical RA (purple). Genes associated with T cell activation are annotated. (G) T cell RNA activation metric in core naïve CD4 T cells over time as CONV progress to clinical RA. Each participant’s longitudinal series is connected by a gray line, with a group trendline and 95% confidence interval in purple. (H) Enriched pathways associated with T cell activation in core naïve CD4 T cells over time as CONV progress to clinical RA. P values were determined by Wilcoxon rank-sum test and FDR were adjusted for all MOFA factors tested (C). P value was determined by linear mixed models (F, G). Normalized enrichment scores (NES) and adjusted P values, by GSEA, are shown (H). Nominal P value is shown in (G). FDR values are indicated for remaining plots. Values below 0.1 were considered significant.

To test whether signatures of activation in naïve T cells accompany progression to clinical RA, we examined longitudinal transcriptome changes in converters. Naïve T cells showed a large number of transcriptional changes, highlighted by downregulation of CD3 complex transcripts and upregulation of cytokine and antigen receptor signaling components (Fig. 5F and G, fig. S11D and E), the latter including STATs, STIM2, and CD28 (fig. S11F and G). Gene set enrichment analysis (GSEA) indicated upregulation of activation pathways, including NFAT and TGF-β signaling, in both CD4 and CD8 naïve T cells (Fig. 5H, fig. S11H, table S31). Together, these results are indicative of an activated state in naïve T cells from ARI during progression to clinical RA.

We next investigated mechanisms responsible for changes in CD4 T cells. We hypothesized that naïve cells are epigenetically poised to preferentially differentiate into pathogenic effector cells. By TEA-seq, there were 3,159 differentially accessible chromatin regions between ARI and controls, including 2,200 near promoters (data file S7, see (25)). Clustering all CD4 T cells based on the ATAC modality of TEA-seq (Fig. 6A) revealed two CD45RA+ naïve clusters (CD4nve-T2, -T5) with a higher frequency in ARI (Fig. 6B) and three CD45RO+ memory clusters (CD4eff-T6, -T9, -T10) with a lower frequency in ARI (fig. S12A). CD4nve-T2 and CD4nve-T5 were epigenetically, though not transcriptionally, distinct (fig. S12B). They exhibited effector-like phenotypes of lower CD62L and CD162 surface protein expression compared with other naive clusters (FDR < 0.01), based on the protein epitope modality of TEA-seq (Fig. 6C) and high chromatin accessibility at the CXCR5 and IL21 loci (Fig. 6D). Only naïve cluster T2 resembled effector cluster T1 with TF motif enrichment for SMADs, STATs, and AP1 (fig. S12C, table S32), suggesting evidence for an activated or pre-activated state. Likewise, CD4nve-T2 had the highest predicted activity for BCL6 and STAT3 (Fig. 6E). Moreover, although minimal IL21 transcript was detected (fig. S12D), an accessible 500bp intronic region in the IL21 locus was present only in naïve CD4 T cells from ARI (Fig. 6F). This region was shown to be more accessible in human tonsil Tfh cells (45) and overlaps a putative enhancer-like structure that contains motifs for key TFs that drive differentiation to effector T cells capable of B cell help, including BCL6 and STAT3 (46). These results identify a putative regulatory region in naïve CD4 T cells that is permissive to induction of IL-21 specifically in ARI and suggest a link between naïve T cell activation in ARI and differentiation to effector cells that promote B cell responses.

Fig. 6: Naïve CD4 T cells exhibit a bias toward B cell help in ARI.

(A) Louvain clusters in CD4 T cells by ATAC modality in TEA-seq. (B) Centered log-ratio (CLR)-transformed frequencies of ATAC clusters CD4nve-T2 and CD4nve-T5 in CD4 T cells. (C) Mean surface protein expression of select markers differentiating CD4 naïve, memory, regulatory T (T_reg), and cytotoxic CD4 T cells (CTL) across ATAC clusters. (D) ATAC UMAP overlaid with inferred gene activity scores for CXCR5 and IL21. (E) ChromVAR TF activity Z scores of BCL6 and STAT3 in CD4 T cells. (F) ATAC signal in ARI (orange) versus HC2 (green), and the delta between the two (red) at the IL21 locus. The gray box highlights a 500bp region containing differentially accessible peaks between ARI and HC2 (chr4: 122,617,500–122,617,999). Black arrows indicate the motif locations of BCL6 and STAT3 binding sites. Gene bodies are displayed on the bottom. Boxplots show median (centerline) and first and third quartiles (lower and upper bound of the box); whiskers show the 1.5x interquartile range of data. P values were determined by linear models (B) or by the zero-inflated Wilcoxon test (F). FDR values are indicated and values below 0.1 were considered significant.

Gene signatures in CD4 T cells of converters reflect Abatacept treatment response

To establish the disease relevance of T cell changes observed in converters, we compared our data to that from clinical trials treating RA patients with Abatacept (ABT; CTLA4-Ig) or, as a control, TNF inhibitor (TNFi) (47, 48) (Fig. 7A). ABT is hypothesized to dampen T cell costimulation in RA (49). Naïve and CM CD4 T cell gene signatures found in converters during progression to clinical RA were significantly modulated by treatment in ABT responders (FDR < 0.05), but not non-responders (Fig. 7B, fig. S13A and B, table S33). Conversely, TNFi treatment did not modulate gene signatures found in converters (Fig. 7C, data file S8, see (25)). Indeed, the majority of genes in the naïve and CM CD4 T cell gene signatures were inversely correlated between converters and RA responders post-ABT (Fig. 7D), but no such correlation existed post-TNFi (Fig. 7E). Further, ABT-driven changes were noted in genes previously implicated in RA-like disease (CD8a, CD99, CDK4, CXCR3, FLT3LG, GZMM, LTB, and TNFRSF14) and those related to the Th17 pathway (IL27RA and NABP1) (Fig. 7F and G, table S33). Naïve and CM CD4 T cell conversion signatures have 20 times higher odds of being reversed by ABT treatment compared with TNFi (P < 0.01) (Fig. 7H). These data suggest that T cell mechanisms relevant to clinical RA could contribute to disease pathogenesis in ARI and may provide mechanistic evidence supporting the role of ABT in delaying onset of clinical RA (10, 11).

Fig 7: Gene signatures in CD4 T cells of converters reflect Abatacept treatment response.

Longitudinal DEGs as CONV progress to clinical RA were assessed within the context of RA patients with efficacious (responders) or non-efficacious (non-responders) clinical response to abatacept (ABT) or TNF inhibitor (TNFi) treatment (from (47, 48)). (A) Overview of the analysis strategy. mo., month. (B and C) Over-representation of CONV cell type-specific longitudinal DEGs amongst ABT (B) or TNFi (C) response DEGs. (D and E) Significant DEGs in ABT (D) or TNFi (E) responders compared to DEG changes over time as CONV progress to clinical RA in core naïve CD4 T cells (left) and CM CD4 T cells (right). Genes (dots) previously implicated in RA-like disease are labeled. (F and G) Normalized RNA expression of NABP1 over time as CONV progress to clinical RA (F) and pre- vs. post-ABT therapy in patients with RA (G). In (F), each participant’s longitudinal series is connected by a gray line, with a group trendline and 95% confidence interval in purple. In (G), each participant’s two samples are connected by a gray line. (H) Odds ratios of the number of longitudinal DEGs in core naïve and CM CD4 T cells from CONV that were reversed by ABT or TNFi treatment. Error bars indicate confidence intervals. P values were determined by hypergeometric enrichment tests (B, C), McNemar’s Chi-squared test (D, E), linear mixed models (F), or Wald test (G). Nominal P values are indicated for (C, D). FDR values are indicated for other panels. Odds ratios were determined by unconditional maximum likelihood estimation method and Z-test was used to compare odds ratios between Abatacept and TNFi treatment (H). Values below 0.1 were considered significant.

Discussion

ACPA, a hallmark of seropositive RA, usually develop prior to the onset of active clinical arthritis and define a pre-clinical, at-risk stage of disease (4, 5). The state of the immune system during the at-risk period and immune mechanisms that underlie progression from at-risk to clinical RA remain unclear. We show immune activation in ARI prior to clinical RA, reflected by elevated circulating cytokines and chemokines, activation signatures of naïve T and B cells, and expansion of an effector T cell population associated with inflammatory responses and B cell help. The inflammatory changes mirror those observed in early and established RA, and in inflamed RA synovial tissue (28–30).

During progression to disease, we were surprised that transcriptional changes were predominantly restricted to naive and CM CD4 T cells. Within the CM compartment, we detected longitudinal expansion of a memory population with a B cell helper phenotype. However, we were unable to distinguish whether they are Tfh or Tph cells based on gene expression. This population also expressed Th17-like features, including MAF, a key TF promoting Tfh and Th17 cell fates (34, 50, 51). This population could be related to Tfh17 cells, an IL-17A-producing subset of Tfh correlated with circulating autoantibodies in juvenile dermatomyositis and Hashimoto’s thyroiditis and known to efficiently promote B cell function (52, 53). We hypothesize this memory population arises from naïve T cells biased towards a B cell helper phenotype (54). This included an epigenetic signature of more accessible BCL6 and STAT3 motifs and increased accessibility around the regulatory region of the IL21 gene (55). Moreover, naïve T cells in ARI exhibited transcriptional and epigenetic signatures that suggest activation through calcium-NFAT. We speculate this may prime signaling downstream of the T cell receptor, lowering the threshold for activation by autoantigens, consistent with previous reports in clinical RA (26).

To evaluate the relevance of the naive and memory CD4 T cell conversion signatures, we leveraged public transcriptomic datasets from clinical trials in active RA (47, 48). This suggests that CTLA4-Ig was more effective than TNFi at modifying the transcriptomic signature of converter CD4 T naive and memory cells as disease approached. Seeing that these signatures were modified only in the subset of RA patients that responded clinically to CTLA4-Ig suggests a pathogenic role for CD4 T cell activation in disease. Intriguingly, the CD4 memory population that increased in converters had the highest expression of CD28, which also marks a subset of Tfh proposed to be a target of CTLA-4 modulation (56). This raises the question of whether integrated multi-omics profiles or longitudinal tracking of patients with complex immune diseases could be used to direct therapies to disease subgroups that have evidence of specific driver mechanisms.

Autoreactive and activated B cell populations contribute to clinical RA symptomology and to disease progression (57, 58). A recent study identified a CD27− B cell subset elevated in ARI ‘progressors’ (59). We identified a CD27-ITGAX+TBX21+ effector memory B cell subset with transcriptional features suggestive of chronic B cell receptor engagement and activation, including elevated class-switching, and a gene program associated with long-lived humoral immunity after vaccination (60). Corresponding populations, termed atypical or DN2 B cells, are known to expand in autoimmunity and chronic infection (31, 42, 61). Cell communication analysis suggested that the expanding memory CD4 T cell population could support activation and maturation of this effector B cell subset and vice versa, suggesting a potential pro-pathogenic feed-forward loop of ongoing mechanistic and clinical interest. Furthermore, in clinical RA, subsets of CD11c+ (encoded by ITGAX) atypical memory B cells have been shown to be clonally related to expanded plasma cells in the periphery and synovial tissue (62, 63), highlighting this population as a potentially interesting target in preventative therapies.

Just prior to conversion, a subset of CD14 monocytes acquired a TNF+IL1B+ phenotype. The transcriptional signature of this population resembled FOLR2+ICAM+ macrophages isolated from RA synovial tissue (30). These cells may contribute to the development of a Th17-like signature in the expanding memory T cell pool, as CD14 monocytes isolated from RA synovium can promote pro-inflammatory Th17 differentiation (64). Whether the circulating TNF-expressing monocytes we identified traffic to the joint prior to clinical symptoms or because of existing joint inflammation remains to be determined and could be informative for understanding initiation of joint pathology. Concurrently, we found elevated CXCL10 expression in ISG+ CD16 monocytes. However, it remains to be determined whether this transcriptional change results from circulating interferons or other mechanisms.

Numerous features of immune dysregulation were found in ARI in our cross-sectional comparisons with healthy controls. Unexpectedly, converters and non-converters had very similar profiles at baseline, with few statistically different biomarkers. Given the molecular and cellular immune heterogeneity in clinical RA (28, 30, 65) and the observed heterogeneity in baseline immune features of ARI, our study is likely not powered to detect subtle changes at baseline in converters and non-converters. This is further confounded by non-converter heterogeneity, as we do not know when or if non-converters will develop clinical RA.

Multiple ‘hits’ are likely required for conversion, consistent with prior studies showing inflamed tissue is associated with citrullinated peptides but is not sufficient for ACPA development (66, 67). Our data suggests these hits include the longitudinal change in the CD4 memory population expressing features associated with B cell help and the acquisition of a TNF+IL1B+ phenotype in CD14 monocytes at conversion. Another contributor to conversion could be the DNA methylome, which distinguishes converters from non-converters at baseline (68). Changes within the synovial tissue could permit or exacerbate inflammatory damage, such as through Granzyme-dependent complement activation (69). Finally, breaks in mucosal barriers driving chronic exposure of immune cells to damaged self-proteins or immunostimulatory pathogen-derived products could promote innate immune cell activation and priming of adaptive cells (70–73). It will be important to understand the spatial and temporal organization of these features in the preclinical joint and how they contribute to clinical pathology.

Our study has limitations. Recruitment through health fairs, social media, community outreach, clinic referrals and first-degree relatives could lead to bias based on race, ethnicity or socioeconomic status. The ability to generalize our findings is not yet clear given that our cohort consisted of primarily non-Hispanic White individuals. Participants differed in ethnicity and age across the groups. We included age as a covariate in our models and we evaluated the effect of ethnicity; however, ethnicity did not impact the models. The onset of clinical RA in converters was assessed by clinical evaluations and was defined as the presence on physical examination of a swollen joint consistent with inflammatory arthritis or synovitis. This is currently the standard for a clinical diagnosis of RA. We acknowledge this may introduce subjective bias; however, we attempted to minimize bias by working with rheumatologists and trained study nurses to evaluate this key outcome. Given the observed heterogeneity and small number of converters, our prospective study may not be powered to detect subtle differences between converters and non-converters. Independent studies of at-risk individuals, ideally longitudinal, are needed to validate these findings. Lastly, the state of cells in the peripheral blood may not fully reflect events occurring in the synovium, and paired synovial tissue would provide important insights. However, the synovium is often less fulminant in individuals with new onset of synovitis and more difficult to biopsy.

Together, these data provide a detailed, systems view of the immunologic features of the ACPA+ at-risk stage of RA and of immune changes that accompany progression to clinical disease. Our results support the concept that RA inflammatory disease begins well before the onset of active synovitis, earlier than clinically appreciated. This has implications for decisions about when to initiate preemptive treatment. Clinical trials for RA prevention have already begun in ARI (9–11, 74), and we anticipate that these data will help guide future therapeutic choices that may be more effective at delaying or preventing RA.

Materials and Methods:

Study Design

The goal of this study was to uncover the immune features associated with the at-risk stage of RA and those immune changes that accompany progression to clinical RA. The following groups were included in this study: (1) At-Risk Individuals (ARI; n = 45), defined as having serum ACPA positivity using the anti-cyclic citrullinated peptide-3 (CCP3) IgG enzyme-linked immunosorbent assay (ELISA; Werfen) with no history or evidence of inflammatory arthritis (IA)/synovitis on physical examination of 66/68 joints at their baseline study visit; (2) Early RA (ERA; n = 11), defined as having on physical examination ≥1 swollen joint consistent with IA and anti-CCP3 positivity, with their initial study visit taking place within 1 year of the initial confirmation of IA by a rheumatologist; (3) ACPA− Controls (HC1; n = 38), defined as being anti-CCP3 negative with no history or evidence of IA and recruited at the University of California San Diego (UCSD) and the University of Colorado Anschutz Medical Campus (CU); (4) Longitudinal healthy ACPA− controls (HC2; n = 29), defined as being anti-CCP3 negative with no history of IA, autoimmune disease, or chronic disease and recruited at the Benaroya Research Institute (BRI) (24). The outcome of clinical RA in ARI (converters; n = 16 ARI) was based on clinical evaluation and a finding on physical examination by a rheumatologist or trained study nurse of ≥1 swollen joint consistent with IA. All ERA met the ACR/EULAR 2010 RA classification criteria. Further details are provided in the Supplementary Materials. Clinical data associated with the study were finalized as of December 18, 2023. The original sample size for the ARI group was based on the expected incidence rate of approximately 30% for clinical RA over approximately 3 years. No randomization or blinding was performed. All studies were approved by ethical review boards: UCSD and CU (Colorado Multiple Institutional Review Board #19-1150), BRI (BRI Institutional Review Board #IRB19-045), and the Allen Institute for Immunology. All participants provided written informed consent prior to participation in these studies. Consistency in sample collection and handling between the three sites was ensured using a common lab manual, common protocols, on-site training prior to the start of work, and regular coordination meetings. Non-clinical assays on all samples were performed at the Allen Institute, except for plasma proteomics, which was outsourced (Olink).

Statistical Analysis

All statistical tests were two-tailed. P values were calculated using linear regression models, linear mixed models, Wald test, Wilcoxon rank-sum test, zero-inflated Wilcoxon test, hypergeometric test, Spearman correlation, Kruskal-Wallis test with Dunn’s posthoc test, ANOVA, McNemar’s chi-squared test, paired Wald test, or paired Wilcoxon test as indicated. P values were corrected for multiple testing using the Storey-Tibshirani, Benjamini-Hochberg, or Bonferroni methods as indicated. False discovery rates (FDR) below 0.1 were considered significant, in order to prioritize hypothesis generation while controlling for false positive rate. For transparency, individual FDR values are reported in each figure panel. Specific details on statistical tests performed for each analysis are described in Supplementary Methods. Individual-level data for n < 20 can be found within the corresponding supplemental table and are presented in data file S9, see (25).

Supplementary Material

List of Supplementary Materials:

Supplementary Materials and Methods

Figs. S1 to S15

MDAR Reproducibility Checklist

Supplied as an auxiliary .xls file.

Tables S1 to S36

Hosted on Dryad:

Data files S1 to S9

DOI: 10.5061/dryad.1rn8pk13z

Data and Materials Availability:

All data associated with this study are in the paper or supplementary materials. Due to ethical considerations, IRB constraints and existing contracts, human source material are not transferable to a third party. Processed scRNA-seq data from human PBMCs generated in this study can be downloaded from GEO under accession number GSE274680. Raw scRNA-seq fastq files were deposited in dbGap (phs003944.v1.p1). Data files S1 to S9 are hosted on Dryad (DOI: 10.5061/dryad.1rn8pk13z). There are no other proprietary materials related to this work. Abatacept treatment data from RA patients was derived from Iwasaki et al. (2024) (47) and can be downloaded from Zenodo (https://zenodo.org/records/8250013). TNF inhibitor treatment data from RA patients was derived from Farutin et al. (2019) (48) and can be downloaded from GEO (GSE129705). T cell TEA-seq data was derived from Thomson et al. (2023) (38) and can be downloaded from Zenodo (https://zenodo.org/records/14783021). R and Python code used to perform analyses and generate figures is provided in Zenodo (DOI: 10.5281/zenodo.17064925) and GitHub (https://github.com/aifimmunology/ALTRA-manuscript/). Data can be accessed and explored interactively at https://apps.allenimmunology.org/aifi/insights/ra-progression/ (DOI: 10.57785/c0pq-s567).