Inflammatory arthritis irAE may represent a unique autoimmune disease primarily driven by T cells but likely not autoantibodies

Abstract

The underlying immunopathogenesis of inflammatory arthritis (IA) immune-related adverse event (irAE) remains obscure. Unlike rheumatoid arthritis (RA), where autoantibodies and B cell dysfunction are central features, the contribution of humoral immunity to IA-irAE is unclear. Here, we performed immunophenotyping of peripheral blood from patients with IA-irAE and compared them with patients with seronegative RA, immune checkpoint inhibition–treated patients without irAE, and healthy controls. IA-irAE was marked with increased cytotoxic gene expression and metabolic activation in T cells and reduced CXCR3 and CCR6 expression in CD4⁺ T cells. Contrary to seronegative RA, patients with IA-irAE displayed no substantial elevation in autoantibody levels or atypical CD11c⁺CD21⁻ B cells. IA-irAE was further characterized by elevated levels of interleukin-6 (IL-6), IL-12, and type I interferon, which correlated with the T cell activation phenotypes. Together, our findings define IA-irAE as a disease with certain immunological features distinctive from RA, representing a potentially T cell–driven, autoantibody-independent autoimmunity. These results offer insights into immune tolerance breakdown and therapeutic targeting in irAEs.

Inflammatory arthritis irAE is a T cell–driven, autoantibody-independent autoimmunity.

INTRODUCTION

While the revolutionary immune checkpoint inhibition (ICI) therapy invigorates antitumor immunity, it also leads to various immune-related adverse events (irAEs). Rheumatic irAEs account for about 5% of all irAEs. Most of the rheumatic irAEs are considered inflammatory arthritis (IA) (1, 2). From an immunological point of view and based on genetic mouse models, the development of irAEs could be considered an expected outcome because ICI therapy blocks key inhibitory receptors, Cytotoxic T-lymphocyte associated protein 4 (CTLA-4) or Programmed cell death protein 1 (PD-1). CTLA-4 deficiency leads to severe systemic autoimmunity and early death (3), whereas PD-1 deficiency results in autoimmune diseases that vary in onset, severity, and affected tissues depending on genetic background (4–6). Antibodies targeting PD-1 or its ligands, Programmed Death-Ligand 1 (PD-L1), are the most used ICI agents in clinical practice. So far, the immunological underpinnings of the irAE development remain elusive. Furthermore, while it is widely assumed that ICI therapy primarily targets CD8⁺ T cells, PD-1 is expressed and functional on CD4⁺ T cells and non–T cells, including B cells (7). However, its impact on patient humoral immunity remains contentious.

The impact of PD-1 inhibition on humoral immunity has been mostly explored in the context of vaccination. While some studies showed that anti–PD-1/PD-L1 agents do not notably affect the titer of anti–severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike protein antibodies following COVID-19 vaccination (compared to non–anti–PD-1/PD-L1 treatment) (8–11), one study found that anti–PD-1/PD-L1 treatment is associated with substantially increased post–COVID-19 vaccination SARS-CoV-2 breakthrough infection (12). An elegant influenza vaccination study found that anti–PD-1 treatment reduces the quality of the influenza-specific antibodies (13). A genetic study on individuals with complete loss of PD-1 demonstrated that B cell–intrinsic PD-1 is critical for memory B cell formation and the production of antibodies against common microbial pathogens (14). Last, one study found that anti–PD-1/PD-L1 treatment is associated with reduced autoantibody profile compared to anti–CTLA-4 treatment (15). Thus, existing data suggest that anti–PD-1/PD-L1 treatment might attenuate vaccine-elicited immune protection, and B cell–intrinsic PD-1 may enhance humoral immunity.

Previous studies using relatively small patient cohorts, and different control groups, have revealed that CD8⁺ T cells from patients with IA-irAE have elevated glucose metabolism at baseline (16), and there is a clonal expansion of CD38^hi cytotoxic, CX3CR1⁺ effector CD8⁺ T cells associated with elevated expression of type I interferon (IFN)–inducible genes and increased T helper 1/2 (T_H1/2)–like T cell transcription signatures in these patients (16–19). However, several important questions remain, including any biomarkers distinguishing ICI-treated patient without irAE from those with IA-irAE, the involvement of humoral immunity, detailed phenotypes of CD4⁺ T cells, and the relationship between IA-irAE and rheumatoid arthritis (RA) with shared clinical and serological phenotypes.

It has been well recognized that systemic breakdown of self-tolerance during the pathogenesis of RA usually involves both autoreactive T cells and autoreactive B cells, which generate autoantibodies, often with T cell help (20, 21). Autoantibodies may lead to arthritis by forming immune complexes (ICs) to promote complement deposition in joint tissues. However, there is growing evidence that IA could be antibody independent because complement fixation or ICs can be absent in some patients with IA (22). Animal models further demonstrate the possibility of T cell–driven and B cell–independent arthritis, including SKG mice (23) and interleukin-1 receptor antagonist (IL-1ra)–deficient mice (24). Nonetheless, the existence of antibody-independent arthritis in humans remains a contentious issue. The immunological basis for possible T cell–mediated antibody-independent arthritis in humans is poorly defined.

Here, we perform comprehensive immunological analyses, integrating single-cell RNA sequencing (scRNA-seq), flow cytometry, proteomics, multiplex assays, and cell culture on a relatively large cohort of patients with IA-irAE, with sex-, age-, and serologically matched patients with RA, patients with ICI-treated cancer without irAE, and healthy controls. Our data demonstrated that IA-irAE has certain distinct immunological features compared with RA, associated with increased cytotoxic T cell signatures, metabolic activities, reduced expression of CXCR3 and CCR6 on CD4⁺ T cells, and highly elevated levels of multiple cytokines and chemokines but not associated with increased autoantibody production or pathological atypical B cells. These phenotypes are partly driven by increased IL-6, IL-12, and type I IFN. Thus, our results indicate that IA-irAE is potentially a T cell–driven and autoantibody-independent disease entity, which may separate it from RA.

RESULTS

Patient characteristics

From 2017 to 2024, we enrolled and performed analysis on 171 patients, who were divided into four groups: rheumatic irAE (irAE; n = 34), ICI control (n = 26), RA control (RAC; n = 47), healthy control (HC; n = 64) (summarized in table S1). The average age of the patients in our cohort was 62. In the irAE and ICI control groups, the majority were male (58.8 and 57.7%, respectively), while in the RAC group, males were a minority (38.3%). Both patients with irAE and ICI controls had a mixture of different cancer diagnoses, and most (97.1% for irAE and 88.5% for ICI) were at tumor stage IV. In terms of ICI medicine, the majority of patients with either irAE (97.1%) or ICI control (88.5%) were treated with anti–PD-1/PD-L1 therapy. Following the immunotherapy treatment, all the patients in the irAE group developed IA (100%, 34/34), while other irAEs included systemic lupus erythematosus (2.9%, 1/34) and antisynthetase syndrome (2.9%, 1/34). The majority of the patients with irAE received steroid treatment (76.5%, 26/34), which was different from the medications in the RAC (14%, 6/43 on steroid treatment). After ICI treatment, more patients with cancer in irAE group had a complete response than those in ICI control group (20/34, 58.8% versus 9/26, 34.6%; P = 0.0742), indicating a possible association with irAE development and better tumor response to treatment. Last, serological tests showed that the majority of patients with irAE are negative for rheumatoid factor (RF) (3/31, 9.7%) and anti–cyclic citrullinated peptide (anti-CCP) antibody (2/30, 6.7%). All patients with RF⁺ and anti-CCP⁺ irAE had relatively low titers of RF or anti-CCP (table S1). Therefore, most of the patients with irAE were considered seronegative, consistent with previous studies (1, 16, 25, 26).

CD8⁺ T cells from patients with irAE have increased cytotoxic T cell features

To define molecular and cellular changes associated with the development of clinical arthritis after immunotherapy, we performed scRNA-seq analysis on the peripheral blood mononuclear cells (PBMCs) from patients with irAE, RAC, ICI controls, and HC. After stringent quality control and filtering steps, approximately 220,000 cells were analyzed with respect to their transcriptomes. We performed supervised cell subset annotation according to the literature (27, 28). Uniform manifold approximation and projection (UMAP) illustrated major immune cell subsets, including B cells, CD4⁺ T cells, CD8⁺ T cells, dendritic cells, monocytes, and natural killer cells (NK cells) (fig. S1A). scRNA-seq showed that the proportion of CD8⁺ effector memory cell reexpressing CD45RA (TEMRA) was increased in patients with irAE and RAC compared to HC and ICI control; meanwhile, both irAE and ICI control exhibited higher CD8⁺ TEM compared to HC and RAC (Fig. 1A). Flow cytometry analysis showed that both patients with irAE and ICI controls had higher CD45RA⁺CCR7⁻ population relative to HC and RAC, although the increase was only significant in patients with irAE (Fig. 1B). The discrepancy in terms of TEMRA in ICI controls between scRNA-seq and flow cytometry may be because sequencing analysis used a set of 21 genes to annotate TEMRA, whereas flow cytometry relied on two markers and thus might miss the heterogeneity of CD45RA⁺CCR7⁻ population.

Fig. 1. CD8⁺ T cells from the patients with irAE are more cytotoxic and metabolic active.

(A) Circos plots showing the percentage of cells from HC (Healthy control, n = 6), ICI (n = 2), RAC (RA control, n = 5), or irAE (n = 5). (B) Expression of CD45RA and CCR7 on CD8⁺ T cells. Right: Summaries of the percentage of cells from HC (n = 53), irAE (n = 29), RAC (n = 41), and ICI (n = 26). (C) Expression of CXCR3 and CCR6 on CD8⁺ T cells. Right: Summaries of T cell subsets. HC (n = 45), irAE (n = 27), RAC (n = 31), and ICI (n = 26). (D) The cytotoxic score was evaluated using the gene list identified previously (83). (E) Specific genes were evaluated on CD8⁺ T cells. (F) Pathways that were significantly enriched in the CD8⁺ T cells between irAE and ICI. NF-κB, nuclear factor κB; STAT5, signal transducer and activator of transcription 5; DN, down. Gene set enrichment analysis (GSEA) plots of the allograft rejection (G), oxidative phosphorylation (H), IFN-α response (I), and IFN-γ response (J) between irAE and ICI. NES, normalized enrichment score (K to M) PBMCs were stimulated with plate-coated anti-CD3 and anti-CD28 (10 μg/ml) for 5 days. Mean fluorescence intensities (MFIs) of MitoTracker Green (MTG) (K), MitoTracker deep red (MTDR) (L) [HC (n = 31), irAE (n = 18), RAC (n = 32), and ICI (n = 16)] or Cy5-linked-1-amino-glucose (GluCy5) (M) [HC (n = 35), irAE (n = 20), RAC (n = 36), and ICI (n = 18)] in CD8⁺ T cells were presented. Expression was normalized to the HC in each experiment. (N) UMAP shows the presence or absence of T cell receptor (TCR) in the major immune cells across all the samples. (O) Pie charts showing the distribution of the top 100 TCR clones across different T cell subsets. Data in graphs represent mean ± SEM. Significance was tested by one-way analysis of variance (ANOVA). [(A) to (E) and (G) to (O)] ICI, ICI control.

CD8⁺ central memory T (TCM) cells (CD45RA⁻CCR7⁺) and naïve CD8⁺ T cells (CD45RA⁺CCR7⁺) were unaltered among groups (fig. S1B). Further analysis revealed that CD45RA⁺CCR7⁻CD8⁺ TEMRA cells had the highest expression of Natural killer cell granule protein 7 (NKG7), a gene critical for T cell cytotoxicity (fig. S1C) (29). We also observed distinct chemokine receptor expression in patients with irAE and ICI controls, where both had increased proportion of CCR6⁻CXCR3⁻CD8⁺ T cells compared to HC or RAC (Fig. 1C), implying that PD-1/PD-L1 blockade might reduce CXCR3 and CCR6 expression on CD8⁺ T cells. Flow cytometry analysis indicated that CCR6⁻CXCR3⁻CD8⁺ T cells from healthy donors contained a higher proportion of CD45RA⁺CCR7⁻CD8⁺ T cells (fig. S1D), with relatively higher NKG7 expression (fig. S1E). Moreover, RNA-seq analysis revealed that CCR6⁻CXCR3⁻CD8⁺ T cells from HC had higher expression of TEMRA signature genes (fig. S1F) (27) and were enriched with NK cell cytotoxicity (fig. S1, G and H), relative to CCR6⁺CXCR3⁻ and CCR6⁻CXCR3⁺ CD8⁺ T cells. Furthermore, effector T cell–associated markers, including the proportion of CD28⁻CD8⁺ T cells (fig. S1I), CD38⁺CD127⁻CD8⁺ T cells (fig. S1J), and expression of a terminal effector T cell marker CX3CR1 (fig. S1K), were all increased in both patients with irAE and ICI control relative to HC and RAC, although the increases were more consistent in irAE than in ICI controls. These markers were interlinked because CD38⁺CD127⁻ cells were significantly enriched in CX3CR1⁺CD8⁺ T cells relative to CX3CR1⁻CD8⁺ T cells (fig. S1L), and CD28⁻CD8⁺ T cells had higher expression of CX3CR1 and a higher proportion of CD38⁺CD127⁻ cells than CD28⁺CD8⁺ T cells (fig. S1, M and N). Furthermore, all these biomarkers were also higher in CXCR3⁻CCR6⁻CD8⁺ T cells (fig. S1, O to Q), indicating an inverse correlation between chemokine receptors, CXCR3, CCR6, and CX3CR1 (30, 31). Thus, our data indicate that while CD8⁺ T cells from both patients with irAE and ICI controls exhibit increased expression of some effector T cell markers, the changes were more consistent in patients with irAE.

Compared to ICI controls, irAE CD8 T cells had higher expression of TEMRA signatures that include many cytotoxic genes, suggesting an increased expression of the cytotoxic T cell program. Bioinformatic analysis showed that irAE CD8⁺ T cells had the highest cytotoxicity score (Fig. 1D) and the highest expression of cytotoxic molecules, such as NKG7, Granulysin (GNLY), Perforin-1 (PRF1) and Granzyme B (GZMB) (Fig. 1E). Gene set enrichment analysis (GSEA) showed that irAE CD8⁺ T cells were enriched with genes involved in tumor necrosis factor (TNF) signaling, IFN-α response, IFN-γ response, allograft rejection, and oxidative phosphorylation compared to ICI controls (Fig. 1, F to J). Consistent with the enriched oxidative phosphorylation related genes, irAE CD8⁺ T cells had highest mitochondrial mass and mitochondrial membrane potential indicated by MitoTracker Green (MTG) and MitoTracker Deep Red (MTDR) stainings, respectively (Fig. 1, K and L), suggesting possibly higher mitochondrial activity in irAE CD8⁺ T cells. In addition, irAE CD8⁺ T cells had highly increased glucose uptake as measured by Cy5-linked-1-amino-glucose (GluCy5) staining (32), suggesting an increased glucose metabolism (Fig. 1M). Last, T cell receptor (TCR) repertoire analysis showed that there was a clonal expansion of CD8 TEM or TEMRA cells in patients with irAE (Fig. 1, N and O). Together, our data indicate that CD8⁺ T cells from patients with irAE are metabolically active and functionally cytotoxic, which distinguishes them from all control patients.

CD4⁺ T cells in the patients with irAE are metabolically active and have increased CXCR3⁻CCR6⁻ frequency and reduced CXCR3⁺CCR6⁺ frequency

Consistent with previous literature (33–35), the CD4/CD8 ratio increased in the control patients with RA, but it remained unaltered in patients with irAE and ICI controls (Fig. 2A). Within CD4⁺ T cells, a significant decrease of the CD127^loCD25⁺CD4⁺ regulatory T cells (T_reg cells) was noted in patients with RA but not in irAE or ICI controls (Fig. 2B), suggesting that ICI therapy did not substantially affect circulating T_reg cell frequency. Human circulating CXCR5⁺CD4⁺ T cells have B cell helper functions analogous to T follicular helper cells (T_FH cells) (36–38). The proportion of CXCR5⁺CD45RA⁻CD4⁺ T cells significantly decreased in the patients with irAE and ICI controls compared to HC (Fig. 2C). However, this reduction was only significant in patients with irAE, but not ICI, compared to patients with RA (Fig. 2C). Within CXCR5⁺CD45RA⁻CD4⁺ T cells, the proportion of CXCR3⁺CCR6⁻, CXCR3⁻CCR6⁺, and CXCR3⁺CCR6⁺ populations remained unaltered among patient groups (fig. S2A). To further investigate other T cell lineages, we use surface molecules CXCR3 and CCR6 to interrogate T_H1-like (CXCR3⁺CCR6⁻), T_H17-like (CXCR3⁻CCR6⁺), T_H1/T_H17-like (CXCR3⁺CCR6⁺), and CXCR3⁻CCR6⁻ cells (39–41). The proportion of T_H1/T_H17-like cells was significantly reduced in patients with irAE compared to HC, RAC, and ICI controls, while CXCR3⁻CCR6⁻ cells increased in patients with irAE relative to HC, RAC and ICI controls (Fig. 2D). Thus, irAE development is uniquely associated with increased CXCR3⁻CCR6⁻ CD4⁺ T cell frequency and reduced CXCR3⁺CCR6⁺ T_H1/T_H17-like cell frequency. Anti–PD-1/PD-L1 therapy might be associated with reduced circulating T_FH-like cell frequency regardless of irAE development.

Fig. 2. CD4⁺ T cells are metabolic active and have reduced CXCR3 and CCR6 expression in patients with irAE.

(A) The CD4/CD8 ratio in PBMCs. HC (n = 53), irAE (n = 23), RAC (n = 42), and ICI (n = 17). (B) Expression of CD25 and CD127 on CD4⁺ T cells. HC (n = 53), irAE (n = 27), RAC (n = 42), and ICI (n = 17). (C) Expression of CXCR5 and CD4 on the CD45RA⁻CD4⁺ T cells. HC (n = 53), irAE (n = 27), RAC (n = 42), and ICI (n = 17). (D) Expression of CCR6 and CXCR3 on CD45RA⁻CD4⁺ T cells. HC (n = 41), irAE (n = 26), RAC (n = 36), and ICI (n = 12). (E) Significantly enriched pathways in CD4⁺ T cells between irAE and ICI. UV, ultraviolet. GSEA plots of IFN-α and IFN-γ response (F), and oxidative phosphorylation (G). (H to M) PBMCs were stimulated with plate-coated anti-CD3/CD28 (10 μg/ml) for 5 days; MFI of MTDR (H; HC, n = 31; irAE, n = 18; RAC, n = 32; ICI, n = 16), tetramethylrhodamine methyl ester (TMRM) (I; HC, n = 34; irAE, n = 19; RAC, n = 35; ICI, n = 17), or GluCy5 (J; HC, n = 35; irAE, n = 20; RAC, n = 36; ICI, n = 18) in CD4⁺ T cells were presented. Expression was normalized to the HC in each experiment. [(K) to (M)] Fresh PBMCs were stimulated with phorbol 12-myristate 13-acetate (PMA), ionomycin, and monensin for 5 hours, and the frequencies of perforin⁺ [(K); HC, n = 27; irAE, n = 13; RAC, n = 12; ICI, n = 8], [TNF-α⁺ (L), and IL-2⁺ (M) (HC, n = 44; irAE, n = 25; RAC, n = 27; ICI, n = 16] CD4⁺ T cells were examined. Data in graphs represent mean ± SEM. Significance was tested by one-way ANOVA. [(A) to (D) and (F) to (M)] ICI, ICI control.

GSEA identified significantly altered pathways between patients with irAE and ICI controls, including IFN-α response, IFN-γ response, oxidative phosphorylation, hypoxia, transforming growth factor–β (TGF-β) signaling, and fatty acid metabolism (Fig. 2E). IFN-α response, IFN-γ response, and oxidative phosphorylation pathways were enriched in CD4⁺ T cells from patients with irAE (Fig. 2, F and G). Moreover, IFN-α response and IFN-γ response were also significantly enriched in irAE CD4⁺ T cells compared to RAC (fig. S2B) or HC (fig. S2C). CD4⁺ T cells from patients with irAE had higher mitochondrial membrane potential measured by MTDR (Fig. 2H) and tetramethylrhodamine methyl ester (TMRM) (Fig. 2I), increased mitochondrial mass measured by MTG (fig. S2D), increased glucose uptake measured by GluCy5 staining (Fig. 2J), and increased reactive oxygen species measured by CellROX (fig. S2E), indicating an overall heightened cellular metabolism. Elevated glucose-dependent metabolism in irAE CD4⁺ T cells was confirmed by SCENITH assay, which uses the energy-intensive protein-synthesis process (measured by puromycin incorporation) as a readout of metabolic activities (fig. S2F). Consistent with higher metabolic signatures in irAE CD4⁺ T cells, RNA-seq results showed that healthy donor CXCR3⁻CCR6⁻CD4⁺ T cell subset was enriched in genes associated with energy-intensive ribosomal biogenesis and translation program (fig. S2, G to I). Glucose metabolism is closely linked to the effector/cytotoxic CD4⁺ T cell function (42, 43). CD4⁺ T cells from patients with irAE had the highest perforin expression among the four groups (Fig. 2K), consistent with the observations in other autoimmune diseases (44, 45) and suggesting increased cytotoxic activity in irAE CD4⁺ T cells. TNF-α production was increased in both irAE and RAC CD4⁺ T cells compared to HC (Fig. 2L). Notably, IL-2 was highly increased on CD4⁺ T cells from patients with RA, but not patients with irAE and ICI controls (Fig. 2M), suggesting a disease-specific cytokine expression pattern in CD4⁺ T cells. Overall, CD4⁺ T cells from patients with irAE were metabolically active, exhibited enhanced effector and cytotoxic functions, and were distinguished by reduced expression of CXCR3 and CCR6.

The humoral immune compartment remains largely unaltered in patients with irAE

It has been established that both T cells and B cells play critical roles in the pathogenesis of RA (20). To gain insight into the B cell compartment in patients with irAE, we analyzed the PBMCs from the patients using flow cytometry (gating strategy was shown in fig. S3A). Total B cell (CD19⁺CD14⁻CD3⁻) frequency trended lower, although not statistically significant, in patients with irAE relative to HC (fig. S3B). We did not observe significant differences in naïve B cells [immunoglobulin D (IgD)⁺CD27⁻CD19⁺; Fig. 3A]; CD27⁺CD38⁺ antibody-secreting cells (ASCs; Fig. 3B), including CD138⁺ ASCs or CD138⁻ ASCs (Fig. 3C); IgD⁻CD27⁺ memory B cells (fig. S3C); and CD27⁺CD38⁻ memory B cells between irAE and other groups (Fig. 3B). Double-negative (DN) B cells have been implicated in the development of various autoimmune diseases including RA (46, 47). The frequency of DN B cells was not altered in patients with irAE (Fig. 3A). However, the atypical DN B cells (CD19⁺CD11c⁺CD21⁻) were significantly elevated only in patients with RA (Fig. 3D). A significant increase of CD21⁻CD19⁺ and CD11c⁺CD21⁻IgD⁻CD27⁻ B cells (fig. S3D) was also observed in patients with RA compared to HC. Other B cell subsets including IgD⁺CD27⁺ or IgM⁺IgD⁺CD27⁺CD38⁻ unswitched memory B cells (fig. S3E) remained largely intact among patients. Moreover, GSEA showed that several pathways were enriched in irAE B cells relative to ICI control (Fig. 3E), including IFN-α response and oxidative phosphorylation (Fig. 3, F and G). These data indicate that IA-irAE may not be associated with any significant alterations in the peripheral B cell subsets, while RA is associated with significantly increased CD11c⁺CD21⁻ atypical B cells.

Fig. 3. Humoral immunity remains largely intact in patients with irAE, while patients with seronegative RA have increased atypical B cell frequency and autoantibodies.

(A) Expression of CD27 and IgD on CD19⁺ B cells. Right: Percentage of IgD⁺CD27⁻ naïve B cells and IgD⁻CD27⁻ DN B cells. (B) Expression of CD27 and CD38 on CD19⁺ B cells. Right: Percentages of CD27^hiCD38^hi ASCs and CD27⁺CD38⁻ memory B cells. (C) Expression of CD138 on CD27^hiCD38^hi ASCs. Right: Percentage of CD138⁺ ASCs and CD138⁻ ASCs. (D) Expression of CD11c and CD21 on CD19⁺ B cells. Right: Percentage of CD11c⁺CD21⁻CD19⁺ B cells. [(A) to (D)] HC (n = 63), irAE (n = 33), RAC (n = 46), and ICI (n = 20). (E to G) GSEA was performed on the B cells between irAE and ICI. (E) Significantly enriched pathways in B cells from irAE and ICI. GSEA plots of IFN-α and IFN-γ response (F), and oxidative phosphorylation (G). (H) The volcano plots of the citrullinated or noncitrullinated relative IgG or IgM isotype autoantigen levels comparing RA versus HC, irAE versus ICI, irAE versus HC, or irAE versus RA. The autoantigens were labeled when P < 0.01. TNF-α reactivities resulted from the administration of anti–TNF-α therapy for the treatment of RA. (I) Immunoglobulin isotype levels in the plasma were measured by multiplex assay. HC (n = 22), irAE (n = 34), RAC (n = 46), and ICI (n = 26). (J) CXCL13 levels in the plasma were measured by enzyme-linked immunosorbent assay (ELISA). HC (n = 20), irAE (n = 34) RAC (n = 47), and ICI (n = 23). (K) B-cell activating factor (BAFF) levels in the plasma were measured by multiplex assay. HC (n = 21), irAE (n = 34), RAC (n = 47), and ICI (n = 18). Data in graphs represent mean ± SEM. Significance was tested by one-way ANOVA [(A) to (D) and (I) to (K)] and logistic regression (H). [(A) to (D) and (F) to (K)] ICI, ICI control.

Because CD11c⁺CD21⁻ atypical B cells are known to produce autoantibodies and contribute to the development of systemic autoimmunity (48–50), we performed two 128-plex autoantigen array experiments using plasma samples from two cohorts of patients (see table S2 for the information on the second cohort). Compared to the HC, patients with RA had significantly increased levels of multiple IgG autoantibodies, both citrullinated (cit) and noncitrullinated (non-cit), some of which targeted complement C3, Ribo phosphoprotein P0, Transcriptional Intermediary Factor 1-gamma (TIF1-γ), and single-stranded DNA (ssDNA; Fig. 3H and fig. S3, F and G). These observations were consistent with previous studies that patients with seronegative RA exhibit increased autoantibodies, including Anti-citrullinated protein antibody (ACPA) fine specificities, although they were negative for anti-CCP or RF in standard clinical tests (tables S1 and S2), highlighting the presence and possible contribution of autoantibodies in patients with seronegative RA (51, 52). In contrast, we observed few substantial increases of either citrullinated or noncitrullinated IgG reactivities in patients with irAE compared to all other control cohorts, although patients with irAE had some slightly increased IgM autoantibodies compared to ICI controls (Fig. 3H and fig. S3, F and H to J). Compared to RAC, patients with irAE had many significantly reduced IgG autoantibodies, such as anti-Elastin, Factor B, Thyroid Peroxidase (TPO), glutamic acid decarboxylase 65 (GAD65), GAD2, Factor B, lipopolysaccharide (LPS), and complement C3 (Fig. 3H and fig. S3, F and J). Furthermore, because overall immunoglobulin isotype concentrations remain comparable between all four cohorts of patients (Fig. 3I), the increased autoantibodies could be the consequence of specific autoreactive B cells in the control patients with RA. Hence, our results suggest that IA-irAE might be a unique subtype of arthritis distinct from seronegative RA, characterized by its lack of strong B cell activation and IgG isotype autoantibody signatures. In contrast, seronegative RA is associated with increased atypical B cells and production of IgG autoantibodies.

We conducted a further analysis of several cytokines associated with B cells. B cell chemoattractant CXCL13 and B cell survival factor, BAFF, were uniquely elevated in patients with irAE (Fig. 3, J and K), whereas sCD40L and April were comparable across four cohorts of patients (fig. S3K). These results suggest that plasma CXCL13 and BAFF levels might not accurately reflect the activity of autoreactive B cells but could be associated with irAE development.

PD-1 blockade does not substantially affect B cell activation and immunoglobulin production ex vivo

To gain insight into the effect of PD-1 blockade specifically on B cells, we activated human naïve B cells with different Toll-like receptor (TLR) ligands or anti-CD40 in the presence of isotype control or pembrolizumab (Keytruda). Pembrolizumab showed no significant impact on human naïve B cell activation as indicated by the percentage of CD27⁺CD38⁺, CD27⁺CD38⁻ (Fig. 4A), and CD138⁺CD27⁺CD38⁺ B cells (Fig. 4B), as well as the B cell activation marker CD86 (fig. S4A). The impacts of pembrolizumab on immunoglobulin production appeared to depend on the type of stimuli. For example, IgA production was reduced by pembrolizumab under anti-CD40 but increased under TLR7/8 ligand R848. None of the immunoglobulins were affected by PD-1 blockade under TLR9 ligand CpG (Fig. 4C). However, under an age-associated B cell (ABC) skewing condition [R848, anti-Ig (M + A + G), anti-CD40, Recombinant human IL-2 (rhIL-2), rhIL-21, BAFF and IFN-γ] (53), pembrolizumab increased the production of all immunoglobulin isotypes. These data were largely consistent with a recent publication using PD-1–deficient patient B cells (14) and indicated a context-dependent effect of PD-1 blockade on antibody production.

Fig. 4. Pembrolizumab does not substantially affect B cell activation and antibody production ex vivo.

(A to C) Human naïve B cells were isolated from PBMCs and cultured under indicated conditions for 7 days; pembrolizumab or isotype control IgG4 (IgG4 Iso) was added on day 2; n = 8. (A) Expression of CD38 and CD27 on B cells. Right: Percentages of CD27⁺CD38⁻ and CD27⁺CD38⁺ B cells. (B) Expression of CD138 on CD27⁺CD38⁺ B cells. Right: Percentage of CD138⁺CD27⁺CD38⁺ B cells. (C) Different immunoglobulin isotypes were measured in the culture supernatant of (A) and (B) by multiplex assay. (D to F) B cells from Humanized PD-1 (HuPD-1) mice were isolated and cultured with LPS, IL-4, BAFF, or ODN 2006, anti-IgM, IL-21, IL-4 or R848, anti-IgM, anti-CD40, IL-21, and IFN-γ for 3 days; pembrolizumab or isotype control was added on day 1. (D) Expression of IgG2c on activated B cells. FSC-H, Forward scatter height. Right: Percentage of IgG2c⁺ B cells; n = 5. (E) Expression of IgG1 on activated B cells. Right: Percentage of IgG1⁺ B cells; n = 5. (F) Different immunoglobulin isotypes in the supernatant were measured by multiplex assay; n = 5. Data in graphs represent mean ± SEM. Significance was tested by two-way ANOVA.

Furthermore, we used a humanized PD-1 (HuPD-1)/PD-L1 mouse model, in which the endogenous mouse exons coding for the extracellular domains of PD-1 (exons 2 and 3) or PD-L1 (exons 3 and 4) were replaced with the coding sequence for the corresponding human exons. It enabled the testing of Food and Drug Administration–approved PD-1 blocking antibodies on mice (54). As expected, pembrolizumab treatment efficiently blocked the detection of surface PD-1 (fig. S4B). However, pembrolizumab did not affect the expression of IgG1⁺, IgG2c⁺ (Fig. 4, D and E), CD138⁺ (fig. S4C), up-regulation of activation markers CD69 and CD86 (fig. S4D), metabolic marker CD71 and CD98 (fig. S4E), cell proliferation (fig. S4F), and the production of immunoglobin levels in the cultural supernatant (Fig. 4F). Together, these data indicate that PD-1 blockade may not substantially affect B cell activation. However, pembrolizumab may modulate B cell immunoglobulin production in a context-dependent manner.

Inflammatory cytokines enriched in the plasma of patients with irAE may negatively regulate B cell antibody production

To assess soluble factors associated with autoimmunity, we measured plasma concentrations of common cytokines and chemokines. IL-6 levels significantly increased in the plasma of patients with irAE compared to all three control groups, while IL-12p70 levels significantly increased in patients with irAE compared to HC and ICI controls (Fig. 5A). IFN-γ, TNF-α, and IL-1β only showed a slight increase in patients with irAE (Fig. 5B). Other cytokines, including IL-2, IL-4, IL-8, and IL-17A, exhibited no clear differences among different groups (fig. S5A). Chemokines Interferon-gamma inducible protein 10 kDa (IP-10), C-X-C motif chemokine ligand 11 (CXCL11), and CXCL9, ligands for CXCR3, were significantly elevated in the patients with irAE compared to HC and RAC, but not ICI controls (Fig. 5C). Similar changes were observed on C-C motif chemokine ligand 20 (CCL20), a ligand for CCR6 (Fig. 5D). While CCL17, one of the ligands for CCR4, did not show any clear differences among the groups of patients (fig. S5B), C-X3-C motif chemokine ligand 1 (CX3CL1) has been reported to be elevated in seropositive patients with RA (55), and reduced level of CX3CL1 in patients was associated with clinical response to treatment with abatacept or infliximab (56). Although CX3CL1 level did not increase in patients with seronegative RA in our cohort, it significantly increased in patients with irAE compared to HC or RAC, but not ICI controls (Fig. 5E). CCL2, a chemokine that plays a key role in attracting monocytes and T cells, also showed significant increases in patients with irAE and ICI controls (Fig. 5F). Other chemokines for recruiting monocytes, NK cells, polymorphonuclear leukocytes, eosinophils, and neutrophils including CCL3, CCL4 (fig. S5C), CXCL1, CXCL5, CCL11, and CXCL8 (fig. S5D) did not show significant changes in patients with irAE, which suggests that T cell trafficking might be preferentially affected in IA-irAE. Hence, plasma IL-6 and IL-12p70 concentrations were significantly increased in patients with irAE. Because of the significantly enriched IFN-α response genes in irAE B cells, CD4⁺ T cells, and CD8⁺ T cells, we concluded that type I IFN was also enhanced in patients with irAE, consistent with previous literature (18).

Fig. 5. Inflammatory signatures enriched in patients with irAE reduce antibody production.

(A to F) Bead-based multiplex assays were used to measure plasma concentration of IL-6, and IL-12p70 (A); TNF-α, IFN-γ, and IL-1β [B; HC (n = 19), irAE (n = 34), RAC (n = 45), and ICI (n = 9)]; IP-10 (CXCL10), CXCL11, and CXCL9 (C; HC, n = 17; irAE, n = 33; RAC, n = 46; ICI, n = 17); CCL20 (D); CX3CL1 (E); and CCL2 (F). (G to J) Human naïve B cells were isolated and cultured with anti–human CD40 (0.5 μg/ml), anti–human Ig (M + G + A) (2.5 μg/ml), and rhIL-21 (20 ng/ml) with IFN-α (100 ng/ml), IL-6 (100 ng/ml), IL-12 (100 ng/ml), control, or the combination of IFN-α, IL-6, and IL-12 for 7 days. Cells and culture supernatants were analyzed. (G) Representative flow plot of CD38 and CD138 expression on CD27^hiCD38^hi ASCs. Right: A summary of the percentage of CD138⁺ ASCs; n = 6. (H) Expression of CD11c and CD27 on CD27⁻IgD⁻ B cells. Right: Percentage of CD11c⁺IgD⁻CD27⁻ B cells; n = 6. (I) Expression of active-caspase-3 in B cells. Right: Percentage of active-caspase-3⁺ B cells from different groups; n = 3. (J) Different immunoglobulin isotype levels in the culture supernatants from (G) to (H) were measured by the multiplex assay; n = 6. Data in graphs represent mean ± SEM. Significance was tested by one-way ANOVA [(A) to (I)] and paired Student’s t test (J). [(A) to (F)] ICI, ICI control.

Next, we investigated the impacts of IL-6, IL-12, and IFN-α on B cells. Human naïve B cells were stimulated with anti-CD40/anti-IgM/G/A and IL-21 in the presence of IL-6, IL-12, IFN-α, or their combination for 7 days. IFN-α and combined IL-6, IL-12, and IFN-α significantly promoted the generation of CD38^hiCD27^hi ASCs and IgD⁻CD27⁻ B cells, while IL-6 and IL-12 had no obvious effects on the B cells (fig. S5, E and F). However, CD138⁺ ASCs were slightly decreased upon IFN-α or a combination of IL-6, IL-12, and IFN-α stimulation (Fig. 5G). IFN-α and combined IL-6, IL-12, and IFN-α significantly down-regulated CD11c expression on IgD⁻CD27⁻ B cells (Fig. 5H). Furthermore, they significantly increased the percentage of active-caspase-3 in B cells (Fig. 5I) and 7-Aminoactinomycin D (7-AAD)⁺annexin V⁺ B cells (fig. S5G), indicating increased cell death. Last, combined IL-6, IL-12, and IFN-α reduced IgM and IgG2 production (Fig. 5J). Therefore, IFN-α, or combined IL-6, IL-12, and IFN-α, increases B cell apoptosis and reduces production of IgM and IgG2. These data are broadly consistent with our observations that patients with irAE exhibit largely normal B cell homeostasis and no clear increase in autoantibody production.

Inflammatory cytokines IL-6, IL-12, and IFN-α promote both CD4⁺ and CD8⁺ T cell effector functions

We next explored the effects of IL-6, IL-12, and IFN-α on both CD4⁺ and CD8⁺ T cells activated by anti-CD3/anti-CD28. IL-6, IL-12, IFN-α, or combined IL-6, IL-12, and IFN-α had no effect on the proportion of CD45RA⁻CD8⁺ T cells on both days 1 and 5 of the culture (fig. S6A) and the Ki67 expression in CD45RA⁻CD8⁺ T cells at day 5 (fig. S6B). However, IFN-α and IL-6, IL-12, and IFN-α combination significantly increased the proportion of CD38⁺CD127⁻CD8⁺ T cells at both days 1 and 5 (Fig. 6A) and CD69 expression (Fig. 6B and fig. S6C). Combined IL-6, IL-12, and IFN-α also increased CD25 expression (Fig. 6C and fig. S6D). Meanwhile, IL-12 alone or combined IL-6, IL-12, and IFN-α markedly increased the proportion of IFN-γ⁺granzyme B⁺ T cells (Fig. 6D) and granzyme B⁺ T cells (fig. S6E). Along with the changes of effector or cytotoxicity, IL-12 alone or combined IL-6, IL-12, and IFN-α increased the mitochondrial membrane potential (Fig. 6E), mitochondrial mass (Fig. 6F), and glucose uptake (Fig. 6G). Together, these data indicate that IL-12 alone or a combination of IL-12, IL-6, and IFN-α promotes cytotoxicity and metabolism of CD8⁺ T cells.

Fig. 6. Inflammatory cytokines enriched in patients with irAE promote both CD4⁺ and CD8⁺ T cell functions.

(A to G) CD8⁺ T cells isolated from healthy donor were cultured in the plate coated with anti–human CD3/CD28 (10 μg/ml) for indicated days in the presence of vehicle control (control), IFN-α (100 ng/ml), IL-6 (100 ng/ml), IL-12 (100 ng/ml), or the combination. (A) Summaries of the percentage of CD38⁺CD127⁻CD8⁺ T cells at days 1 and 5. (B) MFI of CD69 (day 5). (C) MFI of CD25 (day 5). (D) Activated CD8⁺ T cells were restimulated with PMA, ionomycin, and monensin for 5 hours; expression of granzyme B and IFN-γ in CD8⁺ T cells was examined. [(E) to (G)] CD8⁺ T cells were cultured for 5 days. MFIs (all relative to those under control condition) of MTDR (E), MTG (F), and GluCy5 (G) were summarized. (H to L) CD4⁺ T cells were isolated from HC, cultured in the plate coated with anti–human CD3/CD28 (10 μg/ml) for indicated days in the presence of IFN-α (100 ng/ml), IL-6 (100 ng/ml), IL-12 (100 ng/ml), control, or the combination of them. (H) Activated CD4⁺ T cells were restimulated with PMA, ionomycin, and monensin for 5 hours, and IL-21 level was examined in the CD45RA⁻CD4⁺ T cells. (I) Activated CD4⁺ T cells were restimulated with plate-coated anti–human CD3/CD28 (10 μg/ml) with monensin for 6 hours, and expression of CXCL13 was measured in CD45RA⁻CD4⁺ T cells. (J) CD38 and CXCR5 were examined on CD4⁺ T cells at day 5. (K) Percentage of CXCL13⁺ cells in indicated T cells at day 5. (L) Percentage of IL-21⁺ cells in indicated T cells at day 5. Data in graphs represent mean ± SEM. Significance was tested by one-way ANOVA.

CD4⁺ T cells not only promote humoral immune response but also facilitate CD8⁺ T cell function by secreting cytokines such as IL-21 (57–61). IL-12, IL-6, and IFN-α, alone or in combination, did not affect the proportion of CD45RA⁻CD4⁺ cells (fig. S6F) or Ki67 expression on CD45RA⁻CD4⁺ T cells (fig. S6G). However, IL-12 alone or combined IL-6, IL-12, and IFN-α significantly prompted expression of IL-21 (Fig. 6H) and TNF-α (fig. S6H) in CD4⁺ T cells, while IFN-α alone or combined IL-6, IL-12, and IFN-α raised the proportion of CXCL13⁺CD4⁺ T cells (Fig. 6I). Combined IL-6, IL-12, and IFN-α also significantly up-regulated IL-4 levels, while IFN-α alone increased the proportion of IL-2⁺CD45RA⁻CD4⁺ T cells (fig. S6I). IFN-α alone or combined IL-6, IL-12, IFN-α increased the proportion of CD38⁺CXCR5⁻CD4⁺ T cells while reduced CD38⁻CXCR5⁺CD4⁺ T cells (Fig. 6J). Moreover, we further investigated possible differential effects of IL-6, IL-12, and IFN-α on CD38⁺CXCR5⁻ versus CD38⁻CXCR5⁺ CD4⁺ T cells. IFN-α alone or the combination of IL-12, IL-6, and IFN-α elevated CXCL13 expression on CD38⁺CXCR5⁻CD4⁺ T cells, but not CD38⁻CXCR5⁺ CD4⁺ T cells (Fig. 6K). CD38⁺CXCR5⁻CD4⁺ T cells expressed a higher level of IL-21 than CD38⁻CXCR5⁺ CD4⁺ T cells in response to IL-12 alone or a combination of IL-12, IL-6, and IFN-α stimulation (Fig. 6L). CD38⁺CXCR5⁻CD4⁺ T cells also had relatively higher levels of TNF-α (fig. S6J) and IL-4 (fig. S6K) compared to CD38⁻CXCR5⁺ CD4⁺ T cells, while they had comparable levels of IL-2 (fig. S6L). These data suggest that a combination of IL-12, IL-6, and IFN-α expands CD38⁺CXCR5⁻CD4⁺ T cells and promotes the production of IL-21, IL-4, CXCL13, and TNF-α, which could affect other immune cells.

Pharmacological blockade of IL-6, IL-12, and IFN-α attenuates elevated effector T cell phenotypes from patients with irAE

Our above data support a model that IL-6, IL-12, and IFN-α may contribute to some of the immunological aberrations found in patients with irAE, which could be partially reversed by blocking these cytokines or their receptors. To test this hypothesis, we activated irAE PBMCs in the presence of combined anti–IL-6R, anti–IL-12, and anti-Interferon α receptor (IFNAR). Combined IL-6, IL-12, and IFN-α blockade significantly reduced the percentages of CD38⁺CD127⁻ (Fig. 7A), granzyme B⁺IFN-γ⁺ (Fig. 7B), perforin⁺IFN-γ⁺ (Fig. 7C), and granzyme B⁺perforin⁺ (fig. S6M) CD8⁺ T cells from patients with irAE. Furthermore, they significantly down-regulated glucose uptake (Fig. 7D) and mitochondrial membrane potential (Fig. 7E), but not mitochondrial mass (fig. S6N), of irAE CD8⁺ T cells. In terms of CD4⁺ T cells, combined IL-6, IL-12, and IFN-α blockade significantly reduced IL-21 expression in irAE CD4⁺ T cells (Fig. 7F). However, IL-6, IL-12, and IFN-α blockade did not affect the proportion of CD38⁺CXCR5⁻CD4⁺ T cells (Fig. 7G) or did not significantly affect glucose uptake, mitochondrial membrane potential, and mitochondrial mass of irAE CD4⁺ T cells (fig. S6, O to Q), suggesting a stronger impact of these cytokines on CD8⁺ T cells than CD4⁺ T cells. Together, these data indicate that elevated IL-6, IL-12, and IFN-α may contribute to the enhanced expression of some cytotoxicity molecules and metabolic activity in irAE CD8⁺ T cells.

Fig. 7. Pharmacological blockade of IL-6, IL-12, and IFN-α attenuates elevated cytotoxic and metabolic phenotypes of irAE T cells.

(A to G) PBMCs from the patients with irAE were cultured in the plate coated with anti-CD3 and anti-CD28 (10 μg/ml) in the presence of IgG1 isotype control or anti–human IL-6R (50 μg/ml), anti–human IL-12p40 (50 μg/ml), and anti–human IFNAR1 (50 μg/ml) for 3 days; n = 9. (A) Expression of CD38 and CD127 on CD8⁺ T cells. Right: Percentage of CD38⁺CD127⁻CD8⁺ T cells. (B and C) CD8⁺ T cells activated for 5 days were restimulated with PMA, ionomycin, and monensin for 5 hours. (B) Expression of granzyme B and IFN-γ. Right: Percentage of granzyme B⁺IFN-γ⁺CD8⁺ T cells. (C) Expression of perforin and IFN-γ. Right: Percentage of perforin⁺IFN-γ⁺CD8⁺ T cells. [(D) and (E)] MFIs of GluCy5 (D), TMRM, and MTDR (E) in CD8⁺ T cells were presented. [(F) and (G)] CD4⁺ T cells activated for 5 days were restimulated with PMA, ionomycin, and monensin for 5 hours. (F) Expression of IL-21 and IL-2. Right: Percentage of IL-21⁺IL-2⁺CD4⁺ T cells. (G) Expression of CD38 and CXCR5 on CD4⁺ T cells. Right: Percentage of CD38⁺CXCR5⁻CD4⁺ T cells. Data in graphs represent mean ± SEM. Significance was tested paired Student’s t test [(A) to (G)].

DISCUSSION

In this study, we aim to address two major questions: What are the immunological phenotypes that distinguish anti–PD-1/PD-L1–treated patients with IA-irAE from those without irAE, and what are the relationship between IA-irAE and RA? While previous studies have explored the immune cell phenotypes, our current study represents an in-depth immunological investigation on the largest cross-sectional IA-irAE patient cohort with three different control groups so far and thus provides robust information about this fast-growing disease, owing to the increased usage of ICI agents. Our results demonstrate phenotypical changes in IA-irAE CD4⁺ and CD8⁺ T cells. They include reduced T_FH-like CD4⁺ T cells, increased CXCR3⁻CCR6⁻ and reduced CXCR3⁺CCR6⁺ subsets in both CD4⁺ and CD8⁺ T cells, increased expression of effector and cytotoxic molecules, and heightened metabolic activity. Some of these are shared between IA-irAE and ICI control, but many are unique in IA-irAEs, including highly increased cytotoxic program in CD8⁺ T cells, reduced CXCR3 and CCR6 expression on CD4⁺ T cells, and increased metabolic activity in both CD4⁺ and CD8⁺ T cells. These results suggest that increased cytotoxic activity and metabolism but not simply increased expression of effector T cell markers might be a prerequisite to IA-irAE development. This observation partly explains why irAE development is associated with improved tumor response to the ICI therapy (table S1) (62), i.e., those without irAE are more likely to have T cells with insufficient cytotoxic activity. This observation is consistent with the recent observation that CD38^hi cytotoxic CD8⁺ T cells are enriched in ICI-arthritis joints (18). Cytotoxic CD8⁺ T cells may directly contribute to RA pathogenesis (63, 64). The mechanisms through which reduced CXCR3 and CCR6 expression on CD4⁺ T cells may contribute to IA-irAE are unclear. Reduced CXCR3 and CCR6 on CD4⁺ T cells are associated with increased ribosomal activity and protein translation, consistent with the increased metabolism in irAE T cells. Moreover, reduced CXCR3 is largely consistent with previous observation that CXCR3 ligands might promote T cell trafficking to joint tissues (17). Nevertheless, CXCR3⁻CCR6⁻CD4⁺ T cells might serve as a biomarker for IA-irAE. Furthermore, other T cell lineages, such as resident memory T cells, may contribute to IA-irAE (65, 66).

In contrast to the marked phenotypic changes in T cells, B cell compartment remains largely unaltered in patients with IA-irAE, whereas patients with seronegative RA unexpectedly exhibit more changes. Control patients with RA, but not patients with IA-irAE, have significantly increased CD4/CD8 ratio and frequency of CD11c⁺CD21⁻ atypical B cells. These atypical B cells accumulate during aging in healthy individuals, especially in females (67, 68). They are overrepresented in various systemic autoimmune disorders, such as systemic lupus erythematosus and Sjogren’s syndromes. Recent studies found an expansion of these cells in the synovium and peripheral blood of seropositive patients with RA, indicating a possible role in RA pathogenesis (67, 69, 70). Here, we showed that these atypical B cells are also elevated in patients with seronegative RA, consistent with the observation that patients with seronegative RA have elevated levels of multiple autoantibodies (51). These observations suggest a potential loss of tolerance in B cell compartment in patients with seronegative RA, and future in-depth analysis of B cell phenotypes in these understudied patients is warranted (71). However, patients with IA-irAE do not exhibit elevated CD11c⁺CD21⁻ atypical B cells or increased levels of common autoantibodies. Moreover, PD-1 blockade does not consistently promote B cell activation or antibody production. Although a negative argument may require extraordinarily exhaustive proof, our data in its totality suggest that IA-irAE is primarily associated with overactivation of T cell immunity but not B cell immunity. Hence, our study establishes IA-irAE as an immunologically distinct subtype of arthritis and represents the first comprehensive immunological study of a unique T cell–driven and likely autoantibody-independent arthritis in human patients. Further analyses using larger sample size and study on the deposition of ICs and complement in joint tissues from patients with IA-irAE are warranted in future investigations.

Another characteristic of IA-irAE is the elevated levels of selective cytokines and chemokines, some of which (e.g., CCL2) are shared with ICI controls and most of which are absent in RA controls. These data suggest a complex inflammatory environment elicited by PD-1 inhibition and a relatively low inflammatory condition for seronegative RA, although the results could be confounded by the different medications and longer disease duration in the RA controls than patients with IA-irAE. Several biologics have been suggested to treat IA-irAEs, including tocilizumab (72), TNF-α inhibitor (17), and type I IFN blockers (18). Our results indicate that IL-6, IL-12, and type I IFN likely play a role in promoting cytotoxic activity in IA-irAE T cells. ICI treatment can promote IL-6 production, which contributes to irAE development (73). IL-12 is critical for successful anti–PD-1 therapy (74). Type I IFNs may have opposing roles in ICI therapy depending on the timing of its activation, i.e., promoting immunity early while suppressing immunity later on (75). Thus, while targeting combined IL-6, IL-12, IFN-α may benefit IA-irAE, it carries the risk of compromising antitumor immunity. Thoughtful consideration is needed to balance the management of irAE and maintenance of tumor control.

There are several limitations in our study. First, although our cohort is relatively large, more definitive answers on the autoantibody profile will require a much larger sample size and a genome-wide autoantibody scan. Second, our study is a cross-sectional study. A longitudinal study will be needed to identify immunological phenotypes before and after ICI therapy and before and after irAE onset. Third, because we could not match the treatment and disease duration between patients with irAE and RA controls due to practical constraints, the potential influence of different medications and disease duration on irAE versus RA control immune cells cannot be excluded. Last, all the cellular analyses were performed on PBMC samples. Paired joint samples will be needed to identify immune cells responsible for arthritis in situ and provide more definitive answers regarding IC deposition in the synovium.

MATERIALS AND METHODS

Study design

This study was approved by the institutional review board (IRB) at Mayo Clinic (IRB protocol #21-009862), and patients providing written informed consent were eligible for this study. Specifically, four groups of patients were recruited for this study: patients with irAE are patients with cancer on PD-1 or PD-L1 inhibitor (pembrolizumab, nivolumab, cemiplimab, atezolimumab, durvalumab, or avelumab) therapy without preexisting rheumatic diseases, who develop physician-confirmed IA; ICI patients are patients with cancer on PD-1 or PD-L1 inhibitor therapy without preexisting rheumatic diseases, who have not developed any irAE; RAC patients are patients without cancer and with RA that match those with de novo rheumatic irAE; and HC are the sex- and age-matched healthy individuals.

Mice

HuPD-1 and HuPD-L1 mice on C57BL/6 background were a gift from H. Dong at Mayo Clinic (54). Mice were bred and maintained in a specific pathogen–free facility in the Department of Comparative Medicine of the Mayo Clinic. The mice of either sex were used between 8 and 16 weeks old. They were euthanized by carbon dioxide according to the approved protocol. All animal protocols (A00003354-18-R23) were approved by the Institutional Animal Care and Use Committees of the Mayo Clinic Rochester.

Samples collection and processing

Blood samples were collected into sodium heparin tubes, and plasma was collected after centrifuging at 4°C and 1300 rpm for 10 min. PBMCs were isolated from the peripheral blood using Ficoll-Paque gradient centrifugation. Specifically, 13 ml of Ficoll-Paque PLUS density gradient medium (GE Healthcare, catalog no. 17-1440-02) was added to the bottom of 30 ml of diluted peripheral blood and centrifuged at 4°C and 400 g for 25 min with the acceleration and deceleration at lower speed. The PBMC layer was collected and washed with wash buffer [1× phosphate-buffered saline (PBS) + 10 mM Hepes + 0.1% bovine serum albumin (BSA) + 2 mM EDTA], and PBMCs were aliquoted and cryopreserved in a liquid nitrogen tank.

Flow cytometry

Cryopreserved PBMCs were thawed, rested, and washed. For surface staining, cells were stained in PBS containing 1% (w/v) BSA [fluorescence-activated cell sorting (FACS) buffer] with indicated antibodies for 1 hour on ice. Following antibodies have been used: anti-IgD–fluorescein isothiocyanate (BioLegend, IA6-2, catalog no. 348206), anti-CD24–phycoerythrin (PE) (BioLegend, ML5, catalog no. 311106), anti-IgM–PerCP-Cy5.5 (BioLegend, MHM-88, catalog no. 314512), anti-CD21–Allophycocyanin (APC) (BD Pharmingen, B-ly4, catalog no. 561357), anti-CD11c–redFluor 710 (Cytek, 3.9, catalog no. 80-0116-T025), anti-CD138–APC-Cy7 (BioLegend, MI15, catalog no. 356528), anti-CD27–PE-Cy7 (BioLegend, M-T271, catalog no. 356412), anti-CD38–Brilliant Violet 421 (BV421) (BioLegend, HIT2, catalog no. 303525), anti-CD3–Super Bright 600 (Invitrogen, OKT3, catalog no. 63-0037-42), anti-CD19–BV650 (BioLegend, HIB19, catalog no. 302238), anti-CD14–BV785 (BioLegend, M5E2, catalog no. 301840), CXCR5-BV605 (BioLegend, J252D4 catalog no. 356930), anti-CXCR5–Alexa Fluor 647 (AF647) (BioLegend, J252D4, catalog no. 356906), anti-CD45RA–APC-Cy7 (Cytek, HI100, catalog no. 25-0458-T100), anti-CD28–Brilliant Ultra Violet 395 (BD Horizon, CD28.2, catalog no. 569160), anti-CD38 (BioLegend, HIT2, catalog no. 303504), anti-CCR6–PE (BioLegend, G034E3, catalog no. 353410), anti-CX3CR1–PerCP-Cy5.5 (BioLegend, 2A9-1, catalog no. 341614), anti-CD4–AF700 (BioLegend, RPA-T4, catalog no. 300526), anti-CCR7–PE-Cy7 (BioLegend, G043H7, catalog no. 353226), anti-CD8–BV510 (BioLegend, RPA-T8, catalog no. 301048), anti-CD127–BV605 (BioLegend, A019D5, catalog no. 351334), anti-CD3–BV650 (BioLegend, OKT3, catalog no. 317324), anti–PD-1–BV711(BioLegend, EH12.2H7, catalog no. 329928), anti-CD25–BV785 (BioLegend, BC96, catalog no. 302636), anti-CXCR3–BV421 (BioLegend, G025H7, 353716), and anti-CD69 (BioLegend, FN50, catalog no. 310904). For the cytokine intracellular staining, PBMCs were stimulated with phorbol 12-myristate 13-acetate (PMA), ionomycin, and GolgiStop in a complete RPMI 1640 medium [RPMI 1640 + 10% (v/v) fetal bovine serum (FBS) + 1% penicillin–streptomycin–l-glutamine] for 5 hours. Cells were stained for surface molecules, fixed with BD Fixation/Permeabilization solution (BD Cytofix/Cytoperm, catalog no. 51-2090KZ) on ice for 20 min, permeabilized with BD Perm/Wash buffer (BD Perm/Wash solution, catalog no. 51-2091KZ), and stained with the following antibodies: anti–TNF-α–AF488 (BioLegend, MAb11, catalog no. 502915), anti–granzyme B–PE (Invitrogen, GB11, catalog no.12-8899-41), anti–IL-2–PE-Cy7 (BioLegend, MQ1-17H12, catalog no. 500326), anti–IFN-g–APC (BioLegend, 4S.B3, catalog no. 502512), anti–IL-21–PC (BioLegend, 3A3-N2, catalog no.513008), anti–IL-4–PE (BioLegend, MP4-25D2, catalog no. 500810), and anti-perforin–BV510 (BioLegend, dG9, catalog no. 308120). For CXCL13 staining, cells were stimulated with coated anti-CD3 (10 μg/ml), anti-CD28, and Monensin Solution for 5 hours. Cells were stained with surface molecules, fixed with BD Fixation/Permeabilization solution, permeabilized with BD Perm/Wash buffer, and stained with anti-CXCL13–PE (R&D Systems, 53602, catalog no.IC8012P). For the transcriptional factor staining, cells were stained with surface molecules, fixed at room temperature with True-Nuclear Fix buffer for 30 min, permeabilized with True-Nuclear perm buffer, and stained with anti-Ki67–BV421 (BD Horizon, B56, catalog no. 562899). Cell viability was examined by Fixable viability dye (Tonbo Bioscience) or 7-AAD (Thermo Fisher Scientific).

For the staining of metabolic indicators, cells were washed with PBS and stained with viability dye and the indicated metabolic indicators, including 20 nM MTDR (Thermo Fisher Scientific), 20 nM MTG (Thermo Fisher Scientific), 100 nM TMRM (Thermo Fisher Scientific), 500 nM CellROX, or 1 μM GluCy5 in Hanks’ balanced salt solution at 37°C for 20 min, followed by the surface molecule staining. Flow cytometry was performed on a BD LSRFortessa X-20, LSR II instrument, or Attune NxT system (Life Technologies). Data were then analyzed by the FlowJo software (Tree Star).

Cytokine and immunoglobulin measurement

For the cytokine, chemokine, and immunoglobulin levels in the human plasma, LEGENDplex kits from BioLegend were used according to the manufacturer’s instructions. Specifically, plasma samples were diluted two- or 100,000-fold (for immunoglobulin) using assay buffer, and the following kits were used in this study: Human Essential Immune Response Panel (catalog no. 740930), Human Proinflammatory Chemokine Panel 1 (catalog no. 740985), Human B Cell Activator Panel (catalog no. 740535), and Human Immunoglobulin Isotyping Panel (catalog no. 740640). The data were acquired on the Attune NxT system (Life Technologies) and analyzed using cloud-based software from Qognit. For CXCL13, IL-21, and CX3CL1 measurements, the following kits were used: Human CXCL13/BLC/BCA-1 Quantikine Enzyme-Linked Immunosorbent Assay (ELISA) (R&D Systems, catalog no. DCX130), Human IL-21 DuoSet ELISA (R&D Systems, catalog no. DY8879-05), and Human CX3CL1/Fractalkine DuoSet ELISA (R&D Systems, DY365); all steps were performed according to the manufacturer’s instructions. The plates were read at 450 nM with the wavelength correction of 570 nM.

Human naïve B cell pulsed with pembrolizumab

Cryopreserved PBMCs were thawed, rested, and washed. Human naïve B cells were isolated from PBMCs using the EasySep Human Naïve B Cell Isolation Kit (STEMCELL Technologies, catalog no. 17254), and only the samples with the percentage of IgD⁺CD27⁻ more than 95% were used for the culture experiment. Specifically, three conditions were adopted (14, 53, 76, 77) and used in this study. Condition 1: anti–human Ig (M + G + A) (2.5 μg/ml; Jackson ImmunoResearch, catalog no. 109-006-064), CpG oligodeoxynucleotide (ODN) (2.5 μg/ml; Invivogen, catalog no. tlrl-2006-1), anti–human CD40 (10 μg/ml; Bio X Cell, catalog no. BE0189), rhIL-21 (20 ng/ml; Peprotech, catalog no. 200-21-50UG), rhIL-4 (10 ng/ml; BioLegend, catalog no. 574004), and rhIL-2 (10 ng/ml; Peprotech, catalog no. 200-02-250UG). Condition 2: anti–human CD40 (0.5 μg/ml), rhIL-21 (20 ng/ml), and anti-Ig (M + G + A) (2.5 μg/ml). Condition 3: R848 (0.5 μg/ml; Invivogen, catalog no. tlrl-r848-1), rhBAFF (20 ng/ml; BioLegend, catalog no. 7449534), rhIL-2 (10 ng/ml), anti–human CD40 (10 μg/ml), anti–human Ig (M + G + A) (2.5 μg/ml), rhIL-21 (10 ng/ml), rhIFN-γ (20 ng/ml; BioLegend, catalog no. 570204). Pembrolizumab (100 μg/ml; Keytruda) or isotype control IgG4 (Bio X Cell, catalog no. CP147) was added into the culture media 2 days after stimulation, and B cells were further cultured for another 5 days. Surface molecules, including IgD, CD86, CD138, CD27, CD38, CD11c, and PD-1, were examined by flow cytometry, and the supernatant was used to measure immunoglobulin levels (LEGENDplex Human Immunoglobulin Isotyping Panel, BioLegend).

IL-6, IL-12, and IFN-α stimulation on human naïve B cells

Naïve B cells were isolated from healthy donor PBMCs using EasySep Human Naïve B Cell Isolation Kit (STEMCELL Technologies, catalog no. 17254). B cells were stimulated with anti–human CD40 (0.5 μg/ml), anti–human Ig (M + G + A) (2.5 μg/ml), rhIL-21 (20 ng/ml) with or without rhIL-6 (100 ng/ml; BioLegend, catalog no.570806), rhIL-12 (100 ng/ml; BioLegend, catalog no. 573004), rhIFN-α₂ (100 ng/ml; BioLegend, catalog no. 592704), or the combination of rhIL-6 (100 ng/ml), rhIL-12 (100 ng/ml), and rhIFN-α (100 ng/ml) for 7 days. Culture supernatants were collected, and cells were analyzed by flow cytometry.

Human T cell culture

CD4⁺ T cells or CD8⁺ T cells were isolated from PBMCs using the EasySep Human CD4⁺ T Cell Enrichment Kit (STEMCELL Technologies, catalog no. 19052) and EasySep Human CD8⁺ T Cell Isolation Kit (STEMCELL Technologies, catalog no. 17953), respectively. Only the samples with purity of more than 95% were used in the experiment. A total of 0.5 million isolated CD4⁺ T cells or CD8⁺ T cells was stimulated with plate-coated anti-CD3 (10 μg/ml; BioXCell, catalog no. BE0001-2) and anti-CD28 (BioXCell, catalog no. BE0291), rhIL-6 (100 ng/ml), rhIFN-α₂ (100 ng/ml), rhIL-12 (100 ng/ml), or the combination of rhIL-6, rhIFN-α₂, and rhIL-12 for 5 days. Surface molecules CD38, CXCR5, CD127, CD45RA, CD69, and CD25 were measured on the cultured T cells, and Ki67 and cytokine levels were also evaluated on the T cells.

HuPD-1 mouse B cell culture

Mouse B cells were isolated from splenocytes using the EasySep Mouse B Cell Isolation Kit (STEMCELL Technologies, catalog no. 19854). B cells were labeled with CellTrace Violet (Invitrogen) and cultured in RPMI 1640 medium supplemented with 10% (v/v) FBS and 1% penicillin-streptomycin. Three conditions were used in this study. Condition 1: LPS (3 μg/ml; Sigma-Aldrich, catalog no. L2880-25MG) and recombinant mouse IL-4 (10 ng/ml; R&D Systems, catalog no. 404-ML-100/CF) plus recombinant human BAFF (20 ng/ml; BioLegend, catalog no. 7449534). Condition 2: anti-mouse IgM (5 μg/ml; Jackson ImmunoResearch, catalog no. 115-006-075), CpG ODN (2.5 μg/ml), rmIL-21 (100 ng/ml; BioLegend, catalog no. 570502), rmIL-4 (20 ng/ml), and rhIL-2 (100 ng/ml). Condition 3 (53): R848 (1 μg/ml), anti-mouse CD40 (1 μg/ml; Bio X Cell, catalog no. BE0016-2), anti-mouse IgM (1 μg/ml), rmIL-21 (100 ng/ml), rmIFN-γ (10 ng/ml; Peprotech, catalog no. 315-05; 20 μg). B cells were cultured for 3 days, and pembrolizumab or isotype control IgG4 (100 μg/ml)was added on day 1.

Puromycin incorporation

Puromycin incorporation assay was adopted (78). Briefly, cells were treated with control, 100 mM 2-deoxy-d-glucose (2-DG), 1 μM oligomycin (Oligo), or a sequential combination of the drugs at the final concentrations for 30 to 45 min; puromycin (10 μg/ml; Invitrogen) was incubated with cells for 20 min. After puromycin treatment, cells were washed in cold PBS and stained with a combination of Fc receptor blockade and cell viability dye and then primary conjugated antibodies against surface markers for 30 min on ice in FACS buffer. After washing, cells were fixed and permeabilized using True-Nuclear Transcription Factor Buffer Set (BioLegend, catalog no. 424401) following manufacturer instructions. Intracellular puromycin staining was performed with anti-puromycin–AF647 (Sigma-Aldrich, 12D10, catalog no. MABE343-AF647) by incubating cells for 1 hour on ice. The mean fluorescence intensity (MFI) was used to calculate mitochondrial and glucose metabolism. Briefly, mitochondria metabolism = (Con – DGO) – (2-DG – DGO), and glucose metabolism = (Con – DGO) – (2-DG – O), where dimethyl sulfoxide control is Con, DGO is 2-DG + Oligo, and O is Oligo.

Flow cytometric cell sorting for bulk RNA-seq

Human PBMCs were thawed and rested in a complete medium for 1 hour; CD4⁺ or CD8⁺ T cells were enriched using EasySep Human CD4⁺ T Cell Isolation Kit and EasySep Human CD8⁺ T Cell Isolation Kit, respectively. Enriched CD4⁺ T cells were stained with PE–anti–human CCR6, 7-AAD, PE-Cy7–anti–human CD4, and BV421–anti–human CXCR3, while enriched CD8⁺ T cells were stained with PE–anti–human CCR6, 7-AAD, BV510–anti–human CD8. and BV421–anti–human CXCR3. CXCR3⁻CCR6⁻, CXCR3⁺CCR6⁻, CXCR3⁻CCR6⁺, or CXCR3⁺CCR6⁺ cells were sorted from CD4⁺ or CD8⁺ T cells using BD FACSAria III Cell Sorter. The sorted cells were lysed, and total RNA was extracted using Quick-RNA Microprep (Zymoresearch, catalog no. R1051). After quality control, high-quality total RNA was used to generate the RNA-seq library using the DNBSEQ platform at Innomics. Reads with low quality, containing the adapter (adapter pollution), or with high levels of N base were removed to generate clean data. HISAT (V2.2.1) was used to align the clean reads to the Homo_sapiens reference genome, and Homo_sapiens_9606.NCBI.GCF_000001405.39_GRCh38.p13.v2201 was used to align the clean reads to the reference genes. Differentially expressed gene (DEG) analysis was carried out using DESeq2, and genes with log₂ fold change > 0 and false discovery rate < 0.05 were considered for gene cluster analysis.

Anti–IL-6R, anti–IL-12, and anti-IFNAR treatment on PBMCs from patients with irAE

PBMCs from patients with irAE were thawed and rested in a complete medium for 1 hour and then activated with plate-coated anti–human CD3 and anti–human CD28 (10 μg/ml) with IgG1 isotype control (Bio X Cell, catalog no. CP174) or a combination of anti–human IL-6R 50 μg/ml; Bio X Cell, catalog no. SIM0014), anti–human IL-12p40 (50 μg/ml; Bio X Cell, catalog no. SIM0020), and anti–human IFNAR1 (50 μg/ml; Bio X Cell, catalog no. SIM0022) for 3 days. Cells were treated with PMA, ionomycin, and monensin for 5 hours; surface molecules were stained, followed by intracellular cytokine staining using BD Cytofix/Cytoperm Fixation/Permeabilization Kit according to the manufacturer’s instruction.

Sample processing for scRNA-seq and Single cell T cell receptor sequencing (scTCR-seq)

Cryopreserved PBMC samples were thawed and recovered in the RPMI 1640 medium supplemented with 10% (v/v) FBS and 1% penicillin-streptomycin for 2 hours at 37°C in 5% CO₂. PBMCs were filtered with 40 μm of nylon mesh and washed with PBS supplemented with 0.04% (v/v) BSA. The cell concentration was adjusted to 5 × 10⁵/ml, and cell counts and viabilities were determined using trypan blue exclusion and counted on a Countess II FL automated cell counter (Life Technologies), and only the PBMCs with viability of more than 90% were sent for sequencing. Reagents, reaction master mixes, reaction volumes, cycling numbers, cycling conditions, and clean-up steps were performed according to 10x Genomics’ guidelines. cDNA was allocated for preparation of a gene expression library (Chromium Single Cell 5′ Library Construction Kit, catalog no.1000020) or TCR enrichment/library preparation (Chromium Single Cell V(D)J Enrichment Kit, Human T cell, catalog no. 1000005). Sequencing was performed on an Illumina NovaSeq S4 platform.

scRNA-seq + scTCR-seq data processing and analysis

FASTQ files generated from the Gene Expression (GEX) and Feature Barcode (TCR) libraries were processed using the 10x Genomics Cell Ranger multi pipeline (v7.0.0) to create expression matrices for downstream analysis. Integrated single-cell analysis of gene expression and TCR data (scGEX-seq + scTCR-seq) was performed using the Immunopipe package (v1.4.0) (79). Cell type annotation was conducted using the Azimuth package with the Human PBMC reference dataset (28). Azimuth is an extension of the Seurat framework that leverages annotated reference atlases to automate the processing, analysis, and interpretation of new scRNA-seq datasets. Specifically, it applies anchor-based integration to identify shared cell states between the reference and query, enabling automated and reproducible mapping of clusters to known cell types. The Human PBMC reference dataset, described in Hao et al. (28), is a multimodal atlas of human PBMCs, where cell types were manually annotated by the authors based on both RNA and protein markers, and organized into three levels of annotation. For our analysis, we used the level 2 annotation categories, which represent well-described subtypes of human immune cells. Because the original reference did not include a CD8⁺ TEMRA annotation, we further subclustered the CD8⁺ T cell population (initially identified at level 1 of Azimuth) and annotated the subtypes by manual review of RNA marker expression (80, 81). Specifically, CD8⁺ T cells were defined by the gene set shown below: Naïve CD8⁺ T cells: IL7R, SATB1, TCF7, SELL, LEF1, and CCR7; CD8⁺ TEM cells: IL7R, ANXA1, LDHB, NELL2, KRT1, TRAC, and YBX3; CD8⁺ TCM cells: CD69, ANXA2, MYO1F, GZMB, AMICA1, IFNG, CCL3, CCL4, and CXCR6; CD8⁺ TEMRA: KLRD1, GZMH, NKG7, CST7, FGR, GNLY, CX3CR1, FCRL6, S1PR5, CCL5, GZMK, PRF1, GZMA, and PLEK. Individual patient UMAP and TCR distribution were presented in Auxiliary Data 1 and Auxiliary Data 2. To enhance transparency, we also provide a heatmap (Auxiliary Data 3), a supplementary table summarizing the top RNA markers for each cell type (Auxiliary Table 1) and a table summarizing cell counts (Auxiliary Table 2). GSEA was performed using the clusterProfiler package (v4.10.1) in R (82). In addition, analysis of variance (ANOVA) test was applied to compare gene expression scores across patient groups for selected genes involved in the CD8⁺ T cell cytotoxicity pathway (83).

Autoantigen array

Autoantibody profiling was performed using an OmicsArray antigen microarray platform (GeneCopoeia, Rockville, MD). This platform has the capacity to display large number of antigens on a three-dimensional surface of a biochip and thereby serves as a multiplex screening method for the determination of autoantibody specificities. In this study, we used systemic autoimmune-associated Antigen Array (catalog no. PA001) panel that contains 120 autoantigens and eight internal controls. A while-chip citrullination process was performed to citrullinate 120 autoantigens using a peptidyl arginine deiminase (PAD) cocktail that contained a mixture of four PAD isoforms (PAD1, PAD2, PAD3, and PAD4) (84, 85). The noncitrullinated and citrullinated antigen arrays were run in parallel on a cohort of 125 human serum samples for profiling IgG and IgM antibody reactivities. Briefly, 2 μl of serum samples was pretreated with 1 unit of deoxyribonuclease I to remove free DNA, then diluted at 1:100 with Phosphate buffered saline with Tween® 20 (PBST), and hybridized onto antigen arrays. The antibodies binding with the antigens on the array were detected with Cy3-conjugated anti–human IgG and Cy5-conjugated anti–human IgM (1:2000; Jackson ImmunoResearch Laboratories). The fluorescent images were acquired with a GenePix 4000B scanner (Molecular Devices, San Jose, CA), and the signals were converted to signal intensity values using the GenePix 7.0 software (Molecular Devices). Background was subtracted, and the net signal was normalized to internal controls for IgG and IgM, respectively. The final value for each autoantibody was expressed as normalized net signal intensity.

All autoantigens from the arrays were log₁₀ transformed and then standardized by subtracting the overall mean and dividing by the overall SD. Association between each autoantigen and irAE versus healthy, RA, and ICI controls (as well as RA versus healthy controls) was assessed using logistic regression adjusted for age and sex with indicator for irAE as the outcome and log₁₀-transformed, standardized autoantigen value on the right-hand side. The data were obtained from two separate arrays. Five patient samples were run on both arrays and were compared to ensure the data could be combined. No systematic differences between arrays were observed, and the duplicate samples were removed from the first array for analysis. In the first array, there was an imbalance of age and sex between the healthy controls and the other groups. To address this, all comparisons with healthy controls were limited to females less than 64 years of age to match the makeup of the healthy controls for comparability. As the second array was matched on age and sex across all groups, no subsetting was necessary. Results from each of the comparisons were summarized as volcano plots, with –log₁₀ P values plotted on the y axis against the effect estimates on the x axis. In addition, effect estimates were summarized as heatmaps. In all analyses, P values less than 0.05 were considered statistically significant. Analysis was done using SAS version 9.4 M8 (SAS Institute, Cary, NC, USA) and R version 4.2.2 (R Foundation for statistical computing, Vienna, Austria).

Supplementary Materials

This PDF file includes:

Figs. S1 to S6

Tables S1 and S2