Synovial tissue insights into heterogeneity of rheumatoid arthritis

Abstract

Purpose of Review:

Rheumatoid arthritis is one of the most common rheumatic and autoimmune diseases. While it can affect many different organ systems, RA primarily involves inflammation in the synovium, the tissue that lines joints. Patients with RA exhibit significant clinical heterogeneity in terms of presence or absence of autoantibodies, degree of permanent deformities, and most importantly, treatment response. These clinical characteristics point to heterogeneity in the cellular and molecular pathogenesis of RA, an area that several recent studies have begun to address.

Recent Findings:

Single-cell RNA-sequencing initiatives and deeper focused studies have revealed several RA-associated cell populations in synovial tissues, including peripheral helper T cells, autoimmunity-associated B cells (ABCs), and NOTCH3⁺ sublining fibroblasts. Recent large transcriptional studies and translational clinical trials present frameworks to capture cellular and molecular heterogeneity in RA synovium. Technological developments, such as spatial transcriptomics and machine learning, promise to further elucidate the different types of RA synovitis and the biological mechanisms that characterize them, key elements of precision medicine to optimize patient care and outcomes in RA.

Summary:

This review recaps the findings of those recent studies and puts our current knowledge and future challenges into scientific and clinical perspective.

Introduction

Rheumatoid arthritis affects approximately 0.5–1% of the population and as such has been the target of numerous drug therapy trials [1]. Despite the many treatments current approved to treat this disease, only a limited percentage of patients respond well to a given therapy, and some patients to not respond to any available treatments [2]. “Treat to target” guidelines emphasize the need to get RA under good clinical control as quickly as possible after diagnosis in order to improve long-term outcomes [3], but rheumatologists currently have few parameters to guide treatment decisions[4]. Clinical factors and routine blood labs do not provide useful predictive markers except in certain circumstances (e.g., rituximab is more effective for seropositive than seronegative RA [5]). Thus, some patients undergo many months or even years of trials of ineffective treatment, increasing the risk of irreversible joint damage and other sequelae of prolonged active disease.

Since synovial tissue is the core site of RA activity, characterizing the cellular and molecular heterogeneity of synovial inflammation is key to understanding the different axes of inflammation that can occur in RA, a necessary step toward identifying which patients are likely to respond to which treatment. These analyses may also reveal novel treatment targets among patients with poor responses to available treatments. After useful synovial tissue categories are identified, surrogate biomarkers in blood could bring the classification scheme and its clinical treatment response predictions to clinical use. This review will summarize work to date on characterizing RA heterogeneity and outline next steps.

Inflamed synovial tissue

Normal versus inflamed synovium

Normal synovial tissue is composed of a thin lining layer composed of fibroblasts and tissue resident macrophages covering a sublining composed of a loose network of collagen fibers and other extracellular matrix [6]. Fibroblasts in the lining layer produce hyaluronic acid, proteoglycan 4 (lubricin), and other factors that lubricate the joint [6–8]. In contrast, inflamed synovial tissue features hyperplasia of both the lining and sublining layers, often with infiltrating myeloid cells and lymphocytes in either diffuse arrangements or in organized aggregates [1, 6].

Several studies of RA heterogeneity have used these simple histological features, with or without immunohistochemistry for selected cell type markers, to measure synovial inflammation (e.g. Krenn synovitis score [9]) or classify RA samples [10–12]. More recently, the advent of large-scale single-cell RNA-sequencing (scRNA-seq) have provided new insights regarding the cells involved in synovitis in RA [13–15]. Along with a more granular understanding of the players involved, these studies are also beginning to reveal the cellular and molecular heterogeneity of synovial inflammation in RA.

T cells

The importance of CD4 T cells in seropositive RA has long been recognized, thanks to the strong genetic risk factor at the HLA-DRB1 locus, the so-called “shared epitope” [16, 17]. Individuals carrying an HLA-DRB1 risk allele have a 2–4-fold increased risk of developing rheumatoid arthritis [18]. HLA-DRB1 molecules carrying the shared epitope appear to be particularly suited to present citrullinated peptides to CD4 T cells, providing a potential mechanism for the development of anti-citrullinated protein antibodies (ACPAs, also referred to as anti-cyclic citrullinated peptide (anti-CCP) antibodies) in seropositive RA [19, 20].

The T cell type most strongly associated with seropositive RA is peripheral helper T (Tph) cells, a CD4 T cell subset that specializes in promoting B cell recruitment and differentiation [21]. Tph cells produce CXCL13, IL-21, and other factors that recruit and stimulate B cells. Given their ability to provide B cell help, Tph cells likely represent the link between the HLA-DR shared epitope and ACPAs, although this remains to be formally proven [22].

CD8 T cells are also present in inflamed synovium and has been reported to be more frequent in synovium from patients with ACPA+ RA [13, 23, 24]. Most of the CD8 T cells in RA synovium express granzyme K, with either low or no expression of cytotoxic proteins such as granzyme B (GzmB) and perforin [13, 23]. Instead, these cells produce cytokines such as TNF and IFNγ, and GzmK itself has proinflammatory effects on synovial fibroblasts [23]. These cells may also play additional roles in RA. For example, a group working on Sjogrens disease has proposed that GzmK+ CD8 T cells differentiate into resident memory T cells [25].

T cells are abundant in synovial tissue from some patients but rare in others. A recent study from the Accelerating Medicines Partnership RA/SLE Network suggests that there are several categories of T cell-rich synovium, each specifically enriched in a particular T cell subset, as discussed in further detail below [15].

B cells and plasma cells

Like T cells, non-plasma B cells are numerous in synovial tissue from some patients with RA and largely absent in others. Patients with abundant synovial B cells are more likely to be seropositive and to have severe disease and erosions [11, 24, 26]. Synovial tissue that contains abundant B cells also tends to be rich in T cells and macrophages, so the presence of B cells in synovial tissue is a sign of highly inflamed tissue samples with marked leukocyte infiltration [11].

Some patients with RA have formal tertial lymphoid structures with germinal centers, suggesting that B cell responses can develop entirely within the synovial tissue (i.e., independent of lymph nodes). A recent study reported RA synovial tissue enrichment of B cells expressing NR4A1, NR4A2, and NR4A3 with a transcriptomic profile matching germinal center light zone B cells [27]. These NR4A+ B cells exhibited clonal overlap with synovial plasma cells, indicating active differentiation of antibody-secreting cells within synovial tissue and supporting the presence of ectopic lymphoid structures in RA tissues.

Age-associated B cells (ABCs, also known as CD11c+ Tbet+ B cells) accumulate in the blood of patients with RA and other autoimmune diseases regardless of age [28, 29]. ABCs are also enriched in RA synovium compared to OA synovium, although they are not found in synovial tissue from all patients [13, 15]. Their levels correlate with the frequency of Tph cells in a given tissue [15]. ABCs can differentiate into antibody-secreting cells in vitro, and overlapping clonotypes among ABCs and plasma cells in synovial tissue in B-cell-receptor repertoire studies indicate that this occurs in vivo as well [30, 31]. ABCs are more likely to produce autoreactive antibodies than other B cells and thus may play a central role in pathogenesis of RA [30].

Plasma cells are found in many RA synovial tissue samples, including synovium that largely lack non-plasma B cells [15]. Future spatial transcriptomic studies performed on sections of intact synovial tissue may provide further insights into the accumulation of plasma cells in the absence of other B cell populations and the impact these cells have on the overall synovial environment and on patient treatment response.

Myeloid cells

Myeloid cells are the most heterogenous cell type across synovial tissue samples, likely because myeloid cells are uniquely responsive and adaptive to cytokine environments [15, 32]. Several different myeloid populations are enriched in RA synovium, likely reflecting differentiation triggered by the particular cytokine milieu of the synovial tissue [6, 33]. For example, HBEGF+ macrophages differentiate in response to TNF and other factors secreted by synovial fibroblasts [34]. Similarly, SLAMF7+ macrophages differentiate as the result of IFNγ or Toll-like receptor (TLR)-mediated stimulation [35]. In contrast, MERTK⁺ TREM2⁺ and MERTK⁺ LYVE1⁺ myeloid cells are tissue resident macrophages that have anti-inflammatory properties and induce repair mechanisms [6, 36]. These cells are enriched in synovium from patients who are in remission from clinically symptomatic RA.

Given their plasticity, the phenotypes of myeloid cells, in particular, may represent a useful indicator of the molecular environment and perhaps treatment response in RA. Identifying products or other molecular surrogates of these tissue macrophage populations in blood or synovial fluid may be a useful strategy for identifying biomarkers of RA synovial inflammatory phenotypes. Further studies are needed to investigate myeloid cells in RA and their potential role as a barometer of RA synovitis.

Dendritic cells remain a relatively poorly understood cell type in RA. According to a recent cell atlas of RA synovial cells, RA synovium contains several subsets of conventional dendritic cells, including subsets (DC1–4, as defined by Villani et al) [15, 37]. These DC subsets tend to be found in RA synovium containing predominantly T cells and fibroblasts (DC4) or T cells and macrophages (DC1–3) [15]. Plasmacytoid DCs are also detected in many samples. The exact roles in RA pathogenesis of these DC populations remains to be determined.

Fibroblasts

Fibroblasts undergo dramatic transformations in rheumatoid arthritis synovium, from pro-homeostatic producers of lubricin and collagen to aggressive, almost malignant, producers of IL-6 and other pro-inflammatory factors [6–8].

Normal, healthy synovium is composed predominantly of lining fibroblasts, but in inflamed RA synovium, several subsets of sublining fibroblasts (characterized by expression of THY1, encoding CD90) dominate [6, 8, 38]. NOTCH3-mediated signaling is one of the pathways by which this transformation occurs [39]. Arterial endothelial cells express NOTCH3 ligands such as JAG1 and DLL4, which induce the expression of NOTCH3, JAG1, and DLL4 on synovial fibroblasts. In this way, fibroblasts are arranged along a NOTCH3-mediated gradient around arterioles in RA synovium. Blockade of NOTCH3 in a mouse model of inflammatory arthritis blocked the clinical severity [39]. It is not yet known whether NOTCH gradients are imperative in all patients with RA or only in specific patient subsets.

While only cells of the adaptive immune system (i.e. B and T cells) are thought to have formal immunological memory, a recent study by Friscic et al suggests that synovial fibroblasts undergo metabolic changes that confer changes in phenotype [40]. In this study, the authors found that joints of mice treated with TLR ligands like monosodium urate crystals or zymosan experienced more severe joint swelling after a second round of stimulation compared to mice receiving their first stimulation. This priming effect required complement-mediated alterations of the metabolic state of synovial fibroblasts and conferred an increased migratory and invasive phenotype. The authors found that synovial fibroblasts from humans with RA exhibited similar functional behaviors and metabolic changes in vitro. It is not yet known whether inflammatory priming occurs through similar mechanisms in humans and whether it plays an important role in all patients with RA or only specific subgroups.

Other cells

Some cells do not handle tissue disaggregation well due to cell size, membrane morphology, or other factors. Examples include adipocytes, plasma cells, mast cells, and neutrophils, all of which play an important role in inflammatory arthritis according to mouse models [41–43]. Currently, our knowledge about these cells is limited, but hopefully our understanding will grow with the application of new technologies such as spatial transcriptomics, which do not require disaggregation of tissues. Information about the phenotypes and localization of these cell types would add an important new dimension to the study of heterogeneity.

Soluble factors in RA synovium

The central importance of soluble factors such as cytokines in RA is undisputed, given the clinical success of inhibitors of the TNF, IL-6, and Jak/STAT signaling pathways. The vast differences in clinical response that patients experience underscore the biological heterogeneity of RA. On the other hand, the clinical efficacy of targeted cytokine inhibition for many patients argues in favor of placing special focus on defining the cytokine environments of the synovium of patients with poor responses to available RA treatments.

One potential such candidate is LIF, which is a crucial mediator of cytokine production by synovial fibroblasts [44]. LIF is produced by synovial fibroblasts upon stimulation with cytokines such as TNF or IL-17, and it acts in an autocrine manner to amplify secretion of IL-6 as well as other cytokines such as IL-1α, IL-1β, and IL-8. Other potential cytokines of interest among non-responders include TGFβ and VEGF, which are gathering increased attention as drivers of synovial tissue inflammation in the absence of large numbers of leukocytes, as discussed further below.

Some cytokines that were considered and discarded as treatment targets in the past have garnered new interest. Type I (IFNα, IFNβ) and type II (IFNγ) interferons are good examples of this. Recent scRNA-seq data from synovial tissue and fluid have demonstrated populations with strong type I and type II interferon signatures among myeloid cells and fibroblasts [15]. For example, compared to OA synovium, RA synovium is enriched in the overlapping SLAMF7+ and CXCL10+ CXCL11+ macrophage subsets, both of which are induced by IFNγ stimulation [35, 45]. Among fibroblasts, the CXCL10+ CCL19+ fibroblasts subset can be replicated in vitro by culturing synovial fibroblasts with supernatants from stimulated T cells, which contain high levels of IFNγ [46]. Finally, age-associated (or autoimmunity-associated) B cells (ABCs) are induced by the combined action of IFNγ and IL-21 [47]. However, based on studies in mouse models and the results of past human clinical trials with recombinant IFNγ treatment, the effects of IFNγ appear to be pleiotropic [48–53]. Whether IFNγ has different effects in different patient groups (i.e. in some types of RA synovitis but not others) is not clear and could be addressed in future studies.

Extracellular matrix

The extracellular matrix in synovium is composed of collagen fibrils, hyaluronic acid, and numerous other extracellular molecules. In healthy synovium, the ECM is a loose web of collagen fibrils interspersed with cells such as fibroblasts and adipocytes [54]. In RA synovium, breakdown of ECM by matrix metalloproteases enables infiltration of the tissue by pro-inflammatory leukocytes from the blood [54, 55].

ECM plays a more dynamic role in synovial tissue than merely to provide a structural framework. Many cell types express receptors for ECM components, such as LYVE1, a hyaluronan receptor expressed by a subset of synovial tissue MERTK⁺ macrophages enriched in patients with RA in remission [36]. CD44, another receptor for hyaluronan, mediates uptake of this ECM molecule, thereby regulating its local abundance [56]. LAIR1 is an inhibitory receptor that binds to several types of collagen [56]. It is expressed by myeloid cells and some T and NK cells and downregulates cytokine production upon engagement [57]. Thus, the ECM can directly affect the functions of cells in synovial tissue.

The ECM is also a bulletin board of sorts that displays cytokines and chemokines to nearby cells [54, 56]. Heparan sulfate, an abundant membrane-bound and secreted proteoglycan, is negatively charged and thus attracts a number of cytokines and chemokines, including CCL2, CXCL12, IFNγ, IL-8, and VEGF [58]. This binding provides a spatial framework for the cytokine and chemokine gradients that are necessary for recruiting leukocytes into and through synovial tissue. T cells and other cells express heparanase, which can release these cytokine-bound heparan moieties to diffuse away and reach more cells [59].

The architecture of the ECM is not captured by transcriptional studies, but the combination of spatial transcriptomics with traditional histologic staining will hopefully address some of the questions surrounding the ECM in inflamed and remission tissues, including how ECM differs across patients with RA and the roles these differences play in the pathogenesis and treatment response.

Current cellular or molecular classification systems in RA

With the range of targeted treatments available and the promise of more in the pipeline, the push toward categorizing RA synovium is gaining momentum. A few classification matrices have been tested so far.

The most developed classification system, synovial tissue “pathotypes,” comes from Costantino Pitzalis and colleagues, building upon the work of prior groups [4, 10, 11]. This framework classifies synovial tissue into so-called pathotypes using immunohistochemistry for common cell type markers, including CD3 (T cells), CD20 (B cells), CD68 (macrophages), and CD138 (plasma cells) [11]. The three pathotypes are lympho-myeloid (rich in B cells, T cells, and plasma cells), diffuse myeloid (abundant macrophages with few B or T cells), and fibroid/pauciimmune (few leukocytes) (Figure 1). These categories are associated with distinct gene expression patterns at the bulk tissue RNA-seq level, reflecting their different cellular compositions, and they can change with time and/or treatment [60]. Applying the pathotype framework to the Pathobiology of Early Arthritis Cohort (PEAC) of patients with newly-diagnosed RA, Humby et al did not find any association of pathotypes with clinical response to methotrexate alone or in combination with other treatments (e.g. TNF inhibitors) [11]. In R4RA, a clinical trial comparing rituximab and tocilizumab, Rivellese et al find that patients with the diffuse-myeloid pathotype have a higher likelihood of responding to tocilizumab compared to rituximab (81% vs 35%) [61].

Figure 1.

Current Frameworks for Classifying Heterogeneity of Synovitis in RA

*Pie charts display average cell proportions across samples in each CTAP from Zhang et al, Nature 2023.

^¶Cell subsets in parentheses show trends but are not statistically significant.

Portions of this illustration were created using BioRender.

The Pitzalis group has also explored molecular classification systems based on bulk RNA-seq. For example, in the R4RA trial, they found that classifying tissues as B-cell-poor versus B-cell-rich using transcriptomic signatures could identify patients who were less likely to respond to rituximab than to tocilizumab (Figure 1) [12]. They have also performed post hoc analyses of the R4RA trial using signatures of two cell types at once [61]. In these analyses, patients with B-cell-poor but macrophage- and dendritic cell-rich tissues had a high frequency of response to tocilizumab but poor response to rituximab. In the These findings argue in favor of considering multiple cell types at once when investigating factors predictive of treatment response.

In a recent study, the AMP consortium (of which the author is a member) has proposed a classification system that takes six major cell types into account [15]. In this study, the authors investigated the heterogeneity of synovial tissue in RA using CITEseq (single-cell RNAseq and surface protein marker data) from a panel of synovial tissue biopsies and synovectomies from 70 patients with RA. The patient cohort was recruited to include recently diagnosed treatment-naïve patients, patients with inadequate response to methotrexate, and patients with inadequate response to TNF blockers, thus capturing a diversity of clinical phenotypes. The authors found that the patients formed a spectrum that could be divided into six categories based on the abundance of B cells, endothelial cells, fibroblasts, myeloid cells, NK cells, and T cells (Figure 1). They named this classification system Cell Type Abundance Phenotypes (CTAPs), and the six categories are called CTAP-EFM, CTAP-F, CTAP-TF, CTAP-TB, CTAP-TM, and CTAP-M, based on the predominant population(s) present in that CTAP. Importantly, synovial tissue can change from one CTAP to another with time and treatment [15, 60].

While the classification relies upon the frequency of coarse cell types, the authors find that these categories are associated with specific cell subsets and molecular pathways. For example, the T cell subsets associated with the three T-cell-rich CTAPs (i.e. CTAP-TB, -TF, and TM) differed. Follicular helper (Tfh) CD4+ T cells and Tph cells were preferentially enriched in CTAP-TB, whereas naïve and GZMB+ CD4+ and CD8+ T cells were enriched in CTAP-TF, and GZMK+ CD8 T cells trended toward an association with CTAP-TM [15]. The fact that other cell types differed in frequency in these CTAPs as well (e.g. CXCL12+ SFRP1+ sublining fibroblasts, are enriched in CTAP-TF) supports the hypothesis that CTAPs reflect different inflammatory synovial environments.

CTAPs can be determined from several different kinds of data, including single-cell RNAseq, bulk tissue RNA-seq, and flow cytometry [15]. Bulk tissue RNAseq is a particularly useful technology since it can be applied to large numbers of samples with relatively low cost and technical difficulty (beyond obtaining the synovial tissue sample). The CTAP algorithm was applied to bulk synovial tissue RNAseq collected in the context of a clinical trial comparing rituximab to tocilizumab for the treatment of patients with RA with insufficient response to TNF inhibitors. In this analysis, patients with CTAP-F had a poor response to both treatments. Further studies are needed to investigate the clinical utility of the CTAPs, but even before such translational data are available, the CTAPs provide a cellular classification framework that links clinical samples to specific molecular pathways.

Conclusions: Multiple axes of inflammation in autoimmune disease and within RA

In summary, RA synovial tissue is a complex environment of cells, soluble factors, and extracellular matrix that interact in ways that we do not yet fully understand. Based on both clinical factors (e.g., treatment response) and cell populations, there appear to be several different inflammatory phenotypes, or axes, among patients with RA. We need to characterize and classify this heterogeneity, with the goal of associating clinical phenotypes (e.g., clinical responders to a given treatment) with specific cell subsets or molecular pathways in synovium.

Data from recent studies such as AMP: RA/SLE, PEAC, and R4RA have provided crucial insights into RA heterogeneity and datasets upon which to build future work [11, 12, 15, 60, 61]. Recent leaps in technological development, such as machine learning, spatial transcriptomics, and microbiome analysis, mean that we stand on the threshold of an even deeper and more complex understanding of RA heterogeneity [62–66]. The biggest limitation at this moment is the number of clinically annotated synovial tissue samples available for study. One way to meet that need is for clinical trials in RA to include synovial tissue biopsies and other biological specimens for analysis by bulk or single-cell RNA-seq as modeled by PEAC and R4RA. As a robust synovial tissue-based classification system takes shape, associated studies can identify blood biomarkers to correlate with these synovial tissue categories. The time is right for clinical trialists, clinical researchers, and basic scientists to team up to solve the mystery of RA heterogeneity and open the door to precision medicine in RA.