Current treatment strategies for rheumatoid arthritis (RA) standardise treatments across patient groups, but as a heterogeneous disease, differences will exist in the underlying immune mechanisms. As a consequence, not all patients will respond similarly to the same drug with only 60-70% having a response to any biologic drug (1). Given that an increase of pro-inflammatory and/or a decrease of anti-inflammatory CD4⁺ T cell subsets has been reported in some patients, these cell types could represent therapeutic targets and aid patient stratification for precision medicine. However, which CD4⁺ T cell subset is associated with the disease, and in what proportion of patients, remains unclear.

The study of T cells in RA was revolutionised after the Th1/Th2 paradigm was proposed by Mossman et al. in 1986 (2). This paradigm was described as a dichotomy between type 1 T helper cells (Th1), characterised by interferon-γ (IFN-γ) and tumour necrosis factor (TNF), and type 2 T helper cells (Th2). RA pathogenesis predominantly involves Th1 cytokines, and the Th1 immune response is antagonised by Th2 cytokines [reviewed in (3)]. The Th1/Th2 paradigm was modified further when pathogenic IL-17-producing T helper cells (Th17) and regulatory T cells (Tregs) were discovered to play critical roles in initiating and regulating autoimmunity, respectively [reviewed in (4)]. Increases in Th1 and Th17 cells which antagonise, and are antagonised by, Th2 and Treg cells has been the focus of much research in autoimmune diseases including RA (5, 6). More recently, interferon-producing Th1 memory cells were found to be associated with RA using novel techniques of analysing high-dimensional single-cell data (7). A meta-analysis found that circulating Treg cells, as defined by both FOXP3 and CD25, were decreased in RA patients in 9 combined studies (8). However, other studies found no decrease in Treg frequency or function in RA (9). To add to the complexity of the paradigm, there are an ever-growing list of T helper cells [Th9 cells (10); Th22 cells (11); T follicular helper cells (Tfh) (12); peripheral helper T cells (Tph) (13)]. Critically, it has now been shown that many of these T cell types exhibit plasticity. It was previously thought that the mutually exclusive expression of a master transcription factor determined the fate of T cells, with T-bet, GATA3, RORγt and FoxP3 determining the fate of Th1, Th2, Th17 and Treg cells respectively. However, co-expression of these transcription factors may temporarily alter the effector function of T cell subsets [reviewed in (14)] and many of these T cells might also be transitional and therefore appear only in very low frequencies.

The plasticity of T cell subsets and heterogeneity of RA may also explain why clinical trials targeting specific cytokines, for instance IL-17, have not reached their primary endpoint (15). Stratified medicine may allow clinicians to identify RA patients who have predominantly IL-17-driven disease to improve the design of such trials, as researchers have successfully shown in psoriatic arthritis (16). Exploring peripheral blood immunophenotypes by flow cytometry in RA has been shown to mirror findings in the synovial compartment (17) and still represents the most widely used and cost-effective technique to enumerate immune cell populations. The advent of mass cytometry (CyTOF) has vastly increased the number of cellular markers which can be detected simultaneously (18) and together with the development of high-throughput automated methods of data analysis (19–23), rare and novel cell types involved in the aetiology of RA have been recently identified (7, 13, 24, 25). Automated cell clustering algorithms allow for unbiased marker combinations and therefore the hypothesis-free discovery of unanticipated cell subsets, relevant to disease and precision medicine (7, 22, 26, 27). We postulate that Th cell plasticity and overlap between subsets and definitions explain conflicting associations and lack of reproducibility in small sample sizes. Therefore, we developed a 37 marker T cell mass cytometry panel to encompass most definitions for the most studied CD4⁺ T cell subsets to date (Th1, Th2, Th17 and Treg).

The aims of this study are first, to confirm the T helper cell subset imbalance previously described in RA; second, to explore whether this imbalance is detectable in a small sample size, if the standard definition of Th subsets is relaxed in favour of Th cell plasticity (unbiased marker combinations) and third, to test whether innovative techniques (mass cytometry and unsupervised clustering algorithm) facilitate the identification of pro- and anti-inflammatory CD4⁺ T cell subsets over standard techniques (flow cytometry and manual bi-axial gating) in a small sample size.

We show the superiority of innovative immunophenotyping and automated analytical techniques over hypothesis-driven conventional biaxial gating (based on canonical biomarkers) to identify a T cell subset imbalance in RA peripheral blood in small sample sizes. Our strategy allows for unusual combinations of T cell markers, for instance T-bet and IL-17, capturing the plasticity of T helper cell populations, a more difficult task to achieve with standard gating strategies.

Th17 cells are known to be phenotypically unstable (35). In our study, T-bet⁺ IL-17⁺ T cells are increased in RA blood. These cells have been termed Th1-like Th17 cells in the literature, co-produce IL-17 and IFN-γ, and are known to contribute to inflammation in the context of autoimmunity (35–37). Interestingly, IFN-γ production in Th1-like Th17 cells has been shown to be repressed by miR-146a (37) and loss of function polymorphisms in the microRNA-146a (miR-146a) gene have been associated with an increased risk of developing RA and lupus in both European and Asian genetic association studies (38, 39); taken together, these data provide some mechanistic evidence that Th1-like Th17 cells are involved in the pathogenesis of RA. The methods outlined in this paper easily detect an increase of this subset in RA.

We also found a significant decrease in the number of IL-2⁺ FoxP3⁺ T cells (Treg-like cells) in RA samples compared to controls. It is well-established that IL-2 is critical in maintaining Treg function (40) and that some Treg produce IL-2 (41). However, the relative role of paracrine versus autocrine IL-2 secretion from nearby effector T cells (42) versus IL-2⁺ FoxP3 T cells (41) is unclear, and the low level or absence of CD25 expression on the IL-2⁺ FoxP3⁺ subsets in this data appears to conflict with classical Treg definitions (reviewed by (43). Other studies have characterised subpopulations of CD4⁺ FoxP3⁺ memory cells that produce IL-2 (44). Consistent with our findings, the authors show that these cells possess a phenotype that lies between Tregs and other T helper cells, where they exhibit lower expression of classical Treg markers, including CD25, than their non-IL-2-producing counterparts. Therefore, rather than identifying bona fide Tregs, the unsupervised analysis has potentially identified IL-2-producing Treg-like subpopulations that may be more relevant for RA. Regardless, these findings highlight the utility of automated analyses for capturing non-canonical cell types, better reflecting the heterogeneity and plasticity of T cells.

Due to our small sample size, our association results could theoretically represent false positive due to a sampling bias or caused by multiple testing. However, this is very unlikely, as first we confirm a plausible and previously reported imbalance of Th/Treg-like subsets in RA; second the mechanistic evidence supports our findings, and finally all statistical tests have been stringently corrected for multiple testing. These elements reinforce the credibility of our main conclusion: the superiority of hypothesis-free clustering of many markers in small sample sizes over standard hypothesis-driven bi-axial gating.

The hallmark of precision medicine and the primary aim of stratified medicine initiatives in RA (45) consists of stratifying patients into treatment response categories. Our study shows large heterogeneity across patients with some with a predominantly Th1 profile, others a Th17 profile, or a Th1-like Th17 profile, while others are mainly characterised by a low IL-2⁺ Treg-like profile. As an example, stratifying patients by their IL-2⁺ Treg titres may dictate response to low-dose IL-2 therapy, as has now been successfully trialled in lupus (46) and RA (47). Furthermore, a Japanese group found that stratified medicine for psoriatic arthritis patients increased response rates significantly (16). Immunophenotyped patients received ustekinumab if Th1-dominant and secukinumab if either Th17-dominant or with a Th1/Th17-hybrid phenotype resulting in clinically significant improvement in response rates (16). The same group have good one year follow-up data (48), suggesting that the immunophenotype could be the main predictor of drug response in precision medicine. Stratifying RA patients by their Th1/Th17 phenotype, or by the Th1-like Th17 phenotype highlighted in this study, may suggest that, despite the fact that previous trials of secukinumab in RA failed to reach their primary endpoint, it could be a viable treatment option in RA patient subgroups; further work is required to confirm this hypothesis (49).

We have shown that non-biased automated analysis of large immune datasets is successful in identifying Th cell imbalance in RA with implications for precision medicine. We now plan to expand on these findings by testing their role in patient stratification for treatment response studies.

What is already known about this subject?

What does this study add?

How might this impact on clinical practice?

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://flowrepository.org/, FR-FCM-Z5RM.

The studies involving human participants were reviewed and approved by North West Ethics committee approval MREC 99/8/84 Manchester Royal Infirmary (NIHR portfolio ID 7881). The patients/participants provided their written informed consent to participate in this study.

SV conceived the project, SV, SR and AB guided it through its stages of development, ensured progress, and contributed to the final manuscript. BM ran the laboratory experiments with help from SH. MS pre-processed the raw mass cytometry data. BM and LM ran the data analysis (BM: FlowJo; LM: R pipeline) and drafted the manuscript. CF supported the automated pipeline analysis and provided R code for MASC. TH and SR provided immunology expertise. All authors contributed to the article and approved the submitted version.