Single cell sequencing identifies clonally expanded synovial CD4+ TPH cells expressing GPR56 in rheumatoid arthritis

Rheumatoid arthritis (RA) is an autoimmune disease affecting synovial joints where different CD4+ T cell subsets may contribute to pathology. Here, we perform single cell sequencing on synovial CD4+ T cells from anti-citrullinated protein antibodies (ACPA)+ and ACPA- RA patients and identify two peripheral helper T cell (TPH) states and a cytotoxic CD4+ T cell subset. We show that the adhesion G-protein coupled receptor 56 (GPR56) delineates synovial CXCL13high TPH CD4+ T cells expressing LAG-3 and the tissue-resident memory receptors CXCR6 and CD69. In ACPA- SF, TPH cells display lower levels of GPR56 and LAG-3. Further, most expanded T cell clones in the joint are within CXCL13high TPH CD4+ T cells. Finally, RNA-velocity analyses suggest a common differentiation pathway between the two TPH clusters and effector CD4+ T cells. Our study provides comprehensive immunoprofiling of the synovial CD4+ T cell subsets in ACPA+ and ACPA- RA.

R heumatoid arthritis (RA), which affects around 1% of the global population, is an autoimmune disease characterized by articular cartilage and bone erosion leading to physical disability, pain, and decreased quality of life. Joint damage is particularly exacerbated in the subset of RA patients who present with antibodies against citrullinated proteins (ACPA) 1,2 . Although the use of biologics has revolutionized the treatment of RA, 30-40% of patients fail to respond to treatment stressing the need for novel therapeutic targets 3 . The activity of CD4 + T cells in RA pathogenesis is evidenced by the HLA-DRB1 association with ACPA+ RA 4 . Expanded patient-specific T-cell clones have been identified in the synovium and synovial fluid (SF) of RA patients using bulk TCRβ repertoire analysis 5 but without any connection to their effector function or antigen reactivities. Over the years, several CD4 + T-cell subsets, including Th1 and Th17 cells have been described in the synovial joints of RA patients 6 . Recently, PD-1 high MHC-II + (major histocompatibility complex class II) peripheral helper T cells (T PH ) were identified in the synovial fluid and tissue of seropositive RA patients [7][8][9] where they are proposed to facilitate B cell recruitment and activation through CXCL13 and IL-21 production; however, there is still no satisfying marker that defines the T PH subset. Furthermore, it is also unclear how this population is generated in the synovial joint. We have recently identified CD4 + T cells with cytotoxic features in SF of RA patients carrying the PTPN22 1858T risk allele variant 10 . This finding, as well as previous studies reporting expanded cytotoxic CD4 + CD28 null T cells in the blood of RA patients 11,12 , suggests that cytotoxic CD4 + T cells might contribute to RA. However, a transcriptomic and clonality analysis of cytotoxic CD4 + T cells in the context of previously described CD4 + T-cell subsets, is still missing. The identification of T-cell effector functions is deemed important since it can lead to new therapeutic strategies.
Here, we performed 5´single-cell sequencing in combination with TCRαβ sequencing on peripheral blood (PB) and synovial fluid (SF) from ACPA+ and ACPA− RA patients to investigate CD4 + T-cell subsets. We demonstrate that clonal expansions are prominent amongst CXCL13 high T PH , Treg, effector and cytotoxic CD4 + T cells in RA SF. We also describe two T PH CD4 + T-cell clusters expressing different levels of CXCL13. Finally, we find that T PH cells express the G-protein-coupled receptor GPR56, the inhibitory receptor LAG-3, and the tissue-resident memory (T RM ) receptors CXCR6 and CD69 implicating that these T cells are maintained in the ACPA+ RA synovial joint. Our data provide a comprehensive immunoprofiling analysis of pathogenic CD4 + T cells at the site of inflammation in ACPA− and ACPA+ RA.

Results
Cytotoxic CD4 + T cells are enriched in ACPA+ RA synovial fluid. CD4 + T cells with cytotoxic function share common functional characteristics with NK and CD8 + T cells, including the expression of cytolytic proteins PRF1 (perforin-1), GZMB (granzyme B) 13 and NKG7 (natural killer cell granule protein 7) 14 . Both transcription factors Eomes 15,16 and Hobit 17 are expressed in circulating human cytotoxic T cells. GPR56 is an adhesion G-proteincoupled receptor encoded by the ADGRG1 gene expressed on human circulating NK, CD8 + , and CD4 + cytotoxic cells 18 . Synovial fluid mononuclear cells (SFMC) from ACPA− (n = 9) and ACPA+ (n = 12) RA patients were screened for the expression of these cytotoxic effector molecules and transcription factors in both CD4 + and CD8 + T cells by flow cytometry (Fig. 1, Supplementary Figs. 1a and 2, and Supplementary Data 1). ACPA+ SFMC presented with a significantly increased frequency of GZMB + PRF1 + , Hobit + , NKG7 high , and GPR56 + CD4 + cells (Fig. 1a, b). No significant difference was observed for expression of GZMA and Eomes. Among CD8 + T cells, Hobit expression was increased in ACPA+ SFMC and a similar trend was observed for the expression of GZMA, GZMB, and NKG7 without reaching statistical significance ( Supplementary Fig. 2). No significant difference was observed in the expression of PRF1 and GPR56 on CD8 + T cells in ACPA+ versus ACPA− RA. The frequency of GZMB + PRF1 + CD4 + T cells in SF positively correlated with CCP (cyclic citrullinated peptide) positivity (Fig. 1c). In ACPA+ synovial fluid, GPR56 identified a distinct CD4 + T-cell population with a frequency that also correlated with CCP positivity (Fig. 1d).
Single-cell sequencing map of CD4 + T-cell subsets in RA. To further characterize cytotoxic CD4 + T cells in SFMC, we performed 10X 5´single-cell sequencing in combination with TCRαβ sequencing on purified CD4 + T cells from paired SFMC and peripheral blood mononuclear cells (PBMC) from 11 ACPA+ RA patients and 4 ACPA− patients ( Fig. 2a and Supplementary Data 1). After quality control (Supplementary Figs. 3 and 4a), we obtained transcriptomic data from a total of 166,944 cells. Unsupervised analysis of this transcriptomic data generated 12 total clusters that were annotated based on both known cell-type markers and enrichment for cell-type-specific gene modules (Fig. 2b, c, Supplementary Figs. 4b, c and 5a, b, and Supplementary Data 2 and 3). Differentially-expressed genes (DEGs) were also compared to clusters described in PB of healthy donors 19 and the Azimuth reference single-cell mapping database 20,21 using the EnrichR interface 22 ( Supplementary  Fig. 5c). Together, using this approach we identified a naive CD4 + T-cell cluster (cluster 1) expressing high levels of SELL, CCR7, and IL7R 23 , central memory CD4 + T cells (cluster 3) expressing LTB and KLF2; effector CD4 + T cells (cluster 4) expressing KLRB1 and GZMA; Tregs (cluster 5) expressing FOXP3 and IL2Ra 24 ; proliferating CD4 + T cells (cluster 11) expressing MKI67 and STMN1; cytotoxic CD4 + T cells (cluster 6) expressing cytotoxicity related genes including NKG7, GZMH, PRF1 and ZNF683 (Hobit) among others; activated CD4 + T cells (cluster 12) expressing CD38 and HLA-DR; SESN3 high memory/ effector CD4 + T cells (cluster 7) and EGR1 high naive CD4 + T cells (cluster 10). Interestingly, SESN3 high CD4 + T cells presented a cytokine signature matching with IL-2, TNF, or IL-6 signatures depending on the comparative datasets (Supplementary Fig. 5c). Cells from cluster nine expressed mitochondrial encoding genes and were named Humanin CD4 + T cells as previously described in psoriatic arthritis patients 25 . Finally, based on the expression of CXCL13, MAF, TOX, PRDM1, TIGIT and PDCD1 7,9,26 , two T PH cell clusters were subclassified as CXCL13 high T PH (cluster 2) and CXCL13 low T PH (cluster 8). We further conducted our analysis on paired PB and SF compartment from four ACPA− and four ACPA+ RA patients (Fig. 2b, see "Single-cell sequencing of RA samples" in "Methods" and Supplementary Fig. 3). As expected, naive and central memory CD4 + T-cell frequencies were decreased in SF as compared to PB (Fig. 2d). Humanin and EGR1 high CD4 + T-cell cluster frequencies were also decreased in SF. The frequency of Tregs was increased in SF as previously described by us and others 27,28 . Similarly, CXCL13 high T PH , CXCL13 low T PH , effector CD4 + , proliferating CD4 + , and activated CD4 + T-cell clusters frequencies were all increased in SF as compared to PB. Of note, the cytotoxic CD4 + T-cell cluster was equally frequent in PB and SF. These differences remain significant even when exclusively looking at the memory T-cell compartment ( Supplementary Fig. 6a). No significant differences were observed between ACPA− and ACPA+ patients ( Supplementary Fig. 6b, c). Of note, the frequency of the different clusters was not significantly different in patients undergoing corticosteroid, methotrexate, or biologics treatments at the time of sampling ( Supplementary Fig. 7a). Hence, single-cell transcriptomic data identify several CD4 + T-cell subsets, including a subset of cytotoxic CD4 + T cells and two distinct T PH clusters in SF of RA patients.
GPR56 delineates T PH CD4 + T cells in ACPA+ RA synovial fluid. Since GPR56 expression was upregulated on 26% of CD4 + T cells in SF of ACPA+ RA patients (Fig. 1a, b), we further investigated ADGRG1 (encoding GPR56) in the single-cell dataset. In PB, ADGRG1 was mainly expressed by cytotoxic CD4 + T cells (cluster 6), that are also characterized by the expression of GZMA, GZMB, NKG7, PRF1 and ZNF683 (Hobit) (Supplementary Fig. 8a-c). At the protein level, flow cytometry experiments validated that, in circulating CD4 + T cells, GPR56 expression correlates with the expression of cytotoxic effector molecules PRF1, GZMB, GZMA, NKG7, and the transcription factors Eomes and Hobit ( Supplementary Fig. 8d-f) confirming a previous report 18 . Still, a fraction of circulating GPR56 + CD4 + T cells expressed CXCL13 after CD3/CD28 stimulation (Supplementary Fig. 9). Interestingly, in SF, ADGRG1 expression was mainly detected in CXCL13 high T PH (cluster 2), CXCL13 low T PH (cluster 8) and proliferating CD4 + T cells (cluster 11) (Fig. 3a, Supplementary Fig. 10a-c). We further examined the expression of known T PH -associated markers 7,9,26 and observed that ADGRG1 expression overlapped with the expression of PDCD1 (encoding PD-1), CXCL13, HLA-DRB1, MAF, ICOS, PDRM1 (encoding BLIMP-1), and TOX in the two T PH states (Fig. 3a). At the RNA level, the two T PH clusters were characterized by different levels of ADGRG1, PDCD1, CXCL13, HLA-DRB1, MAF, and PDRM1 ( Supplementary Fig. 11a, b). In particular, in ACPA+ SF, CXCL13, PDCD1, and ADGRG1 were increased in CXCL13 high T PH whereas PRDM1 was increased in CXCL13 low T PH cells. The expression of TOX was not significantly different between the two subsets ( Supplementary Fig. 11a). The same differences were also observed in T PH cells from ACPA− SF, in particular for ADGRG1 and PRDM1 ( Supplementary Fig. 11b). Flow cytometry experiments confirmed previous reports 7,9 showing an increased frequency of PD-1 high and PD-1 expression on CD4 + T cells in ACPA+ as compared to ACPA− RA patients ( Fig. 3b and Supplementary Fig. 12). In SF, GPR56 expression defines the subset of T PH cells (Fig. 3b) -BV510  CD4-BV510  CD4-BV510  CD4-BV510  CD4-BV510   0  -10  3  10  3  10  4   ACPA− T PH cells, however, expressed lower levels of GPR56 as their ACPA+ counterparts (Fig. 3b). Of note, these frequencies were not different depending on the treatment at the time of sampling ( Supplementary Fig. 7b). We confirmed that the two T PH subsets could be distinguished based on different levels of GPR56 expression within PD-1 high CD4 + T cells (Fig. 3c, gating strategy in Supplementary Fig. 1b). PD-1 high GPR56 + CD4 + T cells expressed a higher level of MHC-II, PD-1, and CXCL13 as compared to PD-1 high GPR56 low CD4 + T cells (Fig. 3c-e). However, PD-1 high GPR56 + CD4 + T cells also expressed higher levels of BLIMP-1 although PDRM1 gene expression was lower in this subset (Fig. 3e, lower panel and Supplementary Fig. 11a, b). While the frequency of PD-1 high GPR56 + CD4 + T cells was lower in ACPA− SF (Fig. 3c, left panel), they also display increased CXCL13, MHC-II, and PD-1 expression as compared to their PD-1 high GPR56 low counterparts ( Fig. 3c-e). GPR56 was not detected on cytotoxic CD4 + T cells in SF ( Supplementary  Fig. 10d, e) and the expression of ADGRG1 was low in SF cytotoxic CD4 + T cells ( Supplementary Fig. 10b, c). In NK cells, the expression of GPR56 is regulated by cleavage and endocytosis after activation 29 and a similar phenomenon could also contribute to the absence of GPR56 on cytotoxic CD4 + T cells in SF. We, therefore, investigated GPR56 expression after three hours of CD3/CD28 beads or PMA/ionomycin stimulation on PBMC and SFMC from RA patients. While three hours of CD3/ CD28 stimulation had no effect on GPR56 expression, PMA/ ionomycin induced downregulation of GPR56 surface expression on PB and SF CD4 + T cells ( Supplementary Fig. 13). These experiments show that rapid GPR56 downregulation can also occur on GPR56 + CD4 + T cells through TCR-independent mechanisms. Altogether, these data indicate that, while GPR56 is mainly expressed on cytotoxic CD4 + T cells in PB, it also delineates the subset of PD-1 high CXCL13 high T PH cells which is increased in ACPA+ RA SF.
GPR56 + CD4 + T cells display distinct tissue-resident memory receptors. Besides being an inhibitory receptor, PD-1 is also included in the signature associated to T-cell tissue residency 30 and implicated in follicular helper T-cell maintenance in germinal centers 31 . We therefore assessed the expression of known tissueresident memory (T RM ) T-cell receptors: ITGA1 (CD49a), CD69, CXCR6 but also CX3CR1 which is downregulated on T RM cells 30 . We also evaluated the expression of the inhibitory receptor LAG-3 which is often co-expressed with PD-1 on tumorinfiltrating lymphocytes 32 . In ACPA+ SF, we found that LAG-3 was mainly expressed on T PH cells (cluster 2 and 8) and proliferating T cells (cluster 11) (Fig. 4a). A similar tendency was also observed for the expression of CXCR6 in SF. Since we previously observed that GPR56 delineates the subset of T PH cells, we assessed the expression of these receptors in the context of GPR56 by flow cytometry. LAG-3, CXCR6, and CD69 frequencies were enriched on synovial GPR56 + CD4 + T cells (Fig. 4b, c, gating strategy in Supplementary Fig. 1c) in ACPA+ SF. In contrast, CD49a frequency was higher on GPR56 − CD4 + T cells. In ACPA− SF, GPR56 + CD4 + T cells also showed a higher frequency of CXCR6, CD69, and LAG-3 as compared to GPR56 − CD4 + T cells. However, LAG-3 frequency was higher on CD4 + a TPH-associated genes in synovial fluid T cells in ACPA+ SF as compared to ACPA− SF (Fig. 4d) and correlated with CCP positivity (Fig. 4e). In CD8 + T cells, CD49a was the only receptor for which an increased frequency was observed in ACPA+ SF ( Supplementary Fig. 14). These data indicate that, in SF, GPR56 expression correlates with the concomitant expression of a distinct set of tissue-resident memory receptors on CD4 + T cells: CXCR6 and CD69 and the inhibitory receptor LAG-3.
Clonally expanded CXCL13 high T PH CD4 + T cells in RA synovial fluid. Next, we examined the expanded TCR clones in relation to the aforementioned CD4 + T-cell clusters (Supplementary Data 4 and 5). A clone was defined by at least 2 cells sharing the same CDR3 amino acid sequence from both TCRα and TCRβ chains (Supplementary Data 6a). Of note, TCR recovery was low in patient RA12 which was excluded from further clonality analysis. In SF, the most expanded clones were found in CXCL13 high T PH (median frequency 45.7%), Treg (median frequency 28.7%), effector CD4 + (median frequency 11.3%) and cytotoxic CD4 + (median frequency 7.2%) T-cell clusters (Fig. 5a-c and Supplementary Data 6b). In all four ACPA+ RA patients CXCL13 high T PH cells contained most of the expanded clones. In two out of three ACPA− RA patients (RA8, RA9), Tregs represented the most expanded clones in SF (Fig. 5b, left panel, Fig. 5c, left panel and Supplementary Data 6b). This observation prompted us to analyze the frequency of FOXP3 + CD4 + T cells which was slightly increased in ACPA− SF (Fig. 5d). FOXP3 + CD4 + T-cell frequency in SF was not different depending on treatment at the time of sampling ( Supplementary Fig. 7b). In blood, most of the expanded clones originated from central memory CD4 + T cells (median frequency 45.8%) and cytotoxic CD4 + T cells (median frequency 32.6%) (Fig. 5b, right panel, Fig. 5c, right and Supplementary Data 6c) irrespective of their ACPA status. Overall, these data show that in SF of RA patients most of the expanded clones are found within the CXCL13 high T PH , the Tregs, the effector and the cytotoxic CD4 + T-cell subsets. In blood, the expanded T-cell clones were mostly found within cytotoxic and central memory CD4 + T cells.
The two T PH subsets are functionally related. Further, we examined the overlapping of all CDR3 sequences between SF CD4 + T-cell subsets that would be indicative of a possible shared differentiation pathway (Fig. 6a, b). Of note, no common CDR3 sequences were observed across patients. In SF, CXCL13 high T PH showed clonal overlap with CXCL13 low T PH (Jaccard Index (JI):0.038), proliferating (JI:0.031), effector (JI:0.023) and cytotoxic CD4 + T cells (JI:0.013) (Fig. 6b). SF cytotoxic CD4 + T cells showed also clonal overlap with effector (JI:0.016) and proliferating CD4 + T cells (JI:0.013). CXCL13 low T PH presented a clonal overlap with CXCL13 high T PH (JI:0.038), proliferating (JI:0.028), effector (JI:0.014), and cytotoxic CD4 + T cells (JI:0.007). SF Tregs showed clonal overlap with proliferating CD4 + T cells (JI:0.02). We then focused our analysis on expanded clones in CXCL13 high T PH , Tregs, effector CD4 + T cells and cytotoxic CD4 + T cells (Supplementary Data 7). A median of 59% of the CXCL13 high T PH expanded clones shared CDR3 sequences with cells from other clusters (either clones or unique cells) (Supplementary Data 6d). These shared sequences were identified in all patients. A median frequency of 28% of these clones was shared with proliferating CD4 + T cells, 25% with CXCL13 low T PH , 19% with effector CD4 + T cells, and 6% with cytotoxic CD4 + T cells ( Fig. 6c and Supplementary Data 6e). Treg clones shared a median of 42% and 17% of their CDR3 sequences with proliferating CD4 + T cells and CXCL13 high T PH cells, respectively. Effector CD4 + T-cell clones shared a median of 26% and 10% of their CDR3 sequences with CXCL13 high T PH and cytotoxic CD4 + cells, respectively. Cytotoxic CD4 + T-cell clones shared a median of 26% and 6% of their CDR3 sequences with CXCL13 high T PH and effector CD4 + T cells. Of note, for effector and cytotoxic CD4 + T cells, sharing was primarily coming from two patients (RA10, and RA11, Supplementary Data 6e and 7). Cytotoxic CD4 + T cells from PB and SF also presented clonal overlap (JI:0.036). To further understand the cell fate of CXCL13 low T PH cells in the context of cytotoxic CD4 + T cells, we sorted GPR56 low PD-1 high CD4 + T PH cells and stimulated with CD3/CD28 beads (gating strategy in Supplementary Fig. 1D). We observed an increased PD-1 expression and a tendency towards increased GPR56 and BLIMP-1 expression (Fig. 6d). Perforin-1 expression, however, did not dramatically increase. Of note, CXCL13 did not increase after TCR stimulation, possibly due to the lack of additional inducers such as TGFβ and IL-2 neutralizing conditions as previously described 26 . To further evaluate cell differentiation across clusters, we performed an RNA-velocity analysis (scVelo) which estimates future cell states based on the ratio of detected spliced and unspliced mRNAs transcripts 33 . Unsurprisingly, we observed a progression from naive CD4 + T cells into central memory and effector CD4 + T cells validating our approach (Fig. 6e). In ACPA+ SF, both CXCL13 low T PH and effector CD4 + T cells led to CXCL13 high T PH cells (Fig. 6e, right  panel). This transition from CXCL13 low to CXCL13 high T PH was less obvious in ACPA− SF (Fig. 6e, left panel). In addition, the cytotoxic CD4 + T-cell final differentiation stage was originating from SESN3 + and effector CD4 + T cells in both ACPA− and ACPA+ RA. Hence, effector CD4 + T cells project into two      independent branches leading to T PH in one direction and cytotoxic CD4 + T cells in the other direction. Tregs were connected to central memory and proliferative CD4 + T cells. These lines of evidence suggest that CXCL13 low and CXCL13 high T PH in SF are functionally and developmentally related, possibly originating from effector CD4 + T cells while cytotoxic CD4 + T cells potentially arise from other effector CD4 + T cells independent of the T PH state.

Discussion
Leveraging single-cell technology of CD4 + T cells in PB and SF of RA patients, we dissected the T PH CD4 + T-cell subset at the single-cell level showing that it is composed of two independent clusters with different CXCL13 expression levels. So far, the T PH CD4 + T-cell subset was only defined by high expression of PD-1 and MHC-II while more accurate markers were still lacking. We show that GPR56 clearly delineates the CXCL13 high T PH subset and associates with the expression of tissue-resident memory receptors CXCR6 and CD69 and the inhibitory receptor LAG-3. We also identify an increased frequency of cytotoxic CD4 + T cells in ACPA+ RA SF. Finally, we show that CXCL13 high T PH , Treg, T effector, and cytotoxic CD4 + T cells are clonally expanded in RA SF. Together, these data provide a refined characterization of T PH and cytotoxic CD4 + T cells, two important CD4 + T-cell subsets in RA (Fig. 6f).
In synovial tissue of ACPA+ RA patients, T PH CD4 + cells produce CXCL13 7,9,34 , localize with B cells and have the ability to induce in vitro plasma cell differentiation through IL-21 production 7 . T PH cells are defined by high expression of PD-1 and MHC-II, the cytokines CXCL13, IL-21, and IL-10, and the transcription factors BLIMP-1, TOX, MAF, and SOX4 7,9,26 . T PH CD4 + T-cell differentiation is induced by TGFβ in IL-2 neutralizing conditions 26 . Single-cell sequencing identified two distinct clusters of T PH cells: CXCL13 low T PH cells and CXCL13 high T PH cells. We postulated that these two clusters correspond to two states of differentiation: CXCL13 low T PH CD4 + T cells could be the precursors of the more differentiated cytokine-producing CXCL13 high T PH CD4 + T cells. In line with this hypothesis, common TCR sequences were observed between CXCL13 low T PH cells and CXCL13 high T PH cells clusters suggesting a common origin between these two subsets. CD3/CD28 stimulation of GPR56 low PD-1 high T PH cells also induced an increased PD-1 upregulation compatible with this scenario. Finally, RNA velocity and TCR clonality analysis show a connection between effector memory CD4 + T cells and CXCL13 high T PH, indicating that the precursors of T PH cells are probably contained within the effector CD4 + T-cell population which is enriched in synovial fluid.
CXCL13 high T PH CD4 + T cells were also detected in ACPA− RA SF but at a much lower frequency than in ACPA+ patients. In ACPA− patients, these few CXCL13 high T PH also presented an increased expression of bona fide T PH markers (MHC-II, PD-1, CXCL13, BLIMP-1) but lower levels of LAG-3, PD-1, and GPR56 than in ACPA+ patients. The expression of these inhibitory receptors might be a sign of persistent T-cell activation in the ACPA+ synovial joint. We also identified clonally expanded Tregs in SF of two ACPA− RA patients and a slight increase in Treg frequency. Altogether, these data suggest that the differentiation towards CXCL13 high T PH cells is increased in the joint of ACPA+ RA as compared to ACPA− RA patients. In line with this hypothesis, the RNA-velocity analysis of ACPA+ CD4 + T cells revealed a trajectory from CXCL13 low to CXCL13 high T PH . An increased frequency of PD-1 high CD4 + T cells was recently identified in leukocyte-rich RA synovial tissues 8 . We confirm the presence of this subset also in SF with an increased frequency in ACPA+ RA patients. While SF analysis allows a refined analysis of many immune cells and the downstream clonality analysis, a limitation is that these samples are originating from patients with long disease duration and under immunosuppressive drugs. In our dataset, we did not find any specific association with ongoing medication but, interestingly, the only ACPA− patient whose SF presented CXCL13 high T PH cell clonal expansion was not under any immunosuppressive medication at time of sampling. Longitudinal studies will inform about the possible effects of current treatments on the expansion and maintenance of these CD4 + T-cell subsets.
Expanded TCR clones have been identified in synovial tissues and fluid from RA patients using bulk TCRβ sequencing 5 . In blood, expanded cytotoxic CD4 + CD8 null cells present with a bias in their TCRVβ usage 35,36 . We found that expanded clones in the blood were originating from the CD4 + cytotoxic T-cell subset. Clonally expanded cytotoxic CD4 + T cells were also observed in RA SF, although at a lower degree than the CXCL13 high T PH cells . Overlap clonality between cytotoxic clones in PB and SF suggest their recirculation. Expansions of cytotoxic CD4 + T cells are frequently observed in chronic viral infections, and they are capable of directly killing EBV-infected B cells in an MHC-IIdependent manner 37 . Our data show that cytotoxic CD4 + T cells are enriched in SF of ACPA+ RA patients and correlate with CCP titers, suggesting that these cells are mainly implicated in the subset of autoantibody-positive RA. Still, so far, their autoreactive properties have not been investigated in the context of citrulline immunity. The differentiation processes that lead to the differentiation of cytotoxic CD4 + T cells are not completely elucidated, but IL-2 38 , IL-15 39 , and type I IFNs 40 contribute to their development. The RNA-velocity analysis suggests that cytotoxic CD4 + T cells originate from effector CD4 + T cells and from the subset of SESN3 + memory T cells. Although the memory SESN3 + CD4 + T-cell subset has not been extensively studied before, it has been identified in skin biopsies from patients with psoriasis 41 . Interestingly, we detected cytokine-related signatures in SESN3 + CD4 + T cells which could be implicated in their differentiation into cytotoxic CD4 + T cells. Further studies are warranted to dissect the differential pathways leading to cytotoxic CD4 + and CXCL13 high T PH cells, two CD4 + T-cell types enriched in ACPA+ RA.
GPR56 is an adhesion G-protein-coupled receptor encoded by the ADGRG1 gene implicated in numerous migration/adhesion processes including neuron migration 42 and tumor growth inhibition 43 . In NK cells, GPR56 is an inhibitory receptor binding Fig. 6 CD4 + T-cell subsets interconnectivity. a Chord diagrams showing the interconnectivity between CD4 + T cells sharing the same CDR3 (paired TCRα, and TCRβ chains) amino acid sequences within CD4 + T-cell clusters in RA SF (left panel) and between compartments in selected clusters (right panel) (n = 4 ACPA−, n = 4 ACPA+). b Jaccard overlap quantifications for clonal overlap between cell clusters across compartments (n = 4 ACPA−, n = 4 ACPA+). c Percentage of shared expanded clones (n ≥ 2 cells) shared with selected subsets among total cell clones, n = 7 RA patients (n = 4 ACPA+, 3 ACPA−). d Representative flow cytometry dot plots showing the expression of PD-1, GPR56, CXCL13, BLIMP-1, and PRF1 in control (upper panel) or after 48 h CD3/CD28-beads activation of T PH cell states (lower panel) (PD-1 high GPR56 low and PD-1 high GPR56 + from ACPA+ SF), quantified in n = 3 ACPA+ RA patients. d Data are from a pool of three independent experiments where a circle is a single replicate. Line indicates median, two-tailed Wilcoxon paired test, P = 0.0005 (MFI PD-1 on PD-1 high GPR56 low after activation), P = 0.0740 (MFI PD-1 on PD-1 high GPR56 + after activation), P = 0.0869 (MFI GPR56 on PD-1 high GPR56 low after activation), P = 0.0276 (MFI GPR56 on PD-1 high GPR56 + after activation), P = 0.0563 (% BLIMP-1 on PD-1 high GPR56 low after activation), P = 0.0775 (% BLIMP-1 on PD-1 high GPR56 + after activation), P = 0.019 (% PRF1 on PD-1 high GPR56 low after activation). e Velocity plots showing the phenotype directionality between the different CD4 + T-cell clusters in RA split in ACPA− (left panel) and ACPA+ (right panel) (n = 4 ACPA−, n = 11 ACPA+, pool PB and SF). f Graphical summary showing the proposed developmental link between the two T PH subsets, effector and cytotoxic CD4 + T cells in SF as well as the identified receptors.
to the tetraspanin CD81 29 . We confirmed data from a previous report showing that GPR56 is also a marker of peripheral cytotoxic CD4 + cells 18 . At the site of inflammation, however, we discovered that GPR56 was highly expressed on T PH cells in ACPA+ RA patients and could be used to identify T PH CD4 + T cells. In SF CD4 + T cells, GPR56 expression correlated with the expression of the inhibitory receptors LAG-3 and PD-1 as well as CXCR6 and CD69, receptors expressed on tissue-resident memory T cells 30 . Similarly, in multiple sclerosis, CD8 + T cells from active lesions exhibit a tissue-resident memory phenotype (CD69, CD103, PD-1, CD49a) associated with an intermediate expression of GPR56 without granzyme B expression 44 . In mice, Blimp-1 and Hobit mediates the transcriptional program associated with T-cell tissue residency 45 . In ACPA+ SF, BLIMP-1 was expressed on T PH CD4 + cells presenting with tissue-resident memory receptors whereas Hobit was only expressed on cytotoxic CD4 + T cells suggesting that, BLIMP-1 could drive tissue residency of T PH CD4 + T cells in the synovial joint. GPR56 might be implicated in the migration and/or maintenance of T PH CD4 + T cells in the synovial joints where its ligand remains to be determined. In SF, cytotoxic CD4 + T cells expressed low levels of ADGRG1 and had no detectable GPR56 surface expression. We showed that similar to what has been described in NK cells 29 , PMA/Ionomycin stimulation induced a rapid GPR56 downregulation on GPR56 + CD4 + T cells. In vivo, the inflammatory milieu might therefore contribute to GPR56 downregulation on SF cytotoxic CD4 + T cells.
In summary, we identify two clusters of T PH CD4 + T cells and observe clonally expanded CXCL13 high T PH , Tregs, effector, and cytotoxic CD4 + T cells in RA SF. We show that GPR56 is a receptor that should be used to identify CXCL13 high T PH CD4 + cells. Future studies aiming at understanding the function of GPR56 in T PH CD4 + and cytotoxic CD4 + T cells will contribute to evaluating the feasibility of targeting this receptor in ACPA+ RA.

Methods
Patient and samples. This study was approved by the research ethics committee of Karolinska University Hospital, Sweden and all patients signed informed consent according to the Declaration of Helsinki. All patients had established disease, based on the 1987 criteria from the American College of Rheumatology. Synovial fluid from ACPA− (n = 13) and ACPA+ (n = 16) RA patients were collected at several visits at the Rheumatology Clinic of Karolinska University Hospital, Stockholm (Supplementary Data 1). In n = 12 ACPA− and n = 10 ACPA+ RA patients, paired PBMCs were also available. PBMC and SFMC were isolated by Ficoll separation (GE Healthcare) and cryopreserved in fetal bovine serum (FBS) with 10% DMSO in liquid nitrogen.   Fig. 3). In experiment 10X-1 (Supplementary Data 1), for samples from RA1-7, following CD4 + T-cell isolation, cells were counted, hashed with TotalSeq-C0251 (ref: 394661) and TotalSeq-C0253 (ref: 394665) anti-human antibodies, Biolegend) and pooled per patient (i.e., pool 1 = PB and SF from patient 1, pool 2 = PB and SF from patient 2, etc.). The pooled cells were then counted again and immediately loaded onto the chromium chip G using the standard protocol for the Chromium single-cell 5' kit v1 (10x Genomics, Inc). Following Gel Bead-in Emulsion (GEM) generation, samples were processed using the standard manufacturer's protocol. This pipeline was also used to process samples for experiment 10X-2 (Supplementary Data 1) with RA8-15 except for antibody hashing and the chromium kit version; each sample was processed independently (i.e., 1 sample/10x chromium lane; Supplementary Fig. 3) and cells were processed with the Chromium single-cell 5' kit v2 (Dual Index). Once sequencing libraries passed standard quality control metrics, the libraries for RA1-7 were sequenced on Illumina NextSeq500/550 high output 75-cycle v2.5 kits with the following read structure: read1: 26, read2: 58, index 1: 8. Libraries for RA8-15 were sequenced on Illumina NextSeq500/550 high output 150-cycle v2.5 kits with the following read structure: read1: 26, read2: 90, index 1: 10, index 2: 10. Both sets of libraries were sequenced to obtain a read depth greater than 20,000 reads/cell for the gene-expression (GEX) libraries and greater than 5000 reads/cell for the V(D)J-enriched T Cell libraries.
Data preprocessing and quality control. Aligned matrices were first filtered to remove low-quality barcodes, keeping only those with greater than 200 UMIs and less than 25% mitochondrial reads. After removing low-quality barcodes, cell doublets were then computationally removed using DoubletFinder 47 with calculated doublet rates based on cell input for each sample. Using cell-type-defining markers, the cleaned datasets were filtered for non-CD4 T cells (e.g., monocytes, macrophages, CD8 + T, CD20 + B cells using CD14, CD8a, CD8b, and MS4A1). Finally, we attempted to deconvolute the multiplexed PBMC and SFMC cell populations, from RA1-7, using antibody hashes; however, due to sub-optimal tagging, we were unable to separate the compartment cell populations. To prevent this complication from occurring with RA8-15, we omitted antibody hashing from our pipeline and processed compartments separately to retain compartmentspecific cell populations. Samples from RA1-15 were then merged accordingly for the subsequent unsupervised (Fig. 2b, left panel) and velocity analyses ( Fig. 6e and Supplementary Fig. 3).
RNA velocity. The scVelo package, version 0.2.4 33,46 was used to model the RNA velocity of cells from volunteers RA1-15. Briefly, the RNA-velocity analysis leverages the expression relationship between spliced (exonic) and unspliced (intron-containing) RNA across selected variable genes to predict the directionality of transitions between cell types, as well as the putative driver genes of these transitions. Following the unsupervised analysis, the analyzed Seurat object was converted into an H5AD file using the SeuratDisk package [https://mojaveazure.github.io/seurat-disk/index.html], version 0.0.0.9011, and then imported into a jupyter notebook for preprocessing according to the scVelo standard pipeline: (1) genes were first normalized, selecting the top 3000 genes (pp.filter_-and_normalize(adata,n_top_genes = 3000, enforece = T), (2) principal components, nearest neighbors in PCA space, and first and second-order moments of the nearest neighbors were calculated (n_pcs = 70, n_neighbors = 20), (3) using the dynamical model from scVelo, RNA velocities were then estimated (tl.recover_dynamics(adata, mode = ' dynamical')) and (4) used to compute the velocities and velocity graph (tl.velocity and tl.velocity_graph). This approach was used for both ACPA+ and ACPA-RA cohorts.
TCR mapping. Following sequencing, V(D)J libraries for RA8-15 (experiment 10X-2, Supplementary Fig. 3, Supplementary Data 1) were demultiplexed using bc2fastq (v2.17.1.14) to generate FASTQ files for each sample, uploaded to terra, and mapped and quantified using the 10X Genomics Inc software package Cell-Ranger on Cumulus (https://cumulus.readthedocs.io/en/latest/cellranger.html, snapshot 15, Cellranger 5.0.1, default parameters) to generate consensus annotation files for each sample. Using the scRepertoire 51 package, sample-specific consensus annotation files were consolidated into a list of TCR sequencing results and then integrated with the Seurat object, containing RA8-15, using the combineExpression function. Clonotypes were called based on the CDR3 nucleotide sequence and VDJC gene (i.e., scRepertoire nomenclature "gene+nt") and clonotype expansion were assigned based on the relative frequency. The interconnectivity between specific cell types, both within and between tissue compartments, were visualized using chord diagrams.
Reporting summary. Further information on research design and data analysis is available in the Nature Research Reporting Summary linked to this article.

Data availability
Sequence data have been deposited at the European Genome-phenome Archive (EGA), which is hosted by the EBI and the CRG, under accession number EGAS00001005241 (Dataset accession number: EGAD00001007586 https://ega-archive.org/studies/ EGAS00001005241). Further information about EGA can be found on https://egaarchive.org "The European Genome-phenome Archive of human data consented for biomedical research". All other data are available in the article and its Supplementary files or from the corresponding author upon reasonable request. Source data are provided with this paper.