A Pathogenic Th17/CD38⁺ Macrophage Feedback Loop Drives Inflammatory Arthritis through TNF-α

Rheumatoid arthritis (RA) and psoriatic arthritis (PsA) are common chronic inflammatory diseases, and both are characterized by pain and swelling in the joints and have significant systemic manifestations. RA is a debilitating and life-threatening inflammatory autoimmune disease with complex pathobiology that manifests primarily in joint tissues (reviewed in Ref. 1). Through single-cell analysis of primary RA patient samples and relevant mouse models, great efforts have recently focused on the molecular characterization of cellular subsets in RA (2–9). An abundance of evidence in humans and mice points toward the disease relevance of stromal cells, as well as various immune cell types, including T cells and macrophages (3, 5, 9–13). Established mouse models of RA (12, 14–16) recapitulate many aspects of human RA pathobiology and have been useful tools for dissecting the roles played by these cell types in disease. Of these, the GPI-induced arthritis mouse model offers a robust, reproducible, and fast progressing disease with a high rate of incidence. The short time course between the induction, peak, and resolution phases facilitates a comprehensive and in-depth profiling of all stages of disease involving multiple cell-type interactions in the inflammatory joints compared with the collagen-induced arthritis (CIA) model (14). Moreover, many cell types known to drive human RA, including T cells and macrophages, are required during the induction and effector phases of the GPI mouse model (12, 13), whereas lymphocytes play a secondary role in the K/BxN model (15). However, the level of similarity of these murine cell types to their human counterparts at the molecular level needs to be better defined if mouse models are to be further used to validate new therapeutic targets for RA. Furthermore, although much work has been done to classify cell types in RA, it is not fully understood which cell types communicate during disease, and key disease-driving cell–cell interaction networks in RA remain to be elucidated.

CD4⁺ T cells have long been implicated in RA (17) and in general can be further classified as either T regulatory cells (Tregs) or one of many Th subtypes, including Th1, Th2, and Th17 (reviewed in Ref. 18). These various Th subsets are characterized by divergent functions in host immunity and disease. In the synovium of RA patients, Th1 clones, which express TBET and IFN-γ, and Th17 clones, which express RORC (ortholog of murine Rorγt) and IL-17A, are expanded, while Th2 clones that express GATA3 and IL-4 are absent (19–22). Additional work has highlighted a multifaceted role for Th17 cells in driving RA pathobiology (23–25), including the recruitment of neutrophils (26). FOXP3⁺ Tregs, which act as potent suppressors of Th activity and inhibit autoimmunity, are also present in the synovium but may be functionally impaired in RA (27, 28). Despite their opposing functions, Tregs and Th17 cells are linked by a shared requirement for TGF-β, yet their development diverges through IL-6 activity, which enforces the Th17 cell fate (29–31). Th17 functionality is enhanced by IL-1β and IL-23 (32, 33), which are elevated in the synovium of RA patients along with IL-6 and TGF-β (34, 35). These factors work by promoting Rorγt expression, which in turn drives production of IL-17A (36). Although many cell types have the capacity to produce these Th17-promoting cytokines, additional work is needed to fully classify the molecular cell types responsible for enhancing the pathogenicity of Th17 cells in RA.

Macrophages are a heterogeneous group of phagocytic cells that also function to present Ag and produce a variety of cytokines when stimulated. Tissue macrophages exhibit mixed ontology and either arise during primitive hematopoiesis or differentiate from migratory monocytes that differentiate from bone marrow hematopoietic progenitors. In the synovium, macrophages that line the joint have been shown to arise during primitive hematopoiesis, are repopulated from locally renewing progenitors, and are protective from arthritis (3). However, interstitial monocyte-derived macrophages (MDMs) that expand during arthritis are believed to be drivers of disease. In support of this, depletion of macrophages during active disease in mice and humans using clodronate-containing liposomes (37, 38) has been shown to alleviate disease (13). Furthermore, these MDMs are highly plastic and respond to a wide variety of stimuli to adopt various polarization states, which historically have been oversimplified into an inflammatory/anti-inflammatory or M1/M2 paradigm (39). Recent studies have focused on classifying a wide spectrum of polarization states assumed by macrophages, including those induced by a complex milieu of stimuli similar to what is found in the arthritic synovium. More specifically, inflammatory HBEGF⁺ macrophages have been identified in the synovium of patients with RA and are induced by a combination of the cytokine TNF-α and stimuli produced by fibroblasts (5). Complementary cytometry by time of flight (CyTOF) profiling identified a strong correlation between synovial HBEGF⁺ macrophages and those expressing CD38 that are enriched in leukocyte-rich RA (9). CD38 is induced on macrophages in inflammatory conditions (40, 41) and has been demonstrated to play a role in macrophage responses to infection (42, 43). Although Cd38^−/− mice exhibit attenuated CIA through diminished dendritic cell functionality (44), dendritic cells are not abundant in the RA synovium (9), so it is possible that Cd38^−/− macrophages also contribute to the hindered disease phenotype. Overall, the exact role of CD38⁺ macrophages in RA is currently unknown.

TNF-α is a pleiotropic proinflammatory cytokine produced by many stromal and immune cell types, including macrophages and T cells (45). TNF-α is expressed as both a transmembrane and soluble form, the latter of which is released by proteolytic cleavage of the membrane form (46). TNF-α exerts various proinflammatory functions in its target cells by binding to its receptors, TNFR1 and TNFR2, which induces an intracellular signaling cascade to activate the transcription factor NF-κB to induce inflammatory cytokine production (47–49). Expression of TNF-α and its receptors is elevated in the synovium of patients with RA, and cells from RA patients exhibit a TNF-α activation signature (9, 50). Therapies targeting TNF-α have demonstrated efficacy in reducing joint swelling in both mice and humans (51–54), while combining anti–TNF-α with CD4-targeted therapy provides improved efficacy (55, 56). Although anti–TNF-α therapies have been shown to contribute to reduced levels of growth factors, chemokines, and inflammatory cytokines, such as IL-6 and IL-1β (57–59), the full impact on the pathobiology of RA is not fully appreciated.

In this study, we investigate the cellular and molecular interactions between macrophages and Th17 cells in the arthritic joint. We first molecularly classify the cell state changes in the synovium that accompany inflammatory arthritis in mice to identify an expansion of Th17 cells and macrophage subtypes that are similar to those found in RA patients. Next, we uncover macrophage-produced ligands that induce gene expression changes in Th17 cells and determine a requirement for macrophages in supporting the functionality of Th17 cells in the arthritic joint. Using an in vitro coculture system, we characterize a reciprocal interaction between Th17 cells and macrophages, where Th17-derived TNF-α induces a CD38⁺ macrophage polarization state that enhances the pathogenicity of activated Th17 cells. These findings translated to the discovery of an in vivo role for TNF-α in the induction of CD38⁺ synovial macrophages and production of Th17 cytokines. Overall, our results highlight a novel cellular feedback loop between Th17 cells and macrophages, while clarifying the role of TNF-α in the arthritic joint by demonstrating its function to induce CD38⁺ macrophages that further promote a pathogenic Th17 response responsible for driving inflammatory arthritis.

Male 7-wk-old DBA/1 mice were purchased from Envigo. Male 5-wk-old C57BL/6J, B6;129S-Tnftm1Gkl/J (60), and B6.Cg-Tg(TcraTcrb)425Cbn/J (61) mice were purchased from Jackson Laboratory. On arrival, mice were acclimated for 1 wk prior to the start of any procedures. All mice were confirmed to be pathogen free by the vendor and were cohoused at four mice per cage in 70°F rooms with a 12-h light/dark schedule under specific pathogen-free conditions at Eli Lilly and Company (San Diego, CA). All procedures were approved by the Eli Lilly and Company Institutional Animal Care and Use Committee. At the termination of any study, mice were euthanized by cervical dislocation under isoflurane anesthesia.

For GPI-induced arthritis experiments, mice were randomly allocated to groups for the subsequent induction of arthritis as previously described (12), and the investigators were blinded to this allocation during scoring. Briefly, 7 mg of human recombinant GPI protein was diluted with ice-cold DPBS and 1.75 ml CFA (Sigma) to a final volume of 3.5 ml (2 mg/ml) and emulsified using a Bead Ruptor Elite (Omni International). Under isoflurane anesthesia, DBA/1 mice hind flanks were shaved and injected s.c. with 100 μl of emulsion on each side of the base of the tail (total 400 μg GPI per mouse, “day 0”). Clinical arthritis scores were calculated as the cumulative scores from each forelimb and hindlimb based on the following criteria: 0, normal; 1, erythema and slight swelling of major joint; 2, moderate to severe swelling of major joint; 3, severe swelling of entire paw. For IL-17A neutralization experiments, mice were i.p. injected with 10 μg of anti-mouse IL-17A or control (mouse IgG1) Ab in a total volume of 100 μl DPBS twice weekly, beginning on day 8. For macrophage depletion experiments, mice were i.v. injected with 200 μl of Clodrosome (Encapsula NanoSciences) into the tail vein on days 11 and 13 of disease.

Hind paw joints were isolated by gently removing the surrounding skin, tendons, and muscle before cutting on the joint/pannus boundaries. Dissected tissues from both hind paw joints were combined and digested similarly as previously described (62) in RPMI (Fisher) supplemented with 2 mg/ml Liberase TL (Sigma) and 100 U/ml DNase I (Sigma) for 45 min at 37°C while shaking. The concentration of Liberase TL was empirically determined from optimization experiments performed using a titration of enzyme, with or without serum (data not shown). The resulting supernatant was filtered through a 100-μm cell strainer (Fisher), centrifuged, and washed with cold DPBS (Fisher). Live cell numbers were determined using Trypan Blue stain (Fisher) on a Cellometer Auto 2000 cell counter (Nexcelcom Bioscience).

For cytokine analysis, cells were analyzed using ELISpotPLUS (ALP) kit (Mabtech) according to the manufacturer’s recommendations. Briefly, plates were blocked using ELISpot Basal Medium (RPMI supplemented with 10% FBS, 55 μM 2-ME [Life Technologies], 2 mM l-glutamine [Life Technologies], and 100 U/ml penicillin/streptomycin) for 1 h at room temperature prior to beginning culture of 3 × 10⁴ synovial cells in 100 μl/well at 37°C for 2 d. This cell number was empirically determined from optimization experiments performed using a cell number titration (data not shown). Prior to the detection of spots as directed, cells were removed and plates were washed with DPBS. Plates were read using the iSpot Spectrum ELISpot Reader (Autoimmun Diagnostika), and the number of spots per well was calculated using AID ELISpot Software v7.0S (Autoimmun Diagnostika). Wells that did not receive any cells were processed in parallel and for all experiments had counts of zero or one spot per well.

Single-cell suspensions were incubated for 15 min with Fc Block (BD Biosciences) and LIVE/DEAD NearIR (1:1600 dilution, when used) in cold FACS Buffer (DPBS + 2% FBS) on ice, protected from light prior to staining with fluorochrome-conjugated Abs against surface Ags for 30 min on ice, protected from light. Cells were subsequently washed twice with FACS Buffer prior to downstream applications. If cells were analyzed immediately after staining, cells were resuspended in FACS Buffer with 7-aminoactinomycin D (1:100 dilution, when used) and analyzed using an LSRFortessa X-20 flow cytometer (BD Biosciences). For cell sorting, cells were diluted to 2 × 10⁷ cells/ml (500 μl minimum volume) and collected in RPMI (Fisher) supplemented with 50% FBS (Fisher) using a FACSAria III cell sorter (BD Biosciences).

For detection of cytoplasmic Ags, cells were fixed and permeabilized using the Intracellular Fixation & Permeabilization Buffer Set (Fisher) according to manufacturer’s recommendations. Briefly, cells were incubated in IC Fixation Buffer for 30 min on ice, protected from light prior to being washed twice in 1× Permeabilization Buffer. Cells were permeabilized for 30 min on ice, protected from light in 1× Permeabilization Buffer supplemented with 5% normal rat serum (STEMCELL Technologies), 1 μg/ml Fc block, and 1 μg/ml Rat IgG1 (BD Biosciences). Permeabilized cells were subsequently stained with fluorochrome-conjugated Abs for 45 min on ice, protected from light in 1× Permeabilization Buffer. Cells were washed three times with 1× Permeabilization Buffer and resuspended in FACS buffer for analysis.

For detection of nuclear Ags, cells were fixed and permeabilized using the Foxp3/Transcription Factor Staining Buffer Set (Fisher) according to manufacturer’s recommendations. Briefly, cells were incubated in 1× Fixation/Permeabilization Buffer for 45 min on ice, protected from light prior to being washed twice in 1× Permeabilization Buffer. Cells were incubated for 30 min on ice, protected from light in 1× Permeabilization Buffer supplemented with 5% normal rat serum (STEMCELL Technologies), 1 μg/ml Fc block, and 1 μg/ml Rat IgG1 (BD Biosciences) prior to staining with fluorochrome-conjugated Abs for 45 min on ice in 1× Permeabilization Buffer. Cells were washed three times with 1× Permeabilization Buffer and resuspended in FACS buffer for analysis.

For cell division analysis, cells were diluted to 10⁶ cells/ml in room temperature PBS and stained with 5 μM CellTrace Violet (CTV) reagent (Fisher) for 20 min at room temperature protected from light. The reaction was stopped by adding five times the original staining volume of RPMI supplemented with 20% FBS and incubated on ice for 5 min. Cells were washed twice with RPMI supplemented with 20% FBS and counted for downstream applications.

Splenocytes were isolated by disruption of spleens using frosted glass slides into cold Cell Separation Buffer (DPBS + 2% FBS + 1 mM EDTA) and filtered through a 100-μm cell strainer (Fisher). Naive CD4⁺ T cells were enriched similarly as previously described (63). Briefly, cells were incubated in ACK Buffer (Life Technologies) for 5 min at room temperature and washed twice with Cell Separation Buffer. Cells were diluted to 10⁸ cells/ml in Cell Separation Buffer containing a mixture of biotinylated Abs against CD8a, CD45R, CD25, CD11b, CD11c, TER-119, and CD49b (each at 1 μg/ml) and incubated on ice for 15 min. Cells were washed and diluted to 10⁸ cells/ml in Cell Separation Buffer containing 15 μl/ml Streptavidin Microbeads (Miltenyi) and incubated on ice for 15 min. Cells were washed and separated using a Large Depletion Column (Miltenyi) according to manufacturer’s recommendations. The negative fraction was diluted to 10⁸ cells/ml in Cell Separation Buffer containing 6 μl/ml CD62L Microbeads (Miltenyi) and incubated on ice for 15 min. Cells were washed and separated using a Large Selection Column (Miltenyi) according to manufacturer’s recommendations. The positive fraction was counted and used for subsequent cell culture.

To polarize naive CD4⁺ T cells, we cultured isolated cells similarly as previously described (63, 64). Briefly, cells were cultured at 0.4 × 10⁶ cells/ml in RPMI Base Medium (RPMI supplemented with 10% FBS, 55 μM 2-ME [Life Technologies], 1× nonessential amino acids [Life Technologies], 2 mM l-glutamine [Life Technologies], and 100 U/ml penicillin/streptomycin) with 0.5 μg/ml anti-mouse CD28 on CD3e-coated well-plates (precoated overnight at 4°C with 2 μg/ml anti-mouse CD3e diluted in DPBS). Cells were cultured for 4 d at 37°C with additional supplements: Th0: IL-2 (10 ng/ml), anti-mouse IL-4 (1 μg/ml), and anti-mouse IFN-γ (1 μg/ml); Th1: IL-2 (10 ng/ml), IL-12 (10 ng/ml), and anti-mouse IL-4 (1 μg/ml); Th1: IL-2 (10 ng/ml), IL-4 (10 ng/ml), and anti-mouse IFN-γ (1 μg/ml); Th17: TGF-β1 (1 ng/ml), IL-6 (20 ng/ml), anti-mouse IL-2 (1 μg/ml), anti-mouse IL-4 (1 μg/ml), and anti-mouse IFN-γ (1 μg/ml) with or without IL-1β (5 ng/ml) and IL-23 (5 ng/ml); inducible Treg (iTreg): TGF-β1 (1 ng/ml), IL-2 (10 ng/ml), anti-mouse IL-4 (1 μg/ml), and anti-mouse IFN-γ (1 μg/ml).

Prior to FACS analysis of cytoplasmic Ags, cells were stimulated in RPMI Base Medium supplemented with 1× Cell Stimulation Cocktail (plus protein transport inhibitors) (Fisher) for 5 h at 37°C.

Bone marrow cells from hindlimb bones (femurs, tibias, and iliac crests) were isolated by crushing with a mortar and pestle into cold Cell Separation Buffer and filtered through a 100-μm cell strainer. Bone marrow monocytes were enriched using the EasySep Mouse Monocyte Isolation Kit (STEMCELL Technologies) according to manufacturer’s recommendations. The negative fraction was filtered through a 100-μm cell strainer and cultured at 37°C on 10-cm Petri dishes (Fisher) at 4.5 × 10⁶ cells per 10 ml of IMDM Base Medium (IMDM [HyClone] supplemented with 10% FBS, 55 μM 2-ME, 1× nonessential amino acids, 2 mM l-glutamine, 1 mM sodium pyruvate [Fisher], and 100 U/ml penicillin/streptomycin) supplemented with M-CSF (10 ng/ml). After 3 d, the supernatant was removed and replaced with fresh IMDM Base Medium supplemented with M-CSF (10 ng/ml) and cultured for an additional 4 d at 37°C. These culture conditions were decided on by empirically testing various published cell culture media conditions that have been used to support macrophages and a dose titration of M-CSF (Supplemental Fig. 3F, 3G). The resulting differentiated macrophages were collected by replacing the supernatant with Accutase (Innovative Cell Technologies) prewarmed to 37°C and incubating at 37°C for 10 min.

Polarized Th cells and MDMs were cultured at 37°C for 3 d at a 1:4 ratio in IMDM Base Medium supplemented with M-CSF (10 ng/ml) and 0.5 μg/ml anti-mouse CD28 on CD3e-coated well-plates (precoated overnight at 4°C with 2 μg/ml anti-mouse CD3e diluted in DPBS). For neutralization studies, Abs were added at the beginning of the coculture period. Prior to cytokine release analysis, the supernatant was transferred, centrifuged, and carefully removed from the cell pellet before being stored at −80°C until analysis. The adherent cells were washed twice in DPBS that was prewarmed to 37°C and incubated at 37°C for 10 min with Accutase that was prewarmed to 37°C. The collected adherent fraction was combined with the cell pellet obtained from the nonadherent fraction and washed in DPBS prior to downstream analyses.

Prior to FACS analysis of T cell cytoplasmic Ags, cells were stimulated in RPMI Base Medium supplemented with 1× Cell Stimulation Cocktail (plus protein transport inhibitors) (Fisher) for 5 h at 37°C. Cells were then washed and collected as described earlier.

For relevant experiments, LPS (Invivogen), NAD⁺ (Sigma), OVA protein (Sigma), and recombinant cytokine protein (R&D Systems) were each reconstituted in sterile culture-grade water and further diluted in complete media prior to addition to cells. Neutralizing or blocking Abs were added to a suspension of cells in complete media.

Cell culture supernatants were collected as described earlier, thawed at room temperature, and centrifuged prior to analysis using the V-PLEX Mouse Cytokine 29-Plex Kit and U-PLEX TGF-β Combo Kit (Mesoscale Discovery) according to manufacturer’s recommendations. Briefly, 25 μl of supernatant was diluted 2-fold and added to each well for analysis. Cytokine standards were plated in duplicate. Plates were read using the Meso Sector S 600MM System (Mesoscale Discovery), and cytokine concentrations (pg/ml) were calculated using Discovery Workbench v4.0 (Mesoscale Discovery). Values falling outside of the limits of quantitation were excluded from the analysis.

Following digestion of hind paw joints and FACS-purification of live CD45⁺ singlets, cells were counted and diluted to 1 × 10⁶ cells/ml in DPBS, and 10,000 cells from each sample were loaded onto separate lanes of a Chromium Controller instrument (10× Genomics). The resulting Gel Bead-in Emulsions (GEMs) were processed according to the standard 10× Genomics single-cell 3′ v3.1 assay protocol. After library preparation and i7 indexing of individual samples, the final pooled library was composed of equimolar ratios of each individual sample library. Sequencing was performed on the Illumina NovaSeq-6000 platform (GeneWiz).

Following sequencing, demultiplexed fastq files were used for analysis. First, we used Seurat (version 3.1.4) to read a combined gene-barcode matrix of all samples. We removed the low-quality cells based on their unique feature counts and mitochondrial gene content. Low-quality cells often have very few genes or exhibit extensive mitochondrial contamination. For normalization, we employed a global-scaling method (the default setting of Seurat). The combined gene-barcode matrix was scaled by total UMI counts, multiplied by a scale factor, and then transformed to the log space. After checking the “Elbow plot” and comparing the distribution of p values for each principal component with a uniform distribution using the “JackStrawPlot” function provided by Seurat, we decided to include the first 20 principal components in further downstream analysis. We used the nonlinear dimensional reduction technique, Uniform Manifold Approximation and Projection (UMAP), to visualize and explore these datasets. Differential expression analysis was performed by the function “FindMarkers” in Seurat. To increase the speed of marker discovery, we prefiltered features that have fold-change <1.2 between the average expression of two groups.

To generate and classify cell clusters in single-cell RNA sequencing (scRNA-seq) data, we processed filtered data from the Naive-2 sample using ICGS2 (65) within the AltAnalyze software platform (66). The final naive dataset was generated by running cellHarmony (67) within the AltAnalyze software platform on the Naive-1 sample, using Naive-2 as the reference sample (centroid-based alignment, default parameters). The combined naive dataset was used as the reference for subsequent cellHarmony-based label projection (centroid-based alignment, default parameters) onto the three arthritis samples (days 6, 14, and 25) separately.

To perform cross-species macrophage comparisons, we performed human versus mouse gene lift over using HCOP (68) and dbOrtho (69). For human versus naive mouse macrophage correlation analysis, we calculated the differential gene expression across the four naive macrophage clusters. If the absolute value of differential expression of a given gene across the four mouse clusters was <0.1, the gene was excluded from analysis. For each remaining gene, a score of 1 was given to a mouse cluster if it had the maximum value and −1 if it had the minimum value. A correlation score was calculated for each human versus mouse cluster comparison by dividing the sum of scores of the mouse cluster by the total number of genes within the respective human cluster. For human-to-arthritic mouse macrophage correlation analysis, the differential gene expression for each mouse cluster versus its naive counterpart was calculated. If the absolute value of differential expression of a given gene across the four mouse clusters was <0.1, the gene was excluded from analysis. For each remaining gene, a score of 1 was given to a mouse cluster if the differential expression of a given mouse gene (disease versus naive) was in the same direction as the human differential expression (RA versus osteoarthritis), and −1 if in the opposite direction. A correlation score was calculated for each human versus mouse cluster comparison by dividing the sum of scores of the mouse cluster by the total number of genes within the respective human cluster.

NicheNet analysis was performed by starting from a Seurat object using R package nichenetr (version 1.0.0). NicheNet’s ligand–target prior model, ligand–receptor network, and weighted integrated networks were downloaded from Zenodo (https://zenodo.org). To perform ligand activity analysis, we defined a list of potentially active ligands and sets of affected and nonaffected background genes in the receiving cell group. Then we ranked ligands according to how well they predict whether a gene belongs to the gene set of interest compared with the background gene set. For target gene prediction, we inferred active ligand–target links by looking for genes that are affected in the receiver cell group and have a high potential to be regulated by the prioritized ligands.

To identify monocyte-to-macrophage differentiation trajectories, we performed ordering along pseudotime and identified differentially expressed genes across pseudotime using the Monocle3 package from R (Refs. 70–72 and L. McInnes, J. Healy, and J. Melville, manuscript posted on arXiv, DOI:10.21105/joss.00861). Trajectory inference and downstream analysis was done for each time point (Naive1+Naive2, day 6, day 14, and day 25) individually. For differential expression analysis, negative binomial distribution was used, only cells with minimum expression of 0.5 were taken into analysis, and any gene with q < 0.05 and high Moran’s I statistic was considered significant.

Published scRNA-seq data from the AMP-RA consortium were obtained from Kuo et al. (5) or downloaded from the Broad Single Cell portal (https://singlecell.broadinstitute.org/single_cell/study/SCP279/amp-phase- 1#study-visualize).

All in vitro and in vivo experiments were repeated two to five times with consistent results, except for scRNA-seq, which was performed one time using individual mice: two naive (day 0) mice and one mouse for each day 6, 14, and 25 time point. When indicated, n represents individual biological replicates. Data are shown as either mean or mean ± SEM, and the statistical significance was determined by paired or unpaired two-tailed t test as indicated (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).

scRNA-seq data are deposited as GEO Series GSE184609 and can be found at the following Web site: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE184609. All code was obtained from publicly available web sources.

To characterize the cellular composition of the murine synovium, we first FACS-profiled synovial cells at steady state (Supplemental Fig. 1A, 1B). After removing contaminating erythrocytes from analysis, we noted that most resident synoviocytes in naive mice (day 0) were CD45⁻ stromal cells, while the majority of CD45⁺ immune cells were CD11b⁺ F4/80⁺ macrophages (Supplemental Fig. 1B–D). We next investigated the dynamic changes to the synovium during inflammatory arthritis (12), which was accompanied by the emergence of a visible pannus with a relative size that correlated with joint swelling scores (Fig. 1A, Supplemental Fig. 1A). By FACS, synovial immune cells, consisting mostly of neutrophils and macrophages, were drastically increased, peaking at day 14 of disease (Fig. 1A–C, Supplemental Fig. 1B–D). Although neutrophil numbers declined sharply by day 25, macrophage numbers remained elevated (Fig. 1B), and monocyte, B cell, and CD4 T cell numbers continued to increase (Fig. 1C). We noted a complete absence of CD8a⁺ T cells in the synovium at all time points (Fig. 1C, Supplemental Fig. 1E), in discordance with human RA patient synovial biopsies (7, 9).

We next investigated the transcriptional changes in synoviocytes at various stages of inflammatory arthritis by FACS-purifying live nonerythroid synovial cells from naive and arthritic mice on days 6, 14, and 25 for single-cell RNA-seq (scRNA-seq) analysis (Supplemental Fig. 1B). Unsupervised clustering of 11,374 synoviocytes from naive mice using Iterative Clustering and Guide-gene Selection 2 (65) identified 27 unique cell clusters characterized by the differential expression of 1543 genes that were subsequently classified by Gene Ontology analysis (Fig. 1D, Supplemental Fig. 2A–D). Like those identified in patients with RA (6–9), lining fibroblasts were identified by Prg4 and Cd55 expression (Supplemental Fig. 2C) and were separated from a group of seven highly similar clusters of fibroblasts (Supplemental Fig. 2C, 2D). Furthermore, Pdgfrb⁺ and Mcam⁺ mural cells (8) were clustered together, while two similar groups of endothelial cells were combined (Supplemental Fig. 2C, 2D). Other stromal cell types that were identified include Ctsk⁺ osteoclasts (73), Ostn⁺ osteoblasts (74), Acan⁺ chondrocytes (75), and some contaminating Ttn⁺ skeletal muscle cells (76) (Supplemental Fig. 2C). In concordance with our FACS data (Supplemental Fig. 1B, 1D), macrophages comprise a majority of the immune cells analyzed by scRNA-seq and are segregated into four related yet distinct clusters (Fig. 1D, Supplemental Fig. 2C–E). Among these clusters are those that resemble previously described Cx3cr1⁺ lining macrophages, as well as MHC class II⁺ (MHC II⁺) and Retnla⁺ interstitial macrophage subsets (3). In addition, we identified a (to our knowledge) novel cluster of macrophages expressing an NF-κB activation signature marked by differential Cxcl2 and Tnf expression (Supplemental Fig. 2C). Three distinct clusters of neutrophils were identified that exhibit gene expression patterns similar to previously described populations, including neutrophil precursors, immature neutrophils, and synovial neutrophils (77, 78) (named preNeu, immNeu, and synNeu, respectively, to conform with an effort to unify nomenclature in the field) (79). Additional immune cell clusters identified at steady state include Cd19⁺ B cells, Cd3e⁺ T cells, and Cpa3⁺ mast cells (Supplemental Fig. 2C). Given these established reference states, we next used cellHarmony (67) for label projection onto synovial cells captured from arthritic mice (Supplemental Fig. 2F). Similar to our FACS results (Fig. 1B, 1C, Supplemental Fig. 1B–D), immune cells were greatly expanded during arthritis, while the macrophage and granulocyte clusters were disproportionally increased compared with the others (Fig. 1D, Supplemental Fig. 2F). Surprisingly, the B cell cluster grew 5.6-fold at day 6 compared with naive controls, which was not reflected in our FACS analysis (Fig. 1D, Supplemental Fig. 2F). Through these analyses, we have classified the dynamic changes to synovial transcriptional states during inflammatory arthritis.

Because CD4⁺ Th cells include multiple subtypes that are each characterized through divergent functions in immunity, we next sought to further define the identified synovial T cells. Cells of mixed Th1/Th17 phenotype that secrete IL-17A, IFN-γ, and TNF-α are expanded in the draining lymph nodes of arthritic mice (12, 80) and are critical for disease induction in both mice (12, 80) and humans (81–83). We noted an expansion of T cells expressing Rorc, Il17a, Il17f, Il22, and Tnf and a reduction of T cells expressing Tbx21 and Ifng at day 14 of disease (Fig. 1E, 1F), suggesting that the number of Th17 cells is increased in the synovial tissue at the peak of arthritis. To confirm the presence of Th17 cells in the arthritic joint, we performed ELISpot assays on synovial isolates. The number of cells producing IL-17A at the peak of disease was significantly increased, suggesting a direct role for Th17 cells at the site of inflammation (Fig. 1G). Consistent with this observation, the Ab-mediated neutralization of IL-17A initiated on day 8 blunted the onset of disease (Supplemental Fig. 2G, 2H), confirming a previous report that IL-17A is an important driver during the early stages of arthritis (80).

We next investigated the changes in the synovial macrophage subtypes during arthritis. In addition to the drastic increase in the total number of synovial macrophages at the peak of disease (Fig. 1B), we noted a disproportionate expansion of the MHC II⁺ subtype (Fig. 1D, 1H, Supplemental Fig. 2F). In contrast, the frequency of Lining and Retnla⁺ macrophages decreased sharply, while the NF-κB subtype expanded slightly (Fig. 1D, 1H, Supplemental Fig. 2F). To better understand the disease significance of these murine macrophage subtypes, we compared their gene expression profiles with macrophages isolated from the synovium of humans with RA. Using published marker genes for each of the four identified human macrophage clusters (5), we visualized their differential expression across the four murine macrophage clusters in naive mice (Fig. 1I). Compared with the other three clusters, the mouse Lining subtype expressed higher levels of human SC-M2 genes, which is a cluster believed to contain human Lining macrophages, while exhibiting an anticorrelation to the human inflammatory SC-M1 cluster (Fig. 1I, 1J). In contrast, the NF-κB cluster correlated well with human SC-M1 macrophages (Fig. 1I, 1J). The Retnla⁺ subtype was anticorrelated to the poorly defined human SC-M3 population while also exhibiting a weak positive correlation to the other three human clusters (Fig. 1I, 1J). Lastly, the MHC II⁺ macrophages were strongly correlated with the SC-M3 cluster and strongly anticorrelated to the SC-M2 cluster (Fig. 1I, 1J). Next, we investigated how the expressions of these genes change in arthritis. We first noted that the majority of the SC-M1, SC-M3, and SC-M4 human marker genes are elevated in macrophages from patients with RA versus osteoarthritis, while most SC-M2 genes are downregulated (Fig. 1K, arrows). We next determined the differential expression of these genes in each murine cluster at each time point during arthritis compared with day 0 controls. Strikingly, most SC-M3 genes were upregulated, while many SC-M2 genes were downregulated, across all four macrophage subtypes at all stages of disease in mice (Fig. 1K). In contrast, a portion of IFN/STAT-responsive SC-M4 genes (5) were specifically elevated at day 6 in all murine clusters (Fig. 1K), concordant with the noted increase in Ifng gene expression by T cells at this specific time point (Fig. 1F). Finally, many SC-M1 genes were specifically elevated at the peak of disease in all clusters, with the MHC II⁺ cluster displaying the strongest correlation to SC-M1 cells at this time point (Fig. 1K, 1L). Our results identify the presence of conserved synovial macrophage subtypes in murine arthritis and furthermore suggest that the expanded MHC II⁺ subtype at the peak of disease is highly similar to the human inflammatory SC-M1 macrophage subtype found in the synovium of humans with active RA.

Because we noted a considerable expansion of macrophages in the arthritic synovium at the peak of disease (Fig. 1H), we next sought to better characterize these molecular subtypes. We first performed differential gene expression analysis on the four macrophage clusters at each time point of disease. Surprisingly, the number of differentially expressed genes (adjusted p < 0.05) was modest across the dataset, with the majority of expression changes occurring at day 14 of disease in all four populations (Supplemental Fig. 2I). Moreover, the magnitude of differential gene expression was also modest, because only 90 genes per group on average were changed ≥2-fold (Supplemental Fig. 2I). These findings suggest that the core gene expression signatures that define each macrophage population are mostly stable during disease, and that the inherent nature of each macrophage subtype along with their relative frequency may serve as a better indicator of their role in arthritis pathobiology. Because MHC II⁺ macrophages were disproportionally expanded during arthritis (Fig. 1H), more broadly express Il1b (Supplemental Fig. 2J), and correlate strongly with the inflammatory human SC-M1 macrophage population (Fig. 1K, 1L), we focused our analysis on the differentially expressed genes (adjusted p < 0.05) for this subtype at each time point during arthritis. There was a striking upregulation of genes involved in the response to cytokine and exocytosis at all time points, while those genes involved in cell adhesion and migration were downregulated across all time points (Fig. 2A). In addition, marker genes of alternative macrophage lineages (Retnla, Cd209a, Irf4, and Aqp1) were downregulated during disease, whereas those of osteoclasts (Ctsk, Acp5, Jdp2) (74, 84, 85) were upregulated at day 14 (Fig. 2A), suggesting the presence of a switch in developmental trajectory for this macrophage subset. To further investigate this, we performed Monocle pseudotime analysis to order macrophages across an inferred developmental trajectory that begins with infiltrating monocytes. At steady state, the developmental trajectory traverses from monocytes to MHC II⁺ to NF-κB to Retnla⁺ and ending with lining macrophages (Supplemental Fig. 2K). These changes are accompanied by a loss of expression of monocytic marker genes, including Cd52 (86), and gain in expression of canonical macrophage marker genes, including C1qa (87) (Supplemental Fig. 2L). Although this trajectory is maintained at day 6 of arthritis, the inferred pseudotime on day 14 is truncated with cells arresting in the NF-κB macrophage state (Supplemental Fig. 2K), which is accompanied by prolonged Cd52 expression and the partial failure to launch C1qa (Supplemental Fig. 2L). In addition, at day 14 there is a gain in osteoclast marker genes including Acp5 and Ctsk that is absent across the developmental trajectory of macrophages at steady state and day 6 (Supplemental Fig. 2L). The developmental trajectory is only partially restored at day 25 (Supplemental Fig. 2K), with close to normal levels of C1qa and Cd52 expression but a persistence of osteoclast marker gene expression that began at day 14 (Supplemental Fig. 2L). These results suggest that the noted expansion of MHC II⁺ and NF-κB macrophage clusters during arthritis is partially due to a disruption in the normal developmental trajectory of synovial macrophages, which is accompanied by lineage priming of the osteoclast fate and is overall long lasting even as local inflammation resolves.

Given that a larger fraction of MHC II⁺ macrophages express Il1b than other macrophage clusters at steady state (Supplemental Fig. 2J), the upregulation of genes involved in exocytosis and production of IL-1 in MHC II⁺ macrophages during arthritis (Fig. 2A), and the coincident expansion of MHC II⁺ macrophages and Th17 cells at day 14 of disease (Fig. 1D–H), we hypothesized that synovial MHC II⁺ macrophages may directly promote Th17 cell functionality to drive arthritis. To investigate this, we performed NicheNet analysis (88) to identify ligand–receptor pairs enriched between MHC II⁺ macrophages and Th17 cells. We first subclustered the previously classified T cells (Fig. 1D) to identify those that exhibit a Th17 gene signature (positive for Cd3e, Cd4, Rorc, Il17a, Il17f) and found 7 cells in naive mice, 11 cells at day 6, 67 cells at day 14, and 84 cells at day 25 that match this phenotype. To discover potential disease drivers at the peak of arthritis, we focused our analysis on identifying interactions at day 14 between MHC II⁺ macrophages and Th17 cells. Among many receptor–ligand pairs, a strong enrichment for both cell-surface-bound and soluble factors known to promote Th17 polarization, including Icosl-Icos, Il1b-Il1r1, Il1b-Il1r2, Tgfb1-Tgfbr1, and Tgfb1-Tgfbr2 (Fig. 2B) (31, 33, 89), further suggests a functional link between these two cell types. Finally, to determine whether the expansion of MHC II⁺ macrophages is a critical driver of arthritis by promoting Th17 cell function, we depleted macrophages at the peak of disease using clodronate-containing liposomes and observed a significant decline in disease scores (Fig. 2C, 2D), confirming previous reports (13). Importantly, the number of synovial cells producing IL-17A or IFN-γ was significantly decreased after macrophage depletion (Fig. 2E), suggesting a direct link between MHC II⁺ macrophages and pathogenic Th17 cells in the arthritic synovium.

Given the observed link between MHC II⁺ macrophages and Th17 cytokine production in the arthritic synovium (Fig. 2A–E), we next sought to directly assess the ability of macrophages to influence Th cells in vitro. To do so, we polarized Th subtypes from wild-type naive CD4 T cells (Supplemental Fig. 3A–D) and differentiated macrophages from wild-type bone marrow monocytes in vitro (Supplemental Fig. 3A, 3E–G) for downstream coculture experiments (Supplemental Fig. 3A). We validated Th1, Th2, Th17, and iTreg cell identities by confirming the expression of canonical lineage transcription factors and cytokines (Supplemental Fig. 3C, 3D). After 3 d of coculture with MDMs, T cell viability was improved in all groups (Supplemental Fig. 3H, 3I), while the expression of Tbet in Th1 cells, Gata3 in Th2 cells, and Foxp3 in iTregs was significantly lower than in Th subsets cultured alone (Fig. 3A). In contrast, Rorγt expression was significantly increased in Th17 cells cocultured with MDMs compared with Th17 cultured alone, although appreciably lower than Th17 cells at the beginning of culture (Fig. 3A). Coculture with macrophages also increased the proportion of Th17 cells that express IL-17A, IL-17F, and IFN-γ (Fig. 3B), while also increasing secretion of IL-17A and IL-17F (Fig. 3C). The expression of TNF-α was significantly decreased after culture with macrophages (Fig. 3B). These effects were dependent on T cell activation, because the addition of macrophages did not similarly increase Rorγt, IL-17A, or IL-17F expression in Th17 cells in the absence of CD3/CD28 stimulation (Supplemental Fig. 3J, 3K). To determine whether this observed Th17–macrophage interaction occurs in an Ag-dependent manner, we generated Th17 cells that uniformly express α-chain and β-chain TCRs specific for OVA (61) for downstream culture experiments with macrophages. Similar to our previous results, coculture with MDMs enhanced the expression of Rorγt by Ag-specific Th17 cells in a dose-dependent manner (Fig. 4D). Next, to determine whether macrophages directly promote polarization of CD4 cells toward the Th17 phenotype, we tested freshly isolated naive CD4 cells in our coculture system. The addition of MDMs decreased T cell expression of Rorγt and did not induce IL-17A (Supplemental Fig. 3L, 3M). Activated naive CD4 cells expressed IL-17F, as has been documented in primary human Th0 cells (90), while the addition of macrophages reduced IL-17F and TNF-α (Supplemental Fig. 3M). Our results demonstrate that macrophages do not induce Th17 cells, but instead maintain the identity and functionality of activated Th17 cells after polarization.

Because the noted elevation in Rorγt levels in Th17 cells cultured with MDMs may be explained by an accumulation of this protein in nondividing cells, we next determined whether Th17 cell proliferation is altered in coculture with MDMs. To do so, we labeled Th17 cells with CTV dye (Fig. 3E) prior to culture for 3 d with or without MDMs and measured the extent of dye dilution as a surrogate for cell division. Interestingly, the extent of CTV retention was significantly lower in Th17 cultured with MDMs compared with Th17 cells alone (Fig. 3E). Moreover, a majority of cocultured Th17 cells divided more than three times, while most Th17 cells cultured alone underwent three or fewer cell divisions (Fig. 3F), and cocultured Th17 cells expressed more Rorγt irrespective of divisional history (Fig. 3G). Overall, our data suggest that macrophages uniquely support Th17 cell identity while promoting their functionality and expansion during periods of activation.

Because Th17 cells in coculture with MDMs produce polarizing cytokines such as IFN-γ and TNF-α (Fig. 3B, 3C), and the MDM-driven effects to T cells were dependent on T cell stimulation (Fig. 3A, Supplemental Fig. 3J, 3K), we hypothesized that macrophages may be reciprocally affected by Th17 cells. First, we screened macrophages for robust surface markers of polarization states after stimulation with an array of canonical Th cytokines or LPS (Supplemental Fig. 4A, 4B). Surprisingly, Ccr7 and CD209a are not expressed on the surface of murine MDMs in vitro, while CD80, PD-L1, and CD206 exhibit broad expression patterns (Supplemental Fig. 4B). Compared with resting macrophages cultured with M-CSF, those stimulated with Th1 cytokines or LPS uniquely expressed CD86, and those stimulated with Th2 cytokines uniquely expressed PD-L2 (Supplemental Fig. 4B). Conversely, CD38 was expressed at low levels on resting MDMs, was minorly induced by IFN-γ and IL-21, and was robustly induced by TNF-α, IL-10, and LPS (Supplemental Fig. 4B). Given the observed dynamic expression patterns of CD86, PD-L2, and CD38 by macrophages in response to various Th cytokines, we next investigated how coculture with Th cells affected the expression of these surface markers on macrophages. Similar to the single-stimuli results, Th1 cells robustly induced CD86 and minorly induced CD38 on MDMs, while Th2 cells robustly induced PD-L2 (Fig. 4A). Intriguingly, Th17 cells had the unique capacity to induce high levels of CD38 on MDMs (Fig. 4A). Apart from Th1-induced CD38, these changes were dependent on T cell stimulation (Supplemental Fig. 4C). Moreover, Th17-induced CD38 expression on macrophages also occurred through Ag-specific stimulation (Supplemental Fig. 4D). Macrophages in culture with Th17 cells also exhibited a functional change, as reflected by an increase in the secretion of the pro-Th17 cytokines, IL-6, TGF-β1, IL-1β, and IL-23 (Fig. 4B). Ab-mediated cytokine blockade in the coculture system identified a pivotal role for IL-1R and IL-6 in macrophage-induced Rorγt expression in Th17 cells (Fig. 4C). To ascertain whether Th17-induced CD38⁺ macrophages are the fraction of cells responsible for enhancing Th17 cell function in our previous experiments (Fig. 3A–C), we FACS sorted CD38⁺ or CD38⁻ MDMs after culture with Th17 cells for another round of coculture (Fig. 4D). While FACS-sorted CD38⁺ MDMs maintained high CD38 expression and increased the levels of Rorγt in Th17 cells, CD38⁻ MDMs gained CD38 expression and minorly induced Rorγt expression in Th17 cells (Fig. 4D, Supplemental Fig. 4E). Given that CD38 functions as an ectoenzyme to catalyze extracellular NAD⁺ (91) and administration of NAD⁺ protects against Th17-driven experimental autoimmune encephalomyelitis by reducing Th17 cells in the spinal cord, but not the function of splenic Th17 cells, we sought to determine the direct impact of exogenous NAD⁺ on Th17 cells in vitro. We noted a striking dose-dependent decrease in Th17 proliferation in response to NAD⁺ beginning at doses lower than those that caused a decrease in cell viability (Fig. 4E, Supplemental Fig. 4F, 4G). Furthermore, the addition of NAD⁺ did not further decrease Rorγt expression but instead decreased the fraction of highly proliferative Th17 cells (Supplemental Fig. 4H, 4I). Our findings identify (to our knowledge) a novel Th17-induced CD38⁺ macrophage activation state denoted by increased pro-Th17 cytokine production that enhances Th17 cell function and expansion.

To determine the cause of Th17-induced CD38 expression on MDMs, we revisited the cytokine measurements of culture supernatants to identify those that are changed in the context of Th17/MDM coculture. Of those that were changed in coculture compared with either Th17 or MDMs cultured alone, IL-21 modestly induced CD38, while IL-10 and TNF-α strongly induced CD38 on MDMs (Fig. 4F), recapitulating our previous findings (Supplemental Fig. 4B). IL-6 also had the capacity to induce CD38, while all other tested cytokines had no effect (Fig. 4F). Because our results do not exclude the possibility of synergy between multiple cytokines, we functionally assessed the role of these cytokines in our Th17/MDM coculture system. Although Ab-mediated neutralization of IL-21 modestly decreased the Th17-induced gain in CD38 expression on MDMs, neutralization of TNF-α alone or in combination with others almost completely blocked CD38 induction (Fig. 4G), suggesting a major role for TNF-α in driving CD38 surface expression on macrophages. In support of this, we identified a dose-dependent increase in CD38 levels on TNF-α–stimulated MDMs (Supplemental Fig. 4J), which was driven by Tnfr1, but not Tnfr2, signaling (Supplemental Fig. 4K). Because TNF-α is produced by both macrophages and Th17 cells, we next sought to determine the cellular source of TNF-α in the coculture system responsible for inducing CD38⁺ macrophages. Both wild-type and Tnf^−/− macrophages responded to TNF-α (Supplemental Fig. 4L) and gained equivalent levels of CD38 when cocultured with wild-type Th17 cells; however, CD38 levels remained at baseline when MDMs were cultured with Tnf^−/− Th17 cells (Fig. 4H), similar to what was observed when TNF-α was neutralized (Fig. 4G). Finally, we sought to determine whether TNF-α alone is sufficient to augment macrophage function to support Th17 cells. To do so, we stimulated MDMs with TNF-α prior to coculture with CD3/CD28-stimulated Th17 cells in the presence or absence of TNF-α neutralization (Fig. 4I). Although the addition of a TNF-α–neutralizing Ab did not affect Th17 cultured alone, it blocked the maintenance of Rorγt expression in Th17 cocultured with MDMs as seen previously (Fig. 4I). However, MDMs that were prestimulated with TNF-α were able to support higher levels of Rorγt in Th17 even in the presence of TNF-α neutralization (Fig. 4I). These results demonstrate that TNF-α alone induces a functionally distinct CD38⁺ macrophage state that promotes the identity and expansion of Th17 cells, independent of continuous TNF-α stimulation.

Our in vitro results revealed a cellular feedback loop in which activated Th17 cells produce TNF-α to induce CD38⁺ macrophages that further support Th17 function. To determine whether this also occurs in vivo, we visualized macrophage marker gene expression in our synovial scRNA-seq data. Similar to RA patients, all synovial macrophage subsets uniformly expressed Cd86, few macrophages expressed Pdcd1lg2 (encoding PD-L2), and some expressed Cd38 during arthritis (Fig. 5A, 5B). We next profiled the surface immunophenotype of synovial macrophages. We noted that CD86 is found on most naive macrophages and is slightly downregulated during arthritis, while PD-L2 expression was absent on macrophages during all time points apart from a small fraction on day 6 (Fig. 5C). In contrast, the CD38⁺ macrophage subset was significantly higher at all time points during arthritis compared with naive mice, peaking at day 14 (Fig. 5C). The noted induction of CD38 was not simply due to LPS-driven stimulation from CFA administration (Supplemental Fig. 4M). Because the Th17/CD38⁺ macrophage feedback loop was found to be largely driven by TNF-α in vitro (Fig. 4G, 4H), we next sought to determine whether these findings extended to our mouse model of arthritis by dosing mice with a TNF-α–neutralizing Ab beginning either at day 0 or day 8 and analyzing the synovium by FACS and ELISpot on day 14 (Fig. 5D). In contrast with mice that received an isotype control Ab, those that received anti–TNF-α Ab beginning on day 8 exhibited a significant reduction in disease scores, and those that received anti–TNF-α Ab at the time of GPI immunization showed a significant delay in disease with almost no joint swelling (Fig. 5D, 5E). FACS profiling of synovial macrophages at day 14 revealed a significant reduction in CD38 expression in mice treated with TNF-α–neutralizing Abs (Fig. 5F). Although those mice that were dosed beginning at day 8 exhibited a uniform reduction in CD38⁺ macrophages coincident with lower disease scores (Fig. 5F, Supplemental Fig. 4N), those mice that were dosed throughout the entire study were segregated into two groups characterized by either high disease scores and high CD38 expression (named “Non-responders”) or low disease scores and low CD38 expression (named “Responders”) (Fig. 5F, Supplemental Fig. 4O). Moreover, in mice dosed at day 8, there was a modest reduction in the number of synovial cells secreting IL-17A, while IFN-γ production was unaltered (Fig. 5G). Similarly, the frequency of cells secreting IL-17A or IFN-γ in mice from the Non-responder group was similar to the day 8 group (Fig. 5G). In contrast, the Responders had near-baseline numbers of synovial cells that secrete IL-17A or IFN-γ (Fig. 5G), coincident with disease scores and levels of CD38⁺ macrophages. Our results reveal (to our knowledge) a novel mechanism for TNF-α in arthritis by promoting CD38⁺ macrophage expansion to further enhance the functionality of Th17 cells to drive disease and demonstrate how TNF-α–targeted therapies are efficacious in the treatment of arthritis.

Although conventional disease-modifying antirheumatic drugs (cDMARDs) and biologics developed over the past few decades have progressively improved remission rates for RA patients, there remains a risk for relapse that is not fully understood due to an incomplete knowledge of disease pathobiology. To better understand the molecular composition of the joint at various stages of disease, a massive effort has recently identified the cell states present during arthritis in both humans and mice. However, there remains a need to identify the critical disease-driving interactions between key cell types in the arthritic synovium. This information can aid in the discovery of new therapeutics aimed at disrupting pathogenic cell–cell interactions to enhance long-term remission rates for patients.

In this article, we expand on the current knowledge base by temporally profiling the synovium of mice during GPI-induced arthritis at the single-cell level. Importantly, ours is the first study (to our knowledge), to examine the composition of the joint in this model that recapitulates the human disease, and we improve on the current standard of the field by performing a temporal analysis during arthritis. This approach helped to identify shifts in cellular and molecular patterns throughout the span of disease and enabled the measurement of specific changes in the frequency of Th17 cells and macrophage subtypes. Through additional efforts to understand the conservation of macrophage subtypes in mice, our cell-state-level comparisons identified shared molecular responses to disease between human and mouse macrophages and nominated the greatly expanded MHC II⁺ subtype as being the murine counterpart to the pathogenic HBEGF⁺ subtype found in human RA. Similar to HBEGF⁺ macrophages (5), the murine MHC II⁺ macrophages showed an enrichment for Il1b expression, while also upregulating Zbp1, which has been shown to promote the production of IL-1β (92). Our results confirm a recent report that TNF-α can induce IL-1β production by MDMs (93) while also newly revealing a requirement for macrophage-derived IL-1β in maintaining the functionality of Th17 cells that is responsible for driving inflammatory arthritis in vivo. To further investigate potential TNF-driven macrophage dynamics in this model, we used (to our knowledge) novel temporal pseudotime analysis across multiple time points of disease. We first established the normal trajectory of synovial macrophage differentiation, which helped to identify how this development is disrupted in disease. This analysis links MHC II⁺ macrophages directly with the expanded NF-κB subset while demonstrating that macrophage development was arrested at the NF-κB stage at the peak of arthritis. Interestingly, we found evidence of osteoclast lineage priming during this process, which has been reported to be induced in macrophages by TNF-α (94). These findings suggest that Th17 may contribute to joint destruction by skewing macrophages toward the osteoclast fate through TNF-α. Our study nominates this key cell–cell interaction as a potential target for treatment of RA and ultimately furthers our understanding of arthritis pathobiology.

Although anti–TNF-α therapy has represented a substantial leap forward in the treatment of RA and PsA, some patients remain resistant to therapy, suggesting that we do not fully appreciate the role that TNF-α plays in this disease. Although TNF-α is produced by both T cells and macrophages, it has been widely assumed that macrophages are the main source of TNF-α in the arthritic joint. Our experiments have illuminated a new mechanism for TNF-α in driving arthritis, as we identify a role for Th17-derived TNF-α to induce a macrophage polarization state that is marked by CD38 and that is characterized by increased production of pro-Th17 cytokines such as IL-1β and IL-6. This reveals a mechanism to explain how TNF-α can lead to elevated levels of these cytokines in the arthritic joint. Our findings also provide insights into why combination therapy against TNF-α and either IL-1β or IL-6 is more efficacious (95, 96), because this would ultimately block the functionality of both Th17 cells and CD38⁺ macrophages in this cellular feedback loop. CD38 itself is not only acting as an activation marker but is also important in promoting the expansion of Th17 cells by functioning to reduce NAD⁺ levels. Our results provide an alternative explanation for why Cd38^−/− mice are protected from arthritis (44) and implicate CD38⁺ macrophages in other Th17-driven diseases, such as experimental autoimmune encephalitis (commonly known as EAE), in which NAD⁺ is protective (97), as well as in patients with psoriasis, PsA, and ankylosing spondylitis who benefit from either anti–IL-17A or anti–TNF-α therapies (98–105). Importantly, a CD38⁺ macrophage state that correlates with the HBEGF⁺ macrophage population is also found in the inflamed synovium of RA patients, which itself requires TNF-α stimulation for its induction (5, 9). Through our efforts to determine individual factors that can induce CD38⁺ macrophages, we have provided the field with a wide profile of surface markers to identify cytokine-induced mouse macrophage polarization states, aiding in the current effort to expand the classification of macrophages beyond the conventional M1/M2 paradigm (106). These insights nominate CD38⁺ macrophages as key cellular targets for future therapeutics to battle RA or PsA and other TNF-α–driven autoimmune diseases, warranting further investigation of this TNF-α–induced macrophage polarization state, as well as comparing the functionality of macrophages stimulated with other CD38-inducing cytokines.

Although the GPI model is well accepted in the field as a relevant model of human RA, it is limited in that it does not fully recapitulate all aspects of the human disease. Instead, we leverage this model to define the molecular subtypes present in the inflamed synovium to subsequently draw comparisons to their human counterparts in RA patients. Successful identification of the conserved disease-driving cellular populations in the mouse model provides a useful in vivo system for target validation and for approximating the potential impact of investigational therapeutic molecules on the human disease. In support of this, the conserved CD38⁺ macrophage population is impacted by anti–TNF-α pharmacological intervention in mice similarly to the human HBEGF⁺ macrophage population in vitro (5). However, if a certain cell type found in the mouse model lacks a disease-relevant counterpart in human RA patients, less focus should be placed on this cell population. Such is the case for Th17 cells that play a critical role in the GPI model but less so in RA patients, resulting in a disconnect in the strength of response to IL-17A neutralization between species (80, 107, 108). This extends to neutrophils, which are downstream of IL-17A and are similarly more important for disease in mice than in humans (109, 110). Alternatively, because anti–IL-17A therapy is highly efficacious in treating human PsA, this implies the possible mechanistic connection between the GPI-induced arthritis model and inflamed joints in human PsA patients. Through these analyses, molecular populations significant to human disease may be identified, while others can be triaged. Our study suggests that understanding murine arthritis at a cell-state level (tied back to human) may offer an efficient approach for in vivo validation of the next generation of successful therapeutics for RA and PsA.

In summary, our findings identify a (to our knowledge) novel cellular feedback loop that exists between macrophages and Th17 cells but no other Th subtypes, where Th17 cells can enhance their own pathogenicity through TNF-α–induced CD38⁺ macrophages. Given that both macrophages and Th17 cells are well-established drivers of multiple autoimmune diseases for which anti–TNF-α therapies are used, our findings may have far-reaching implications that could impact our understanding and future treatment strategies to combat an array of autoimmune indications.

We thank David Kuo and Laura Donlin for kindly sharing published source data from their paper (5), John Calley for technical assistance with CellRanger software, and Nathan Salomonis for technical assistance with AltAnalyze software.

D.E.M., Z.S., A.S., C.T., J.S.C., C.L., K.K., W.C., and S.N. are employed by Eli Lilly and Company.