The flow cytometry analyses and functional assays performed as part of this study included a total of 28 oligo JIA patients, defined by International League of Associations for Rheumatology (ILAR) criteria ( Table 1 ; Table S1 ) (29). Fifteen patients were ANA⁺, defined as a detectable ANA titer at any time during the disease course. Five pediatric and 11 adult controls also contributed peripheral blood samples.

Synovial fluid (SF) and peripheral blood (PB) samples from oligo JIA patients and controls were analyzed with flow cytometry to evaluate T cell populations known to provide help to B cells, notably T peripheral helper (Tph, CD4⁺PD-1^hiCXCR5^-) and T follicular helper (Tfh, CD4⁺PD-1⁺CXCR5⁺) cells ( Figure 1 ; S1 ). While the frequency of Tph cells rarely exceeded 2% of CD4⁺ T cells in the PB of oligo JIA and control subjects, we found a dramatic enrichment of this population in the SF of oligo JIA patients ( Figures 1A, B ). In addition, Tph cells were significantly enriched in the SF of ANA⁺ oligo JIA (45.3 ± 4.6% of CD4⁺ T cells, mean ± standard error) compared to the SF of ANA^- oligo JIA patients (33.6 ± 2.6%, p-value = 0.03). Regardless of ANA status, there was no significant difference in the levels of Tfh cells between SF and PB, which remained low (<5%) in the vast majority of subjects.

To compare to Tph cells, we evaluated CD4⁺ T cells with intermediate PD-1 expression. CD4⁺PD-1^intCXCR5^- T cells are observed at increased frequencies in patients with autoimmunity, but unlike Tph cells, they are not associated with measures of disease activity or end-organ involvement in autoimmune diseases such as systemic lupus erythematosus (SLE) (36). The frequency of CD4⁺PD-1^intCXCR5^- cells was increased in the SF of oligo JIA patients compared to the PB of patients and controls, but no significant difference was detected between ANA⁺ (42.4 ± 4.2%) and ANA^- (48.3 ± 3.1%) SF samples (p-value = 0.75) ( Figures 1A, B ).

Taken together, these results suggest that Tph cells, and not Tfh cells, are the primary B cell-helper T cell subset in the joints of oligo JIA patients. Further, Tph cells are especially enriched in the SF of oligo JIA patients who are ANA⁺.

PD-1 expression in T cells from an inflammatory environment may indicate exhaustion or the capacity to promote B cell maturation. To determine whether Tph cells in oligo JIA SF present characteristics typical of B cell-helper T cells, we assessed the expression of markers that promote the B cell chemotactic response (CXCL13) and T-B cell interactions in GCs (ICOS) (3, 37, 38). Following stimulation with CD3/CD28-coated beads, the SF of oligo JIA patients showed a significant enrichment in CD4⁺PD-1^hiCXCL13⁺ T cells and in CD4⁺PD-1^hiICOS⁺ T cells, which were absent in PB ( Figures 1C, D ; S2 ). The frequency of CD4⁺PD-1^hiCXCL13⁺ T cells was significantly enriched in ANA⁺ SF (15.5 ± 3.1%) compared to ANA^- SF (7.4 ± 1.6%, p-value = 0.02). When gating on different sub-populations of CD4⁺ T cells, we confirmed that CXCL13 secretion was exclusive to SF Tph cells ( Figure 1E ). In contrast, ICOS was found in SF Tph but also in SF CD4⁺PD-1^intCXCR5^- and PB CD4⁺PD-1^intCXCR5^- T cells. ANA positivity in oligo JIA is most common in children with early-onset disease (26, 27). Interestingly, CD4⁺PD-1^hiCXCL13⁺ T cells were also significantly increased in the SF of oligo JIA patients with disease onset at or before 6 years of age (13.8 ± 2.3%) compared to those who were older at the initial presentation of disease (5.8 ± 2.1%, p-value = 0.04) ( Figure 2 ). There was a trend towards lower frequencies of CXCL13⁺ Tph cells in the small number of evaluated patients with extended oligo JIA (n=4); however, this did not reach statistical significance (p-value = 0.17).

We then assessed cytokine production in CD4⁺ T cell subsets from oligo JIA patients and controls, focusing on IL-21 due to its role in promoting B cell survival and maturation (39, 40). IL-21⁺ cells were largely enriched in oligo JIA SF Tph cells (CD4⁺PD-1^hiCXCR5^-) compared to other CD4⁺ T cell populations (PD-1^intCXCR5^-) in the joint or blood ( Figures 1F, G ). IL-21⁺ Tph cells were further enriched in ANA⁺ SF (68.1 ± 4.2%) compared to ANA^- SF (42.8 ± 5.6%, p-value = 0.0003). IFNγ⁺IL-21⁺ dual expressing cells were also most enriched in SF Tph cells. By contrast, the fraction of IFNγ⁺ cells did not differ between SF PD-1^intCXCR5^- and SF Tph cells, suggesting that both populations contribute to IFNγ secretion in oligo JIA SF.

Finally, we assessed the expression of HLA-DR, a marker of activation associated with a Tph cell phenotype in RA (9). As previously described, we found that Tph cells in oligo JIA SF co-express HLA-DR to a greater extent than CD4⁺PD-1^intCXCR5^- T cells; however, we found no significant difference in the frequency of HLA-DR⁺ Tph cells between ANA⁺ and ANA^- patients ( Figures 2D, E ) (41).

Altogether, these studies show that CD4⁺PD-1^hiCXCR5^- (Tph) cells in oligo JIA SF express effector molecules that provide help to B cells, notably CXCL13 and IL-21. Further, both CXCL13 and IL-21 were most frequently expressed by SF Tph in ANA⁺ compared to ANA^- patients. This effector protein expression profile supports the hypothesis that PD-1 expression by Tph cells in oligo JIA does not simply reflect activation or exhaustion in the inflammatory environment of the arthritic joint.

We next leveraged flow cytometry to evaluate the phenotype of B cells in oligo JIA SF ( Figures 3A-G ; S3 ). Switched memory B cells (CD27⁺IgD^-, Bmem) were significantly increased among CD19⁺ B cells in the SF of oligo JIA patients compared to the PB of controls ( Figures 3A, B ). Plasmablasts were not enriched in oligo JIA SF, regardless of ANA status ( Figures 3C, D ). However, CD19⁺ B cells in oligo JIA SF displayed unusual characteristics: transitional (CD38^hiCD24⁺) and classical (CD38⁺CD21⁺) B cells were depleted in oligo JIA SF compared to control PB, while CD21^lo B cells, a potentially autoreactive subclass of B cells that has been identified in other autoimmune and immune dysregulatory conditions, were enriched ( Figures 3E-G ) (42, 43).

To assess the functionality of SF Tph cells, we studied their capacity to promote the differentiation of Bmem cells into antibody-producing plasmablasts in vitro ( Figures 3H-J , S3 ). Different populations of CD4⁺ T cells were sorted from the SF of oligo JIA patients and co-cultured with Bmem cells (CD27⁺IgD^-) from the PB of a third-party control. After 5 days of culture in the presence of the superantigen staphylococcal enterotoxin B (SEB), the frequency of CD27⁺CD38^hi plasmablasts was assessed. In co-cultures with double negative CD4⁺PD-1^-CXCR5^- T cells or with Tfh cells from oligo JIA SF, the fraction of plasmablasts generally remained low (<5%). In contrast, co-culture of Bmem with SF Tph cells induced the production of plasmablasts with 18.2 ± 0.3% of CD27⁺CD38^hi detected on average ( Figures 3H, I ). Accordingly, the concentration of IgG trended higher in the supernatant of co-cultures with SF Tph cells compared to co-cultures with other SF CD4⁺ T cell populations, although this effect did not achieve statistical significance ( Figure 3J ).

Taken together, these results indicate that SF Tph cells are a functional population with the capacity to promote the differentiation of antibody-producing B cells.

In lymphoid organs, the GC reaction is regulated in part by T follicular regulatory (Tfr) cells, a subset of CD4⁺ T cells that share features of regulatory T (Treg) and Tfh cells (17–21, 44). We sought to determine if a similar population of Treg cells with Tph features might exist in peripheral tissues such as the joints of children with oligo JIA. Flow cytometry studies demonstrated that PD-1^hiCXCR5^- cells were observed within the SF Treg population (CD4⁺CD127^loCD25⁺FOXP3⁺ cells, Figures 4A, B ; S1 ). PD-1^hiCXCR5^- Tregs, hereafter referred to as Tph-like Tregs, were enriched in the SF of ANA⁺ and ANA^- patients compared to the PB of patients and controls. Overall, these Tph-like Tregs accounted for 5.2 ± 0.5% of SF CD4⁺ T cells and were found at similar frequencies in ANA⁺ and ANA^- samples ( Figure 4C ). Similar to our observations in total CD4⁺ T cells, PD-1^intCXCR5^- cells were also enriched in oligo JIA SF Tregs compared to PB Tregs, while PD-1⁺CXCR5⁺ cells were not.

To confirm that PD-1^hiCXCR5^- Tregs express markers associated with T-B cell interactions, we evaluated CXCL13 and ICOS expression in Tregs ( Figures 4D, E , S2 ). PD-1^hiCXCL13⁺ cells were not detected in PB Tregs of oligo JIA or control subjects but were modestly yet significantly enriched in ANA⁺ SF Tregs (1.2 ± 0.3%) and ANA^- SF Tregs (0.9 ± 0.2%). Similarly, frequencies of PD-1^hiICOS⁺ cells remained low in PB Tregs (<2%), and were significantly elevated in ANA⁺ SF Tregs (12.6 ± 2.4%) and ANA^- SF Tregs (10.9 ± 2.1%).

In total, these studies suggest the presence of a Treg population with Tph features that reside in inflammatory environments outside of lymphoid organs, including a subset that express the B cell help factors CXCL13 and ICOS.

To complement our flow cytometry studies and validate key markers of Tph and Tph-like Tregs, we studied the transcriptomic profile of SF Teff (CD4⁺CD25^-) and SF Treg (CD4⁺CD127^loCD25⁺) cells at the single-cell level. We combined and re-analyzed 2 previously published datasets, enabling us to assess 17,436 SF Teffs and 12,156 SF Tregs from 4 ANA⁺ and 4 ANA^- oligo JIA patients ( Figures 5 - 7 , S4 ) (34, 45).

The combined analysis of 8 oligo JIA patients revealed similar SF Teff clusters as identified in Julé et al. where only 2 patients were studied while also providing a much greater number of transcripts to evaluate ( Figures 5A, B ; Supplemental File S1 ) (34). Notably, 2 clusters (4 and 5) expressed robust levels of the B cell help factor CXCL13 alongside activation markers typical of B cell-helper T cells, including PDCD1 and CD40LG ( Supplemental File S1 ). In addition, the transcription factor BATF, which is important for the differentiation of B cell-helper T cells, was increased in these clusters ( Figures 5E, F ) (46). Thus, clusters 4 and 5 were delineated as Tph cells. Cluster 5 was further characterized by elevated levels of HLA-DR transcripts ( Figure 5B ). Both Tph clusters encompassed cells from the SF of ANA⁺ and ANA^- patients ( Figures S4A, B ); however, Tph cells originating from ANA⁺ SF samples showed substantially higher scores for the Tph transcriptomic signature as described in RA ( Figures 5C, D ) (47). Interestingly, the transcriptomic signature of Tph cells in oligo JIA SF differed from the transcriptomic signature of Tfh cells ( Figure S4C ) (44).

To further delineate the transcriptomic signature of Tph cells in oligo JIA, we performed a differential gene expression analysis (DEA) of cells in Tph clusters versus cells in other Teff clusters ( Figures 5E, F ; Supplemental File S1 ). This DEA highlighted 482 significantly upregulated genes in oligo JIA SF Tph and confirmed the increased expression of multiple T cell activation markers in Tph cells, including HLA-DR genes, PDCD1, CD40LG and KLRB1 (encoding CD161). CXCL13 appeared as the most upregulated gene in oligo JIA SF Tph cells. Other B cell help factors with increased expression in SF Tph cells included BATF and SH2D1A/SAP. In contrast, ICOS was not identified as a differentially expressed gene. Transcription factors typically expressed in Tfh or Tph cells, namely BCL6 and PRDM1, were also not robustly expressed in oligo JIA SF Tph cells. However, our DEA highlighted other transcription factors that may play a role in shaping the transcriptomic landscape of Tph cells, such as BHLHE40. A putative central role for this basic helix-loop-helix transcription factor in oligo JIA SF Tph was supported by the simultaneous upregulation of BHLHE40-AS1 and ID2 in these cells, both of which might act as regulators of BHLHE40. In addition, ALOX5AP, which plays a role in leukotriene biosynthesis, was highly upregulated in Tph cells compared to other Teff populations in the joint.

We then ran a gene set enrichment analysis (GSEA) on the genes identified as significantly upregulated in oligo JIA SF Tph ( Figure S4D ). This GSEA showed that the transcriptomic signature of Tph cells in oligo JIA is enriched in genes associated with IFNγ signaling, antigen presentation and TCR signaling, suggestive of cytokine-producing, antigen-activated T cells. In the absence of stimulation, cytokine expression is barely detectable in 10X studies; therefore, we could not quantitatively assess IL-21 production by SF Tph at the transcriptomic level, although the few IL21 expressing cells were primarily observed in Tph clusters ( Figure 5F ).

During the combined analysis of SF Tregs, a distinct cluster of Tph-like Tregs (Cluster 1) emerged from this re-analysis with a larger number of cells and patients ( Figure 6 ; Supplemental File S1 ) (34, 45). The Tph transcriptomic signature derived from Zhang et al. was expressed at the highest level in cells falling within this cluster ( Figures 6B, C ) (47). DEA of Tph-like Tregs versus other Tregs confirmed that these cells in oligo JIA SF share key markers with Tph and other CD4⁺ T cells that help B cells, including PDCD1, CD40LG, CXCL13, IL21, SH2D1A/SAP and MAF ( Figures 6D, E ). KLRB1 (encoding CD161), ID2, BHLHE40, and ALOX5AP also emerged as key markers of SF Tph-like Tregs. In contrast, Tph-like Tregs did not express elevated levels of ICOS or BATF. As has been demonstrated in Tfr cells in secondary lymphoid organs, Tph-like Tregs expressed lower levels of FOXP3 and IL2RA than other SF Tregs but did not present transcriptomic features of induced Tregs ( Figure S4E ) (48). Furthermore, when directly contrasted to SF Teffs, Tph-like Tregs maintained high expression of key Treg associated genes. Among the top significant differentially expressed genes (DEGs), appeared FOXP3 (average log2 fold change between SF Tph-like Tregs and total SF Teff LFC = 1.28), IL2RA (LFC = 1.31), CTLA4 (LFC = 1.18) TNFRSF18 (encoding GITR; LFC = 1.85), TNFRSF4 (encoding OX40; LFC = 1.62), and LAG3 (LFC = 1.50) ( Supplemental File S1 ). These single-cell studies demonstrate a transcriptional program in a subset of Tregs consistent with B cell-helper T cells.

The combined SF Teff single-cell dataset included 11,604 cells with available T cell receptor (TCR) sequencing data for both the alpha and beta chains. Key repertoire metrics for each patient, including the number of total and unique clones, are detailed in Table 2 . Compared to other SF Teff clusters, Tph clusters were significantly enriched in expanded clones (defined as clones with at least 3 copies across the dataset), regardless of ANA status ( Figures 7A, B ). SF Tph had a distinct clonotypic profile: while the 2 Tph clusters shared up to 30% of clonotypes with each other; they shared only few with other Teff clusters ( Figure 7C ). A small fraction (2-6%) of SF Tph clonotypes were also found in the blood ( Figure 7D ). Other SF Teff cells showed similar levels of overlap with PB Teff, suggesting that Tph and other Teff cells in oligo JIA SF recirculate in the blood to a similar extent ( Figure 7D ). In summary, Tph cells in the arthritic joint are expanded and demonstrate a TCR repertoire that is distinct from other SF effector T cells.

Patients with oligo JIA defined by the ILAR criteria were recruited from the Rheumatology Clinic BCH (29). Medical record review was conducted by a board-certified pediatric rheumatologist to confirm eligibility and collect relevant clinical information. Patients with a positive readout for ANA at any timepoint and titer throughout their disease course were categorized as ANA⁺. Synovial fluid and/PB samples obtained within 6 months of symptom onset as reported by the patient were considered as new-onset samples. Medical records were reviewed closer to the time of publication to determine the disease course (persistent or extended oligo JIA). Table 1 and Supplemental Table S1 detail the patients and samples included in this study.

PB and SF samples from oligo JIA patients were collected at the same time as a clinically indicated blood draw or joint procedure. Children seen in the Rheumatology Clinic at BCH for non-inflammatory causes of joint pain were recruited as pediatric controls. Healthy adults also volunteered to provide PB samples. Peripheral blood was collected in tubes with ethylenediaminetetraacetic (EDTA). In addition, discard samples were obtained from heathy adults who donated blood to the BCH Blood Conation Center as leukopaks. In all cases, Ficoll density gradient centrifugation (GE Healthcare, now Cytiva) was used to isolate mononuclear cells (MCs). Resulting PBMCs and SFMCs were used fresh or cryopreserved in liquid nitrogen before thaw, as indicated for each application.

For flow cytometry analyses of CD4⁺ T cells, samples were thawed and washed in PBS (HyClone) and resuspended into complete media (RPMI (HyClone) with 10% FBS (Genesee Scientific), penicillin 100 IU/mL (Corning Cellgro), streptomycin 100 μL/mL (Corning Cellgro), nonessential amino acids (Corning Cellgro), sodium pyruvate 1mM (Corning Cellgro), and HEPES 10mM (ThermoFisher Scientific)) for overnight incubation. For Tph quantification, CXCL13 and ICOS staining, cells were incubated with CD3/CD28 Dynabeads (Gibco) for 18 to 20 hours, in the presence of brefeldin A (GolgiPlug, BD Biosciences) in the last 4 hours. For HLA-DR and cytokines staining, cells were rested overnight and stimulated with a PMA, ionomycin and brefeldin A cocktail (BD Biosciences) for 4 hours. For all samples and panels, some cells were also left unstimulated as a control condition. Cells were harvested in PBS, incubated in the presence of Live/Dead Fixable Yellow dye (Invitrogen) for 20 min at 4°C, and incubated with surface antibody markers in FACS buffer (PBS with 1.5% FBS (Genesee Scientific) and 2.5 mM EDTA (Invitrogen)) for 40 min at 4°C. Cells were then fixed and permeabilized per the FOXP3 staining kit protocol (eBioscience), before staining for intracellular markers for 30 min at 4°C.

For flow cytometry analyses of CD19⁺ B cells, total SFMCs or PBMCs were thawed as detailed above and directly stained for surface markers. Alternately, the CD4-depleted fraction of samples processed for functional assays was used. Antibodies used in flow cytometry studies are listed in Table S2 . All flow cytometry data were acquired immediately on an LSRFortessa (BD Biosciences) and analyzed in FlowJo.

Cryopreserved SFMCs were thawed in plain RPMI (HyClone) and processed immediately for CD4⁺ T cell enrichment through negative selection (Miltenyi Biotec). Tfh (CD4⁺CD45RA^-PD-1⁺CXCR5⁺), Tph (CD4⁺CD45RA^-PD-1^hiCXCR5^-) and double-negative (CD4⁺CD45RA^-PD-1^-CXCR5^-) cells were FACS-sorted on a FACS Aria II. PBMCs from third-party donors were freshly isolated from leukopaks, processed for CD19⁺ cell enrichment through positive selection (Miltenyi Biotec), and then memory B cells (CD3^-CD19⁺IgD^-CD27⁺) cells were isolated by FACS.

Sorted T and B cells were co-cultured in complete media (contents as described above) at a ratio of 1 T cell to 10 B cells and in the presence of staphylococcal enterotoxin B (SEB, Fisher Scientific LLC) at a concentration of 1 μg/mL. After 5 days of incubation, culture supernatants were collected for assessment of total IgG production with an ELISA kit (ThermoFisher Scientific) and cells were stained for assessment of viability (7-AAD viability dye) and plasmablast differentiation (CD19⁺CD27⁺CD38^hi). To be considered in the final analysis, co-culture readouts had to meet the following criteria, suggestive of a successful co-culture with sufficient cell viability: greater than 10% lymphocytes and visible lymphocyte proliferation in FSC/SSC gate, greater than 40% live B cells in CD19/Live-Dead gate, and at least one control population meeting the previous criteria for the concerned experiment.

Single-cell data from 2 previously published studies (34, 45) were integrated and re-analyzed in R version 4.1.0 using Seurat version 4.0.3 (67, 68). Both studies initially used 10X Genomics to analyze FACS-sorted Treg (CD3⁺CD4⁺CD25⁺CD127^lo) and Teff (CD3⁺CD4⁺CD25^-) cells from JIA SF. Maschmeyer et al. further sorted those populations based on expression of CD45RO, which is expressed on the vast majority of CD4⁺ T cells in oligo JIA SF (34). One patient with disease onset at 17.5 years did not meet the ILAR criteria for oligo JIA and was therefore excluded from this re-analysis.

Teff and Treg transcriptomic data were re-analyzed independently. For each T cell population, relevant filtered matrices of raw features, as generated with CellRanger in the initial studies, were imported in R using Seurat’s Read10X() function. Transcriptomic data from all runs were combined into a single dataset for each T cell population, and re-examined for quality control. Droplets with <2000 unique molecular identifier (UMIs), <1000 genes, a log10 of genes per UMI <0.8, or a mitochondrial to nuclear gene ratio >0.1 were excluded from the Teff dataset; and those with <2400 UMIs, <1200 genes, <0.8 log10 genes per UMI, or >0.1 mitochondrial to nuclear gene ratio were excluded from the Treg dataset. Further, cells that did not express CD3E, representing <2% of each dataset, were excluded. Transcriptomic data were then normalized for each 10X run using Seurat SCTransform() while regressing out the effect of the number of UMIs, number of genes and ratio of mitochondrial to nuclear transcripts per cell. Finally, normalized data were integrated across runs using the standard Seurat integration workflow and exclusion of TRAV, TRAJ, TRBV and TRBJ genes from the integration anchors to avoid clustering by clonotype (67, 68).

The first 40 principal components (PCs) were calculated using Seurat’s RunPCA() function and used to define the uniform manifold approximation and projection (UMAP) coordinates. Louvain clustering analysis was run on integrated data using the first 30 (Treg dataset) to 40 PCs (Teff) and following the standard Seurat clustering analysis workflow (67, 68). We opted for a resolution of 0.4 (Teff) to 0.6 (Treg) to delineate the clusters. Cluster identities were determined by running Seurat’s FindAllMarkers() using the default Wilcoxon rank sum test to identify key markers of each cluster. A contaminant cluster of Treg cells (expressing FOXP3 and clustering apart on the UMAP) was identified in the Teff data and excluded from downstream analyses.

Cells were scored for the Tph, Tfh and/or induced Treg (iTreg) signatures using Seurat’s AddModuleScore() (44, 47, 69). Differential gene expression analyses (DEA) of Tph and Tph-like Tregs were run using Seurat’s FindMarkers() with the receiving operating curve (ROC) classifier method and assessing genes expressed in >5% of the cells in either of the contrasted groups. For DEA, the clusters of mitotic cells were excluded from the contrast group. Ranked log2 fold change (LFC) values were used as input for gene set enrichment analyses (GSEA) using clusterProfiler v4.0.2 and querying the Reactome database (70, 71).

To contrast the Tph-like Tregs (identified in the Treg dataset) with Teff cells, the quality-controlled Treg and Teff datasets were eventually combined, normalized and re-integrated across patients as described above. A DEA of Tph-like Tregs versus Teff (excluding mitotic cells) was run using Seurat’s FindMarkers() function with the Wilcoxon rank sum test. Gene lists resulting from all marker identification and differential gene expression analyses are provided in Supplemental File S1 .

For TCR repertoire data derived from 10X studies of SF samples, only the subset of cells included in transcriptomic analyses was considered to enable matching of each clonotype to its cluster identity. For PB samples, the complete TCR repertoire data as compiled by CellRanger vdj was considered. In both cases, clonotypes were defined by their paired CDR3 alpha and beta amino acid sequences and TRAV, TRAJ, TRBV and TRBJ genes. The percentage of overlap between 2 repertoires from different clusters or compartments (PB and SF) was obtained by averaging the fractions of unique overlapping clonotypes among all unique clonotypes in each of the 2 repertoires.

One-way ANOVA with Turkey correction or two-way ANOVA with Šídák correction for multiple testing were used to assess for differences across >2 groups, and two-tailed Student’s t-test were used to assess for differences between exactly 2 groups. A P-value of <0.05 was considered significant. All statistical analyses were run in GraphPad Prism v9.2.0.

This study was conducted in accordance with IRB protocols 07-09-0375 and P00005723. Written informed consent (and assent when appropriate) was obtained for the study participants.