Tumor-Infiltrating Regulatory T-cell Accumulation in the Tumor Microenvironment Is Mediated by IL33/ST2 Signaling

Regulatory T cells (Treg) are a prominent immunosuppressive cell type that associates with tumor progression (1, 2). Increased Tregs in the tumor microenvironment (TME) impair antitumor immunity (3). Treg ablation in DEREG mice delays tumor growth and, in some cases, induces complete tumor clearance (2, 4). However, Treg depletion is often accompanied by life-threatening autoimmunity (5). Therefore, distinguishing and selectively targeting tumor-infiltrating Tregs is required for effective and safe cancer immunotherapy.

Three potential sources for Tregs in the TME are trafficking from the periphery, in situ conversion from conventional T cells (Tconv), and local proliferation (6). Certain chemokines facilitate Treg recruitment into the TME (7–9). The chemokine receptor, CCR8, is a unique characteristic of tumor-infiltrating Tregs in colorectal cancers, non–small cell lung cancers (NSCLC), and breast cancers (10, 11). Tumor-infiltrating Tregs are generated in TME through the TGFβ-dependent conversion of Tconvs (12). Tregs proliferate in response to IL2 and TGFβ in the TME (5), yet the contribution of these signals has not yet been measured experimentally in the TME.

IL33 is normally released after tissue stress or damage to function as an alarmin. During inflammation and cancer, IL33 exerts multiple regulatory functions on Th2 cells, ILC2s, and Tregs, all which express the unique receptor of IL33, ST2 (13). ST2 is regulated via a positive feedback loop in response to IL33 (14). IL33 facilitates the expansion of adipose Tregs or colonic Tregs (15, 16). However, the role of IL33/ST2 signaling in tumor-infiltrating Tregs is not clear, although IL33 in the TME is protumorigenic (17).

Here, we showed that Tregs preferentially accumulated in the TME of various tumor models. We found that tumor-infiltrating Tregs had a distinct transcriptional feature involved in proliferation and were clonally expanded. ST2 was significantly upregulated in tumor-infiltrating Tregs and IL33 promoted the selective proliferation of ST2-expressing tumor-infiltrating Tregs in vitro and in vivo. Conversely, genetic deficiency of ST2 in Tregs led to their limited accumulation in the TME and in vivo blockade of ST2 delayed tumor growth. These data advance our understanding of tumor-infiltrating Tregs, suggesting an opportunity for therapeutic targeting of Tregs in the TME.

Five- to eight-week-old female C57BL/6, C57BL/6-CD45.1, BALB/c, C57BL/6-Foxp3-GFP, and Rag2^−/− mice were purchased from the Jackson Laboratory. ST2-knockout (KO) mice were a generous gift from Dr. Andrew N. Mckenzie (Medical Research Council Laboratory of Molecular Biology, Cambridge, United Kingdom). All mice were maintained in a specific pathogen-free facility in accordance with Institutional Animal Care and Use Committee (IACUC) guidelines at Yonsei University (Seoul, Korea). Animal experiments were approved by the IACUC of Yonsei University (approval number: 201507–356–02, oversight number: 201710–410–06).

The murine lung adenocarcinoma cell line TC-1, Lewis lung carcinoma cell line LLC1, and colon carcinoma cell line CT26 were purchased from the ATCC in 2010 and cultured two passages before injection. Tumor cells were tested yearly for Mycoplasma contamination using the e-Myco Mycoplasma PCR Detection Kit (iNtRON Biotechnology) and reauthentication was not performed in the past years. C57BL/6 mice were injected subcutaneously on the right flank or intravenously into the tail vein with 5 × 10⁵ TC-1 or LLC1 tumor cells diluted in 500 μL of PBS. BALB/c mice were injected subcutaneously on the right flank or intravenously into the tail vein with 5 × 10⁵ CT26 tumor cells. At 3 weeks following tumor cell inoculation, tumor-bearing mice and age-matched normal mice were sacrificed. The spleen, lung, and tumor were harvested and processed into single cells.

To study whether ST2 deficiency can affect tumor-infiltrating Treg accumulation, the lung tumors were harvested from TC-1 tumor–bearing wild-type (WT) or ST2-deficient mice on day 21 following tumor cell inoculation. The number of tumor nodules on left upper lobe of the lung were counted using a hemocytometer.

For treatment with neutralizing antibody, TC-1–tumor bearing WT mice were established as described above and each mouse was treated intravenously with 100 μg of anti-ST2 antibody (DIH4, BioLegend) or isotype control antibody (BioLegend) per dose every 2 days, which started on day 5, and sacrificed on day 15.

At 3 weeks following tumor inoculation, the spleen and tumor from tumor-bearing mice and the spleen and lung from normal mice were harvested and processed into single cells. Tumors and lung tissues were chopped to 1-mm³ pieces and digested in RPMI medium supplemented with 10% FBS (Gibco), 1% penicillin–streptomycin, 1% Collagenase IV (Worthington Biochemical Corp.), and 0.01% DNase (Sigma Aldrich) for 20 minutes at 37°C. The fragments were dissociated and passed through a 70-μm strainer. Lymphocytes were collected from the interphase of the 44% and 67% Percoll (Sigma Aldrich) layers after density-gradient centrifugation.

Cells were stained with the following fluorochrome-conjugated antibodies in FACS buffer (PBS containing 2% FBS): murine CD4 (RM4–5), PD-1 (29F.1A12), CD103 (2E7), Ly5.1 (A20), CCR4 (2G12), and CCR8 (SA214G2; BioLegend); CD25 (PC61.5), CTLA-4 (UC10–4B9), GATA3 (TWAJ), Foxp3 (FJK-16s), Ly5.2 (104), and ST2 (RMST2–33; eBiosciences); GITR (DTA-1), BrdU (3D4), Ki67 (B56), Ly-6C (AL-21), KLRG1 (2F1; BD Biosciences); TIM-3 (215008), and CCR2 (475031; R&D Systems); CCR5 (HM-CCR5; Thermo Fisher Scientific). For ST2 detection, cells were incubated with anti-ST2 biotin for 1 hour at 37°C, washed twice with FACS buffer, and then incubated with streptavidin-APC (Thermo Fisher Scientific, S868) during surface staining. The intracellular staining was conducted using Foxp3 Fixation/Permeabilization solution (Thermo Fisher Scientific, 00–5523–00). For staining cells with antibodies against chemokine receptors, cells were incubated with anti-CCR2, anti-CCR4, and anti-CCR8 antibody for 30 minutes at room temperature and anti-CCR5 antibody for 1 hour at 37°C and then washed twice with FACS buffer before surface staining. To remove dead cell population, the Live/Dead Fixable Dead Cell Stain Kit (Thermo Fisher Scientific, L10119) was used during the staining procedures. Flow cytometry was performed using the FACSCanto II (BD Biosciences) and CytoFLEX LX (Beckman Coulter Life Sciences). Data were analyzed using FlowJo software (Treestar).

A total of 5 × 10⁵ TC-1 cells were intravenously injected into the tail vein of mice. The spleen and tumor from tumor-bearing mice and spleen from normal mice were harvested at 3 weeks following tumor inoculation. CD4⁺ T cells were enriched using a CD4⁺ T Cell Isolation Kit by negative selection (Miltenyi Biotec, 130–104–454) followed by manufacturer's protocol. Pure CD4⁺GFP⁺ Treg and CD4⁺GFP⁻ Tconv populations were obtained through sorting using a FACSAria II (BD Biosciences). The purity of isolated Tregs was greater than 95%. Total RNA was extracted using a QIAGEN RNeasy Plus Micro Kit (Qiagen, 74034) according to the manufacturer's methods. Libraries were prepared using an Illumina TruSeq Stranded mRNA Library Prep Kit (Illumina, 20020594) and sequenced using an Illumina NextSeq500. For the RNA sequencing (RNA-seq) data analyses, we aligned short sequence reads to the mouse reference genome (UCSC mm10; ref. 18) using STAR-2.5.2a (19) followed by quantification using the FeatureCounts (20). We selected candidate genes based on the consensus of four different combinations of read alignment and quantification/differential analysis methods: (i) Tophat v2.0.13 and Cuffdiff v2.2.0, (ii) STAR and EdgeR, (iii) STAR and DESeq2, and (iv) Kallisto 0.42.5 and DESeq2.

Gene-set enrichment analysis (GSEA) was performed for using gene sets representing immunosuppression, proliferation, chemotaxis, and apoptosis pathways. A total of 325, 853, 122, and 465 genes (Supplementary Table S1) were compiled from 159, 164, 39, and 113 gene ontology (GO) biological process terms (Supplementary Table S2) relevant to immunosuppression, proliferation, chemotaxis, and apoptosis categories, respectively. Only genes annotated by GO, evidence of experimental or literature curated data, were used.

CD4⁺ T cells were enriched from pooled lung tissues of normal mice (n = 10) and pooled TC-1 tumors of tumor-bearing mice (n = 7) as described previously. Single-cell sequencing libraries were prepared by using Chromium 10x Genomics platform with a V(D)J Enrichment Kit and 5′ Library & Gel Bead Kit (10x Genomics, PN-1000005, PN-1000006) according to the manufacturer's instructions. The single-cell suspensions of live CD4⁺ T cells sorted by FACSAria II were diluted in nuclease-free water and then combined with Single Cell 5′ Gel Beads, a master mix, and Partitioning Oil on Chromium Chip A, found in the aforementioned kit. RNA transcripts were uniquely barcoded and reverse-transcribed within nanoliter-scale droplets containing uniquely barcoded beads. 10x barcoded full-length cDNA were then pooled and enriched via PCR. For V(D)J Enriched Library, the enriched cDNA pool was amplified using primers specific to T-cell receptor (TCR). For 5′ Gene Expression (GEX) library, the cDNA pool went through an end repair process, A-tailing, adaptor ligation, and sample index PCR. Both libraries for single-cell RNA-seq (scRNA-seq) and single-cell TCR sequencing (scTCR-seq) were sequenced using HiSeqX platform (Illumina).

Cell Ranger analysis pipeline (v3.1.0) was applied to the raw sequencing data to obtain a gene count and a clonotype information for each cell. Using Seurat version 3.1 pipeline, we filtered out cells based on proportion of mitochondrial genes and feature count. Next, we selected cells that have both scRNA-seq and scTCR-seq data. Raw expression data was log normalized with scale factor of 10,000. Downstream analysis was conducted on the basis of top 2,000 highly variable genes only. To avoid batch effect, we integrated sequencing data from normal and tumor samples using default data integration pipeline of Seurat v3.1. Dimension reduction and clustering analysis were performed by UMAP and Louvain clustering (dimension: 50 and resolution: 0.21) algorithms available from Seurat package.

Compositional changes between cellular group of single clonal TCR (singlet) and that of multiple clonal TCR (multiplet) were assessed on the basis of the Pearson residual: |{d_{ij}}\! = {\frac{{{n_{ij}}\ - \ {m_{ij}}\ }}{{\sqrt {{m_{ij}}\ } }}$|⁠, where i and j represent a group (singlet or multiplet) and cell subsets, respectively, and n and m are the observed cell number and the expected cell count, respectively. The significance of the compositional changes was evaluated on the basis of χ² test. The results of compositional analysis were summarized as a mosaic plot.

For BrdU labeling, TC-1 tumor–bearing mice and normal mice were injected intraperitoneally (i.p.) with 2 mg BrdU (Sigma) for 6 consecutive days prior to sacrifice. BrdU incorporation was detected by intracellular staining using the BrdU Flow Kit (BD Biosciences, 559619) according to the manufacturer's methods.

To detect apoptosis, cells were stained with Annexin V and 7-AAD using an Annexin V Apoptosis Detection Kit (BD Biosciences, 556547) according to the manufacturer's methods.

Mouse coding genes from the Consensus Coding sequence (CCDS) database (21) were filtered for cell membrane and cell surface genes annotated in Swiss-Prot (22) and for external side of plasma membrane genes annotated in GO (23) with GO evidence code of IDA, IPI, IMP, IGI, IEP, TAS, and EXP. Next, 2,613 genes passed the filtration step were further filtered for tumor-bearing mice-TM Treg–specific differentially expressed genes (DEG) compared with five other cellular and tissue contexts, resulting in five sets of DEGs: tumor-bearing mice-TM Tregs versus tumor-bearing mice-TM Tconvs [fold change (FC) > 4], tumor-bearing mice-SP Tregs (FC > 2), tumor-bearing mice-SP Tconvs (FC>2), normal mice-SP Tregs (FC>2), or normal mice-SP Tconvs (FC>2) with P < 0.05. For highly conservative candidate selection, we used DEGs by consensus of four different analysis pipelines as described above for each of the five comparisons. The intersection of the five DEG sets resulted in 37 genes. We excluded the genes that had low expression, assuming an expression value of lower than 10% of the CD4 (a marker for CD4⁺ T cells) expression level might be insufficient for the therapeutic effect. The expression profile of the 17 genes that passed the low expression filtration across other immune cell types was examined using the Gene Skyline data browser of Immgen Project (24). The filtering of genes that were highly expressed in other immune cell types resulted in only three candidate genes. Foxp3 was used as reference gene because it is a specific marker for Tregs, having an expression level of approximately 700 in CD4⁺ memory T cells. Thus, we applied a strict threshold expression level of 500.

Naïve lung and tumor on days 8 and 12 following inoculation were harvested. Sections (4.5 μm) of tissues were cut from formalin-fixed paraffin-embedded (FFPE) blocks and dried for at least 1 hour in 60°C dry incubator. The sections were dewaxed using xylene, then followed by multiplex immunofluorescence staining with a Leica Bond Rx Automated Stainer (Leica Biosystems).

Briefly, the slides were baked for 30 minutes and dewaxed with Leica Bond Dewax solution (Leica Biosystems, AR9222), followed by antigen retrieval with Bond Epitope Retrieval 2 (Leica Biosystems, AR9640) in a pH 9.0 solution for 30 minutes. Then, the slides were incubated with first primary antibody for CD45 (clone 30-F11, Thermo Fisher Scientific, 14–0451–81) for 30 minutes. Visualization of CD45 was accomplished using Opal 570 TSA Plus for 10 minutes, after which the slide was treated with Bond Epitope Retrieval 1 (Leica Biosystems, AR9961) for 20 minutes to remove bound antibodies before the next step in the sequence. Then, the sections were stained with CD31 (polyclonal, Abcam, ab124432, Opal 520), IL33 (Clone 396118, R&D Systems, MAB3626, Opal 690), and PDPN (polyclonal, Abcam, ab109059, Opal 620), sequentially. Nuclei were subsequently visualized with DAPI (Akoya Biosciences, FP1490), and the section was coverslipped using HIGHDEF IHC fluoromount (Enzo Life Sciences, ADI-950–260–0025). The slides were scanned using the Vectra Polaris Automated Quantitative Pathology Imaging System (Akoya Biosciences) and images were analyzed using the inform 2.4 software and TIBCO Spotfire (Akoya Biosciences).

At each time point, lungs were collected, weighed, and homogenized in 1 mL of DMEM supplemented with 1% FBS. After homogenate being centrifuged, the supernatant was analyzed for IL33 production using Mouse IL33 DuoSet ELISA (R&D Systems, DY3626–05) according to the manufacturer's protocol.

For in vitro proliferation assays, single-cell suspensions from the tumor of TC-1 tumor–bearing mice at 21 days following tumor inoculation were enriched for CD4⁺ cells using the MagniSort Mouse CD4 T Cell Enrichment Kit (Thermo Fisher Scientific, 8804–6821). A total of 10⁵ CD4⁺ T cells labeled with CellTrace-Violet dye (CTV, Thermo Fisher Scientific) were resuspended in expansion buffer (RPMI, 10% FBS, 1% penicillin/streptomycin). CTV-labeled CD4⁺ T cells were left unstimulated or stimulated with 0.5 μg/mL soluble anti-CD3/CD28 (BD Biosciences) supplemented with 10 ng/mL rhIL33 (PeproTech) and cultured for 3 days.

For blocking of ST2 on Tregs, CTV-labeled CD4⁺ T cells were incubated for 1 hour with 10 μg/mL of anti-ST2 antibody (R&D Systems) or isotype control antibody (BioXCell) at 4°C, washed twice, and stimulated with 0.5 μg/mL soluble anti-CD3/CD28 supplemented with 10 ng/mL rhIL33 for 3 days.

To expand Tregs in vivo, TC-1 tumor–bearing mice either WT mice or ST2-deficient mice were intraperitoneally injected with 2 μg rmIL33 (BioLegend) diluted in 200 μL of PBS on day 4 or days 4, 7, and 10 following tumor inoculation and sacrificed on day 13. The number of tumor nodules were counted as described above.

For in vitro suppression assay, TC-1 tumor–bearing mice were intraperitoneally treated with either PBS or rmIL33 in the manner as described above and sacrificed on day 13 following tumor inoculation. We isolated the Tregs from PBS-treated tumor-bearing mice and IL33-treated tumor-bearing mice using CD4⁺CD25⁺ Regulatory T Cell Isolation Kit (Miltenyi Biotec, 130–091–041). We also harvested the spleen from naïve mouse, isolated CD8⁺ T cells with CD8a⁺ T Cell Isolation Kit (Miltenyi Biotec, 130–104–075), and labeled them with CTV dye. Then, PBS-treated Tregs (PBS Treg) or IL33-treated Tregs (IL33 Treg) were cocultured with CD8⁺ T cells as ratio of 1:1 or 1:3 supplemented with Dynabeads Mouse T-Activator CD3/CD28 for T-Cell Expansion and Activation (Gibco, 11452D) for 72 hours.

To study the mechanism for the accumulation of tumor-infiltrating Tregs, CD4⁺CD25⁺ Tregs were isolated from the spleen of either WT or ST2-KO mice using a CD4⁺CD25⁺ Regulatory T Cell Isolation Kit (Miltenyi Biotec). An equal number of WT and ST2-KO Tregs (1.5 × 10⁵ cells per group) were cotransferred or seperately transferred intravenously into the tail vein of Rag2^−/− mice along with 1.2 × 10⁶ WT CD8⁺ T cells (500 μL of PBS). Two weeks after the transfer, the Rag2^−/− mice were injected intravenously into the tail vein with 5 × 10⁵ TC-1 tumor cells. At day 21 following tumor cell inoculation, tumors were isolated form each mouse for counting of tumor nodules on left upper lobe of the lung and the two different populations of donor Tregs in the tissues were analyzed.

Experimental data were analyzed using GraphPad Prism software (GraphPad). Significant difference between two groups was analyzed by two-tailed unpaired Student t test. For comparison between more than two groups, one-way ANOVA was used with Bonferroni post hoc test. Results are presented as mean ± SEM. The difference between two experimental groups with a P value less than 0.05 was considered statistically significant. The Kolmogorov–Smirnov two-tailed test was performed to analyze the statistical significance in empirical cumulative distribution functions (ECDF). The statistical significance of difference between two Pearson correlation coefficient (PCC) distributions was tested by the Wilcoxon rank-sum test.

All original bulk RNA-seq data and scRNA-seq data including TCR-seq data were deposited in the NCBI's Gene Expression Omnibus (GEO) database (GEO: GSE120280, GSE152022).

To examine the population of Tregs at the tumor site, we took advantage of various tumor models in which mice were subcutaneously or intravenously injected with murine TC-1 lung adenocarcinoma, LLC1 Lewis lung carcinoma, or CT26 colon carcinoma cells. Under homeostatic conditions, Tregs represent 5% to 10% of CD4⁺ T cells in the lung and 10% to 15% in the spleen. Regardless of tumor model or route of tumor cell injection, we found the frequencies of Tregs in tumor-bearing mice significantly increased, up to a maximum of 40%, in the TME compared with those in the spleen (Fig. 1A and B; Supplementary Fig. S1A). Both frequency and absolute number of Tregs in the tumor lung increased compared with those in the normal lung (Fig. 1B).

Analysis of the expression of functional molecules by Tregs in TC-1 lung tumor model, CTLA-4, PD-1, TIM-3, and GITR, demonstrated that tumor-infiltrating Tregs were phenotypically activated, suggesting their enhanced suppressive function (Fig. 1C and D). Increased PD-1 expression in tumor-infiltrating Tregs was also observed in other tumor models (Supplementary Fig. S1B and S1C). We found the monocyte/macrophage and endothelial cell differentiation antigen, Ly-6C, was downregulated in tumor-infiltrating Tregs (Fig. 1E). As Ly-6C downregulation is under the control of TCR signaling (25), enrichment of Ly-6C^lo Tregs may be due to sustained exposure to tumor or self-antigens present in the TME. These results suggested the TME instigated the accumulation of tumor-infiltrating Tregs with distinguishable phenotype.

To identify the molecular mechanisms leading to tumor-infiltrating Treg accumulation, transcriptional profiles of Tregs and Tconvs derived from either tumor of tumor-bearing mice or spleen of normal mice and tumor-bearing mice were compared using RNA-seq analysis (Fig. 2A). Comparison of our mouse RNA-seq data to that of human tumor-infiltrating Tregs from patients with NSCLC (10) indicated the transcriptional profile of tumor-infiltrating Tregs correlates each other (Supplementary Fig. S2). Activated Treg genes for which we had already determined protein expression (CTLA-4 (Ctla4), GITR (Tnfrsf18), TIM-3 (Havcr2), and PD-1 (Pdcd1)) were upregulated in tumor-infiltrating Tregs. Tumor-infiltrating Tregs also highly expressed genes associated with immunosuppressive function including immunosuppressive soluble factors Il10 and Gzmb, ectonucleotidase Entpd1 (CD39), and coinhibitory receptors Ctla4, Pdcd1, Havcr2, and Tigit. In addition, the genes for several cytokine and chemokine receptors including Il1rl1, Ccr2, Ccr4, Ccr5, and Ccr8 were upregulated (Fig. 2B). Protein expressions of CCR2, CCR5, and CCR8, but not CCR4, were highly upregulated by tumor-infiltrating Tregs in our TC-1 lung cancer model (Supplementary Fig. S3).

GSEA (26) showed genes involved in immunosuppression, proliferation, chemotaxis, and apoptosis were significantly upregulated in tumor-infiltrating Tregs compared to tumor-infiltrating Tconvs and splenic Tregs of tumor-bearing mice (Fig. 2C). In addition, the Kolmogorov–Smirnov test revealed a significant enrichment of genes associated with proliferation in tumor-infiltrating T cells compared with splenic T cells of tumor-bearing mice, especially in Tregs rather than Tconvs (Fig. 2D). In contrast, the enrichment was not observed in splenic T cells from between tumor-bearing mice and normal mice (Supplementary Fig. S4). Collectively, these data suggested that both enhanced migration and proliferation activity led to the increased population of Tregs in the TME.

For direct comparison of T cells from lung tumor and normal lung, we conducted scRNA-seq and scTCR-seq for CD4⁺ T cells from TC-1 lung tumor of tumor-bearing mice and those from normal lung of normal mice (Fig. 3A). Single-cell transcriptome data from normal and tumor tissue samples were integrated with batch correction by “multi dataset integration” function of Seurat v.3.1 (Supplementary Fig. S5A; ref. 27). Unsupervised clustering of the cells based on transcriptome similarity revealed seven clusters, all which expressed CD4⁺ T-cell markers for T cells (Supplementary Fig. S5B), representing subsets of CD4⁺ T cells (Supplementary Fig. S6A). Figure 3B shows the transcriptome landscape of CD4⁺ T cells organized by tissue origin (left; tumor lung or normal lung), by cluster (middle), and by clonality (right; multiplet or singlet). Tregs were annotated by the expression of Foxp3 and Il2ra and the rest of the cells were annotated as Tconvs (Supplementary Fig. S6B). We further examined the profiles of signature genes to characterize the Tconv clusters (Supplementary Fig. S6B and S6C). As a result, cluster 1 showed higher expression of exhaustion markers (Ctla4, Pdcd1, and Tigit), whereas the rest of the Tconvs showed higher expression of naive signature (Sell and Ccr7). Therefore, we annotated cluster 1 as exhausted Tconv (exhTconv) and the rest of Tconvs clusters as naïve Tconv (nTconv). For Tregs, we annotated cluster 2 as activated Treg (aTreg) and cluster 5 as resting Treg (rTreg) based on the distribution of clonality (Fig. 3B) and the expression of activated Treg signature (Sell^lo, Ccr7^lo, Itgae^hi, Klrg1^hi; Supplementary Fig. S6B and S6C). As expected, more Tregs were observed in lung tumor than in normal lung. In addition, most of tumor-infiltrating Tregs showed aTreg phenotype, whereas normal lung Tregs seemed to be composed of more rTregs than aTregs (Fig. 3B). Treg ratio was much higher in lung tumor than normal lung (Fig. 3C), which was consistent with the results from flow cytometric analysis (Fig. 1B). In addition, most of the multiple-clonotype Tregs were found in tumor-infiltrating Tregs but not in normal lung Tregs (Fig. 3B). As shown in the mosaic plot, which represents the deviation from the expected cell count for each group (singlet or multiplet) for each CD4⁺ T-cell subset, most of the multiple-clonotype Tregs were enriched in tumor-infiltrating aTregs, whereas there was no multiple-clonotype Tregs in tumor-infiltrating rTregs (Fig. 3D). This observation indicated that clonally expanded Tregs were enriched in tumor tissue upon exposure to antigen. Interestingly, the same clonotype between Tregs and Tconvs was only 1% among tumor-infiltrating Tregs, which was similar to the proportion among normal lung tissue Tregs (Supplementary Fig. S7). Moreover, we observed that the expression profile of tumor-infiltrating rTregs was more similar to that of normal lung tissue Tregs or splenic Tregs in Fig. 2B, whereas the expression pattern of tumor-infiltrating aTregs was comparable with that of tumor-infiltrating Tregs in Fig. 2B (Fig. 3E). Taken together, these results indicated that Tregs in lung tumor were far more proliferative compared with Tregs in normal lung.

It has been suggested Treg expansion in situ contributes to their accumulation in tumors (6). In addition, we showed that clonally expanded Tregs were enriched in tumor-infiltrating Tregs. To examine whether tumor-infiltrating Tregs proliferated in the TME, BrdU incorporation analysis was performed. Under normal steady-state condition, both Tregs in the spleen and lung incorporated a higher amount of BrdU than Tconvs as described previously (28). In the TME, tumor-infiltrating Tregs proliferated much faster than tumor-infiltrating Tconvs and steady-state Tregs (Fig. 4A and B). Robust proliferation of tumor-infiltrating Tregs was also verified by Ki67 staining (Fig. 4C and D). These findings indicated a higher turnover rate of Tregs in the TME promotes their accumulation therein. To investigate the apoptotic rates of tumor-infiltrating Tregs, CD4⁺CD25⁺ Tregs were stained with Annexin V (AV) and 7-AAD. The numbers of both early apoptotic (AV⁺7-AAD⁻) and late apoptotic/dead (AV⁺7-AAD⁺) Tregs increased in the TME compared with the lung of normal mice and the spleen of normal mice and tumor-bearing mice (Fig. 4E and F). Collectively, these data implied that although tumor-infiltrating Tregs displayed a high rate of early and late apoptosis, Tregs in the TME proliferated rapidly enough to overcome the rate of apoptosis of tumor-infiltrating Tregs, resulting in Treg accumulation in the TME.

A multistep candidate gene filtration was used to identify the key regulators of Tregs proliferation in the TME (Fig. 5A). We hypothesized the key regulators are overexpressed in tumor-infiltrating Tregs. Hence, we filtered the differentially expressed surface protein genes in tumor-infiltrating Tregs versus Tconvs and compared it with five other cellular and tissue contexts (see Fig. 2A). A higher level of FC threshold was used to avoid nonspecific targeting of the tumor-infiltrating Tconvs. To obtain high-confidence candidate genes, we generated DEGs based on the consensus of four distinct DEG analysis pipelines (see Methods) and obtained 37 DEGs. We selected 17 genes that express more than 10% of the level of CD4 expression because sufficiently high levels of the drug target molecules was needed to observe the therapeutic effect. To specifically target tumor-infiltrating Tregs, we then filtered out the genes that are highly expressed in other immune cell types using the Immgen database (24). We finally obtained three candidate genes, Il1rl1, Ncmap, and Sytl1, that passed all the filtration criteria (Fig. 5A). The expression level of Ncmap in other immune cell types could not be tested as this data was not available in the Immgen database. The protein encoded by Sytl1 had no extracellular domain, whereas that encoded by Il1rl1 (ST2) contained an extracellular domain based on the UniProtKB/Swiss-Prot database (22). Only Il1rl1 was annotated as an immune-related gene by GO, and the volcano plot of expression changes exhibited similar expression profiles for Il1rl1 and other tumor-infiltrating Tregs signature genes (Fig. 5B). Thus, we selected Il1rl1 (ST2) as the candidate gene for further analysis.

We observed the expression of Il1rl1 significantly correlated with that of genes involved in the proliferation of Tregs isolated from normal and tumor tissues (P = 6.72E−12; Fig. 5C). Consistent with the hypothesis that Il1rl1 is involved in tumor-infiltrating Tregs proliferation, an increased number of ST2⁺ Tregs was previously reported in the immune-tolerant environment of the intestine (15). These results suggest ST2 may be a key factor for Treg proliferation in the TME.

We confirmed an enhanced ST2 protein expression in tumor-infiltrating Tregs during tumor progression and nearly 30% of ST2⁺ tumor-infiltrating Tregs on day 21 in TC-1 tumor model (Fig. 5D and E). ST2 expression in tumor-infiltrating Tregs was higher than that in tumor-infiltrating Tconvs not only in TC-1 tumor model but also in other tumor models including LLC1 and CT26 (Fig. 5F and G), indicating high level of ST2 expression is a general feature of tumor-infiltrating Tregs.

In various human cancer types, the Il1rl1 expression in tumor-infiltrating Tregs was significantly higher than that in Tregs in peripheral blood or adjacent normal tissue (Supplementary Fig. S8A; refs. 10, 11). In addition, scRNA-seq data from NSCLC revealed an increased Il1rl1 in tumor-infiltrating Tregs but not in the blood or adjacent normal tissue Tregs and in tumor-infiltrating Tconvs (Supplementary Fig. S8B; ref. 29). ST2 staining of human Tregs in the blood and tumor tissue obtained from patients with NSCLC confirmed a highly enriched Tregs in tumor tissue and their ST2 expression (Supplementary Fig. S8C). These data indicated ST2 upregulation in Tregs was a distinctive feature of tumor-infiltrating Tregs but not in peripheral tissue Tregs or tumor-infiltrating Tconvs.

To further characterize tumor-infiltrating ST2⁺ Tregs, we compared ST2⁻ and ST2⁺ Tregs in the tissues of normal mice and tumor-bearing mice. As reported in colonic ST2⁺ Tregs and tissue ST2⁺ Tregs (15, 30), enhanced expressions of KLRG1 and CD103 were observed in tumor-infiltrating ST2⁺ Tregs. PD-1 expression in ST2⁺ tumor-infiltrating Tregs was also higher than that in ST2⁻ tumor-infiltrating Tregs on day 16 following tumor inoculation, even though the difference was not apparent thereafter (Fig. 5H; Supplementary Fig. S9). In addition, Ki67 expression in ST2⁺ tumor-infiltrating Tregs was significantly higher than in ST2⁻ tumor-infiltrating Tregs (Fig. 5I; Supplementary Fig. S9B). Similarly, the enhanced Ki67 expression was also observed in human ST2⁺ tumor-infiltrating Tregs in human NSCLC (Supplementary Fig. S8C). This suggests a proportion of Tregs in the TME exhibits enhanced ST2 expression and strong proliferative capacity.

IL33 binding to ST2 enhances the expression of ST2 through positive feedback loop (14). IL33 expression in the TME was significantly higher than that in the normal lung (Fig. 5J). To examine the cellular source of IL33 in the tumor, we performed multiplex immunofluorescence staining. Consistent with the previous reports showing endogenous mouse IL33 expression (31), various cells including CD31⁺ endothelial cells, PDPN⁺ epithelial cells, and CD45⁺ lymphocytes, expressed IL33 in the normal lung (Supplementary Fig. S10A–S10C). We also observed of the other IL33-positive cells in the TME. A previous report in human colon cancer (32) showed that CD31⁺ endothelial cells expressed IL33 in the TME. However, in our mouse lung tumor model, most of populations were IL33-negative and only small portion of CD45⁺ cells were IL33-positive (Supplementary Fig. S10B). This discrepancy surrounding the cellular source of IL33 may be due to different tissue or different TME. However, ELISA data showed that IL33 secretion quickly increased in the TME at early time points (day 8) following tumor inoculation and decreased thereafter (Supplementary Fig. S10D). Also, Ki67 level of ST2⁺ tumor-infiltrating Tregs gradually decreased as lung tumor progresses (Supplementary Fig. S9B). These data suggest that a rapid secretion of IL33 accumulated in the lung rather than upregulated gene expression of IL33 occurs upon an inflammatory signal generated at early stage of lung tumor progression and subsequently, IL33 as an alarmin initiates ST2 upregulation on Tregs in the TME, leading to proliferation of tumor-infiltrating Tregs.

Next, we examined whether in vitro cytokine treatment stimulated the expansion of tumor-infiltrating Tregs in the presence of TCR activation. Given tumor-infiltrating Tregs showed upregulation of Il2ra, Il10ra, Il12rb1, and Il1rl1 (Fig. 2B), we added the corresponding cytokines to tumor-infiltrating CD4⁺ T cells. IL33 treatment predominantly increased tumor-infiltrating Tregs (Supplementary Fig. S11). In addition, proliferation of Tregs upon IL33 treatment was greater than that of Tconvs (Fig. 6A and B). Upon inhibition of ST2, Tregs proliferation was significantly reduced, indicating the IL33-mediated proliferation results from direct interaction with ST2, but, Tconvs proliferation was slightly modulated by IL33 or anti-ST2 antibody treatment. Accordingly, Treg numbers increased upon IL33 treatment and decreased when ST2 was blocked (Fig. 6C) or genetically depleted (Supplementary Fig. S12). We also found Tregs proliferation upon IL33 treatment was restricted to ST2-expressing Tregs. Conversely, blocking ST2 decreased the IL33-mediated proliferation of ST2-expressing Tregs (Fig. 6D and E). Meanwhile, IL33 treatment did not promote the expansion of splenic Tregs from tumor-bearing mice, in which Tregs had limited ST2 expression (Supplementary Fig. S13). This suggests that IL33 promotes the proliferation of ST2⁺ tumor-infiltrating Tregs through direct interaction with ST2 in vivo.

Next, we investigated the potential impact of IL33 on tumor progression in vivo. When tumor-bearing mice were treated with IL33 on days 4, 7, and 10 following inoculation, the mice increased number of tumor nodules (Fig. 6F). Even single injection with IL33 increased tumor nodules (Supplementary Fig. S14A and S14B). Upon IL33 treatment, frequency of CD8⁺ T-cell and ratio of CD8⁺ T cells to Tregs within tumor decreased, but frequency of tumor-infiltrating Tregs significantly increased, suggesting IL33 may play an important role in impairing antitumor immunity (Fig. 6G; Supplementary Fig. S14C). Specifically, IL33 promoted the upregulation of Ki67 and ST2 in tumor-infiltrating Tregs in vivo (Fig. 6H and I; Supplementary Fig. S14D and S14E). This resulted in a significantly increased population of the ST2⁺Ki67⁺ tumor-infiltrating Tregs (Fig. 6I; Supplementary Fig. S14E). As triggering ST2 signaling by IL33 upregulated ST2 via positive feedback loop, upregulation of Ki67 and ST2 was observed in tumor-infiltrating Tconvs to some extent (Fig. 6H and I; Supplementary Fig. S14D and S14E).

To test whether these phenotypes were connected to IL33-mediated accelerated tumor growth, we performed in vitro suppression assays. As expected, IL33-treated Tregs were more effective at suppressing CD8⁺ T-cell proliferation than PBS-treated Tregs, when Tregs were administered at 1:1 ratio or 3:1 ratio (CD8⁺: Treg; Fig. 6J). Therefore, IL33 in the TME augments not only Tregs quantity but their function, thus contributing to immune suppression and tumor progression.

We examined whether ST2 deficiency could affect tumor-infiltrating Treg accumulation. In ST2-deficient compared with WT mice, the numbers of tumor nodules and tumor-infiltrating Tregs were significantly reduced (Fig. 7A and B). A significant decrease of tumor-infiltrating Tregs subsequently led to an increase in the ratio of CD8⁺ T cells to Tregs within the tumor (Fig. 7B) and enhanced IFNγ production in tumor-infiltrating CD8⁺ T cells (Fig. 7C). These results suggest that ST2 signaling plays an essential role in accumulating tumor-infiltrating Tregs and inhibiting CD8⁺ T cells, thereby contributing to tumor progression.

To confirm whether IL33-mediated Treg proliferation is mediated by ST2 in vivo, either WT or ST2-KO mice were treated with IL33 on days 4, 7, and 10 following tumor inoculation (Supplementary Fig. S15A). Consistent with our in vitro data, ST2 deficiency did not display Treg accumulation in the TME upon in vivo IL33 treatment, resulting in decreased tumor nodules (Supplementary Fig. S15B and S15C). ST2-deficient mice did not also exhibit an enhanced Treg proliferation after in vivo IL33 treatment, unlike WT mice (Supplementary Fig. S15D), further supporting that ST2 was critical for IL33-mediated Treg expansion.

To assess whether ST2 deletion in only Tregs could sufficiently inhibit Treg accumulation in the TME, we used coadoptive transfer system, in which Tregs of both WT and ST2-KO were cotransferred along with CD8⁺ T cells into Rag2^−/− mice. The recipient mice were subsequently inoculated with tumors (Fig. 7D). In the spleen, there was no difference in the frequency of WT and ST2-KO Tregs. In contrast, within the tumor, the frequency of ST2-KO Tregs markedly decreased compared with that of WT Tregs (Fig. 7E). The PD-1–expressing population was significantly less among ST2-KO Tregs than among WT Tregs, implying the association of ST2 with Treg activation (Fig. 7F). To test whether ST2-expressing Tregs contributed to an enhanced tumor progression in vivo, we performed single-adoptive transfer experiment, in which WT or ST2-KO Tregs were transferred separately (Fig. 7G). ST2-KO Treg–transferred mice showed significantly smaller numbers of tumor nodules than WT Treg–transferred mice, indicating that ST2 expression empowered tumor-infiltrating Tregs to suppress antitumor immunity (Fig. 7H).

We further tested whether neutralization of ST2 could affect tumor control once tumors were established (Fig. 7I). In vivo treatment with ST2 antibody decreased tumor nodules (Fig. 7J), in which tumor-infiltrating Tregs exhibited reduced Ki67 and ST2 expression (Fig. 7K), indicating that ST2 was critical for tumor-infiltrating Treg proliferation.

Taken together, these data indicated IL33/ST2 axis–mediated intrinsic signaling in Tregs promoted their proliferation, which was one of the critical pathways for the preferential accumulation of Tregs in the TME, thus contributing to the inhibition of antitumor immune response and subsequent tumor progression.

In this study, we investigated the molecular mechanisms by which Tregs accumulate in the TME. Our transcriptome analysis of tumor-infiltrating Tregs revealed functional molecules and cellular processes involved in Treg accumulation in the TME. Recent reports human tumor-infiltrating Tregs exhibit a different gene expression compared with blood Tregs or tumor-infiltrating Tconvs support the concept for TME-dependent changes of tumor-infiltrating Tregs (10, 33). In particular, the genes involved in proliferation, cell cycle, and chemotaxis were significantly enriched in tumor-infiltrating Tregs.

One possible explanation for the high density of Tregs in the TME is that the enhanced proliferation of tumor-infiltrating Tregs leads to their accumulation. Using multistep candidate gene-filtering process, we identified ST2 was one of significantly upregulated receptors in tumor-infiltrating Tregs. Tregs observed in adipose tissue and skin express high level of ST2 as determined by methylome and transcriptome patterns (30). In addition, ST2 expression was required for colonic or adipose Treg-specific expansion (15, 16, 34). While IL33 promotes tumor not only in mouse tumor models (35, 36) but also in clinical tumor patients (37, 38), few studies reported regarding the direct role of IL33 in tumor-infiltrating Treg expansion. Here, we demonstrated both in vitro– and in vivo–treated IL33 showed a growth factor-like effect on the expansion of tumor-infiltrating Tregs. IL33-dependent preferential proliferation of Tregs in the TME might contribute to the inhibition of tumor antigen–specific T cell responses, thereby promoting tumor progression. Indeed, IL33 treatment accelerates tumor growth in two different cancer models, 4T1 and CT26 (35, 39). However, we cannot exclude the possibility of Treg-independent mechanism for IL33-mediated tumor progression considering not only the ST2 expression by various types of immune cells but also the direct and indirect effects of IL33 contribute to this process (40). More comprehensive studies using ST2 conditional knockout in various immune cell types might address this issue.

IL33 increased in the TME at the early stage of tumor progression, which suggests a rapid IL33 secretion induces ST2 upregulation on tumor-infiltrating Tregs via a positive feedback loop, resulting in IL33/ST2-dependent vigorous proliferation. Although IL33/ST2 signaling remains to be explored, blocking the IL33/ST2 axis on tumor-infiltrating Tregs could be a potential therapeutic strategy. Indeed, ST2 blockade controlled tumor burden in our study and IL33 blockade suppressed tumor growth in another preclinical model (41). In addition, soluble form of ST2 (sST2) as a decoy receptor led to tumor growth inhibition by modulating TME (42). More convincingly, our data using ST2-KO Tregs clearly demonstrated that ST2 expression in Tregs was critical in their proliferation and accumulation in the TME. One interesting observation was that PD-1 is expressed much less in ST2-KO Tregs compared with WT Tregs, suggesting ST2 signaling in Tregs associated with an activation pathway.

Second possibility for Treg enrichment is the selective trafficking of Tregs into the TME. Tumor cells and tumor-associated macrophages secrete CCL22 to facilitate CCR4-expressing Tregs in human ovarian cancer (43). CCR8 was upregulated in tumor-infiltrating Tregs in breast cancer, colorectal carcinoma, melanoma, and lung adenocarcinoma, suggesting CCR8 as a potential therapeutic target for Treg modulation (10, 11). Blocking multiple chemokine receptors may block more efficiently Treg migration into the TME due to redundancy between modules of chemokine ligand/receptors. For example, blocking both CCL2 and CCL12 in combination with the Ad.E7 vaccine significantly decreased tumor-infiltrating Treg frequency and augmented antitumor immunity (44). Combination of CCL1 blocker and CpG-ODN resulted in not only reduction of tumor-infiltrating Tregs but also complete rejection of the tumor (45).

Another possibility for Treg enrichment is conversion of Tconvs to Tregs in the TME. On the basis of the observation that Tregs usually express a high Nrp1, most tumor-infiltrating Tregs are peripherally derived from Tconvs (46). However, it now appears that Nrp1 may not be an appropriate marker to distinguish Treg origin (37). As shown in Supplementary Fig. S7, tumor-infiltrating Tregs and tumor-infiltrating Tconvs shared only 1% of TCR repertoire, which suggests tumor-infiltrating Tregs are dominantly derived from pre-existing Tregs rather than conversion from Tconvs. Regarding the origin of tumor-infiltrating Tregs, previous studies support our observation. Epigenetic (47) and TCR repertoire analysis (48) suggest tumor-infiltrating Tregs originate from naturally occurring Tregs. In addition, tumor antigen–specific Tregs are developed in the thymus in an autoimmune regulator (Aire)–dependent manner (49).

The major mechanism underlying immune suppression by tumor-infiltrating Tregs has not yet been clarified. Analysis of gene sets found molecules related to immune suppression were enhanced in the TME. This finding provides an intriguing hypothesis that immune checkpoint antibodies given to patients with cancer might bind more strongly to tumor-infiltrating Tregs than to tumor-infiltrating Tconvs. In accordance with this hypothesis, anti–CTLA-4 given to melanoma patients bound to CTLA4 expressed by Tregs and induced the depletion of CTLA-4–expressing Tregs (50, 51). Thus, the action mechanism of immune checkpoint inhibitors might be mediated by functional neutralization and/or depletion of the corresponding immune checkpoint–expressing Tregs in the TME.

In conclusion, our study delineated the cellular processes and the key molecule involved in Treg accumulation in the TME. Furthermore, our work provided novel insights and strategies for therapeutic targeting of tumor-infiltrating Tregs via modulating the IL33/ST2 axis.

J. Son reports a patent for “Compositions and Methods for Treatment of Cancers” pending. J.-W. Cho reports a patent for “Compositions and Methods for Treatment of Cancers” pending. J. Moon reports a patent for “Compositions and Methods for Treatment of Cancers” pending. S. Park reports a patent for “Compositions and Methods for Treatment of Cancers” pending. I. Lee reports a patent for “Compositions and Methods for Treatment of Cancers” pending. S.-J. Ha reports a patent for “Compositions and Methods for Treatment of Cancers” pending. No potential conflicts of interest were disclosed by the other authors.

J. Son: Conceptualization, data curation, formal analysis, funding acquisition, validation, investigation, methodology, writing–original draft, writing–review and editing. J.-W. Cho: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. H.J. Park: Conceptualization, data curation, formal analysis, investigation, methodology. J. Moon: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft. S. Park: Resources, data curation, formal analysis, validation, investigation. H. Lee: Resources, investigation, methodology. J. Lee: Resources, data curation, investigation, methodology. G. Kim: Resources, formal analysis, investigation, methodology. S.-M. Park: Resources, formal analysis, investigation, methodology. S.A. Lira: Resources, methodology. A.N. Mckenzie: Resources, methodology. H.Y. Kim: Conceptualization, resources, methodology. C.Y. Choi: Conceptualization, resources, methodology. Y.T. Lim: Conceptualization, resources, methodology. S.Y. Park: Resources, investigation, methodology. H.R. Kim: Conceptualization, resources, data curation, funding acquisition, investigation, methodology, writing–original draft. S.-H. Park: Conceptualization, resources, investigation, methodology. E.-C. Shin: Conceptualization, resources, investigation, methodology. I. Lee: Conceptualization, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, writing–original draft, project administration, writing–review and editing. S.-J. Ha: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This study was supported by a National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT; 2012H1A2A1016220, to J. Son; 2017R1A5A1014560, 2018M3A9H3024850, 2018R1A2A1A05076997, and 2019M3A9B6065221, to S-.J. Ha; 2019M3A9B6065192 and 2018R1A5A2025079, to I. Lee; and 2019M3A9B6065231, to H.R. Kim). J. Son, J.-W. Cho, H.J. Park, J. Moon, and S. Park are fellowship awardees by the BK21 PLUS program.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.