Tumor Fibroblast–Derived FGF2 Regulates Expression of SPRY1 in Esophageal Tumor–Infiltrating T Cells and Plays a Role in T-cell Exhaustion

During chronic viral infection and cancer, CD8⁺ T cells undergo a differentiation process commonly referred to as T-cell exhaustion (1). During the progression to exhaustion, T cells gradually lose proliferative potential and cytotoxic function, express multiple inhibitory receptors, and are defined by an altered transcriptional program (1). To date, persistent antigenic stimulation has been viewed as the main driver of T-cell exhaustion in both infection and tumor models (1, 2). Antigenic stimulation via T-cell receptor (TCR) engagement can induce the expression of diverse transcription factors, such as NFAT, T-bet, IRF4, Blimp-1, BATF, and TOX, which have been implicated in determining CD8⁺ T-cell exhaustion (1, 3, 4). These transcriptional regulators promote the expression of inhibitory receptors, including programmed cell death protein 1 (PD-1), and mediate impaired T-cell effector function. In cancer, soluble molecules are the secondary drivers of negative regulatory signals that induce T-cell exhaustion, including IL10, TGFβ, and type I IFN (5). For example, TGFβ upregulates CD70 expression on effector memory T (Tem) cells and induces Tem-cell exhaustion (6). IL10 and IL35 cooperatively promote Blimp-1–dependent intratumoral CD8⁺ T-cell exhaustion (7).

A hallmark of exhausted CD8⁺ T cells is the coexpression of multiple checkpoint molecules, including PD-1, LAG3, TIM-3, CTLA4, TIGIT, and 2B4 (1). These inhibitory molecules protect T cells from uncontrolled immune activation, but they can also be used by tumors, making them resistant to T-cell attack (8). Checkpoint blockade therapy seeks to reinvigorate T-cell responses by targeting these immune checkpoint receptors. The currently approved immune checkpoint blockers are mAbs that target the cytotoxic T lymphocyte-associated protein 4 (CTLA-4) or PD-1 pathway (9). However, many patients fail to respond to and some cancers are refractory to these therapies. Furthermore, some treated patients initially respond and show positive treatment potential, but relapse (10). Thus, the extrinsic and intrinsic negative regulatory mechanisms of CD8⁺ T cells in cancer immunotherapy need to be further investigated.

Esophageal cell cancer (ESCC) is the world's sixth leading cause of cancer-related death after lung, liver, gastric, colorectal, and breast cancers and is characterized by early and frequent metastasis. Esophageal cancers develop various mechanisms to evade antitumor immune responses. Furthermore, there are limited treatment options, and the responses to these therapies are poor (11). Immunotherapy, such as immune checkpoint blockade therapy, represents a possible treatment option for patients with ESCC. Single-cell RNA sequencing (scRNA-seq) of infiltrating T lymphocytes allows a better understanding of tumor-infiltrated T cells; thus, it can help develop successful therapeutic interventions. Recently, scRNA-seq was applied to immune cells to reveal the tumor-infiltrated T-cell landscape in hepatocellular carcinoma, breast cancer, and lung cancer (12–15). However, the complexity of tumor-infiltrating T cells in ESCC is still unknown.

In this study, scRNA-seq was applied to analyze >8,000 single T cells isolated from primary ESCC tumors. We identified 16 unique T-cell subsets, including 12 CD8⁺ T-cell clusters and four CD4⁺ T-cell clusters. Here, we showed that Sprouty RTK signaling antagonist 1 (SPRY1) was significantly upregulated in two subsets of exhausted CD8⁺ T cells. SPRY1 is a member of the SPRY gene family, which is composed of four genes, namely SPRY1, SPRY2, SPRY3, and SPRY4, all of which are homologs of Drosophila Spry, which was initially identified as an inhibitor of FGFR signaling (16). Both SPRY1 and SPRY2 act as negative regulators of Ras-MAPK signaling and the transcription factors, NF-κB, NFAT, and AP-1, and limit T-cell activation (17–19). The absence of SPRY1 and SPRY2 enhances the survival of effector CD8⁺ T cells and results in more polyfunctional memory cells (17). Mice with selective deletion of SPRY1 in T cells demonstrate enhanced tumor rejection (18). High levels of SPRY2 are observed in human immunodeficiency virus (HIV)-specific T cells, and inhibition of SPRY2 expression upregulates HIV-specific T-cell polyfunctionality (20). In this study, we found that SPRY1 acted as a negative feedback inhibitor of TCR-triggered signaling by interacting with Cbl proto-oncogene (CBL). A positive correlation between FGF2 and SPRY1 was demonstrated in esophageal cancer on the basis of data from the Tumor Immune Estimation Resource (TIMER). In the ESCC tumor microenvironment, FGF2 expression was evident in αSMA⁺ fibroblasts, which closely contacted tumor-infiltrating CD8⁺ T cells. More importantly, IHC scoring for FGF2 in tumor tissue from 97 patients with ESCC revealed a correlation between poor overall survival and high FGF2 expression. All these findings related to FGF2 prompted us to study its role in T-cell antitumor immunity. In vitro and animal studies showed that FGF2 administration upregulated SPRY1 expression in CD8⁺ T cells and attenuated TCR-triggered T-cell activation, leading to compromised tumor control. These results indicated that FGF2 plays an important role in establishing T-cell exhaustion in ESCC. This effect is due, in part, to its effect on SPRY1 induction in tumor-infiltrating CD8⁺ T cells.

Human tissue specimens for scRNA-seq were obtained with informed written consent from 5 male patients and 3 female patients with ESCC who underwent surgery at the Sun Yat-sen University Cancer Center (Guangzhou, Guangdong, P.R. China). None of these patients were treated with chemotherapy or radiation prior to tumor resection. The collection of all the samples used in this study was approved by the Committees for the Ethical Review of Research Involving Human Subjects at the Sun Yat-sen University Cancer Center (Guangzhou, Guangdong, P.R. China). Peripheral blood mononuclear cells (PBMC) were collected from buffy coats, which were obtained from the Hong Kong Red Cross. Collection was approved by the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster (Hong Kong, China).

Detailed methods of scRNA-seq data processing and single-cell trajectories construction are provided in Supplementary Materials and Methods.

PBMCs from normal healthy donors were collected from buffy coats by Ficoll gradient centrifugation. CD8⁺ T cells were isolated from PBMCs by negative selection (CD8⁺ T Cell Isolation Kit, Miltenyi Biotec) and cultured in complete RPMI medium (RPMI1640, 10% FBS, 10 mmol/L HEPES, 1 mmol/L sodium pyruvate, 2 mmol/L l-glutamine, 0.05 mmol/L 2-mercaptoethanol, and streptomycin/penicillin). The MEC2 cell line was provided by the laboratory of L. Fu (The University of Shenzhen, Shenzhen, Guangdong, P.R. China) and was authenticated using short tandem repeat profiling in November 2019. The NIH3T3, Jurkat, and 293FT cell lines were purchased from the ATCC. All cell lines tested negative for Mycoplasma contamination.

Paraffin-embedded tumor samples were provided by the Sun Yat-sen University Cancer Center under a protocol approved by the institutional review board. The IHC study was performed using a standard streptavidin-biotin-peroxidase complex method as described in a previous publication (21). An immunoreactivity scoring system was applied to analyze FGF2 IHC staining. The intensity of FGF2-positive staining was scored as: 0, negative; 1, weak; 2, moderate; or 3, strong. Opal multicolor immunofluorescence (IF) staining was performed using a PANO 4-plex IHC Kit (Panovue).

Immunoprecipitation was performed according to the manufacturer's protocol (Pierce Direct Magnetic IP/Co-IP Kit). Western blot analysis was performed using a standard method (21).

Detailed methods of SPRY1 overexpression and knockout in Jurkat cells are described in Supplementary Materials and Methods.

Electroporation was used to transiently transduce CBL siRNA and control siRNA into SPRY1-overexpressing or -knockout Jurkat cells. In brief, Jurkat cells were washed in PBS and then suspended in Gene Pulser Electroporation Buffer (Bio-Rad) at a density of 5 × 10⁷/mL for electroporation. The suspended cells were electroporated at 250 V and 350 μF in a 0.4-cm gap cuvette using Gene Pulser (Bio-Rad). After electroporation, the cells were incubated in a cuvette at room temperature for 10 minutes and transferred to prewarmed complete RPMI medium. siRNAs for knockdown experiments were ordered from RiboBio. The three target sequences of siRNAs are CTACCAGCATCTCCGTACT; GAGCTTTCGACAGGCTCTA; and TCGGATTACTAAAGCAGAT. The product number of nonspecific control is siN0000001-1-5.

For SPRY1 overexpression, CD8⁺ T cells were cultured with IL7 (10 ng/mL) and IL15 (100 ng/mL) in complete RPMI medium for 2–3 days. CD8⁺ T cells (2 × 10⁶/mL) were then cultured with a concentrated lentivirus (centrifuged at 27,000 rpm for 2 hours at 4°C) supplemented with polybrene (8 μg/mL) in 96-well plates precoated with RetroNectin (1 mg/mL, T100B, Takara) and centrifuged at a speed of 500 × g for 90 minutes at 32°C. The lentiviral supernatant was discarded 6 hours later, and CD8⁺ T cells were cultured in complete RPMI medium with IL2 (10 ng/mL) for 3 days. Infection efficiency was analyzed by flow cytometry, and GFP-positive CD8⁺ T cells were sorted using a flow cytometer.

All animal experiments were approved by the Animal Ethics Committee of Sun Yat-sen University Cancer Center. Female C57/bl6 mice (ages 4–5 weeks) were purchased from Guangdong Medical Experimental Animal Center. NIH3T3 cells were stably transfected with lentivirus-control and lentivirus-FGF2 as mentioned above. For FGF2 overexpression, 5 × 10⁶ MEC2 cells were subcutaneously injected with 25 × 10⁵ NIH3T3-control cells or 25 × 10⁵ NIH3T3-FGF2 cells into the right flank of the mice. For FGF2 inhibition, 5 × 10⁶ MEC2 cells and 25 × 10⁵ NIH3T3-FGF2 cells were subcutaneously injected into the right flank of the mice. Two weeks later, tumor-bearing mice with a tumor size ranging from 100 to 200 mm³ were randomly divided into the vehicle control and AZD4547 treatment groups. The animals were given AZD4547 (6 mg/kg) or vehicle control three times per week by oral gavage. AZD4547 was formulated in a 1% (v/v) solution of Tween-80 in deionized water. Tumor size was recorded every week. Tumor size was calculated as: volume (mm³) = length × width² × 0.5.

Mice were euthanized on day 25 after tumor cell inoculation. Tumor tissues were cut into small pieces and digested with collagenase type IV (0.5 mg/mL) and DNase I (50 μg/mL) in RPMI1640. Leukocytes were enriched by density gradient centrifugation using 40% and 80% Percoll (GE Healthcare). Cell surface antigens and intracellular proteins were measured by flow cytometry using the method described in a previous publication (22). To evaluate CTL activity, MEC2 cells were seeded in 24-well plates 1 day in advance and cultured with CD8⁺ tumor-infiltrating lymphocytes (TIL) isolated from tumor single-cell suspensions by CD8 TIL Microbeads (Miltenyi Biotec). An anti-CD107 antibody (BioLegend) was added to the coculture. MEC2-specific death was detected by 7AAD staining, and T-cell surface mobilization of CD107 was quantified.

Kaplan–Meier analysis was performed using SPSS. All other statistical analyses were performed with GraphPad Prism. P values below 0.05 were considered significant. All error bars represent the SD of the mean.

We randomly divided 8 treatment-naïve patients with ESCC (Supplementary Fig. S1A) into two groups, each containing four pooled ESCC patient samples. Then, we sorted CD3⁺CD4⁺ and CD3⁺CD8⁺ T cells from a single-cell suspension prepared from the ESCC tumors. Single-cell transcriptome data were obtained for 2,880 and 6,032 T cells from the first and second groups, respectively. The mean reads per cell were 127,277 for the first pooled sample and 105,020 for the second pooled sample, with median numbers of genes detected per cell of 1,038 and 1,052, respectively. Sequencing saturation was >90% for each group. After preprocessing, normalization, and batch correction, unsupervised clustering was performed on a total of 8,912 T cells. We visualized these clusters in two dimensions using t-distributed stochastic neighbor embedding (t-SNE; Fig. 1A; ref. 23).

A total of 12 clusters for CD8⁺ T cells and four clusters for CD4⁺ T cells were identified on the basis of scRNA-seq profiles (Fig. 1A). Focusing on CD8⁺ T cells, we found that three different clusters, including CD8_C1- EOMES, CD8_C7-PRDM1, and CD8_C8-SPRY1, all of which expressed high levels of immune checkpoint genes, including PDCD1, CTLA4, HAVCR2, ENTPD1, LAG3, EOMES, and LAYN (Fig. 1B–D), which could be defined as exhausted T cells (T_EX; ref. 1). Notably, consistent with a previous study, these T_EX cells were not inert. They expressed high levels of effector molecules, such as IFNG, GZMB, and PRF1 (Fig. 1B). Other CD8⁺ T cells, for example, CD8_C2-ZNF683 cells, expressed tissue-resident memory T-cell markers, such as ITGAE and CXCR6 (14, 24). CD8_C4-KLRG1 cells were characterized by high expression of Tem cell markers, such as KLRG1, GZMK, LYAR, and EOMES (25), but low expression of IL7R (CD127) (Fig. 1B; Supplementary Fig. S1B), and thus, represented KLRG1^hiIL7R^low short-lived Tem cells (26). CD8_C11-IL7R cells expressed memory T-cell signature genes, such as IL7R and LYAR (13), and showed much lower KLRG1 expression than CD8_C4-KLRG1 cells (Fig. 1B; Supplementary Fig. S1B), suggesting that KLRG1^lowIL7R^hi memory precursor cells are capable of generating long-lived memory T cells. CD8_C12-FGFBP2 cells highly expressed TBX21, CX3CR1, GNLY, and PRF1 (Fig. 1B; Supplementary Fig. S1B), which are generally associated with T cells with cytotoxic effector functions (27). CD8_C9-CCR7 contained cells that exhibited high levels of ribosomal subunit transcripts and specifically expressed “naïve” marker genes, such as SELL and CCR7 (Fig. 1B; Supplementary Fig. S1B; ref. 28). CD8_C3-KLF2 shared a few common genes with cluster 9, such as CCR7, SELL, and IL7R, but also expressed high levels of JUNB, FOSB, TCF7, and KLF2 (Fig. 1B; Supplementary Fig. S1B), suggesting a central memory T-cell phenotype (29). Of interest, our analyses identified the presence of two other transcriptionally distinct clusters. One of them was CD8_C14-IFIT1, which uniquely expressed IFN-inducible genes, including IFIT1, IFIT3, IFI44L, IFI6, and ISG15 (Fig. 1B; Supplementary Fig. S1B). The other cluster, CD8_C15-MKI67, expressed signature genes, including STMN1, PCLAF, PCNA, and MKI67 (Fig. 1B; Supplementary Fig. S1B), and thus, exhibited a signature that is profoundly associated with mitotic division. The potential role of these two CD8⁺ T-cell clusters in the ESCC tumor microenvironment needs to be further explored.

To evaluate T-cell development, the pseudotime developmental trajectory of 12 CD8⁺ T-cell clusters was established by Monocle 3 (Fig. 1E; ref. 30). Three clusters of T_EX cells were highly enriched on one side of the branched structure positioned at the opposite end as CD8_C9-CCR7–naïve T cells, demonstrating the T-cell state transition from activation to exhaustion. Although three clusters of T_EX cells that were neighboring along the path, exhibited distinct expression patterns. CD8-C7-PRDM1 cells expressed high levels of chemokines (CCL4L2, XCL1, XCL2, and CCL4), but relatively low levels of GZMA and GNLY, whereas CD8-C8-SPRY1 T cells exhibited the opposite pattern (Fig. 1B). In addition, genes involved in the “antigen processing and presentation” and “TCR signaling” pathways were dramatically enriched in the CD8_C1-EOMES and CD8_C7-PRDM1 clusters compared with the CD8_C8-SPRY1 cluster based on Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis (Fig. 1F).

Similarly, we identified four major CD4⁺ T-cell clusters. CD4_ C0-FOXP3 contained CD4⁺ T cells exhibiting high expression of regulatory T-cell–related genes, such as FOXP3 and IL2RA (Supplementary Fig. S1B and S1C; ref. 31). CD4_C6-TXNIP cells were characterized by high expression of the memory cell markers, TXNIP and IL7R (Supplementary Fig. S1B and S1C; ref. 14). CD4_C10-CXCL13 showed T follicular helper–like features with high expression of CXCL13, PDCD1, IL6R, and CXCR5 (Supplementary Fig. S1B and S1C; ref. 32). Interestingly, a comparably small proportion of CD4⁺ T cells, CD4_C13-IL17A cells, was found. This group of CD4⁺ T cells uniquely expressed the cytokines IL17A and IL17F, the chemokine CCL20, and the transcriptional factors RORA and RORC (Supplementary Fig. S1C), suggesting that they were Th17 cells (33).

Our study revealed that SPRY1 exhibited a highly specific expression pattern in CD8_C7-PRDM1 and CD8_C8-SPRY1 T_EX cells (Fig. 1B and C). The association of SPRY1 with tumor-infiltrating CD8⁺ T cells has not been studied. To verify the expression of SPRY1 in tumor-infiltrating T cells, we detected SPRY1 expression in PD-1⁺Tim-3⁺CD8⁺ and PD-1⁻Tim-3⁻CD8⁺ TILs isolated from 3 additional patients with ESCC. Coexpression of Tim-3 and PD-1 correlates with more severe exhaustion of CD8⁺ T cells in cancer (34). These T_EX cells showed much higher SPRY1 expression than PD-1⁻Tim-3⁻CD8⁺ T cells (Fig. 2A–C). Next, we confirmed that SPRY1 was expressed on CD8⁺ T cells in esophageal tumor tissue by IF and IHC (Fig. 2D and E). Next, to directly test the role of SPRY1 in regulating CD8⁺ T-cell function, we used lentiviral-mediated overexpression of SPRY1 to mimic the high expression of SPRY1 on tumor-infiltrated CD8⁺ T_EX cells. Infection efficiency was analyzed by flow cytometry (Supplementary Fig. S2A), and GFP-positive CD8⁺ T cells were sorted. Interestingly, the SPRY1-overexpressing CD8⁺ T cells displayed a dramatic reduction in IFNγ, granzyme B, and TNFα production after stimulation with anti-CD3 and anti-CD28 antibodies (hereafter referred to as anti-CD3/CD28; Fig. 2F–J). To further determine the role of SPRY1 in TCR signal transduction, we measured NFATc1 expression and p65 phosphorylation levels. We found that overexpression of SPRY1 in CD8⁺ T cells decreased NFATc1 expression and p65 phosphorylation after TCR stimulation (Fig. 2K–L). However, SPRY1 overexpression did not affect CD8⁺ T-cell proliferation upon treatment with anti-CD3/CD28 (Supplementary Fig. S2B). In summary, these data support a role for SPRY1 as a suppressor of TCR-triggered CD8⁺ T-cell activation.

The negative regulatory role of SPRY1 in TCR signaling has been investigated using CD8⁺ T cells isolated from PBMCs from healthy donors. To further explore the mechanisms underlying this regulatory role, we overexpressed SPRY1 in Jurkat cells and treated the cells with anti-CD3/CD28. We found that overexpression of SPRY1 inhibited TCR-induced ZAP-70, LAT, and SLP-76 phosphorylation (Fig. 3A). In contrast, SPRY1 knockout caused higher phosphorylation of ZAP-70, LAT, and SLP-76 after TCR stimulation (Fig. 3B). The overexpression and knockout efficiencies are shown in Supplementary Fig. S2C and S2D. Collectively, these data suggest that SPRY1 acts as an inhibitor of TCR signaling. We then used coimmunoprecipitation to identify the potential proteins that interact with SPRY1. Endogenous CBL and ZAP-70 were pulled down when SPRY1 was immunoprecipitated using an anti-SPRY1 antibody (Fig. 3C). CBL is one of the earliest proteins to be tyrosine-phosphorylated upon TCR stimulation, and it acts as a negative regulator of ZAP-70 tyrosine phosphorylation (35, 36). Supplementary Fig. S2E shows the expression of CBL as detected by scRNA-seq. We suspected that SPRY1 might function as an adaptor to bring together CBL and ZAP-70. However, we observed that SPRY1 overexpression or knockout did not affect the interaction between CBL and ZAP-70 (Fig. 3D and E). Nevertheless, SPRY1 overexpression indeed downregulated ZAP-70 phosphorylation in cell lysates immunoprecipitated with an anti-CBL antibody (Fig. 3D). In contrast, SPRY1 deficiency upregulated ZAP-70 phosphorylation in cell lysates immunoprecipitated with an anti-CBL antibody (Fig. 3E). Next, we wanted to determine whether the SPRY1-mediated inhibition of ZAP-70 phosphorylation depends on CBL. In CBL-knockdown T cells (knockdown efficiency is shown in Supplementary Fig. S2F), ZAP-70 phosphorylation was enhanced by overexpression of SPRY1 (Fig. 3F). In contrast, in CBL-knockdown T cells, ZAP-70 and LAT phosphorylation were decreased by SPRY1 knockout (Fig. 3G). This experiment revealed the negative effect of SPRY1 on TCR signaling, which indicated that SPRY1-mediated inhibition of TCR signaling depends on CBL. In summary, our data suggest that SPRY1 can augment the CBL-mediated inhibition of ZAP-70 phosphorylation and thus, plays a negative role in TCR-triggered T-cell activation.

Next, we wanted to explore the mechanism by which SPRY1 is upregulated in CD8⁺ T_EX cells. SPRY1 is a known antagonist of FGF signaling, and its expression is induced by the signaling pathways that it inhibits (37–39). In addition, a significant correlation between SPRY1 and FGF2 expression in esophageal cancer was identified on the basis of data from the TIMER (Supplementary Fig. S2G). Therefore, we stimulated CD8⁺ T cells isolated from PBMCs with FGF2 and detected the upregulated expression of SPRY1 at both the mRNA and protein levels (Fig. 4A and B). However, FGF2 did not further induce SPRY1 expression after TCR stimulation (Supplementary Fig. S2H). Multicolor IF staining indicated that FGF2 was mainly expressed by αSMA⁺ fibroblasts in ESCC tissue and that infiltrated CD8⁺ T cells closely contacted the FGF2-expressing αSMA⁺ fibroblasts in the tumor stroma (Fig. 4C). Our data also demonstrate that CD68⁺ macrophages were not the primary sources of FGF2 (Supplementary Fig. S2I). We further assessed the FGF2 expression pattern by IHC staining in tumor tissue from 97 patients with ESCC. Notably, IHC scoring for FGF2 indicated that patients with low FGF2 expression had better overall survival (Fig. 4D and E). These data prompted us to study the role of FGF2 in tumor-infiltrating CD8⁺ T-cell function.

First, we analyzed several common T-cell activation markers on CD8⁺ T cells stimulated with or without FGF2. FGF2 did not change PD-1, Fas, CD27, CD44, or CD69 expression (Supplementary Fig. S2J) and did not affect CD8⁺ T-cell apoptosis (Supplementary Fig. S2K). By performing intracellular staining, we found that the production of IFNγ, granzyme B, and TNFα was dramatically decreased after anti-CD3/CD28 stimulation in CD8⁺ T cells pretreated with FGF2 (Fig. 4F and G). We obtained the same results by detecting IFNγ, granzyme B, TNFα, and IL2 secretion in the supernatant of CD8⁺ T cells stimulated via TCR in the presence or absence of FGF2 (Fig. 4H). In addition, pretreatment with FGF2 reduced ZAP-70 and LAT phosphorylation and enhanced LCK Try505 phosphorylation following TCR stimulation (Fig. 4I). FGF2 pretreatment also decreased NFATc1 expression and p65 phosphorylation in CD8⁺ T cells after TCR stimulation (Supplementary Fig S3A and S3B). These data suggest that FGF2 abrogates the TCR-triggered T-cell response.

FGF2 exerts its activity by binding to surface receptor tyrosine kinases and FGFR1 and FGFR2 (40). Here, we demonstrated that FGF2 induced much higher levels of FGFR1 mRNA than FGFR2 mRNA in CD8⁺ T cells (Supplementary Fig. S3C and S3D). To further determine whether FGF2 mediates the inhibitory effect of TCR on naïve T cells or antigen-experienced T cells, we isolated CD45RA⁺CD8⁺ T cells and CD45RO⁺CD8⁺ T cells from PBMCs and treated them with FGF2 before stimulating them with anti-CD3/CD28. Our data show that FGF2 induced much higher SPRY1 (Fig. 5A–C) and FGFR1 (Supplementary Fig. S3E and S3F) expression in CD45RO⁺CD8⁺ T cells than in CD45RA⁺CD8⁺ T cells. In addition, an FGF2-induced decrease in ZAP-70 and LAT phosphorylation was observed in CD45RO⁺ memory CD8⁺ T cells, but not in CD45RA⁺CD8⁺ T cells in the same cultures (Fig. 5D). Similarly, FGF2 significantly reduced IFNγ, granzyme B, TNFα, and IL2 production from CD45RO⁺CD8⁺ T cells following TCR stimulation (Fig. 5E). To elucidate the transcriptional changes in CD45RO⁺CD8⁺ memory T cells upon FGF2 treatment, we measured the transcriptional profiles of CD45RO⁺CD8⁺ T cells treated with or without FGF2 (Supplementary Fig. S3G). The results revealed that memory CD8⁺ T cells not treated with FGF2 showed much higher IFN-related gene expression than FGF2-treated memory CD8⁺ T cells. In summary, our data suggest that CD45RO⁺CD8⁺ memory T cells are the target cells that mediate the inhibitory effect of FGF2 on TCR stimulation.

To determine whether SPRY1 is specifically upregulated in exhausted CD8⁺ T cells in a mouse tumor model, we evaluated the expression levels of SPRY1 in mouse CD8⁺ T cells isolated from mice bearing tumors induced by the mouse esophageal tumor cell line MEC-2. Much higher SPRY1 expression was detected in exhausted PD-1⁺Tim-3⁺CD8⁺ TILs than in PD-1⁻Tim-3⁻CD8⁺ TILs (Supplementary Fig. S4A). SPRY1 was also expressed at much higher levels in antigen-specific CD39^hiCD8⁺ TILs than in CD39^lowCD8⁺ TILs (Supplementary Fig. S4B). To further investigate the role of FGF2 in regulating T-cell dysfunction and tumor growth, we utilized two mouse models. As FGF2 was mainly expressed in fibroblasts within the tumor stroma (Fig. 4C), we overexpressed FGF2 in the mouse fibroblast cell line NIH3T3 (NIH3T3-FGF2) and subcutaneously injected MEC2 cells with NIH3T3-FGF2 cells or control NIH3T3 cells into C57/bl6 mice (Fig. 6A). The results showed that tumors induced by MEC2 and NIH3T3-FGF2 cells grew much faster than those induced by MEC2 and NIH3T3-control cells (Fig. 6B and C). Moreover, NIH3T3-FGF2 mice showed higher SPRY1 expression in PD-1⁺Tim-3⁺CD8⁺ TILs than NIH3T3-control mice (Fig. 6D and E). However, there was no difference in the ratio between PD-1⁺Tim-3⁺CD8⁺ TILs and PD-1⁻Tim-3⁻CD8⁺ TILs between NIH3T3-control and NIH3T3-FGF2 mice (Supplementary Fig. S4C). NIH3T3-FGF2 mice also exhibited higher FGFR1 expression in CD44^hiCD8⁺ antigen–experienced TILs than in CD44^lowCD8⁺ TILs (Supplementary Fig. S4D). Furthermore, CD8⁺ T cells isolated from the tumors of NIH3T3-FGF2 mice exhibited lower IFNγ, TNFα, and granzyme B production in response to anti-CD3/CD28 stimulation than those isolated from NIH3T3-control mice (Fig. 6F and G), indicating a decreased response to TCR stimulation. As another hallmark of CD8⁺ T-cell function, cytolytic capacity was compared between NIH3T3-FGF2 CD8⁺ TILs and NIH3T3-control CD8⁺ TILs. The results showed that NIH3T3-control CD8⁺ TILs efficiently killed target cells, killing nearly 30% of target cells at a 2:1 effector:target ratio in an overnight assay. However, NIH3T3-FGF2 CD8⁺ TILs killed much fewer target cells under the same conditions (Fig. 6H and I). In addition, CD107 mobilization on the T-cell surface in NIH3T3-FGF2 CD8⁺ TILs was significantly reduced compared with that in NIH3T3-control CD8⁺ TILs (Fig. 6J and K). These data suggest that chronic FGF2 stimulation significantly reduces the cytotoxic potential of CD8⁺ TILs against tumor cells. Collectively, these results indicate that FGF2 upregulates SPRY1 expression in exhausted CD8⁺ T cells and acts as a suppressor of CD8⁺ T-cell activation thus, playing a critical role in suppressing antitumor CD8⁺ T-cell responses.

To further dissect the effect of FGF2 on tumor growth, we inoculated mice with MEC2 cells and NIH3T3-FGF2 cells and then treated them with the FGFR1/FGFR2/FGFR3 inhibitor, AZD4547, which has shown anticancer activity in phase II/III clinical trials focusing on the regression of breast and gastric tumors (41). Two weeks later, tumor-bearing mice with a tumor size ranging from 100 to 200 mm³ were randomly grouped and dosed orally with vehicle or 6 mg/kg AZD4547 for 6 weeks (Fig. 7A). Robust tumor regression was observed in the AZD4547 treatment group (Fig. 7B and C). AZD4547 treatment downregulated SPRY1 expression on PD-1⁺Tim-3⁺CD8⁺ TILs, as well as FGFR1 expression on CD44^hiCD8⁺ TILs (Fig. 7D and E; Supplementary Fig. S4E). Consistent with the retarded tumor growth, the effector function of CD8⁺ TILs from AZD4547-treated mice was significantly enhanced, with AZD4547-treated mice exhibiting higher production of INFγ, TNFα, and granzyme B than control mice (Fig. 7F and G). We compared the ability of CD8⁺ TILs isolated from control mice and AZD4547-treated mice to kill target cells. We found that AZD4547 administration significantly improved the cytotoxicity of CD8⁺ TILs toward MEC2 cells, which was reflected by increased CD107 mobilization on the T-cell surface and an increase in the number of 7AAD-positive tumor cells (Fig. 7H–K). These data indicate that FGF2 inhibition significantly reduces tumor growth in mice and may have nonredundant effects by improving the cytotoxic potential of CD8⁺ TILs against tumor cells.

In summary, by analyzing scRNA-seq data from more than 8,000 single T cells, we characterized the ESCC tumor-infiltrated T-cell landscape. We showed that SPRY1 was highly expressed in immune checkpoint–expressing CD8⁺ T cells. Abundant SPRY1 expression in CD8⁺ T cells impaired the TCR signaling pathway and cytokine production. Our study also highlighted the potential role of FGF2 in inducing SPRY1 expression and decreasing T-cell activation following TCR stimulation. Furthermore, we used an animal model to confirm that FGF2 upregulation in the tumor microenvironment correlated with SRPY1 upregulation and loss of effector function in exhausted CD8⁺ T cells.

Here, we provided a clear illustration of the ESCC tumor-infiltrating T-cell landscape using scRNA-seq and identified three heterogeneous clusters of exhausted CD8⁺ T cells. SPRY1 was mainly upregulated in two clusters of T_EX cells, suggesting that SPRY1 might act as a novel T_EX cell marker. The other important Sprouty gene family member, SPRY2, was barely detectable by scRNA-seq (Supplementary Fig. S4F), so we focused on SPRY1 in this study. Our data show that SPRY1 overexpression in primary CD8⁺ T cells downregulated cytolytic cytokine production and inhibited TCR-triggered NFATc1 expression and p65 phosphorylation. These data corroborated previous findings that SPRY1 regulates CD4⁺ and CD8⁺ T-cell effector generation and function by inhibiting NF-AT activation and NF-κB signaling (42). Mechanistically, SPRY1 translocates to the immune synapse upon TCR engagement and inhibits LAT and PLC-γ activation (17, 42). Here, we expanded on this mechanism and demonstrated that SPRY1 inhibited the phosphorylation of ZAP-70 and LAT by interacting with CBL and thus, downregulating TCR signaling strength. The amplitude and duration of TCR signaling are critical for T-cell effector function. In tumors, T_EX cells show much lower TCR signaling strength than functional T cells (43). Therefore, upregulated SPRY1 expression in T_EX cells and the inhibitory effect of SPRY1 on TCR signaling strength may represent new T-cell exhaustion mechanisms in tumors.

Previous studies have found that SPRY1 has the opposite effect on TCR signaling in fully differentiated Th1 cells and naïve T cells (44). Overexpression of SPRY1 inhibits TCR signaling in Th1 cells, but enhances TCR signaling in naïve CD4⁺ T cells (44). In this study, we found that the inhibitory effect of SPRY1 was dependent on CBL because CBL knockdown reversed the negative effect of SPRY1 on TCR-triggered activation. This finding helps illustrate that the dual effects of SPRY1 on TCR signaling depend on the level of CBL. CBL is a member of the mammalian Cbl family of proteins, including CBL, CBLB, and CBLC, which act as negative regulators to attenuate intracellular signaling induced by the engagement of cell surface receptors. CBL physically interacts with the Tyr²⁹² negative regulatory site in the SH2-kinase linker of ZAP-70 and decreases the TCR-induced phosphorylation of ZAP-70 (45). In our study, we provide evidence that SPRY1 overexpression augments CBL-mediated inhibition of ZAP-70 phosphorylation; thus, it acts as a novel immune checkpoint gene to inhibit TCR signaling.

The most important finding of this study is that FGF2 can induce the expression of the immune checkpoint gene SPRY1 independent of TCR signaling. While persistent TCR signaling is undoubtedly responsible for T-cell exhaustion, it is probably not the only factor that causes high immune checkpoint gene expression on tumor-infiltrating CD8⁺ T cells. In the tumor microenvironment, the activation of TCR signaling via T cell–tumor cell interactions is hindered because of the downregulation of MHC class I, the upregulation of immunosuppressive molecules in tumor cells, and the abundance of immunosuppressive cells and cytokines in the tumor microenvironment, which attenuate TCR signaling (46). Thus, TCR signaling–independent pathways must exist to induce immune checkpoint gene expression in tumor-infiltrating CD8⁺ T cells. FGF2 belongs to the FGF family and is involved in many biological processes, including early embryonic development, angiogenesis, wound healing, and tumor development (47). Upregulation of FGF2 in the tumor stroma is associated with more aggressive phenotypes and a high number of infiltrating M2 macrophages in ESCC (48). FGF2 also suppresses lymphocyte emigration in the angiogenic area, which promotes cancer cells to escape from host immune attack (49). In our study, FGF2 treatment significantly upregulated SPRY1 expression and reduced TCR-triggered CD8⁺ T-cell activation. Previous studies have demonstrated that SPRY1 negatively regulates IFN production and IFN-mediated signaling, which exerts essential antiviral and antitumor effects (16). Our data corroborated and extended the findings that FGF2 is a potent negative regulator of IFN signaling in memory CD8⁺ T cells and that this effect might be mediated by SPRY1 induction. FGF2 overexpression in fibroblasts significantly increased tumor size and reduced infiltrated T-cell cytotoxic function. In contrast, FGF2 inhibition dramatically reduced tumor burden and unleashed the effector function of CD8⁺ T cells. Therefore, considering the powerful immunosuppressive tumor microenvironments that dampen TCR signaling, we believe that SPRY1 induction via FGF2 plays an essential role in inducing T-cell exhaustion.

Many efforts have been made to study the lineage origin and developmental pathways that give rise to exhausted CD8⁺ T cells. In chronic infection, exhausted CD8⁺ T cells arise from KLRG1^lo memory precursors and not terminally differentiated effector CD8⁺ T cells (2). In a cancer study, TCR repertoire analysis revealed that exhausted CD8⁺ T cells originate from tumor resident Tem cells (12). Here, we demonstrated that FGF2 induced much higher SPRY1 expression in CD45RO⁺CD8⁺ memory T cells than in CD45RA⁺CD8⁺ T cells. Consistent with this finding, CD45RO⁺CD8⁺ memory T cells rather than CD45RA⁺CD8⁺-naïve T cells were the target cells that mediated the inhibitory effect of FGF2. An in vivo study showed that FGF2 dramatically increased FGFR1 expression on CD44^hiCD8⁺ antigen–experienced T cells rather than CD44^lowCD8⁺ T cells. On the basis of these data, we suspected that FGF2 might have an unrecognized role in altering the memory T-cell differentiation program and promoting them to differentiate into T_EX cells in the tumor microenvironment.

In summary, by performing scRNA-seq analysis of ESCC-infiltrating T cells, we provided a clear illustration of the T-cell landscape in ESCC tumors and identified a novel T_EX cell marker, SPRY1, the overexpression of which contributes to dysfunctional TCR signaling in T_EX cells. We also found that FGF2, which is highly expressed in the tumor stroma, can upregulate SPRY1 expression in CD8⁺ T cells and dampen TCR-triggered T-cell activation. Although this study focused on the role of the FGF2–SPRY1 axis in regulating TCR signaling, we cannot exclude that other FGF2-responsive genes are also crucial for the inhibitory effect of FGF2 on T cells. However, given that FGF2 and SPRY1 expression were positively correlated in esophageal cancer and that both of these proteins have negative regulatory effect on T-cell activation, the data provide a rationale for blocking the FGF2–SPRY1 axis to improve T-cell function and tumor-killing activity.

No disclosures were reported.

Q.-y. Chen: Conceptualization, funding acquisition, methodology, writing-original draft, writing-review and editing. Y.-n. Li: Methodology. X.-y. Wang: Methodology. X. Zhang: Resources. Y. Hu: Resources. L. Li: Methodology. D.-q. Suo: Methodology. K. Ni: Methodology. Z. Li: Methodology. J.-r. Zhan: Methodology. T.-t. Zeng: Methodology. Y.-h. Zhu: Visualization. Y. Li: Visualization. L.-j. Ma: Conceptualization, writing-review and editing. X.-Y. Guan: Conceptualization, supervision, funding acquisition, writing-original draft, writing-review and editing.

This work was supported by grants from the project funded by the National Basic Research Program of China (2012CB967001), the Hong Kong Research Grant Council Collaborative Research fund (C7065-18GF), RGC Research Impact Fund (R4017-18), Shenzhen Science and Technology programs (KQTD2018041118502879 and KQDT2015033117210153), Shenzhen Fundamental Research Program (JCYJ20180508153249223), the National Natural Science Foundation of China (81802410, 81772554, and 81872007), the Hong Kong Research Grant Council General Research Fund (17143716), the Natural Science Foundation of Guangdong in China (2018A030313034), the Young Talent Teachers Plan of Sun Yat-sen University (15ykpy33), the Guangdong Esophageal Cancer Institute Science and Technology Program (M201801), and the Pearl River Talent Plan-Overseas Youth Talent Introduction Plan (postdoctoral programme). X.-Y. Guan is the Sophie YM Chan Professor in Cancer Research.

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.