Female C57BL/6 wild-type mice and Rag1^−/− mice were purchased from GemPharmatech. CXCR6^gfp/gfp mice (Stock No. 005693), homozygous CXCR6-deficient mice, were purchased from the Jackson Laboratory. Cd8^−/− mice were obtained from Professor Lilin Ye (Third Military Medical University). B6.SJL-Ptprc^a Pepc^b /Boy (CD45.1) were obtained from Professor Xuetao Cao (Nankai University). CXCR6^gfp/gfp OT-I mice were obtained by crossing OT-I TCR transgenic mice with CXCR6^gfp/gfp mice. Mice were used in experiments at 6–8 weeks of age. All mice were housed in animal facility under specific pathogen-free conditions.

For CD8^Cxcr6−/− mice, total bone marrow cells from Cd8^−/− mice and from Cxcr6^−/− or wild-type mice were mixed and adoptively transferred intravenously at a 7:3 ratio into lethally irradiated (9 Gry) naive Cd8^−/− mice. A total of 5×10⁶ bone marrow cells were transferred per mouse. For Cxcr6^+/+:Cxcr6^−/− chimeras, bone marrow cells from WT (CD45.2) or Cxcr6^gfp/gfp (CD45.2) and bone marrow cells from C57BL/6 (CD45.1) mice were mixed and adoptively transferred intravenously at a 1:2 ratio into lethally irradiated Rag1^−/− mice. Recipient mice were fed antibiotics for 2 weeks and allowed to reconstitute for at least 8 weeks before challenged with tumor cells.

MC38 colon cancer cell line was obtained from Cell Resource Center of the Institutes of Biomedical 387 Sciences at Fudan University (Shanghai, China). CT26.WT colon cancer and B16F10 melanoma cell lines were purchased from American Type Culture Collection. B16 OVA melanoma cell line was kindly provided by Shengdian Wang (Institute of Biophysics, Chinese Academy of Sciences). All cell lines were cultured in Dulbecco’s modified Eagle medium or RPMI-1640 medium supplemented with 10% fetal bovine serum, 2 mmol/L glutamax, 100 U/mL penicillin, and 100 mg/mL streptomycin.

Paraffin-embedded slides of samples were deparaffinized, rehydrated, antigen retrieved with sodium citrate and blocked with 5% bovine albumin. Primary antibodies used to stain the sections included: rabbit antimouse CD8 (Abcam Cat# ab217344, RRID:AB_2890649), goat antimouse CXCR6 (Abcam Cat# ab125115, RRID:AB_10974584), mouse antihuman CD8 (Cell Signaling Technology Cat# 70306, RRID:AB_2799781), rabbit antihuman CXCR6 (Novus Cat# NLS1102, RRID:AB_10000951), goat antimouse CXCL16 (R&D, Cat# AF503-SP), goat antirabbit IgG (H+L) Alexa Fluor 647 (Thermo Fisher Scientific Cat# A-21244, RRID:AB_2535812), goat antirabbit IgG (H+L) Alexa Fluor 488 (Thermo Fisher Scientific Cat# A-11008, RRID:AB_143165), goat antimouse IgG (H+L) Alexa Fluor 594 (Thermo Fisher Scientific Cat# A-11005, RRID:AB_2534073) and donkey antigoat Alexa Fluor 647 (Thermo Fisher Scientific Cat# A32849TR, RRID:AB_2866498).

Flow cytometry was performed on Attune NxT Flow Cytometer and analyzed using FlowJo software. Cell-sorting experiments were performed on BD Aria II. Prepared single cells were stained by fixable viability stain reagents (BD Biosciences Cat# 565388, RRID: AB_2869673). Staining was performed at 4°C in the presence of FACS buffer (PBS, 0.5% BSA, 2 mM EDTA, and 0.1% sodium azide) after Fc block (BD Biosciences Cat# 558636, RRID: AB_1645217). In the case of intracellular cytokine staining, cell stimulation cocktail (plus protein transport inhibitors) (500×) (eBioscience Cat# 00-4975-93) was added for 4 hours before staining with the fixation/permeabilization solution kit (BD Biosciences Cat# 554714, RRID: AB_2869008). Antibodies used for staining were: antimouse IFN-γ-PE (Thermo Fisher Scientific Cat# 12-7311-81, RRID:AB_466192), antimouse IFN-γ-BV421 (BD Biosciences Cat# 563376, RRID: AB_2738165), antimouse Perforin-APC (Thermo Fisher Scientific Cat# 17-9392-80, RRID:AB_469514), antimouse Granzyme B-AF647 (BioLegend Cat# 515405, RRID:AB_2294995), antimouse CXCR6-BV421 (BioLegend Cat# 151109, RRID:AB_2616760), antimouse CXCR6-PE (BioLegend Cat# 151103, RRID:AB_2566545), antimouse CD8a-BV605 (BD Biosciences Cat# 563152, RRID:AB_2738030), antimouse CD45-AF700 (Thermo Fisher Scientific Cat# 56-0451-82, RRID:AB_891454), antimouse CD3-AF488 (BioLegend Cat# 100210, RRID:AB_389301), antimouse TNF-α-BV711 (BioLegend Cat# 502939, RRID:AB_2562740), antimouse CD8a-PE-Cy7 (BD Biosciences Cat# 552877, RRID:AB_394506), and antimouse NK1.1-PE-Cy7 (Thermo Fisher Scientific Cat# 25-5941-81, RRID:AB_469664)

For induction in vivo, naïve T cells from OT-I mice were transferred intravenously into mice challenged with B16OVA tumor cells in advance. OT-I⁺CXCR6⁺CD8⁺ T cells were detected on days 2, 4 and 7, respectively. For induction in vitro by tumor cells, MC38 tumor cells and naïve T cells were mixed at a ratio of 1:100 and plated in 24-well plates. Recombinant murine 10 ng/mL IL-2 (Peprotech Cat# 212–12) and 5 µg/mL CD3 (Thermo Fisher Scientific Cat# 16-0032-38, RRID: AB_2865578) and 5 µg/mL CD28 (Thermo Fisher Scientific Cat# 16-0281-81, RRID: AB_468920) functional antibodies were added. For induction in vitro by tumor tissues, shredded tumor tissues were in upper wells and naïve T cells in lower wells with the presence of IL-2 recombinant cytokine and CD3/CD28 functional antibodies. The application of transwell was to eliminate the interference of CXCR6⁺ T cells that already existed in tumor tissues. In the experiment of ACT, the chamber was withdrawn when large batches of CXCR6⁺ T cells were induced and cultured.

Tumor tissues were lysed in radio immunoprecipitation assay buffer supplemented with protease and phosphatase inhibitor (MedChemExpress). Proteins were quantified by the Pierce BCA Protein Assay Kit (ThermoFisher Scientific, Cat# 23227). The proteins were then separated by SDS-polyacrylamide gel electrophoresis and electrophoretically transferred onto polyvinylidene difluoride membranes. The membranes were probed with antibodies overnight at 4°C, and then incubated with a horseradish peroxidase-coupled secondary antibody. Detection was performed using a LumiGLO chemiluminescent substrate system. Antibodies were: goat antimouse CXCL16 (R&D, Cat# AF503-SP) and rabbit anti-β-actin (Abmart, Cat# M20011).

For anti-PD-1 treatment, MC38 cells (1×10⁶) or B16F10 cells (2×10⁵) were injected subcutaneously into female C57BL/6 mice at 6–8 weeks old. On day 10, tumor-bearing mice with similar size were randomly divided into two groups (n=8–10). Mice were treated with 200 µg isotype (Bio X Cell Cat# BE0089, RRID: AB_1107769) or anti-PD-1 antibody (Bio X Cell Cat# BE0146, RRID: AB_10949053) intraperitoneally every 3 days. For combination therapy strategy, the administration dose was reduced to 100 µg. For CXCR6⁺ cell deletion experiment, rabbit antimouse CXCR6 monoclonal antibody (Edelweiss Immune, clone: 19A5) or isotype was administrated 1 day before MC38 cells (1×10⁶) were injected subcutaneously. Mice were treated with 200 µg anti-CXCR6 or isotype intraperitoneally every 3 days. Tumor volume was analyzed by V=1/2 (a × b²).

For CXCR6⁺CD8⁺ T cells induced in vitro transfer, MC38 cells (1×10⁶) were injected subcutaneously into female C57BL/6 mice at age of 6–8 weeks. On day 7, 5×10⁶ induced CXCR6⁺CD8⁺ T cells or CXCR6^-CD8⁺ T isolated by FACS were transferred intravenously into tumor load mice. For OT-I CTL transfer, splenocytes isolated from Cxcr6^−/− OT-I mice or wild-type OT-I mice were stimulated with 2 nM OVA_257–264 (SIINFEKL) for 3 days with the presence of 10 ng/mL IL-2. Cells were centrifuged and cultured in fresh medium containing 10 ng/mL IL-2 for two more days, then the cells in the culture were CTLs. 5×10⁶ Cxcr6^−/− OT-I CTLs or WT OT-I CTLs were injected intravenously into B16OVA bearing mice.

Statistical analyses were performed using GraphPad Prism (GraphPad Software). Statistical significance was determined as indicated in the figure legends. Data are generally shown as mean±SEM unless otherwise stated. P<0.05 was considered significant: *p<0.05; **p<0.01; ***p<0.001. All t-test analyses are two-tailed unpaired t-tests. Statistical significance in multiple comparison determined by one-way analysis of variance with Tukey’s multiple comparison test.

To understand the expression atlas of CXCR6 in tumor, we first analyzed the single cell RNA sequencing data of colon cancer from patients and mice reported by Zhang et al.17 We found that CXCR6 was mainly expressed on T cells, especially on CD8⁺ T cells (figure 1A–D). Compared with peripheral tissues like peripheral blood (P), lymph node (LN) and adjacent normal (N) tissues, CXCR6 was significantly enriched in tumors (T) (figure 1E,F). Spatial transcriptome sequencing of clinical samples we performed previously also showed that the expression of CXCR6 in tumors was higher than that in normal tissues (data not shown). To verify the single cell RNA sequencing results, we also analyzed transcriptome sequencing datasets in TCGA and identified an increased tumor expression of CXCR6 similarly (online supplemental figure S1A). In addition, high expression of CXCR6 in tumor showed a good prognosis and positively correlated with the expression of CD8 (figure 1G,H, online supplemental figure S1C,D).

We validated our findings using immunofluorescence staining in colon cancer slides, which exhibited high colocalization of CXCR6 and CD8 (figure 1I, online supplemental figure S1B). Next, we detected the expression profile of CXCR6 in mice from peripheral blood (PB), LN, spleen and tumor tissues in MC38 and CT26 mouse model (figure 1J,K, online supplemental figure S1E–G). We observed that CXCR6 almost exclusively presented in tumor (figure 1L). Paust et al18 proposed that CXCR6 expressing NK cell could mediate antigen-specific memory of haptens and viruses. Hence, we examined the expression of CXCR6 on NK cell. As shown in online supplemental figure S1H, intratumoral NK cells rarely expressed CXCR6. Taken together, we identified that CXCR6 was predominantly expressed on CD8⁺ T cell in tumor tissue.

To characterize CXCR6⁺CD8⁺ T cell, we mapped CXCR6 expression on the CD8⁺ T cell subclusters using the single-cell sequencing data. We revealed that CXCR6 was mainly enriched in cytotoxic CD8⁺ T cell populations such as CD6⁺CD8⁺ T, GZMK⁺CD8⁺ T and CD160⁺CD8⁺ T cells (figure 2A). Furthermore, there was a strong correlation between the expression of CXCR6 and the infiltration level of central memory CD8⁺ T (TCM) (figure 2B). CXCR6⁺CD8⁺ T cells in tumor also expressed cytotoxic and activation markers such as IFN-γ, PRF1, GZMB, CD69 and CD44 but were negative for CD62L, a maker naïve T cell highly expressed (figure 2C,D, online supplemental figure S2A). These phenotypic characteristics indicated that CXCR6⁺CD8⁺ T cell subset possessed a higher degree of activation and killing ability.

To investigate whether CXCR6 expression on intratumoral CD8⁺ T cells was required for their antitumor efficacy, we constructed bone marrow chimeric mice in which CD8⁺ T cells were deficient for CXCR6 (named Cd8^CXCR6−/− mice), while in most of the other immune cells, CXCR6 expression was intact (figure 2E, online supplemental figure S2B,C). After tumor cells inoculation, survivals were reduced, and tumor growth aggravated significantly in Cd8^CXCR6−/− mice (figure 2F,G). Compared with mice reconstituted with wild-type and Cd8^−/− mice BM, the percentage of intratumoral CD8⁺ T cell in recipients reconstituted with BM derived from Cxcr6^−/− and Cd8^−/− mice was only half. This suggested that CXCR6 was required for CD8⁺ T cells infiltration into tumor (figure 2H). Strikingly, when we evaluated the function of infiltrating CD8⁺ T cells, we found that CXCR6-deficient CD8⁺ T cells exhibited lower IFN-γ, PRF1 and TNF-α secretion than wild-type counterparts (figure 2I, online supplemental figure S2D). We applied anti-CXCR6 monoclonal antibody to delete CXCR6⁺CD8⁺ cells in vivo (online supplemental figure S2E) and found tumor progression aggravated (online supplemental figure S2F). These data demonstrated CXCR6 was crucial for CD8⁺ T cells to enrich in tumor tissues and exert antitumor function.

Given that after anti-CD40 treatment the CXCR6⁺CD8⁺ T cell subpopulation was significantly increased (figure 3A,B), we speculated that CXCR6⁺CD8⁺ T cells might be the main CD8⁺ T cell subset that responded to immune checkpoint therapy. To validate our hypothesis, we used an anti-PD-1-responsive transplantable tumor model: the MC38 tumor cell line19 20 (figure 3C). We observed that in vivo recall responses to anti-PD-1 therapy in Cxcr6^−/− mice were markedly compromised (figure 3D). The percentages of intratumoral CD8⁺ T cells and cytokines production were rapidly decreased in Cxcr6^−/− mice, even treated with PD-1 blockade therapy (figure 3E, online supplemental figure S3A). To determine whether the effects of CXCR6 expression could be extended to other cancers, we also examined the effect of anti-PD-1 treatment strategy in the melanoma model. A similar result was observed that anti-PD-1 treatment significantly increased CXCR6 expression on infiltrating CD8⁺ T cells (online supplemental figure S3B). Moreover, intracellular cytokine staining showed an increase in CD8⁺ T cells expressing GZMB and IFN-γ, especially in CXCR6⁺CD8⁺ T cells, which had a greater variation (online supplemental figure S3C). Taken together, CXCR6 played an essential role in mediating the efficacy of immune checkpoint therapy and CXCR6⁺CD8⁺ T cell was the main subset response to this treatment strategy.

To further elucidate the role of the CXCR6 expression on CD8⁺ T cell, we transferred bone marrow from naïve CD45.1 mixed with WT or CXCR6^−/− mice in a ratio of 1:2 to lethally irradiated Rag1^−/− recipients (figure 4A, online supplemental figure S4A). Thus, in hosts that received bone marrow from both CD45.1 and CXCR6^−/− mice, CD45.1 positive cells were intact, while CD45.2 positive cells were CXCR6 deficient. After inoculated with tumor cells, the percentages of CD8⁺ T cells from both chimeras were in line with the initial adoptive ratio in peripheral blood. In contrast, tumor tissues exhibited the opposite proportions or more, suggesting the key role of CXCR6 for intratumoral CD8⁺ T cell enrichment (figure 4B). Similar results were observed in melanoma model (online supplemental figure S4B). Moreover, the cytolytic capacity and cytokine production of the transferred CD45.2 positive (CXCR6 deficiency) cells was also compromised in recipients (figure 4C, online supplemental figure S4C). These results together indicated that CXCR6⁺CD8⁺ T cells represented tumor preference of CD8⁺ T cells and preserves their effector functions during cancer progression.

According to mismatch repair, colon cancer can be classified into microsatellite instability (MSI) and microsatellite stability (MSS) subtypes.21 Patients with MSI tended to have a better prognosis and higher infiltration of activated CD8⁺ CTL.22–24 Therefore, we applied immunofluorescence staining to detect the colocalization of CXCR6 and CD8 in colon cancer from patients with MSI and MSS, respectively. Consistent with CD8⁺ T cell infiltration, both the frequency of CXCR6 and colocalization with CD8 were significantly higher in MSI group than MSS group (figure 4D, online supplemental figure S4D). This suggested that the abundance of CD8⁺ T cells in MSI colon cancer closely related to the CXCR6 expression. To determine the differential gene expression between CXCR6⁺CD8⁺ T and CXCR6⁻CD8⁺ T cells, we performed transcriptome sequencing on both subsets of CD8⁺ T cells isolated from tumor by FACS (figure 4E, online supplemental figure S4E,F). Transcriptional profiling revealed that CXCR6⁺CD8⁺ T cells expressed higher levels of several effector molecules (Gzmk and Gzmb), activation molecules (Ptx3 and Cd48), chemokines and chemokine receptors (Ccl24, Ccr4, and Cxcl12), classical inflammatory factors or inflammatory factor receptors (Il17f, Il1a, Il23r, and Il6st), compared with CXCR6^-CD8⁺ T cells (figure 4E). Taken together, transcriptional profiling indicated a more terminally differentiated state and innate immune responses of CXCR6⁺CD8⁺ T cells.

We hypothesized that it might be chemotaxis by its unique ligand, CXCL16, that was response for enrichment of CXCR6 in tumor. Thus, we checked the expression of CXCL16 in TCGA database and found high expression of CXCL16 predicted a good prognosis (online supplemental figure S5A). Especially interesting, when we tested the timing course of proportions of CXCR6⁺CD8⁺ T cells in PB and tumor in mice, we observed that CXCR6⁺CD8⁺ T cells were rarely detected in PB and almost maintained no changes, while intratumoral CXCR6⁺CD8⁺ T cells increased dramatically with cancer progression (figure 5A, online supplemental figure S5B,C). Furthermore, after treatment with anti-CD40, the expression of CXCL16 remained unchanged, even slightly lower than control group, while CXCR6 increased significantly (figures 3A,B and 5B). To time-dependently evaluate intratumoral CXCL16 expression, we examined its expression from day 10 to 30 after tumor cells inoculation. We found that either the mRNA expression level (online supplemental figure S5D,E) or protein level (figure 5C, online supplemental figure S5F) rarely changed with tumor progression. So we inferred that high tumor expression of CXCR6 was dependent on ligand-receptor chemotaxis and induction, the latter being even more primary.

To explore what induced high expression of CXCR6 in tumor, we cocultured tumor cells with naïve T cells in vitro. Unexpectedly, the expression of CXCR6 on CD8⁺ T cells was not induced successfully (figure 5D). Next, we constructed conditional coculture of the entire disrupted tumor tissue and naïve T cells. To avoid confusion with infiltrating CXCR6⁺CD8⁺ T cells in the tumor tissue, we used a cell culture chamber in which appropriate amount of tumor tissue was placed on the upper layer, while naïve T cells on lower layer. Gratifyingly, CXCR6⁺CD8⁺ T cells were successfully induced and the later stage of tumor, the higher of CXCR6 expression (figure 5E,F).

To determine the dynamic changes of CXCR6 abduction in vivo, we transferred naïve T cells from normal OT-I mice to WT mice challenged with B16OVA tumor cells (figure 5G). We revealed that on day 4 after T cells injection, CXCR6⁺CD8⁺ T cells in tumor tissue accounted for 40% of OT-I T cells but was poorly maintained in PB (figure 5H). Taken together, the high expression of CXCR6 in tumor tissue was mainly induced by the tumor tissue itself, not just chemotaxis.

We isolated CXCR6⁺CD8⁺ T and CXCR6^-CD8⁺ T cells induced in vitro and transferred them to tumor-bearing mice, respectively (figure 6A). Recipient mice showed significantly alleviative tumor volume (figure 6B, online supplemental figure S6A). Nevertheless, body weight of recipient mice decreased, (online supplemental figure S6B) due to the excessive inflammatory response that damages the body.25

We crossed CXCR6^gfp/gfp mice with OT-I TCR transgenic mice to generate CXCR6 deletion OT-I mice (named CXCR6^−/− OT-I mice) (figure 6C, online supplemental figure S6C). To confirm the intrinsic role of CXCR6 in CD8⁺ T cell function further, we treated B16OVA melanoma with adoptive T cell transfer therapy, which was stimulated in advance by peptide OVA_257-264 (figure 6C). Compared with wild-type, recipient mice transferred with CXCR6^−/− OT-I CTLs showed weaker antitumor activity, evidenced by larger tumor volume and a shorter survival time (figure 6D,E). These data indicated that CXCR6 was required for antitumor efficacy of intratumoral CD8⁺ T cells. When combined CXCR6⁺CD8⁺ T cells and anti-PD-1 antibody treatment strategies, we found that CXCR6⁺ T cells could enhance the effect of anti-PD-1 treatment (figure 6F,G). Even one mouse was completely cured and the tumor disappeared (figure 6F, red circle). The proportion of intratumoral CD8⁺ T cells also rose in combination group although slightly (figure 6H).