CX3CR1–CD8+ T cells are critical in antitumor efficacy but functionally suppressed in the tumor microenvironment

While blockade of PD-1/PD-L1 immune checkpoint revolutionized cancer treatment, how it works on tumor-infiltrating CD8+ T cells recognizing the same antigen at various differentiation stages remains elusive. Here, we found that the chemokine receptor CX3CR1 identified three distinct differentiation states of intratumoral CD8+ T-cell subsets. Adoptively transferred antigen-specific CX3CR1neg CD8+ T cells generated phenotypically and functionally distinct CX3CR1int and CX3CR1hi subsets in the periphery. Notably, expression of co-inhibitory receptors and Tcf1 inversely correlated with the degree of T-cell differentiation defined by CX3CR1. Despite significantly lower expression of co-inhibitory receptors and potent cytolytic activity, in vivo depletion of the CX3CR1hi subset did not alter the antitumor efficacy of adoptively transferred CD8+ T cells. Furthermore, differentiated CX3CR1int and CX3CR1hi subsets were impaired in their ability to undergo proliferation upon re-stimulation, and had no impact on established tumors upon second adoptive transfer compared with the CX3CR1neg subset that remained effective. Accordingly, anti-PD-L1 therapy preferentially rescued proliferation and cytokine production of the CX3CR1neg subset, and significantly enhanced antitumor efficacy of adoptively transferred CD8+ T cells. These findings provide a better understanding of the phenotypic and functional heterogeneity of tumor-infiltrating CD8+ T cells, and can be exploited to develop more effective immunotherapy.


Introduction
CD8 + T cells are a critical component of adaptive immunity against intracellular pathogens and cancerous cells (1). Upon activation through their T cell receptor (TCR) by peptide presentation on major histocompatibility complex class I molecules, naive CD8 + T cells undergo clonal expansion and differentiate to diverse phenotypically, functionally, and anatomically distinct effector CD8 + T cell subsets (2,3). The presence of CD8 + T cells in the tumor correlates with improved survival in multiple solid malignancies (4); however, these cells often fail to mediate tumor regression (5,6).
The paradoxical coexistence of intratumoral CD8 + T cells and tumor progression indicates that tumors establish immune-privileged microenvironments in patients with advanced cancer (7). Much progress has been made in the last 2 decades in uncovering immune checkpoints in tumor-infiltrating CD8 + T cells, which display a spectrum of dysfunctional states induced by persistent TCR stimulation, engagement of inhibitory receptors, the presence of immunosuppressive cell populations and cytokines, and metabolic factors (8). T cell exhaustion is an acquired state of T cell dysfunction, associated with defective proliferative capacities and reduced cytokine production (9), and therapeutic reinvigoration of tumor-specific T cells has uncovered the potential of immunotherapy for diverse tumor types (10,11).
Although blockade of the programmed cell death 1/programmed cell death ligand 1 (PD-1/PD-L1) immune checkpoint has revolutionized cancer treatment, how it works on tumor-infiltrating CD8 + T cells recognizing the same antigen at various differentiation stages remains elusive. Here, we found that the chemokine receptor CX3CR1 identified 3 distinct differentiation states of intratumor CD8 + T cell subsets. Adoptively transferred antigen-specific CX3CR1 -CD8 + T cells generated phenotypically and functionally distinct CX3CR1 int and CX3CR1 hi subsets in the periphery. Notably, expression of coinhibitory receptors and T cell factor 1 (Tcf1) inversely correlated with the degree of T cell differentiation defined by CX3CR1. Despite lower expression of coinhibitory receptors and potent cytolytic activity, in vivo depletion of the CX3CR1 hi subset did not alter the antitumor efficacy of adoptively transferred CD8 + T cells. Furthermore, differentiated CX3CR1 int and CX3CR1 hi subsets were impaired in their ability to undergo proliferation upon restimulation and had no impact on established tumors upon second adoptive transfer compared with the CX3CR1subset that remained effective. Accordingly, anti-PD-L1 therapy preferentially rescued proliferation and cytokine production of the CX3CR1subset and enhanced antitumor efficacy of adoptively transferred CD8 + T cells. These findings provide a better understanding of the phenotypic and functional heterogeneity of tumor-infiltrating CD8 + T cells and can be exploited to develop more effective immunotherapy.
Although CD8 + T cells with the same epitope specificity can display substantial heterogeneity in chronic viral infection and cancer (12)(13)(14)(15), their functions and roles are ill-defined. Mounting evidence has revealed that less-differentiated CD8 + T cells have superior proliferative capacity, longer persistence, and greater antitumor efficacy in vivo compared with fully differentiated effector T cells in the context of adoptive cell therapy (ACT) (12,13). However, the fate of these less-differentiated tumor-specific CD8 + T cells in the tumor microenvironment (TME) remains unclear, and the relationship between T cell exhaustion and differentiation status has not been clearly delineated.
The present study was prompted by the recent findings that expression of CX3C chemokine receptor 1 (CX3CR1) correlates with the degree of effector CD8 + T cell differentiation and identifies 3 functionally distinct populations of pathogen-specific effector and memory CD8 + T cells in the periphery in infectious models of vaccinia and lymphocytic choriomeningitis virus (16). Here, we set out to elucidate phenotypic and functional heterogeneity of antigen-specific effector CD8 + T cells using a preclinical model of melanoma. Similar to antiviral CD8 + T cells (16), CX3CR1 identified 3 distinct effector CD8 + T cell subsets: CX3CR1 negative, intermediate (int), and high (hi), in blood, spleen, lymph nodes (LNs), and the tumor.
Our study reveals that adoptively transferred antigen-specific CX3CR1 -CD8 + T cells generate CX3CR1 int and CX3CR1 hi subsets in the periphery upon vaccination with the cognate antigen, gp100, and CD40/TLR7 stimulation. Expression of coinhibitory receptors, cytokine production and cytolytic capacity, and response to immune checkpoint blockade were different among 3 subsets of CD8 + tumor-infiltrating lymphocytes (TILs). In vivo antitumor reactivity of the CX3CR1 hi subset that exhibited potent cytolytic capacity and lower levels of coinhibitory receptors in the tumor was evaluated by a potentially novel mouse model where CX3CR1 hi CD8 + T cells can be selectively depleted in vivo using diphtheria toxin (DT). Last, differential impact of PD-L1 blockade on 3 subsets of CD8 + TILs was determined.

CX3CR1 expression levels identify 3 phenotypically distinct subsets of adoptively transferred tumor-specific CD8 + T cells in vivo.
CD8 + TILs comprise heterogeneous cell populations with unique differentiation and functional states (8,9). To elucidate heterogeneity of CD8 + T cells recognizing the same antigen in the TME, we used Pmel-1 TCR-transgenic CD8 + T cells specific for the gp100 melanocyte differentiation antigen expressed on B16 melanomas (17). Pmel-1 T cells were activated with human (h) gp100 peptide, IL-7, and IL-15 for 6 days in vitro (18). These in vitro-activated Pmel-1 T cells have increased fluorescence intensity of CD27 and CD44 compared with T cells before activation but do not express CX3CR1 ( Figure 1A and Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/jci. insight.133920DS1). These less-differentiated Pmel-1 T cells were adoptively transferred to mice with large established s.c. B16 melanomas ( Figure 1B). Mice received 500 cGy of sublethal irradiation prior to ACT for lymphodepletion. To maximize antitumor efficacy of adoptively transferred T cells, mice received vaccination with hgp100, agonistic CD40 antibody (Ab), and TLR7 agonist and systemic administration of IL-2 (18,19). Adoptively transferred T cells expressing CD90.1 can be distinguished from endogenous T cells (CD90.2) in C57BL/6 mice. Blood, spleen, LNs, and tumors were harvested at different time points for phenotypic and/or functional analysis of infused Pmel-1 T cells.
Adoptively transferred Pmel-1 T cells started to express CX3CR1 4 days after infusion, and 3 distinct CX3CR1 -, CX3CR1 int , and CX3CR1 hi subsets of CD8 + CD90.1 + T cells were identified in blood, spleen, LNs, and the tumor by day 7 ( Figure 1C and Supplemental Figure 1B). The frequency of the CX3CR1 int subset was maintained in the tumor compared with other tissues whereas CX3CR1and CX3CR1 hi subsets became dominant in LNs and blood on day 25, respectively ( Figure 1C).
Next, we performed phenotypic analysis of 3 subsets of Pmel-1 T cells in spleen and the tumor. In both spleen and the tumor, the CX3CR1subset contained more CD62L + , CD127 + , and KLRG1populations, suggesting less-differentiated T cells while the CX3CR1 hi subset comprised more CD62L -, CD127 -, and KLRG1 + populations consistent with terminally differentiated effector T cells ( Figure 1D) (12,20,21). Transcription factor T cell factor 1 (Tcf1), encoded by Tcf7, is a critical regulator of T cell development and memory formation (22)(23)(24)(25). Emergent evidence indicates that intratumor Tcf1 + CD8 + T cells have increased self-renewal, expansion, and differentiation capacity and play an important role in controlling tumor growth (26)(27)(28)(29)(30). We examined Tcf1 expression in the 3 subsets of tumor-infiltrating Pmel-1 CD8 + T cells and found that the CXC3R1subset expressed significantly higher levels of Tcf1 compared with CXC3R1 int and CXC3R1 hi subsets ( Figure 1E). Of note, we did not observe the expression of a type I transmembrane gly- coprotein CD101 in tumor-infiltrating Pmel-1 CD8 + T cells (Supplemental Figure 1C), which was recently identified in the terminally differentiated state of CD8 + T cells in chronic viral infection (31).
Expression of coinhibitory receptors on 3 subsets of CD8 + TILs inversely correlated with the degree of T cell differentiation defined by CX3CR1. To elucidate the exhaustion state of 3 subsets of intratumor CD8 + T cells with the same epitope specificity, we evaluated expression of coinhibitory receptors on Pmel-1 T cells in B16 melanoma. Interestingly, programmed cell death 1 (PD-1) expression on tumor-infiltrating Pmel-1 T cells inversely correlated with the degree of differentiation defined by CX3CR1 (Figure 2A). Furthermore, despite their terminally differentiated status, the CX3CR1 hi subset expressed significantly lower levels of coinhibitory receptors, PD-1, LAG-3, and TIGIT in the tumor compared with CX3CR1and CX3CR1 int subsets ( Figure 2A). We evaluated kinetics of coinhibitory receptor expression on intratumor Pmel-1 T cells at 7, 10, 13, and 19 days after ACT. This revealed that the CX3CR1 hi subset persistently expressed significantly lower levels of coinhibitory receptors (Figure 2, B-D). Of note, these coinhibitory receptors were not expressed on Pmel-1 T cells in the blood, spleen, or LNs (data not shown).
Functional heterogeneity of 3 subsets of tumor-infiltrating antigen-specific CD8 + T cells defined by CX3CR1. Functional heterogeneity of CD8 + TILs in the context of differentiation status remains elusive. To this end, we harvested splenocytes and TILs 7 days after ACT, cocultured them with hgp100 peptide, and evaluated intracellular expression of IL-2, IFN-γ, TNF-α, granzyme B (GZMB), and granzyme A (GZMA) in Pmel-1 T cells. We found the CX3CR1subset in spleen contained more cytokine-producing CD8 + T cells compared with CX3CR1 int and CX3CR1 hi subsets ( Figure 4A), consistent with observations from infectious models evaluating 3 subsets of virus-specific CD8 + T cells defined by CX3CR1 (16). In the tumor, however, we observed a dramatic reduction in the number of cytokine-producing CX3CR1cells, and the CX3CR1 int subset was found to contain more polyfunctional cells producing IL-2, IFN-γ, and TNF-α compared with CX3CR1and CX3CR1 hi subsets ( Figure 4A). Both in spleen and the tumor, expression of GZMA correlated with T cell differentiation status whereas no difference in GZMB expression was found among all 3 Pmel-1 T cell subsets ( Figure 4A).
Although these results provide insight into the functional heterogeneity among 3 subsets in the periphery and the tumor, cytokines from 1 subset might affect function of the other subsets during coculture with hgp100 peptide. To evaluate functional exhaustion of 3 subsets of tumor-infiltrating effector CD8 + T cells compared with their peripheral counterparts, we flow-isolated each subset of splenic and intratumor effector (CD44 + CD62L -) CD8 + CD90.1 + T cells, cocultured them with B16 melanoma cells, and measured cytokine production by ELISA. We found production of effector cytokines, IL-2, IFN-γ, and TNF-α against B16 melanomas was severely impaired in the CX3CR1subset while cytokine production from intratumor CX3CR1 int and CX3CR1 hi subsets was relatively preserved compared with their splenic counterparts ( Figure 4B).
Next, to evaluate the cytolytic activity of 3 subsets of CD8 + T cells, we cocultured and flow-isolated each subset of tumor-infiltrating Pmel-1 T cells with luciferase-transfected B16 melanoma cells. As measured using a bioluminescence-based cytotoxicity assay, CX3CR1 hi CD8 + TILs exhibited significantly higher cytotoxicity than CX3CR1and CX3CR1 int subsets ( Figure 4C). Furthermore, real-time reverse transcription PCR (real-time RT-PCR) analysis showed Gzma, Gzmb, Gzmk, and perforin (Prf1) were enriched in the CX3CR1 hi subset compared with CX3CR1and CX3CR1 int subsets ( Figure 4D). Consistent with these results, we found increased perforin, GZMA, GZMB, and GZMK expression on CX3CR1 hi TILs from a patient with melanoma ( Figure 4E).
Previous preclinical studies indicated decreased antitumor responses in CX3CR1 -/mice (32, 33) and a critical role of the CX3CR1/CX3CL1 pathway in T cell/NK cell-mediated antitumor immunity (34)(35)(36); however, these studies did not distinguish antitumor efficacy between phenotypically and functionally distinct CX3CR1 int and CX3CR1 hi subsets. Moreover, CX3CR1 is expressed not only on T cells but also on monocytes, dendritic cells (DCs), γδ T cells, and NK cells (37). Therefore, we sought to generate a mouse model where we could selectively deplete tumor antigen-specific CX3CR1 hi CD8 + T cells in vivo.
Diehl et al. developed inducible CX3CR1-diphtheria toxin receptor (DTR) mice by introducing a loxPfloxed stop cassette followed by DTR into the Cx3cr1 locus (38). In their study, CX3CR1-DTR mice were crossed to CD11c-Cre mice, allowing deletion of the stop cassette in CD11c + cells and induction of DTR in CX3CR1 + cells. They showed expression of CD11c-Cre excises the stop cassette, allowing for DTR expression and selective depletion of CX3CR1 hi CD11c + cells upon administration of DT, with no effect on CX3CR1 int CD11c + cells (38).  To determine in vivo antitumor efficacy of the Pmel-1 CX3CR1 hi CD8 + T cell subset, we crossed Pmel-1 mice with CD2-Cre mice (39), followed by CX3CR1-DTR mice (38), and generated Pmel-1 Cd2-cre/ Cx3cr1 +/DTR mice. Expression of Cd2-cre excises the loxP-floxed stop cassette upstream of the DTR-coding region, allowing for DTR expression and selective depletion of the Pmel-1 CX3CR1 hi CD8 + T cell subset upon administration of DT, with no effect on Pmel-1 CX3CR1or CX3CR1 int cells ( Figure 5, A and B).
B16 tumor-bearing mice were treated with adoptive transfer of in vitro-activated Pmel-1 Cd2-cre/ Cx3cr1 +/DTR CD8 + T cells followed by vaccination and IL-2. These mice received phosphate-buffered saline or DT injection every other day for 12 days starting 1 day before ACT ( Figure 5A). Adoptive transfer of Pmel-1 Cd2-cre/Cx3cr1 +/DTR CD8 + T cells significantly delayed tumor growth and increased survival; however, depletion of Pmel-1 CX3CR1 hi CD8 + T cells did not alter antitumor efficacy of ACT ( Figure 5C).
Response of CX3CR1 -, CX3CR1 int , and CX3CR1 hi CD8 + T cells to antigenic rechallenge. To gain further insight into the antitumor efficacy of 3 subsets of CD8 + T cells, we sorted CX3CR1 -, CX3CR1 int , and CX3CR1 hi subsets from blood and spleen of B16 tumor-bearing mice treated with adoptive transfer of Pmel-1 T cells and restimulated with the gp100 antigen in vitro ( Figure 6A). The isolated CX3CR1subset expressed significantly higher levels of Tcf7 and expanded vigorously upon secondary antigenic encounter, compared with CX3CR1 int and CX3CR1 hi subsets ( Figure 6, B and C). Accordingly, isolated CX3CR1cells revealed superior proliferative capability in vivo compared with CX3CR1 + (CX3CR1 int and CX3CR1 hi ) subsets upon second adoptive transfer into B16 tumor-bearing mice ( Figure 6, D and E). Notably, CX3CR1 expression was stable in the infused CX3CR1 + subsets while CX3CR1cells gave rise to CX3CR1 + subsets in vivo ( Figure 6F), demonstrating that antigen rechallenge of effector CD8 + T cells reinvokes a unidirectional differentiation program, consistent with observations using virus-specific CD8 + T cells (16). Importantly, adoptively transferred CX3CR1but not CX3CR1 + cells significantly delayed tumor growth and increased survival ( Figure 6G). Taken together, the CX3CR1subset is composed of less-differentiated Tcf7 + CD8 + T cells that maintain proliferative and differentiation capacity and mediate antitumor efficacy upon antigenic rechallenge.
Less-differentiated CD8 + T cells are preferentially rescued by programmed cell death ligand 1 blockade in the tumor. Although the blockade of immune checkpoints, such as PD-1/programmed cell death ligand 1 (PD-1/PD-L1), reinvigorates tumor-specific CD8 + T cells and can durably control advanced solid malignancies (10), less is known regarding how it works on heterogeneous CD8 + TILs. Evidence from chronic viral infection models suggests that there are differential responses to PD-1/PD-L1 blockade between exhausted CD8 + T cell subpopulations that are characterized by the level of PD-1 expression (40). Our results of different levels of PD-1 expression in CX3CR1 -, CX3CR1 int , and CX3CR1 hi CD8 + T cells indicate the potential of reinvigoration by PD-1/PD-L1 blockade differs among the 3 subsets of CD8 + TILs.
To evaluate the differential impact of PD-L1 blockade on the 3 subsets of tumor-infiltrating CD8 + T cells, we treated B16 tumor-bearing mice with anti-PD-L1 Ab or isotype control Ab after adoptive transfer of Pmel-1 T cells, vaccination, and IL-2 ( Figure 7A). PD-L1 blockade caused a significant delay in tumor growth and improved survival in combination with adoptive transfer of Pmel-1 CD8 + T cells while PD-L1 blockade alone had no therapeutic effect ( Figure 7B). Although there were no changes in terms of the ratios of each CX3CR1 (neg/int/hi) subset, anti-PD-L1 therapy significantly increased the Ki-67-expressing CX3CR1 -CD8 + T subset (Figure 7, C and D). To determine functional reinvigoration of the 3 subsets of CD8 + TILs, we isolated Pmel-1 TILs 7 days after adoptive transfer, cocultured them with hgp100 peptide, and evaluated intracellular expression of effector cytokines, IL-2, TNF-α, IFN-γ, and CD107a. Although we found levels of IL-2, TNF-α, IFN-γ, and CD107a improved in all 3 subsets, a greater impact of PD-L1 blockade was seen on less-differentiated CX3CR1 -CD8 + T cell subsets ( Figure 7E).

Discussion
The blockade of immune checkpoint molecules, such as PD-1, PD-L1, and cytotoxic T lymphocyte associated protein 4, to unleash T cell-mediated antitumor immunity has demonstrated unprecedented success for the treatment of a variety of malignancies; however, only a minority of patients receive durable clinical benefit (10,11). Although the targets of immune checkpoint blockade are defined, how it works remains incompletely understood. Elucidating the phenotypic and functional heterogeneity of CD8 + TILs might lead to the development of more efficacious immunotherapies against cancers. The TME comprises effector CD8 + T cells with various states of dysfunction and differentiation, and the relationship between T cell exhaustion and differentiation status has not been defined clearly. In the present study, we sought to delineate phenotypic and functional heterogeneity of CD8 + TILs recognizing the same antigen and the relationship between T cell exhaustion and the differentiation status using a marker of T cell differentiation, CX3CR1 (16,41,42).
In a preclinical model of melanoma, we have shown that adoptively transferred antigen-specific CD8 + T cells mediate effective regression of a large established tumor and generate heterogeneous cell populations with unique differentiation states defined by the chemokine receptor CX3CR1. The less-differentiated CX3CR1subset in spleen contained the most polyfunctional cells that produced IL-2, IFN-γ, and TNF-α, in line with a recent study using viral infection models (16); however, we found this subset expressed significantly higher levels of coinhibitory receptors in murine and human melanomas. In contrast, the tumor-infiltrating CX3CR1 hi CD8 + T cell subset expressed significantly lower levels of coinhibitory receptors. Direct comparison of the 3 subsets of Pmel-1 T cells in spleen and the tumor further demonstrated the intratumor CX3CR1subset exhibited severely impaired production of these cytokines compared with their counterparts in spleen while there was no difference in cytokine production from the CX3CR1 + subsets in spleen and the tumor.
Although findings of markedly lower expression of coinhibitory receptors and potent cytotoxicity of the intratumor CX3CR1 hi subset are intriguing, our data showed this subset expressed reduced levels of CD127 along with higher levels of KLRG1 and underwent little to no expansion in response to antigen rechallenge. Furthermore, depletion of the CX3CR1 hi subset with DT injection did not decrease antitumor efficacy of adoptively transferred Pmel-1 Cd2-cre/Cx3cr1 +/DTR CD8 + T cells in vivo. These findings were rather unexpected because of the previous studies showing decreased antitumor immunity in CX3CR1 -/mice or with the use of the CX3CR1 antagonist (32)(33)(34)(35)43). A key difference between previous studies and ours, however, is that our model allows us to selectively deplete the antigen-specific CX3CR1 hi CD8 + T cell subset and does not affect other immune cells, including monocytes, DCs, γδ T cells, endogenous CD8 + T cells, and NK cells (37), that may express CX3CR1 and contribute to antitumor immunity.
This discrepancy might also be associated with the lack of CX3CL1 expression in a B16 mouse melanoma model (34), where function of CX3CR1 as a chemokine receptor might not play a significant role in antitumor efficacy. Xin et al. reported that intratumor injection of an adenoviral vector expressing mouse fractalkine (CX3CL1) gene into B16 cells significantly suppressed tumor growth in a T cell/NK cell-dependent fashion (34). Other studies using Lewis lung carcinoma also revealed that tumor growth was significantly inhibited by the increased expression of CX3CL1 in the tumor (44,45), demonstrating a key role of the CX3CR1/CX3CL1 axis in antitumor immunity in CX3CL1-expressing tumors.
Another possibility is that previous studies using CX3CR1 -/mice or a CX3CR1 antagonist affect both CX3CR1 int CD8 + and CX3CR1 hi CD8 + T cells and cannot distinguish therapeutic efficacy between these 2 functionally distinct subsets, and our results may suggest an important role of CX3CR1 int CD8 + T cells in antitumor T cell immunity. We found that the CX3CR1 int subset showed characteristics of memory and effector T cells, expressing CD27, CD127, and KLRG1. Moreover, the CX3CR1 int subset contained a substantially higher frequency of cytokine-producing cells compared with CX3CR1 -CD8 + and CX3CR1 hi CD8 + T cell subsets in the tumor. Although antitumor reactivity of the CX3CR1 int subset arising in the effector phase remains to be elucidated, 2 previous studies revealed an important role played in the memory phase by antigen-specific CX3CR1 int CD8 + T cells in infection models. Gerlach et al. revealed that virus-specific CX3CR1 int CD8 + T cells not only possess the highest self-renewal capacity but also are the dominant subset surveying peripheral tissues among all memory T cells (16). In another recent study, it was shown that CX3CR1 int CD8 + T cells form a substantial component of the memory pool in response to persistent viruses and vaccines in both mice and humans (42). In line with this, we observed that the frequency of the CX3CR1 int CD8 + T cell subsets remained elevated in the tumor but not in blood, spleen, and LNs.
The molecular and cellular mechanisms that regulate CX3CR1 -CD8 + T cells to produce CX3CR1 + subsets remain to be elucidated. In our preclinical model of ACT, we found that CD40/TLR7 stimulation facilitated adoptively transferred CX3CR1 -CD8 + T cells to generate CX3CR1 + CD8 + T cells, in line with recent reports showing a critical role of CD4 + T cell help or CD40 signaling for the generation of CX3CR1 + CD8 + T cells (36,43). Differentiation of CD8 + T cells from CX3CR1 int to CX3CR1 hi subsets is unidirectional (16,42), and CX3CR1 is stably expressed on effector CD8 + T cells, unlike some activation and proliferation markers, such as CD25, CD69, and Ki-67, which are transiently expressed after activation, suggesting potential utility of CX3CR1 on CD8 + T cells as a marker of response to immunotherapy. Indeed, emerging evidence has shown an increased frequency of peripheral blood CX3CR1 + CD8 + T cells in patients responding to immune checkpoint blockade (33,46).
Our experiments using Pmel-1 Cd2-cre/Cx3cr1 +/DTR CD8 + T cells and second adoptive transfer of sorted CX3CR1and CX3CR1 + (CX3CR1 int and CX3CR1 hi ) subsets suggest a critical role of tumor-specific CX3CR1 -CD8 + T cells in the context of ACT. This finding is in line with a growing body of evidence (G) Tumor growth curves and survival curves in C57BL/6 mice bearing B16 melanomas established for 12 days in different treatment groups. (n ≥ 6 mice per group.) Mean (±SEM). **P < 0.01, and ***P < 0.005 by 1-way ANOVA with Tukey's test (B and C), unpaired 2-tailed t test (D-F), or log-rank test (G).
indicating less-differentiated polyfunctional CD8 + T cells capable of producing IL-2, IFN-γ, and TNF-α preferentially survive and provide greater protection in viral infection (47)(48)(49) and mediate robust antitumor immunity compared with fully differentiated effector T cells secreting only IFN-γ and/or TNF-α in vivo (12,13). A potentially novel finding in the current study was that less-differentiated T cells remain the crucial CD8 + T cell subset in mediating effective regression of an established tumor among heterogeneous populations after ACT.
Phenotypic and functional characteristics of T cell exhaustion have been extensively studied in murine models of viral infections and cancer; however, the lineage origin of exhausted CD8 + TILs remains elusive. Our results are in agreement with a previous study showing that exhausted virus-specific CD8 + T cells arise from the memory precursor KLRG1 lo CD127 hi subset, rather than from a terminally differentiated KLRG1 hi subset of effector CD8 + T cells in a chronic viral infection model (50), and support the notion that T cell exhaustion is not a terminal differentiation, but rather develops progressively over time from a skewed pattern of memory T cell differentiation during chronic stimulation (9,50,51). Our findings of a crucial role played by the CX3CR1subset expressing Tcf1 in antitumor efficacy also align with recent studies showing importance of Tcf1 + CD8 + T cells with stemlike properties in the setting of chronic infection and cancer (15,(26)(27)(28)(29)(30).
Although therapeutic reinvigoration of tumor-specific CD8 + T cells revolutionized treatment for a number of malignancies, gaps in understanding the exact mechanisms that regulate CD8 + T cell responses in the TME remain a major impediment in efforts to improve immunotherapy. We found anti-PD-L1 therapy augments antitumor efficacy of adoptively transferred CD8 + T cells, and the impact of anti-PD-L1 therapy on proliferation and cytokine production was more pronounced in tumor-infiltrating CX3CR1 -CD8 + T cells expressing Tcf1 and PD-1. These findings are in agreement with emergent evidence showing that intratumor Tcf1 + PD-1 + CD8 + T cells play critical roles in tumor control and respond to anti-PD-1/ PD-L1 therapy in mouse models and patients (28,29). Collectively, these results suggest that PD-L1 blockade improves antitumor efficacy of adoptively transferred CD8 + T cells by preferentially rescuing subsets that are playing a critical role in mediating tumor regression.
In summary, the current study provides insight into the phenotypic and functional heterogeneity of tumor-specific effector CD8 + T cells, identifies distinct subsets exhibiting a high degree of exhaustion and mediating antitumor efficacy in the TME, and suggests differential impact of PD-L1 blockade on 3 subsets of CD8 + TILs defined by CX3CR1.
RT and real-time PCR analysis. Total RNA samples were isolated with TRIzol reagent (Invitrogen, Thermo Fisher Scientific) and purified according to the manufacturer's instructions. Total RNA was used for cDNA synthesis by RT-PCR with ImProm-II Reverse Transcription System (Promega) with random primers according to the manufacturer's instructions. cDNA was amplified by Maxima SYBR Green/ROX qPCR Master Mix (Thermo Fisher Scientific), with the Bio-Rad CFX96 Real-Time Detection System (Bio-Rad) and the sets of primers shown in Table 2.
In vitro cytotoxicity assay. Luciferase-expressing target cells were seeded in triplicate in 96-well plates at 1 × 10 4 cells/well at various effector-to-target ratios. Effector and tumor cells were allowed to interact at 37°C for 24 hours. A final concentration of 0.15 mg/mL D-Luciferin (Gold Biotechnology) was added to each well. Luciferase bioluminescence was measured by the Xenogen In Vivo Imaging System 100 (Perki-nElmer) and Living Image software (Caliper Life Sciences).
Patient tumor cell samples. Patients were eligible for this study if they were 18 years of age or older and were undergoing resection of metastasis and/or regional LN dissection for clinically evident regional or distant metastases of melanoma. Freshly resected tumors were sent from the surgery suite to the tissue procurement core in sterile containers. The tumor material not required for histopathologic diagnosis was placed in collecting medium, RPMI 1640 supplemented with 200 μg/mL gentamicin, 5 μg/mL ciprofloxacin, 20 μg/mL metronidazole, 100 U/mL penicillin, 50 μg/mL streptomycin, 25 μg/mL vancomycin, and 2.5 μg/mL fungizone (all from Life Technologies), as described before (54). Under sterile conditions, tumors were dissected away from adjacent normal tissue and stroma. Single-cell suspensions were obtained by mechanical dispersion consisting of two 30-minute incubations of 2.5 g minced melanoma tumor at 37°C, 5% CO 2 , in 5 mL RPMI 1640 (Life Technologies) and Tumor Dissociation Kit (Miltenyi Biotec) in C Tubes (Miltenyi Biotec) interspersed with 3 mechanical dispersions on a GentleMACS dissociator (Miltenyi Biotec). The tumor cell suspensions were then filtered through a cell strainer (70 μm; BD Biosciences). TILs and tumor cells were isolated with a 2-step density-gradient centrifugation (54). Tumor suspension in RPMI was layered onto a 2-step gradient with a lower step of 100% Ficoll and a middle step of 75% Ficoll and 25% complete medium. After 20 minutes' centrifugation at 400 g, the interfaces were collected for flow cytometric analysis (54). Statistics. Statistical analysis was performed using t test or Mann-Whitney U test for comparisons between 2 groups, 1-way repeated-measures ANOVA with Tukey-adjusted multiple comparisons for comparisons of more than 2 groups, or the Mantel-Cox method (log-rank test) for survival analysis using GraphPad Prism 7.03 (GraphPad Software). All tests were 2 sided and P < 0.05 was considered statistically significant. Data are presented as mean ± SEM.
Study approval. Human melanoma cell samples were obtained from the patients who signed the informed consent document under the protocol (HUM00054459) approved by the Institutional Review Board of the University of Michigan. All animal studies were reviewed and approved by the Roswell Park institutional animal care and use program and facilities (protocols 1316M and 1356M). All aspects of animal research and husbandry were conducted in accordance with the federal Animal Welfare Act and the NIH's Guide for the Care and Use of Laboratory Animals (National Academies Press, 2011).

Author contributions
TY contributed development of methodology, acquisition of data, analysis and interpretation of data, and writing, review, and revision of the manuscript. TH contributed development of methodology, acquisition of data, analysis and interpretation of data, and review and revision of the manuscript. TO contributed acquisition of data, analysis and interpretation of data, and review and revision of the manuscript. HS contributed acquisition of material and review and revision of the manuscript. KA contributed analysis and interpretation of data and review and revision of the manuscript. MSS contributed acquisition of material and review and revision of the manuscript. AEC contributed acquisition of material and review and revision of the manuscript. KO contributed acquisition of material and review of the manuscript. FI contributed conception and design, development of methodology, acquisition of data, analysis and interpretation of data, and writing, reviewing, and finalizing the manuscript.