OX40 agonism enhances PD-L1 checkpoint blockade by shifting the cytotoxic T cell differentiation spectrum

Summary Immune checkpoint therapy (ICT) has the power to eradicate cancer, but the mechanisms that determine effective therapy-induced immune responses are not fully understood. Here, using high-dimensional single-cell profiling, we interrogate whether the landscape of T cell states in the peripheral blood predict responses to combinatorial targeting of the OX40 costimulatory and PD-1 inhibitory pathways. Single-cell RNA sequencing and mass cytometry expose systemic and dynamic activation states of therapy-responsive CD4+ and CD8+ T cells in tumor-bearing mice with expression of distinct natural killer (NK) cell receptors, granzymes, and chemokines/chemokine receptors. Moreover, similar NK cell receptor-expressing CD8+ T cells are also detected in the blood of immunotherapy-responsive cancer patients. Targeting the NK cell and chemokine receptors in tumor-bearing mice shows the functional importance of these receptors for therapy-induced anti-tumor immunity. These findings provide a better understanding of ICT and highlight the use and targeting of dynamic biomarkers on T cells to improve cancer immunotherapy.


INTRODUCTION
Immunotherapy has become an important treatment option for cancer patients but is only effective in a minority of patients. Therefore, a deeper understanding of factors governing immune responses upon immunotherapy is required to extend clinical efficacy to the majority of patients. 1 Many studies have focused on characterizing intratumoral CD8 + T cells, 2 but system-wide profiling studies have demonstrated that systemic anti-tumor immune responses are essential for immunotherapeutic efficacy. 3 A comprehensive description of how effective cancer immunotherapy affects T cell states in the blood circulation is currently lacking. only be achieved in a minority of patients, 10 which warrants determination of the probability of a clinical response and development of more efficacious treatment options. In this respect, a better understanding of the cellular and molecular mechanisms that mediate tumor rejection could support the design of optimal treatment modalities. 11 Moreover, predictive biomarkers related to effective therapy are highly desired, especially in light of numerous clinical trials with novel (combinatorial) immunotherapeutic approaches that are ongoing. [12][13][14] In addition, methods directly targeting costimulatory receptors, such as members of the tumor necrosis factor receptor (TNFR) superfamily (e.g., CD27, CD134 [OX40], and CD137 ) expressed on tumor-specific T cells, have been developed and shown potential by itself and combined with immune checkpoint blockade. 15,16 However, knowledge to support rational application of combinatorial ICT is lacking.
Emerging single-cell technologies, such as single-cell RNA sequencing (scRNA-seq) and high-dimensional flow and mass cytometry, have provided unprecedented insight into the heterogeneity of the tumor micro-environment (TME) and its modulation by immunotherapy. [17][18][19][20][21] For example, these single-cell technologies highlight the identification of intratumoral T cells with different states of functionalities, ranging from cytotoxic to dysfunctional, 22 and the existence of biomarkers in CD8 + T cells that are associated with responsive tumor regression. 23 Predictive biomarkers in patients treated with ICT, such PD-1 and CTLA-4 blockade, have also been investigated in the systemic circulation, showing key roles of T cells, [24][25][26][27][28][29][30][31][32][33] natural killer (NK) cells, 34 and monocytes. 35,36 Here, we performed deep profiling of the systemic T cell response induced by immunotherapeutic regimens built on driving agonist signals via OX40-mediated costimulation in conjunction with blocking of the inhibitory PD-1-PD-L1 pathway. The additive effects of combination therapy over monotherapy were assessed by studying the transcriptional and proteomic changes of therapy-responsive T cell populations in tumorbearing mice using two complementary high-dimensional single-cell profiling techniques: scRNA-seq 20 and mass cytometry. 37 We found that combined ICT elicited the most profound impact on effector T cell states in the blood, characterized by dynamic kinetics and upregulation of specific biomarkers, including NK cell markers, cytotoxic molecules, and chemokine receptors. System-wide analysis revealed that therapy-responsive T cells were not limited to the blood but connected to other key immune compartments like the spleen, bone marrow, and tumor-draining lymph nodes. Analysis of human peripheral blood mononuclear cells obtained shortly after PD-1 therapy revealed similar effector T cell states in the blood that correlated with the clinical response rate. The identified biomarkers were functionally associated and implicated in treatment efficacy. This study reveals dynamic cellular changes occurring during effective ICT and the role of a set of biomarkers connected to the cytotoxic potential of CD8 + T cells, which are instrumental in tumor immunity and could be used to assess the level of immunotherapy efficiency.

RESULTS
Identification of circulating immunotherapy-responsive T cell subsets by single-cell transcriptional profiling To examine whether stimulating costimulatory receptors can improve PD-(L)1 checkpoint blockade, we challenged wildtype mice with syngeneic MC-38 tumors, which represents an ICT-sensitive colorectal cancer model. Tumor-bearing mice were subsequently treated with anti-PD-L1 antibodies, blocking the inhibitory PD-1/PD-L1 pathway, and with agonistic antibodies targeting the costimulatory receptor OX40 ( Figure 1A). Blockade of the PD-1/PD-L1 axis resulted in delayed tumor outgrowth ( Figure 1B), whereas anti-OX40 treatment did not delay tumor outgrowth (Figures S1A and S1B). Of note, addition of the TLR9 ligand CpG augmented the anti-tumoral activity of anti-OX40 (Figures S1A and S1B), which is in line with a previous report. 38 On the other hand, CpG supplementation did not improve PD-L1 blockade (Figures S1C and S1D). Strikingly, the combination of PD-L1 blockade and anti-OX40/CpG treatment, referred to hereafter as PDOX, cured the majority of mice bearing MC-38 tumors ( Figure 1B). This combination of immunotherapeutics was also most effective against established syngeneic HCmel12 melanoma tumors ( Figure 1B).
On day 18 post MC-38 tumor challenge, the percentage of CD8 + and CD4 + T cells within the circulating leukocytes in blood and spleen increased after PDOX treatment compared with no treatment and anti-PD-L1 treatment ( Figure S1E). The percentage of NK cells in the blood, however, remained similar but decreased in the spleen after PDOX treatment, while NK T cell percentages remained similar in the blood and spleen after any ICT. To identify the circulating T cell subsets associated with effective checkpoint therapy, we isolated CD4 + and CD8 + T cells from the peripheral blood on day 18 post MC-38 tumor challenge of treated (anti-PD-L1, anti-OX40/CpG, and PDOX) and untreated mice. Per condition, more than 1,000 cells were analyzed by scRNA-seq with a coverage of 60,000 reads per cell. The subpopulation structure of the circulating T cells was defined by pooling data from the different treatment groups, representing 5,600 cells in total, and using Seurat package analysis to identify transcriptional clusters. Six distinct T cell clusters could be identified, consisting of three CD4 + and three CD8 + T cell clusters ( Figures 1C-1F). Two clusters (CD4-T3 and CD8-T3) were over-represented in the PDOX group ( Figures 1D-1H), and both of these T3 clusters were characterized by expression of Id2 and Lgals1 transcripts encoding for the transcription factor ID2 and Galectin-1, respectively ( Figures 1F and 1G). Other gene transcripts over-represented in the CD4-T3 and CD8-T3 clusters were Cxcr3 (coding for the chemokine receptor CXCR3) and Ly6a (coding for Sca-1) (Figure 1G). Transcripts linked to NK cell receptors and cytotoxicity, including Klrk1 (coding for NKG2D protein), Klrc1 (coding for NKG2A), Klrg1 (coding for KLRG1), and Gzma, Gzmb, and Gzmk (coding for Granzyme A, B, and K, respectively) were mostly enriched in CD8-T3 ( Figure 1G).
Circulating therapy-responsive T cell subsets display effector cell properties with increased cytotoxic and migratory capacity To validate the association of the Id2 transcripts with transcripts of the Klr genes in subsets of the Id2 + cells at the protein level, the KLR family members and other effector T cell markers were co-stained with ID2 in circulating T cells obtained from tumor-challenged PDOX-treated mice. Within the ID2 + CD8 + T cells, expression of KLRG1, NKG2A, and NKG2D was substantial, whereas ID2 À CD8 + T cells lacked expression of these KLR family members. ID2 + CD4 + T cells also expressed KLRG1 and NKG2A, albeit to a lesser extent as ID2 + CD8 + T cells. ID2 À CD4 + T cells were devoid of KLRG1 or NKG2A, while NKG2D was absent on ID2 + CD4 + and ID2 À CD4 + T cells (Figures 2A and S2A).
To further characterize the total ID2 + subset in CD4 + and CD8 + T cells, we examined other surface markers associated with NK cell receptors on effector T cells, including the O-glycan form of CD43 (sialoforine), which is known as a CD43 isoform expressed transiently by effector T cells. 42 This activation-associated isoform of CD43 is generated by posttranslational glycosylation; hence, this form cannot be distinctively identified in scRNA analysis. However, the glycosylated form of CD43 can be visualized by the 1B11 antibody clone, while the antibody clone S11 recognizes CD43 regardless of glycosylation and reacts with virtually all T cells, activated or not ( Figure S2B). Strikingly, the majority of ID2 + CD8 + T cells expressed the hyperglycosylated CD43 1B11 isoform, in contrast to ID2 À CD8 + T cells, and similar results were found for ID2 + CD4 + and ID2 À CD4 + T cells ( Figure 2A).

Dynamic induction of therapy-responsive T cell subsets
To gain insight into the dynamics of the therapy-responsive T cell subsets, we longitudinally followed the CD43 1B11+ , NKG2A + , and KLRG1 + T cell subsets in the blood circulation of ICT-treated MC-38 tumor-bearing mice. Anti-OX40/CpG treatment, but not anti-PD-L1 treatment, increased the CD43 1B11+ CD8 + T cell subset on day 13 post tumor challenge (6 days after the start of the anti-OX40/CpG treatment), while at later time points these treatments resulted in a comparable increase compared with untreated mice. PDOX treatment elicited a much stronger increase in CD43 1B11+ CD8 + T cells on day 13 compared with anti-OX40/ CpG and anti-PD-L1 treatment, and this increase was even more pronounced on day 18 ( Figure 3A). On day 25 post tumor challenge, the percentage of CD43 1B11+ CD8 + T cells was decreased but still higher than in anti-OX40/CpG-and anti-PD-L1-treated mice. As projected based on the expression profile, the NKG2A + CD8 + and KLRG1 + CD8 + T cell subsets showed similar kinetics ( Figure 3A). PDOX treatment also induced the highest level A B C D F E Figure 2. Circulating therapy-responsive T cell subsets display effector cell properties with increased cytotoxic and migratory capacity (A) Representative histogram plots of NKG2A, NKG2D, KLRG1, and CD43 1B11 expression on gated ID2 À CD8 + or ID2 + CD8 + and gated ID2 À CD4 + or ID2 + CD4 + T cell populations residing in the blood circulation of MC-38-challenged PDOX-treated mice. Numbers indicate average mean fluorescence intensity. (B) Percentage of CD43 1B11+ cells within the total CD8 + and CD4 + T cell population in the blood of untreated and ICT-treated groups. (C) tSNE plots of flow cytometric data visualizing NKG2A, NKG2D, KLRG1, and CD43 1B11 expression (red) on CD8 + and CD4 + T cells in the blood from untreated and PDOX-treated groups. The blue/red tSNE plot indicates cell origin for CD8 + and CD4 + T cells of the untreated and PDOX-treated group, respectively. (D) Representative histograms (left) and quantification of fluorescence intensity (right) of granzyme B expression in blood circulating CD43 1B11À CD8 + and CD43 1B11+ CD8 + and CD43 1B11À CD4 + and CD43 1B11+ CD4 + T cell populations of MC-38-challenged PDOX-treated mice. To dissect the influence of the tumor on the dynamics of CD43 1B11+ CD8 + and CD43 1B11+ CD4 + T cells, we treated non-tumor-bearing mice (mock challenged) with anti-OX40/CpG, anti-PD-L1, or PDOX. While induction of CD43 1B11+ T cells induced by the single therapies in the blood circulation were comparable between tumor-bearing and non-tumor-bearing mice, PDOX treatment amplified the response in tumor-bearing mice, indicating that tumor antigens and/or tumor-associated inflammation drives the synergy between anti-OX40/CpG and anti-PD-L1 blockade ( Figure 3C).
A correlation between the level of CD43 1B11+ CD8 + T cells circulating in the blood and MC-38 and HCmel12 tumor survival could be established ( Figure 3D). Together, these data indicate that the therapy-elicited effector T cells, identified by the activation markers CD43 1B11 , NKG2A, and KLRG1, are characterized by dynamic expansion/contraction kinetics resembling those of vaccine-or infection-provoked T cell responses.

System-wide induction of immunotherapy-responsive T cell subsets
To interrogate whether the PDOX therapy-responsive effector T cell states were elicited system wide and whether these were interconnected, we assessed the phenotype of ICT-induced T cell states in lymphoid tissues on day 18 post MC-38 tumor challenge by CyTOF mass cytometry with 34 cell-surface markers, which allowed identification of T cell signatures in depth ( Figure 4A). The marker panel included markers for effector T cell activation, differentiation, and migration, such as CD43 1B11 , NKG2A, KLRG1, CXCR3, and CD62L and the ectoenzymes CD38 and CD39 ( Figure 4B). Viable CD4 + and CD8 + T cells were analyzed by hierarchical stochastic neighbor embedding (HSNE) using Cytosplore 44 and Cytofast. 45,46 We selected clusters based on the significant difference (p < 0.05) and abundance (>1%). Remarkably, in the blood compartment, all CD8 + and CD4 + T cell clusters that were significantly higher in PDOX-treated mice expressed CD43 1B11 (Figures 4C and S3). Two of the four CD43 1B11+ CD8 + T cell clusters were most abundant in the PDOX-treated group compared with all other groups and co-expressed KLRG1, CD38, CD39, PD-1, or LAG-3 (blood cluster CD8-5) or co-expressed the same markers and NKG2A and ICOS (CD278) (blood cluster CD8-11). Three CD43 1B11+ CD4 + T cell clusters, which were more abundant in the PDOX-treated group compared with all other groups, co-expressed CXCR3 and ICOS and differentially expressed PD-1, CD38, and CD39. In the spleen, one CD8 + T cell cluster expressing CD43 1B11 (spleen CD8-13) was most abundant in PDOXtreated mice, and, similar to blood cluster CD8-5, these cells Article ll OPEN ACCESS co-expressed KLRG1, CD38, CD39, PD-1, and LAG-3. Four splenic CD4 + T cell clusters expressing CD43 1B11 were higher in PDOX-treated mice and highly similar to the CD43 1B11 -expressing CD4 + T cells in the blood. In the tumor-draining lymph node (tdLN) and bone marrow, significantly higher frequencies of CD43 1B11+ CD4 + T cells but not CD43 1B11+ CD8 + T cells in the PDOX treated mice were detected, and these cells differentially expressed ICOS, CD38, CD39, and KLRG1 ( Figures 4C and  S3). Thus, besides also residing in the blood circulation in other lymphoid compartments, CD8 + and CD4 + T cell clusters expressing CD43 1B11 and co-expressing NK cell receptors, chemokine receptors, and ectoenzymes are elicited that are connected to effective immunotherapy.
To determine the correlation between the identified therapyresponsive T cell clusters across all lymphoid tissues via an unbiased approach, correlation analyses were performed system wide ( Figures 4D and S3). The therapy-responsive CD4 + and CD8 + T cell clusters in the blood are closely related to those in the spleen (r > 0. 70), and blood CD4 + T cell clusters are also connected to the bone marrow CD4 + T cell clusters. Moreover, lymph node CD4 + T cell cluster 7 relates to splenic CD4 + T cell clusters. Together, these data show an interconnectivity between therapy-responsive T cell clusters residing in different lymphoid tissues, indicating induction of system-wide effects of ICT enabling efficient tumor immunity.
Identification of NK cell receptors expressing CD8 + T cell subsets in the blood circulation of PD-1 therapyresponsive patients To determine whether corresponding therapy-responsive T cell subsets could be identified in the blood circulation of patients receiving ICT, we evaluated anti-PD-1-responding and non-responding patients with melanoma or non-small cell lung cancer (NSCLC) by mass cytometry (Figures 5A and S4A-S4C). Peripheral blood was collected before and 2 weeks after treatment, and peripheral blood mononuclear cells (PBMCs) were stained with a panel of antibodies that incorporated detection of NK cell receptors expressed by activated human CD8 + T cells; i.e., KLRB1, KLRG1, and CD56 ( Figure 5B). Data analysis by Cytosplore revealed distinct CD8 + T cell clusters that were increased in the anti-PD-1 responding group compared with the anti-PD-1 nonresponders ( Figures 5B-5D and S4C-S4F). These clusters expressed KLRG1, CD29, and CD44 and exclusively expressed KLRB1, CD56, or CD45RO. It is also worth noting that an additional CD8 + T cell cluster, also expressing the NK cell receptors KLRG1 and KLRB1, was elevated in 4 of 8 PD-1 responders (Figures S4D-S4F). To validate these observations, we performed FlowSOM clustering and projected these on opt-SNE dimensionality reduction plots (Figures S5A and S5B). Corroborating the Cytosplore-based data analysis, identical CD8 + T cell clusters were identified (i.e., positive for KLRG1, CD44, CD29, and CD56 or KLRB1) that were increased in the blood of PD-1-treated patients (Figures S5C and S5D).
Next, we evaluated the survival and correlation of the therapyresponsive gene signatures and marker genes discovered here in large cohorts of patients with skin cutaneous melanoma using Gene Expression Profiling Interactive Analysis. 47,48 Higher expression of the gene signatures of clusters 2, 7, and 8 were related to better survival ( Figure 5E). Higher expression of the KLR family members (KLRB1, KLRG1, KLRC1, and KLRK1), granzymes (GZMB and GZMK), as well as CXCR3 and ID2 also associated with a higher survival rate ( Figure S5E). Correlation analysis of these genes indicated a strong association of the granzyme family with the KLR family members and with CD8 ( Figures 5F and S5F). Moreover, non-biased protein interaction analysis confirmed the connection between the therapy-responsive markers, indicating an underlying transcriptional program ( Figure 5G). Thus, comparable with the findings in experimental settings, effective ICT in patients correlates with increases in CD8 + T cell subsets characterized by programming for cytotoxic effector function.
Expansion of functional therapy-responsive CD8 T cell subsets in the TME and draining lymph nodes We next analyzed the therapy-responsive T cells in the TME of MC-38-challenged mice. Because PDOX treatment is very effective, treatment started at later time points (around day 10 with anti-OX40/CpG followed by PD-L1 blockade) to obtain sufficient tumor material for analysis. Compared with untreated animals, PDOX treatment increased the percentage of leukocytes in the TME, which was mainly caused by an increase in CD8 + T cells, and this correlated with an increase in tumor-specific CD8 + T cells (Figures 6A and S6A). In the TME of HCmel12-challenged mice, similar data were obtained ( Figure S6B). Detection of CD8 + T cells by immunofluorescence showed that PDOX treatment promoted higher numbers of tumor-infiltrating CD8 + T cells (Figure 6B). The increase in CD8 + T cells coincided with a decrease in FOXP3 + CD4 + regulatory T (Treg) cells; hence, the CD8 + /Treg ratio was profoundly increased in the TME because of PDOX treatment ( Figure 6A). The Treg cells in the peripheral blood were, however, relatively increased by the PDOX treatment. OX40 expression was not detected on intratumoral Treg cells upon PDOX therapy ( Figure S6C), which can be attributed to obstruction of the injected anti-OX40 antibody and/or depletion Article ll OPEN ACCESS of OX40 high -expressing Treg cells. 49,50 Like CD8 + T cells, helper CD4 + T cells (FOXP3 À ) were significantly increased in the TME ( Figures 6A and S6A).
To assess the effector phenotype and potential of the tumor-infiltrated T cells, cell-surface expression, proliferation, granzyme expression, and cytokine production were evaluated. In untreated and PDOX-treated mice, the majority of tumor-infiltrated CD8 + T cells expressed CD43 1B11 , NKG2A, and/or KLRG1 ( Figure S6D). The percentage of CD8 + T cells lacking any of these markers was decreased by PDOX treatment (p = 0.02, Mann-Whitney U test), indicating more activated CD8 + T cells in the TME. CD43 1B11 and KLRG1 were also expressed by CD4 + T cells in the TME, and PDOX therapy mainly elevated CD43 1B11+ CD4 + T cells ( Figure S6A). Moreover, the proliferation marker Ki-67 was abundantly expressed by tumor-infiltrating CD8 + T cells expressing CD43 1B11 , KLRG1, and NKG2A, whereas PDOX therapy only significantly increased KLRG1/Ki-67 double-positive cells (Figures 6C and  S6E). Co-staining of Ki-67 with CD43 1B11 , KLRG1, and NKG2A of blood circulating CD8 + T cells, however, indicated a particular increase in Ki-67-co-expressing cells after PDOX therapy, suggesting that such marker combinations could be used as robust biomarkers (Figures 6C and S6E). A higher increase in Ki-67 + CD43 1B11+ and Ki-67 + CD43 1B11+ co-expression after therapy was also observed within circulating FoxP3 À CD4 + T cell subsets compared with tumor-infiltrating FoxP3 À CD4 + T cells ( Figure S6E). Consistent with increased granzyme B levels in circulating CD8 + T cells after therapy ( Figure 2D), granzyme B expression in CD43 1B11+ tumor-infiltrating CD8 + T cells was increased by PDOX treatment ( Figure 6D). PDOX therapy also elicited higher percentages of polyfunctional cytokine-producing CD8 + and CD4 + T cells (Figures 6E  and S6F).
To better understand local and systemic immune responses, we compared tdLNs with non-draining lymph nodes (ndLNs). Here, we noticed that, in untreated animals, the tdLNs were enlarged compared with ndLNs ( Figures 6F and S6G-S6I). PDOX therapy enhanced the absolute numbers in tdLNs and ndLNs and in particular increased the magnitude and percentages of CD43 1B11 -and KLRG1-expressing CD8 + and CD4 + T cells in the tdLNs (Figures 6F, S6G, and S6I-S6K). Accordingly, PDOX treatment also enhanced the magnitude of NKG2A-and NKG2D-expressing CD8 + T cells in tdLNs ( Figures S6G, S6I, and S6J). In line with the increase in cytotoxic and cytokine polyfunctional CD8 + T cells, depletion of CD8 + T cells completely dismantled the efficacy of PDOX treatment ( Figure 6G). Depletion of CD4 + T cells, despite being present in the TME and having an activated phenotype, did not impact tumor control in PDOX-treated mice ( Figure 6G). Thus, PDOX-therapy-responsive CD8 + T cells are functionally effective and proliferate in the peripheral blood, TME, and draining lymph nodes.

Functional receptor expression on therapy-responsive T cell subsets affects expansion and tumor infiltration
To functionally assess the relevance of the elevated levels of NK cell receptor expression on therapy-responsive CD8 + T cells, we targeted NKG2D, which is in contrast to the inhibitory NKG2A molecule known as a molecule with the capacity to provide costimulatory signals to T cells ( Figure 7A). 51 Like NKG2A, NKG2D is also upregulated in the TME ( Figure 7B). Blockade of NKG2D reduced the effectiveness of the PDOX treatment in controlling tumor outgrowth, whereas NKG2D blockade in untreated mice had no implication ( Figure 7C). This effect of NKG2D blockade on PDOX treatment was related to a diminution of NKG2A + CD8 + T cells in the blood circulation and TME (Figure 7D). Moreover, NK cell depletion did not impact tumor control of PDOX therapy, indicating that PDOX therapy relates primarily to CD8 + T cell-mediated effects ( Figure S7A). We conclude that NK cell receptor-expressing T cell subsets are instrumental for PDOX therapeutic efficacy with NKG2D, supporting systemic stimulation of the therapy-responsive effector CD8 + T cells.
To functionally assess whether CD43 expression is critical for the efficacy of ICT, we examined PDOX responsiveness in settings of CD43 availability and absence. For this, wild-type mice and mice deficient in the Spn gene (coding for CD43) were challenged with MC-38 tumor cells and left untreated or treated with PDOX ( Figures 7E and 7F). Whereas CD43 proficient mice showed the anticipated therapeutic efficacy of PDOX treatment upon tumor challenge, mice deficient in CD43 could not control MC-38 tumor outgrowth despite PDOX treatment ( Figure 7G). Remarkably, CD43 deficiency did not affect the percentage of NKG2A + CD8 + T cells in the blood but rather resulted in a diminution of NKG2A + CD8 + T cells in the TME ( Figure 7H), indicating that tumor migration of therapy-responsive T cells is regulated by CD43, which is in line with the ability of the CD43 1B11 isoform to function as a ligand for the cell adhesion molecule E-selectin (CD62E). 52 ID2 expression in tumor-infiltrating CD8 + T cells was not affected by PDOX treatment in wild-type or CD43-deficient mice ( Figure S7B).
To test whether the chemokine receptor CXCR3, expressed on the CD43 1B11+ T cell subset and known to mediate adhesion induction, 53 was also implicated in tumor migration, we blocked this chemokine receptor by antibodies provided during PDOX treatment of tumor-challenged mice ( Figure 7I). Obstruction of CXCR3 resulted in reduced efficacy of PDOX to control tumor outgrowth ( Figure 7J). This effect of CXCR3 blockade was related to decreased infiltration of CD43 1B11+ CD8 + T cells in the TME as well as reduced tumor-infiltrating NKG2A + CD8 + T cells and tumor-specific CD8 + T cells, while the CD43 1B11+ CD8 + and NKG2A + CD8 + T cells were unaffected in the blood circulation ( Figures 7K and S7C). Moreover, CXCR3 blockade prevented NKG2D + CD8 + T cell entry into to the TME (Figures 7K and 7L). Altogether, we conclude that CD43 and CXCR3 mediate tumor migration of therapy-responsive CD8 + T cells, which is critical for these cells to achieve tumor control.

DISCUSSION
A major challenge for ICT is to overcome the substantial variability of this therapy through identification of predictive biomarkers. Ideally, such biomarkers can be interrogated in easily accessible compartments, such as the peripheral blood, while accurately reporting therapy responses in the TME. Here, we demonstrated, in two different murine tumor models, MC-38 (colorectal carcinoma) and HCmel12 (melanoma), that a dynamic and systemic T cell response develops upon efficient immunotherapy, which is characterized by an interconnected gene signature related to cytotoxicity and migration. Analyses of peripheral blood samples from PD-1 blockade therapyresponsive and unresponsive patients highlight the potential clinical utility of the cytotoxic gene signature consisting of several NK cell markers expressed by CD8 + T cells.
Combination of a PD-1/PD-L1 pathway antagonist and an OX40 agonist with CpG was remarkably efficient in eradicating developing tumors. These data show the possible benefit of using combinatorial treatment of already used therapeutics in patients (e.g., anti-OX40 54 and anti-PD-L1 55 ) and emphasize the potential of this combination, as has also been observed in other mouse tumor models. 56,57 This harmonizing effect of the combinatorial treatment was deciphered by complementary high-dimensional single-cell technology platforms. scRNA-seq and mass cytometry highlighted functionally dynamic CD4 + and CD8 + T cell states that were characterized by NK cell receptor expression and expression of adhesion/ migration receptors. The kinetics of the therapy-responsive T cells (i.e., sharp expansion followed by a contraction phase, which is typical for acute infection) may reflect temporal systemic activation, but T cell activation may be ongoing in the TME. Although less impressive, this expansion was also observed following OX40/CpG treatment, while PD-L1 blockade seems to mainly facilitate this T cell expansion in a combination setting. Additional PD-1 upregulation occurring following OX40 triggering may thus be efficiently counteracted and allows T cells to rapidly expand and differentiate to cytotoxic effector T cells. Although expression of OX40 on CD4 + T cells is higher compared with CD8 + T cells, this mechanism is likely to occur in CD4 + and CD8 + T cells because direct triggering of OX40 on CD4 + and CD8 + T cells results in effector T cell formation. 58,59 OX40-activated CD4 + T cells may additionally help the CD8 + T cells with their expansion and differentiation. 60 Nevertheless, we observed that depletion of CD4 + T cells had no effect on tumor control by PDOX treatment, which could be explained by depleting CD4 + helper T cells and inhibitory Treg cells. Another mechanism that could potentiate the PDOX combination may be related to intensification in OX40 expression on responding T cells after targeting of the PD-1/PD-L1 pathway. [61][62][63] The addition of CpG to OX40 was synergistic, and this is likely related to enhanced upregulation of OX40 on T cells because of enhanced cytokine secretion by macrophages and dendritic cells. 38 CpG is also able to upregulate costimulatory molecules such as CD70, CD80, and CD86, 64 which may empower OX40-mediated costimulation. 65 Although CpG did not enhance PD-L1 blockade in the subcutaneous (s.c.) administration setting we used, intratumorally provided CpG may synergize with blockade of the PD-1-PD-L1 pathway 66 by enhancing dendritic-cell-mediated cross-presentation of tumor antigens. 67 Whether triggering of OX40related receptors, such as 4-1BB or CD27, belonging to the costimulatory members of the TNFR superfamily, has similar effects as described here remains to be determined.
Our study also indicated that the efficiency of the immunotherapeutic treatment is mirrored by the induction of peripheral T cells that could be identified by cell surface markers. The cell-surface expression of the hyperglycosylated form of CD43 and the NK cell receptors NKG2A, NKG2D, and KLRG1 on CD8 + T cells was a strong signature for these cells as an indicator for cytotoxic effector function based on the co-expression with granzymes. In line with this are findings in the TME of melanoma patients, where KLRG1 was found to be expressed in the cytotoxic T cell compartment. 22 Moreover, fate-tracking studies in mice indicated that KLRG1 + CD8 + T cells display developmental plasticity and that basically these cells can differentiate into all memory T cell lineages, which underscores the value of this  68 Indeed, KLRG1 + CD8 + T cells are excellent predictors of the effectivity of cancer vaccines. 69 The combination of KLRG1 with the proliferation marker Ki-67 may provide an even better biomarker for circulating CD8 + T cells responding to immunotherapy. In the PBMC compartment of patients we also observed KLRB1 (CD161), which was recently discovered as a functional marker on human CD8 + T cells in the TME of glioma. 70 Besides effects on CD8 + T cells, we also observed system-wide expansion and contraction of CD4 + T cells expressing CD43 1B11 , KLRG1, and CXCR3, which is in line with studies of human peripheral blood of anti-PD-1-treated patients in which CXCR3 + CD4 + T cells correlated with a positive clinical outcome. 71 Remarkably, transcripts of encoding NK cell receptors (Klrc1, Klrk1, and Klrg1) were observed in CD4 + T cells, but at the protein level only KLRG1 was found to be highly expressed, suggesting differential posttranscriptional regulation. Whether NKG2A and NKG2D, which likely have opposing functions with respect to T cell activation, are both concurrently functional is complex, given that the ligands of these receptors (i.e., HLA-E, and MICA/B, ULBP1-6, respectively) can be inducibly expressed in healthy and cancer tissue. 72,73 The blocking studies targeting the NK cell receptor NKG2D and the chemokine receptor CXCR3, together with the CD43deficient setting, identified the importance of simultaneous induction of cytotoxic and migratory properties. CD43 has also been involved in T cell activation, where its effects could be either costimulatory or negative regulatory; [74][75][76] it may thus be determined spatiotemporally when and where CD43 exerts its pleiotropic effects. Although the combination therapy we used was already effective, blockade of NKG2A in more resistant tumors, which is known to synergize with PD-1 blockade, 77 may further improve anti-tumoral responses. Another upregulated molecule upon PDOX treatment that could be targeted to improve outcome is the costimulatory molecule ICOS, which is linked to enhancing PD-1-targeted immunotherapy in mice 78 but also in responsive patients. 79 Additional clinical studies are required to determine the predictive significance of our findings. Studies with agonistic antibodies targeting costimulatory receptors such as OX40 in combination with CpG and inhibitory immune checkpoint blockade may be of particular interest. Recent studies already indicated the correlation between effector/effector-memory CD8 + T cell responses in the peripheral blood and clinical responses to immune checkpoint blockade. 29,32,33 Our work highlights these studies and additionally proposes that effective combinatorial therapy is more powerful in induction of such peripheral T cell responses. Moreover, we show that NK cell receptors and the chemokine receptor CXCR3 are, besides biomarkers, also functional markers, and the targeting thereof may further improve the clinical response. Detection and targeting of glycosylated CD43 isoforms in ICT-treated patients may have therapeutic potential as well.
In conclusion, we provided evidence of an immune signature that relates to effective immunotherapy. Future studies entailing a systematic and multicenter cohort of patients with different cancer types for which a combinatorial anti-TNFR family member with anti-PD-1/PD-L1 treatment is approved remains needed. A prediction signature might then be directly used in clinical practice to stratify different levels of effectiveness of treatments.
Limitations of the study Although the PDOX treatment has been analyzed extensively here in the mouse models, this treatment was not tested in a clinical trial. In addition, the number of PD-1-treated patients included in the study is limited, and kinetics of human CD8 + T cell responses were not assessed. Furthermore, it remains unclear whether the impact of OX40 agonism can be solely attributed to enhancing the CD8 + T cells because depletion of CD4 + T cells did not abrogate PDOX efficacy. A direct impact on helper CD4 + T cells was observed but could have been counteracted by depletion of Treg cells in the circulation and in the tumor. The reduction in Treg cells in the tumor after PDOX therapy could be partly caused by the depletion effect of OX86, which may enhance the effectiveness of the PDOX therapy.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following: