IL-10 secretion by CD8 T cells orchestrates early NK cell recruitment and antitumor immunity after anti-VEGFR-2/PD-1 therapy.

Tumor angiogenesis is essential for tumor growth, therefore, targeting of the vascular endothelial growth factor (VEGF) and its receptor (VEGFR) signaling pathway was promoted as a means to develop effective antitumor therapies. Early development of anti-angiogenic agents (AAs) was aimed primarily at blocking tumor blood vessel formation. However, it is now understood that VEGF signaling pathways also affect how immune cells behave in tumors. VEGF has been found to impair CD8 T cell proliferation and function, promote T cell exhaustion, and increase the expression of inhibitory checkpoint molecules.^[1][3] Conversely, antiangiogenic therapy improves the response of mouse tumors to anti-programmed cell death protein-1 (anti-PD-1), further substantiating the rationale for combining anti-VEGF and anti-PD-1 therapies in mice.^[4] The above observations provided the rationale to evaluate the combination of antiangiogenic therapies and immune checkpoint inhibitors in patients. These combination therapies demonstrate clinical activity across multiple tumor types and have been approved to treat hepatocellular carcinoma, renal cell carcinoma, and endometrial cancer. However, we do not fully understand how immune mechanisms mediate this therapeutic response. This is particularly important for colorectal cancer (CRC), where immune checkpoint inhibition has limited effectiveness outside of the microsatellite-instable subsets.^[5] Recently published clinical data supporting a benefit from combined anti-VEGF and PD-1 inhibition for patients with CRC illustrate how crucial it is to fully understand how anti-angiogenesis therapy modifies the antitumor immune response in this disease.^[6]

Natural killer (NK) cells and CD8 T cells are key cytotoxic immune cell populations that contribute critically to tumor immune surveillance and eradication. However, there is still little clarity regarding how CD8 T cells and NK cells are recruited and functionally coordinated within the tumor microenvironment (TME).

Interleukin-10 (IL-10) has generally been regarded as an immunosuppressive cytokine in TMEs; it is associated with inhibition of both T cells and NK cells’ function and tumor immune evasion.[7 8] However, IL-10 may stimulate immune activity under certain conditions, particularly when it is transiently produced by activated immune cells.^[9][11] The role of IL-10 in facilitating immune stimulation when employed with combination immunotherapy and the potential impact it has on early immune events associated with antiangiogenic therapies that impact the TME, are not yet fully understood.

In this study, we examined the immune responses induced by combined VEGFR-2 and PD-1 blockade in two preclinical models of CRC, with a focus on early intratumor changes. We show that VEGFR-2 inhibition directly affects CD8 T cells, leading to their rapid expansion and the induction of IL-10 before the development of full effector function. CD8 T cell-derived IL-10 unexpectedly promoted the recruitment of NK cells into tumors, initiating a cytokine cascade that supported macrophage maturation and secondary CD8 T-cell infiltration. This immune sequence was required for the antitumor activity of combined VEGFR-2 and PD-1 blockade. These findings identify an early immune program triggered by angiogenesis inhibition and provide insight into how VEGF-targeted therapy can support effective antitumor immunity in preclinical CRCs.

Results

CD8 T cells are needed for anti-VEGFR-2/anti-PD-1 efficacy against CRC tumor model

To investigate the immune events occurring in tumors following combination treatment (Bith) with anti-VEGFR-2 (αR2) and anti-PD-1 (αPD1) antibodies, we assessed the response of CT26 and MC38 CRC models. Both models showed modest efficacy of the anti-PD-1 monoclonal antibody therapy alone, with a slight slowdown in tumor growth. Anti-VEGFR-2 had no efficacy in CT26 while it slightly slowed tumor growth in the MC38 model. In contrast, significant responses were observed with the combination of anti-VEGFR-2 and anti-PD-1 antibodies (bith) showing a synergistic effect in the CT26 model and an additive effect in the MC38 model (figure 1 A; online supplementary figure S1, A-B). We also evaluated tumor growth in an orthotopic implantation model of CT26-luciferase cells in the caecum, where the combination therapy also demonstrated significant efficacy (figure 1C; online supplementary figure S1B). To confirm our data, we also studied the response using anti-VEGF-A and found that it worked to a similar extent (online supplementary figure S1D). Consequently, using nude mice, we demonstrated that T lymphocytes are crucial for the efficacy of the bitherapy, as it had no effect on tumor growth in the absence of T lymphocytes in both CT26 and MC38 tumors (online supplementary figure S1E and F).

To broadly assess the impact of the bitherapy on the immune profile of the TME, we performed a NanoString analysis to analyze the transcriptome of CT26 tumor bulk samples collected 9 days after the initiation of the combination therapy. When looking at genes associated with T cell functions, more than 38% of them were significantly upregulated in the bitherapy group (figure 1C; Online supplemental table S1), suggesting activation of T cells induced by the bitherapy.

Accordingly, flow cytometry analysis of CT26 tumors revealed a very rapid and significant increase in CD8 T cell numbers 1 and 9 days after the first injection of anti-VEGFR-2 treatment (figure 1, D and E), whereas the CD4 T cell population remained unchanged (online supplementary figure S2, A and B). To explain such a rapid increase in response to anti-VEGFR-2, we assessed the CD8 T cell proliferation 1 day after the treatment. We found that CD8 T cells exhibited significantly greater expression of Ki-67 in treated tumors (figure 1F). Interestingly, 10% of CD8 T cells from CT26 tumors expressed VEGFR-2, suggesting a direct on-target effect of VEGFR-2 inhibition on CD8 T cells (figure 1G). Additionally, most tumor antigen-specific CD8 T cells in CT26 tumors expressed VEGFR-2 (figure 1H) and were increased after the first anti-VEGFR2 treatment (online supplementary figure S2C). Accordingly, only the transfer of CD8 T cells expressing VEGFR-2 into CT26 tumor-bearing Nude mice allowed an effect of the bitherapy on tumor growth whereas the transfer of VEGFR-2− CD8 T cells had no impact (online supplementary figure S2D). As the VEGF/VEGFR axis is known to suppress T cell proliferation, these data strongly suggest a direct effect of the anti-VEGFR-2 onto the CD8 T cell proliferation.^[12]

Interestingly, although the CD8 T cell population increased in the tumor as early as 1 day post-treatment, these cells did not show at this time point enhanced activation (as measured by CD69 expression) or increased cytotoxicity (as indicated by granzyme B expression) (figure 1, I and J). However, 9 days post-treatment, CD8 T cells exhibited increased activation and higher levels of granzyme B after the bitherapy (figure 1, K and L), suggesting that different mechanisms regulate both their proliferation and activation over time.### CD8 T cells play an important role early after treatment initiation

To further confirm the role of CD8 T cells in the immune response to the combination therapy, we conducted tumor growth assays using a CD8β-depleting monoclonal antibody that efficiently targets and eliminates CD8 T cells within 24 hours (online supplementary figure S2E). CD8 T cell depletion completely abrogated the efficacy of the combination therapy in both the CT26 and MC38 models, underscoring the crucial role of CD8 T cells in the response to treatment (figure 1M and online supplementary figure 2F). We also investigated the temporal importance of CD8 T cells in the efficacy of the combination therapy by depleting CD8 T cells either at the start of treatment (D0) or 3 days after treatment initiation (D3). While CD8 T cell depletion 3 days after treatment initiation had a very limited impact on tumor growth inhibition, depletion at the start of treatment drastically impaired the ability of the therapy to control tumor growth in both CT26 and MC38 tumor models (figure 1N; Online supplementary figure S2 G and H). These findings suggest that CD8 T cells play a critical antitumor role within the first 3 days after anti-VEGFR-2 treatment.

From these findings, we aimed to investigate the early role of CD8 T cells. We found by flow cytometry that early CD8 T cell depletion reduced the number of intratumoral NK cells 1 day after the treatment in both CT26 and MC38 tumors (figure 1O and online supplementary figure S2 I). NK cells do not express VEGFR-2, eliminating the possibility of a direct effect of anti-VEGFR-2 on these cells (online supplementary figure S2J). CD8 T cell depletion did not affect NK cell proliferation, suggesting that CD8 T cells might facilitate NK cell recruitment into the tumor (online supplementary figure S2K). Taken together, these data further establish CD8 T cells as direct targets of the anti-VEGFR-2 treatment and suggest that they play a pivotal role in the subsequent early increase in NK cells following treatment.

In summary, our data demonstrate that the combination therapy of anti-PD-1 and anti-VEGFR-2 is dependent on CD8 T cells for effective tumor growth inhibition in preclinical models of colorectal carcinoma. Moreover, we reveal an unexpected relationship between CD8 T cells and NK cells within the TME as early as 1 day after treatment.### Characterization of CD8 T cells and NK cells’ putative interactions after anti-VEGFR-2

To gain a deeper understanding of the relationship between CD8 T cells and NK cells in CT26 tumors, we performed a single-cell RNA sequencing (scRNA-seq) analysis 1 day after the first anti-VEGFR-2 treatment (figure 2A). After quality control, cells clustered into 16 distinct cell types (online supplementary figure S3).

We focused on the four clusters of NK cells (NK1, NK2, NK3 and NK4) and the two clusters containing CD8 T cells (Tcells1 and Tcells2). We showed that the four NK cell clusters correspond to cycling NK cells (Mki67, Birc5 and Nusap1) (NK4), secreting NK cells (Ccl3, Ccl4, Ccl5) (NK3), cytotoxic NK cells (Gzmd, Gzme, Gzmc) (NK2) and finally NK cells with no defining features except expression of the classical NK cells markers Eomes and Ncr1 hereafter referred to as resting NK cells (NK1) (figure 2, B and C; Online supplementary figure S4C). Regarding T cells, the Tcells2 cluster corresponds to cycling CD8 T cells (Mki67, Birc5 and Cd8a) and Tcells1 cluster corresponds to mixed T cell population (Cd4, Cd8a, Cd3e) (figure 2, B and C).

We used CellChat to identify putative intercellular communication pathways between CD8 T cells and NK cells and compare two scRNA-seq datasets, established using cells from Ig-treated mice and from anti-VEGFR-2 treated mice 1 day after the first anti-VEGFR-2 treatment (figure 2A). Based on this analysis, we identified several possible interactions from mixed T cells to NK cells and (figure 2D) from cycling T cells to NK cells (figure 2E). As we were looking for interactions triggered by the treatment, we focused on interactions found solely in the anti-VEGFR-2 treated dataset (figure 2F) and identified seven possible signaling crosstalk pathways induced by the treatment. The expression of Spp1 messenger RNA (mRNA) by CD8 T cells was matched with its receptors Cd44, ItgavItgb1, Itga4Itg1b and ItgavItgb3 mRNA in NK cells. The expression of Cxcl10, Il10 and Tnsfs9 mRNA in CD8 T cells was associated with the presence of their receptors Cxcr3, Il10ra-Il10rb and Tnfrsf9_ mRNA in NK cells.

Together our scRNA-seq data underline potential interactions between CD8 T cells and NK cells induced by anti-VEGFR-2 therapy.

We then assessed the protein expression of the candidate pathways identified on CD8 T cells and NK cells. Using flow cytometry, we found that CD8 T cells do not express significant levels of SPP1 nor TNFSF9 proteins 1 day after anti-VEGFR-2 treatment (online supplementary figure S4, D and E), ruling out these pathways. The CXCL10-CXCR3 pathway displayed a low probability, a high p value, and appeared only once, making that pathway unlikely. Hence, the IL-10/IL-10 receptor (IL-10R) axis remained the only plausible pathway.### IL-10 is produced by CD8 T cells after anti-VEGFR-2 treatment

To explore whether the IL-10/IL-10R pathway is involved in CD8 T cell-mediated NK cell recruitment, we first assessed the levels of IL-10 in the tumor 3 days after the first anti-VEGFR2 treatment and found that both the anti-VEGFR-2 alone and the bitherapy were able to trigger an increase in tumorous IL-10 in CT26 tumors (online supplementary figure S5A). We were also able to detect an increase in IL-10 in MC38 tumors treated with the bitherapy 3 days after the first anti-VEGFR2 treatment (online supplementary figure S5B) and in a MC38 tumor growth assay, we observed that using a blocking antibody directed against IL-10 reduced the antitumor efficacy of the bitherapy against MC38 tumor growth (online supplementary figure S5C). Then, we tested the ability of intratumoral CD8 T cells to produce IL-10 1 day following the anti-VEGFR-2 treatment. We found that CD8 T cells isolated from CT26 tumors produced more than twice the amount of IL-10 compared with untreated controls while other immune cell types did not show increased IL-10 production (figure 3A and online supplementary figure 5 D to O). Therefore, the intratumoral increase in IL-10 was mostly due to CD8 T cells and our scRNA-seq data confirmed the presence of cells coexpressing Cd8a and Il10 as well as cells co-expressing Foxp3 and Il10; however, no increase in Il10 expression was observed in regulatory T cells (Tregs) (online supplementary figure 5P-Q). Accordingly, CD8 T cells isolated from the spleen of naive mice and activated for 3 days produced increasing amounts of IL-10 (figure 3B). This observation aligns with previous studies demonstrating that highly activated CD8 T cells can produce IL-10 following viral infections.^[13]

To investigate whether VEGF-A could modulate IL-10 production by CD8 T cells, we incubated CD8 T cells purified from the spleen of naive mice with increasing concentrations of VEGF-A, within the range observed in vivo in CT26 tumors (online supplementary figure S5 R), for 72 hours ex vivo. We observed that VEGF-A significantly inhibited IL-10 production by CD8 T cells (figure 3C). Next, we evaluated the role of IL-10 in the efficacy of the anti-VEGFR-2 and anti-PD-1 combination therapy in CT26 tumors. By using an IL-10-neutralizing monoclonal antibody, we found that IL-10 blockade significantly impaired the therapeutic efficacy, leading to faster tumor growth (figure 3D; Online supplementary figure S6A), highlighting an immunostimulatory role of IL-10. These results indicate that IL-10 production by CD8 T cells is rapidly induced by anti-VEGFR-2 treatment and plays a crucial role in enhancing the efficacy of this bitherapy.### IL-10 can directly drive NK cells recruitment

To evaluate the potential effect of IL-10 on NK cells, we first measured IL-10R expression on their surface. We found that nearly all NK cells from the blood of both naive and CT26 tumor-bearing mice expressed IL-10R (figure 3E). In vivo, we observed a significant decrease in the percentage of NK cells expressing IL-10R within the tumor 1 day after anti-VEGFR-2 treatment (figure 3F), consistent with the known internalization of the IL-10R on binding to IL-10 (3). Using NK cells sorted from naive mice spleen and cultured in vitro with increasing doses of IL-10, we confirmed the downregulation of IL-10R expression following IL-10 signaling (figure 3G). These findings suggest that NK cells in the TME were exposed to significantly active IL-10 levels following anti-VEGFR-2 treatment. We also observed that the increase in NK cell numbers induced by anti-VEGFR-2 treatment in CT26 tumors was abolished on IL-10 blockade (figure 3H), suggesting a critical role for IL-10 in NK cells’ increase after treatment. Additionally, this is consistent with the absence of increase in NK cell number in tumors lacking CD8 T cells, as shown in figure 1O. The observed increase in NK cells 1 day after anti-VEGFR-2 treatment, which is both CD8 T cell-dependent and IL-10-dependent, and the increased IL-10 production by CD8 T cells suggested a potential direct effect of IL-10 on NK cell recruitment.

To explore the direct impact of IL-10 on NK cell recruitment, we sorted NK cells from the spleen of naive mice and cultured them with increasing doses of IL-10. After 24 hours, we observed no significant change in NK cell proliferation, as measured by the Ki-67 expression (figure 3I). We then performed migration assays using a transwell system, exposing splenic NK cells from naive mice to increasing concentrations of IL-10 for 4 hours. Here, we demonstrated that IL-10 significantly attracted NK cells in a dose-dependent manner (figure 3J). Moreover, a similar effect was observed with NK cells isolated from untreated CT26 tumors (online supplementary figure S6B). As IL-10 signals through the mTOR pathway (4), we assessed the role of mTOR in IL-10-mediated NK cell migration. To address this, we treated NK cells with rapamycin, a potent inhibitor of mTOR complex 1, resulting in complete inhibition of NK cell migration (figure 3J). Taken together, these results confirm that IL-10, in an mTOR-dependent manner, displays chemotactic capacity and facilitates NK cell migration.

Finally, to show that the antitumor role of IL-10 during treatment is dependent on NK cell function, and in the absence of IL10RA knockout mice, we performed a CT26 tumor growth experiment in WT mice using the bitherapy along with the anti-asialo-GM1 (to deplete NK cells) and an anti-IL-10R monoclonal antibody. We observed that both NK cell depletion and IL-10R blockade limited the efficacy of the bitherapy to a similar degree and that the combination of these two antibodies does not have an additional effect, strongly suggesting that NK cells and IL-10R are part of the same cascade of events, reinforcing the link between IL-10 signaling and NK cells (figure 3K and online supplementary figure S6C-D).

In conclusion, our data demonstrate that CD8 T cells produce IL-10 in response to anti-VEGFR-2 treatment, and this IL-10 signaling plays a critical role in recruiting NK cells to the tumor.### NK cells are necessary for the bitherapy-induced macrophage maturation

To elucidate the role of NK cells in the efficacy of the bitherapy, we used a monoclonal antibody targeting asialo-GM1 that depletes NK cells (online supplementary figure S7A). NK cell depletion significantly impaired tumor control in both CT26 and MC38 tumors, demonstrating that NK cells are essential for the therapeutic efficacy of the bitherapy (figure 4A and online supplementary figure S7B). Notably, flow cytometry analysis of the TME showed that NK cell depletion led to a significant reduction in the proportion of mature macrophages within the TME 9 days after treatment initiation (figure 4B).

Flow cytometry analysis of the immune infiltrate in a kinetic assay revealed that the treatment induced enhanced macrophage maturation, as indicated by the increase in the percentage of macrophages expressing high levels of major histocompatibility complex (MHC) II 3 days after treatment initiation (figure 4C).

Although granulocyte-macrophage colony-stimulating factor (GM-CSF) is well known for its ability to drive macrophage maturation,^[14] interferon-gamma (IFN-γ)^[15] and tumor necrosis factor-alpha (TNF-α)^[16] can also induce macrophage maturation and are produced by NK cells.^[17] Kinetic analysis of these cytokine levels in CT26 tumors revealed that while TNF-α levels remained unchanged following bitherapy (figure 4D), IFN-γ levels were elevated only 9 days post-treatment, when GM-CSF levels were significantly upregulated in tumors treated with the bitherapy at 3 days post-treatment and remained elevated at D9 (figure 4D). The increase in GM-CSF coincided with the observed rise in macrophage maturation at days 3 and 9, suggesting a pivotal role for GM-CSF in this process.

To directly assess the contribution of GM-CSF to the efficacy of the bitherapy, we used a monoclonal antibody against GM-CSF and found that GM-CSF blockade significantly diminished the bitherapy’s effectiveness in controlling CT26 tumor growth (figure 4E). Blocking GM-CSF disrupted bitherapy-induced macrophage maturation, as shown by reduced MHC II expression (figure 4F and online supplementary figure 7C). This highlights the indispensable role of GM-CSF in facilitating macrophage maturation in vivo in the context of this combination therapy.### GM-CSF production by NK cells increases macrophage maturation

To pinpoint the source of GM-CSF production in the TME 3 days after treatment, we sorted the main immune cell populations known to produce GM-CSF from CT26 tumors treated with either the bitherapy or vehicle and cultured them ex vivo for 24 hours. While macrophages, dendritic cells (DCs), and CD8 T cells did not show an increase in Csf2 mRNA, coding for GM-CSF, NK cells exhibited a significant increase in Csf2 mRNA and GM-CSF protein production after bitherapy treatment (figure 4G; online supplementary figure S7D and E). We further validated the in vivo role of NK cells in GM-CSF production by assessing GM-CSF levels in CT26 tumors treated with bitherapy, either with or without NK cell depletion. The increase in GM-CSF levels observed with bitherapy was abrogated in the absence of NK cells, confirming that NK cells are the primary source of GM-CSF following the bitherapy (figure 4H). Additionally, as CD8 T cells mediate NK cell recruitment in response to the bitherapy (figure 3), we observed that both GM-CSF production and macrophage maturation were absent when CD8 T cells were depleted (figure 4, I and J and online supplementary figure S7F). Finally, in Ncr1cre Csf2flox mice in which NK cells cannot produce GM-CSF, bearing MC38 tumors and treated with the bitherapy, we observed that the absence of NK-cell-derived GM-CSF blocked the expansion of MHC II high macrophages (online supplementary figure 7G).

Collectively, these data demonstrate that the bitherapy promotes NK cells to produce GM-CSF, which is essential for macrophage maturation in the TME. This macrophage maturation is crucial for the overall efficacy of the bitherapy in controlling tumor progression.### CXCL11 recruits CD8 T cells to the tumor at later time points after treatment

We observed an increase in the CD8 T cell population at day nine following the bitherapy with anti-VEGFR-2 and anti-PD-1 (figure 1E). To better understand the role of anti-PD-1, we discriminated different subpopulations of CD8 T cells using PD1, Tim3 and SlamF6 to define non-exhausted (triple negative), progenitor exhausted (PD1+SlamF6+) and terminally exhausted CD8 T cells (PD1+Tim3+). At this time point, terminally exhausted CD8 T cells were preferentially increased in the combination treatment group, further supporting a role for anti-PD-1 in shaping the functional state of tumor-infiltrating CD8 T cells (Extended Data, online supplementary figure 8A). While anti-VEGFR-2 promoted CD8 T cell accumulation, anti-PD-1 is required for optimal acquisition of effector functions, including granzyme B and IFNγ production across multiple CD8 T cell subsets (Extended Data, online supplementary figure 8 A, B and C). To better understand the underlying mechanisms driving this increase, we assessed CD8 T cell proliferation by measuring Ki-67 expression. No significant difference in Ki-67 expression was observed, suggesting that the increase in CD8 T cells was not due to enhanced proliferation but rather to increased recruitment of these cells to the tumor (figure 5A). We next investigated the molecular mechanisms underlying CD8 T cell recruitment by evaluating the expression of CXCL9, CXCL10, and CXCL11, which are ligands of CXCR3 and are known to attract CD8 T cells.^[18] All three chemokines showed increased mRNA expression compared with controls in tumor bulk analyses (online supplementary figure S8 C to E). Interestingly, only CXCL11 protein levels were significantly increased in CT26 tumors following bitherapy treatment at day 9 (figure 5B, online supplementary figure S8, F and G). While our data support CXCL11 as the predominant functional CXCR3 ligand in this setting, we cannot exclude that CXCL9 and CXCL10 may also contribute to the overall CXCR3-dependent response, potentially through transient production and rapid local consumption within the TME. Although CCL5, a ligand of CCR5, is also capable of attracting CD8 T cells,^[19] it was not significantly increased 9 days after treatment (online supplementary figure S8H).

To determine when CXCL11 is increased in tumors, we performed a kinetic analysis of this protein and showed that the bitherapy significantly increases CXCL11 secretion in CT26 tumors from day 5 in vivo (figure 5C). To test the importance of CXCL11 in the efficacy of the bitherapy, we used a monoclonal antibody targeting CXCR3. A tumor growth assay demonstrated that blocking CXCR3 significantly impaired the efficacy of the bitherapy in CT26 tumors (figure 5D), highlighting the importance of the CXCR3 axis. Flow cytometry analysis showed that CXCR3 blockade inhibited the increase in CD8 T cells in tumors at day 9 (figure 5E), confirming that CXCR3 is crucial for CD8 T cell recruitment in response to the bitherapy.### Macrophages’ secretion of CXCL11 is GM-CSF dependent

To identify the source of CXCL11 in the TME after treatment, we sorted DCs and macrophages from CT26 tumors treated or not with the bitherapy at day 5 post-treatment initiation. This specific time point was chosen as it matches the observed increase in CXCL11 secretion in response to the bitherapy (figure 5C). Macrophages, which are among the known producers of CXCL11, were the only immune population that showed a significant increase in Cxcl11 mRNA expression after treatment (figure 5F; Online supplementary figure S8I-J).

Mature macrophages secrete CXCL11, and GM-CSF is known to promote macrophage maturation. In vivo, GM-CSF production triggered by the bitherapy was critical for the increase in CXCL11, as blockade of GM-CSF completely prevented the upregulation of CXCL11 in the TME 9 days after treatment (figure 5G). Finally, given that CXCL11 recruits CD8 T cells, we investigated whether GM-CSF plays a role in T cell recruitment after the bitherapy. We observed that GM-CSF blockade abolished the increase in CD8 T cells observed at day 9 post-treatment in vivo in CT26 tumors (figure 5H), confirming the role of GM-CSF in this process.

As GM-CSF is produced by NK cells, we examined the role of NK cells in CXCL11 production and CD8 T cells accumulation at day 9. We show that NK cell depletion using the anti-asialo-GM1 monoclonal antibody prevented the increase in both CXCL11 and CD8 T cells (figure 5, I and J), confirming that NK cells are essential for CXCL11 production and therefore for the recruitment of CD8 T cells to the tumor. Finally, given that CD8 T cells promote NK cell recruitment 1 day after treatment, we investigated whether CD8 T cell depletion could impact CXCL11 production in tumors. CD8 T cell depletion resulted in a loss of the CXCL11 increase in the tumor after treatment (figure 5K), indicating that CD8 T cells play a role in sustaining CXCL11 production.

In summary, our data demonstrate that the increase in CD8 T cells within the tumor at day 9 post-treatment is primarily driven by their recruitment via CXCL11, produced by macrophages. This recruitment is dependent on GM-CSF, NK cells, and CD8 T cells, and plays a crucial role in the antitumor efficacy of the bitherapy.### Treg recruitment by the bitherapy hinders its efficacy

CD4 Tregs are known to express high levels of CXCR3 and can be recruited by its ligands: CXCL9, CXCL10, and CXCL11.^[20] In our study, we observed increased levels of CXCL11 in tumors treated with the bitherapy (figure 5, B and C). To assess whether the Treg population in CT26 tumors was similarly affected by the treatment, we performed flow cytometry analyses. As anticipated, we found that the bitherapy led to a significant increase in Tregs in CT26 tumors 9 days after treatment but not at day 1 (figure 6A and B). Interestingly, Treg did not produce increased levels of IL-10 on anti-VEGFR-2 treatment (online supplementary figure S9A). Furthermore, we confirmed that Tregs express CXCR3, and we observed that the bitherapy reduced CXCR3 expression on the Treg surface, consistent with prior reports indicating that CXCR3 is internalized and degraded following ligand engagement^[21] (figure 6C).

The increase in Tregs was completely abolished when CXCR3 signaling was blocked using a monoclonal antibody against CXCR3 (figure 6D), confirming that Tregs are recruited to the tumor by the bitherapy via the CXCR3 pathway. Additionally, a NanoString analysis of CT26 tumors at day 9 following the treatment showed overexpression of genes associated with Tregs (figure 6E), which corroborated the findings from flow cytometry.

Tregs are known to foster an immunosuppressive environment that limits the antitumor immune response. To explore whether targeting Tregs could improve the efficacy of the bitherapy, we used a monoclonal antibody targeting CD25, which is the alpha subunit of the IL-2 receptor required for Treg survival, allowing for specific Treg inactivation.^[22] In a CT26 tumor growth assay, the combination of the bitherapy along with the anti-CD25 antibody, which targets Treg cells without affecting CD8 T cells or NK cells, resulted in a dramatic improvement in tumor control, with nearly 100% of mice achieving complete tumor regression (figure 6, F and G and online supplementary figure S9 B to F). Rechallenge of the cured mice on the contralateral leg with CT26 tumor cells revealed no tumor development, suggesting the establishment of robust, long-lasting antitumor immunity (figure 6H).

Our previous work demonstrated that CXCL11 is induced by GM-CSF production from NK cells (figure 4). In line with this, we found by flow cytometry that GM-CSF blockade prevented the bitherapy-induced increase in Tregs in CT26 tumors (figure 6I). Similarly, NK cell depletion also abrogated the increase in Tregs observed after bitherapy treatment (figure 6J). Given that NK cells are recruited by CD8 T cells (figure 3), we next investigated whether CD8 T cell depletion would impact Treg numbers in CT26 tumors post-treatment. As expected, depletion of CD8 T cells completely prevented the increase in Tregs in CT26 tumors observed 9 days after treatment (figure 6K), further confirming the sequential cascade of events that leads to CXCL11 production and Treg recruitment.

Together, these findings suggest that Tregs are recruited to the tumor by the bitherapy, involving CXCL11 and GM-CSF production as well as the activation of NK and CD8 T cells. Despite the observed beneficial role of IL-10 in NK cell recruitment, Tregs still exert pro-tumor functions, and their elimination allows for the full antitumor efficacy of the bitherapy.

CD8 T cells are needed for anti-VEGFR-2/anti-PD-1 efficacy against CRC tumor model

To investigate the immune events occurring in tumors following combination treatment (Bith) with anti-VEGFR-2 (αR2) and anti-PD-1 (αPD1) antibodies, we assessed the response of CT26 and MC38 CRC models. Both models showed modest efficacy of the anti-PD-1 monoclonal antibody therapy alone, with a slight slowdown in tumor growth. Anti-VEGFR-2 had no efficacy in CT26 while it slightly slowed tumor growth in the MC38 model. In contrast, significant responses were observed with the combination of anti-VEGFR-2 and anti-PD-1 antibodies (bith) showing a synergistic effect in the CT26 model and an additive effect in the MC38 model (figure 1 A; online supplementary figure S1, A-B). We also evaluated tumor growth in an orthotopic implantation model of CT26-luciferase cells in the caecum, where the combination therapy also demonstrated significant efficacy (figure 1C; online supplementary figure S1B). To confirm our data, we also studied the response using anti-VEGF-A and found that it worked to a similar extent (online supplementary figure S1D). Consequently, using nude mice, we demonstrated that T lymphocytes are crucial for the efficacy of the bitherapy, as it had no effect on tumor growth in the absence of T lymphocytes in both CT26 and MC38 tumors (online supplementary figure S1E and F).

To broadly assess the impact of the bitherapy on the immune profile of the TME, we performed a NanoString analysis to analyze the transcriptome of CT26 tumor bulk samples collected 9 days after the initiation of the combination therapy. When looking at genes associated with T cell functions, more than 38% of them were significantly upregulated in the bitherapy group (figure 1C; Online supplemental table S1), suggesting activation of T cells induced by the bitherapy.

Accordingly, flow cytometry analysis of CT26 tumors revealed a very rapid and significant increase in CD8 T cell numbers 1 and 9 days after the first injection of anti-VEGFR-2 treatment (figure 1, D and E), whereas the CD4 T cell population remained unchanged (online supplementary figure S2, A and B). To explain such a rapid increase in response to anti-VEGFR-2, we assessed the CD8 T cell proliferation 1 day after the treatment. We found that CD8 T cells exhibited significantly greater expression of Ki-67 in treated tumors (figure 1F). Interestingly, 10% of CD8 T cells from CT26 tumors expressed VEGFR-2, suggesting a direct on-target effect of VEGFR-2 inhibition on CD8 T cells (figure 1G). Additionally, most tumor antigen-specific CD8 T cells in CT26 tumors expressed VEGFR-2 (figure 1H) and were increased after the first anti-VEGFR2 treatment (online supplementary figure S2C). Accordingly, only the transfer of CD8 T cells expressing VEGFR-2 into CT26 tumor-bearing Nude mice allowed an effect of the bitherapy on tumor growth whereas the transfer of VEGFR-2− CD8 T cells had no impact (online supplementary figure S2D). As the VEGF/VEGFR axis is known to suppress T cell proliferation, these data strongly suggest a direct effect of the anti-VEGFR-2 onto the CD8 T cell proliferation.^[12]

Interestingly, although the CD8 T cell population increased in the tumor as early as 1 day post-treatment, these cells did not show at this time point enhanced activation (as measured by CD69 expression) or increased cytotoxicity (as indicated by granzyme B expression) (figure 1, I and J). However, 9 days post-treatment, CD8 T cells exhibited increased activation and higher levels of granzyme B after the bitherapy (figure 1, K and L), suggesting that different mechanisms regulate both their proliferation and activation over time.

CD8 T cells play an important role early after treatment initiation

To further confirm the role of CD8 T cells in the immune response to the combination therapy, we conducted tumor growth assays using a CD8β-depleting monoclonal antibody that efficiently targets and eliminates CD8 T cells within 24 hours (online supplementary figure S2E). CD8 T cell depletion completely abrogated the efficacy of the combination therapy in both the CT26 and MC38 models, underscoring the crucial role of CD8 T cells in the response to treatment (figure 1M and online supplementary figure 2F). We also investigated the temporal importance of CD8 T cells in the efficacy of the combination therapy by depleting CD8 T cells either at the start of treatment (D0) or 3 days after treatment initiation (D3). While CD8 T cell depletion 3 days after treatment initiation had a very limited impact on tumor growth inhibition, depletion at the start of treatment drastically impaired the ability of the therapy to control tumor growth in both CT26 and MC38 tumor models (figure 1N; Online supplementary figure S2 G and H). These findings suggest that CD8 T cells play a critical antitumor role within the first 3 days after anti-VEGFR-2 treatment.

From these findings, we aimed to investigate the early role of CD8 T cells. We found by flow cytometry that early CD8 T cell depletion reduced the number of intratumoral NK cells 1 day after the treatment in both CT26 and MC38 tumors (figure 1O and online supplementary figure S2 I). NK cells do not express VEGFR-2, eliminating the possibility of a direct effect of anti-VEGFR-2 on these cells (online supplementary figure S2J). CD8 T cell depletion did not affect NK cell proliferation, suggesting that CD8 T cells might facilitate NK cell recruitment into the tumor (online supplementary figure S2K). Taken together, these data further establish CD8 T cells as direct targets of the anti-VEGFR-2 treatment and suggest that they play a pivotal role in the subsequent early increase in NK cells following treatment.

In summary, our data demonstrate that the combination therapy of anti-PD-1 and anti-VEGFR-2 is dependent on CD8 T cells for effective tumor growth inhibition in preclinical models of colorectal carcinoma. Moreover, we reveal an unexpected relationship between CD8 T cells and NK cells within the TME as early as 1 day after treatment.

Characterization of CD8 T cells and NK cells’ putative interactions after anti-VEGFR-2

To gain a deeper understanding of the relationship between CD8 T cells and NK cells in CT26 tumors, we performed a single-cell RNA sequencing (scRNA-seq) analysis 1 day after the first anti-VEGFR-2 treatment (figure 2A). After quality control, cells clustered into 16 distinct cell types (online supplementary figure S3).

We focused on the four clusters of NK cells (NK1, NK2, NK3 and NK4) and the two clusters containing CD8 T cells (Tcells1 and Tcells2). We showed that the four NK cell clusters correspond to cycling NK cells (Mki67, Birc5 and Nusap1) (NK4), secreting NK cells (Ccl3, Ccl4, Ccl5) (NK3), cytotoxic NK cells (Gzmd, Gzme, Gzmc) (NK2) and finally NK cells with no defining features except expression of the classical NK cells markers Eomes and Ncr1 hereafter referred to as resting NK cells (NK1) (figure 2, B and C; Online supplementary figure S4C). Regarding T cells, the Tcells2 cluster corresponds to cycling CD8 T cells (Mki67, Birc5 and Cd8a) and Tcells1 cluster corresponds to mixed T cell population (Cd4, Cd8a, Cd3e) (figure 2, B and C).

We used CellChat to identify putative intercellular communication pathways between CD8 T cells and NK cells and compare two scRNA-seq datasets, established using cells from Ig-treated mice and from anti-VEGFR-2 treated mice 1 day after the first anti-VEGFR-2 treatment (figure 2A). Based on this analysis, we identified several possible interactions from mixed T cells to NK cells and (figure 2D) from cycling T cells to NK cells (figure 2E). As we were looking for interactions triggered by the treatment, we focused on interactions found solely in the anti-VEGFR-2 treated dataset (figure 2F) and identified seven possible signaling crosstalk pathways induced by the treatment. The expression of Spp1 messenger RNA (mRNA) by CD8 T cells was matched with its receptors Cd44, ItgavItgb1, Itga4Itg1b and ItgavItgb3 mRNA in NK cells. The expression of Cxcl10, Il10 and Tnsfs9 mRNA in CD8 T cells was associated with the presence of their receptors Cxcr3, Il10ra-Il10rb and Tnfrsf9_ mRNA in NK cells.

Together our scRNA-seq data underline potential interactions between CD8 T cells and NK cells induced by anti-VEGFR-2 therapy.

We then assessed the protein expression of the candidate pathways identified on CD8 T cells and NK cells. Using flow cytometry, we found that CD8 T cells do not express significant levels of SPP1 nor TNFSF9 proteins 1 day after anti-VEGFR-2 treatment (online supplementary figure S4, D and E), ruling out these pathways. The CXCL10-CXCR3 pathway displayed a low probability, a high p value, and appeared only once, making that pathway unlikely. Hence, the IL-10/IL-10 receptor (IL-10R) axis remained the only plausible pathway.

IL-10 is produced by CD8 T cells after anti-VEGFR-2 treatment

To explore whether the IL-10/IL-10R pathway is involved in CD8 T cell-mediated NK cell recruitment, we first assessed the levels of IL-10 in the tumor 3 days after the first anti-VEGFR2 treatment and found that both the anti-VEGFR-2 alone and the bitherapy were able to trigger an increase in tumorous IL-10 in CT26 tumors (online supplementary figure S5A). We were also able to detect an increase in IL-10 in MC38 tumors treated with the bitherapy 3 days after the first anti-VEGFR2 treatment (online supplementary figure S5B) and in a MC38 tumor growth assay, we observed that using a blocking antibody directed against IL-10 reduced the antitumor efficacy of the bitherapy against MC38 tumor growth (online supplementary figure S5C). Then, we tested the ability of intratumoral CD8 T cells to produce IL-10 1 day following the anti-VEGFR-2 treatment. We found that CD8 T cells isolated from CT26 tumors produced more than twice the amount of IL-10 compared with untreated controls while other immune cell types did not show increased IL-10 production (figure 3A and online supplementary figure 5 D to O). Therefore, the intratumoral increase in IL-10 was mostly due to CD8 T cells and our scRNA-seq data confirmed the presence of cells coexpressing Cd8a and Il10 as well as cells co-expressing Foxp3 and Il10; however, no increase in Il10 expression was observed in regulatory T cells (Tregs) (online supplementary figure 5P-Q). Accordingly, CD8 T cells isolated from the spleen of naive mice and activated for 3 days produced increasing amounts of IL-10 (figure 3B). This observation aligns with previous studies demonstrating that highly activated CD8 T cells can produce IL-10 following viral infections.^[13]

To investigate whether VEGF-A could modulate IL-10 production by CD8 T cells, we incubated CD8 T cells purified from the spleen of naive mice with increasing concentrations of VEGF-A, within the range observed in vivo in CT26 tumors (online supplementary figure S5 R), for 72 hours ex vivo. We observed that VEGF-A significantly inhibited IL-10 production by CD8 T cells (figure 3C). Next, we evaluated the role of IL-10 in the efficacy of the anti-VEGFR-2 and anti-PD-1 combination therapy in CT26 tumors. By using an IL-10-neutralizing monoclonal antibody, we found that IL-10 blockade significantly impaired the therapeutic efficacy, leading to faster tumor growth (figure 3D; Online supplementary figure S6A), highlighting an immunostimulatory role of IL-10. These results indicate that IL-10 production by CD8 T cells is rapidly induced by anti-VEGFR-2 treatment and plays a crucial role in enhancing the efficacy of this bitherapy.

IL-10 can directly drive NK cells recruitment

To evaluate the potential effect of IL-10 on NK cells, we first measured IL-10R expression on their surface. We found that nearly all NK cells from the blood of both naive and CT26 tumor-bearing mice expressed IL-10R (figure 3E). In vivo, we observed a significant decrease in the percentage of NK cells expressing IL-10R within the tumor 1 day after anti-VEGFR-2 treatment (figure 3F), consistent with the known internalization of the IL-10R on binding to IL-10 (3). Using NK cells sorted from naive mice spleen and cultured in vitro with increasing doses of IL-10, we confirmed the downregulation of IL-10R expression following IL-10 signaling (figure 3G). These findings suggest that NK cells in the TME were exposed to significantly active IL-10 levels following anti-VEGFR-2 treatment. We also observed that the increase in NK cell numbers induced by anti-VEGFR-2 treatment in CT26 tumors was abolished on IL-10 blockade (figure 3H), suggesting a critical role for IL-10 in NK cells’ increase after treatment. Additionally, this is consistent with the absence of increase in NK cell number in tumors lacking CD8 T cells, as shown in figure 1O. The observed increase in NK cells 1 day after anti-VEGFR-2 treatment, which is both CD8 T cell-dependent and IL-10-dependent, and the increased IL-10 production by CD8 T cells suggested a potential direct effect of IL-10 on NK cell recruitment.

To explore the direct impact of IL-10 on NK cell recruitment, we sorted NK cells from the spleen of naive mice and cultured them with increasing doses of IL-10. After 24 hours, we observed no significant change in NK cell proliferation, as measured by the Ki-67 expression (figure 3I). We then performed migration assays using a transwell system, exposing splenic NK cells from naive mice to increasing concentrations of IL-10 for 4 hours. Here, we demonstrated that IL-10 significantly attracted NK cells in a dose-dependent manner (figure 3J). Moreover, a similar effect was observed with NK cells isolated from untreated CT26 tumors (online supplementary figure S6B). As IL-10 signals through the mTOR pathway (4), we assessed the role of mTOR in IL-10-mediated NK cell migration. To address this, we treated NK cells with rapamycin, a potent inhibitor of mTOR complex 1, resulting in complete inhibition of NK cell migration (figure 3J). Taken together, these results confirm that IL-10, in an mTOR-dependent manner, displays chemotactic capacity and facilitates NK cell migration.

Finally, to show that the antitumor role of IL-10 during treatment is dependent on NK cell function, and in the absence of IL10RA knockout mice, we performed a CT26 tumor growth experiment in WT mice using the bitherapy along with the anti-asialo-GM1 (to deplete NK cells) and an anti-IL-10R monoclonal antibody. We observed that both NK cell depletion and IL-10R blockade limited the efficacy of the bitherapy to a similar degree and that the combination of these two antibodies does not have an additional effect, strongly suggesting that NK cells and IL-10R are part of the same cascade of events, reinforcing the link between IL-10 signaling and NK cells (figure 3K and online supplementary figure S6C-D).

In conclusion, our data demonstrate that CD8 T cells produce IL-10 in response to anti-VEGFR-2 treatment, and this IL-10 signaling plays a critical role in recruiting NK cells to the tumor.

NK cells are necessary for the bitherapy-induced macrophage maturation

To elucidate the role of NK cells in the efficacy of the bitherapy, we used a monoclonal antibody targeting asialo-GM1 that depletes NK cells (online supplementary figure S7A). NK cell depletion significantly impaired tumor control in both CT26 and MC38 tumors, demonstrating that NK cells are essential for the therapeutic efficacy of the bitherapy (figure 4A and online supplementary figure S7B). Notably, flow cytometry analysis of the TME showed that NK cell depletion led to a significant reduction in the proportion of mature macrophages within the TME 9 days after treatment initiation (figure 4B).

Flow cytometry analysis of the immune infiltrate in a kinetic assay revealed that the treatment induced enhanced macrophage maturation, as indicated by the increase in the percentage of macrophages expressing high levels of major histocompatibility complex (MHC) II 3 days after treatment initiation (figure 4C).

Although granulocyte-macrophage colony-stimulating factor (GM-CSF) is well known for its ability to drive macrophage maturation,^[14] interferon-gamma (IFN-γ)^[15] and tumor necrosis factor-alpha (TNF-α)^[16] can also induce macrophage maturation and are produced by NK cells.^[17] Kinetic analysis of these cytokine levels in CT26 tumors revealed that while TNF-α levels remained unchanged following bitherapy (figure 4D), IFN-γ levels were elevated only 9 days post-treatment, when GM-CSF levels were significantly upregulated in tumors treated with the bitherapy at 3 days post-treatment and remained elevated at D9 (figure 4D). The increase in GM-CSF coincided with the observed rise in macrophage maturation at days 3 and 9, suggesting a pivotal role for GM-CSF in this process.

To directly assess the contribution of GM-CSF to the efficacy of the bitherapy, we used a monoclonal antibody against GM-CSF and found that GM-CSF blockade significantly diminished the bitherapy’s effectiveness in controlling CT26 tumor growth (figure 4E). Blocking GM-CSF disrupted bitherapy-induced macrophage maturation, as shown by reduced MHC II expression (figure 4F and online supplementary figure 7C). This highlights the indispensable role of GM-CSF in facilitating macrophage maturation in vivo in the context of this combination therapy.

GM-CSF production by NK cells increases macrophage maturation

To pinpoint the source of GM-CSF production in the TME 3 days after treatment, we sorted the main immune cell populations known to produce GM-CSF from CT26 tumors treated with either the bitherapy or vehicle and cultured them ex vivo for 24 hours. While macrophages, dendritic cells (DCs), and CD8 T cells did not show an increase in Csf2 mRNA, coding for GM-CSF, NK cells exhibited a significant increase in Csf2 mRNA and GM-CSF protein production after bitherapy treatment (figure 4G; online supplementary figure S7D and E). We further validated the in vivo role of NK cells in GM-CSF production by assessing GM-CSF levels in CT26 tumors treated with bitherapy, either with or without NK cell depletion. The increase in GM-CSF levels observed with bitherapy was abrogated in the absence of NK cells, confirming that NK cells are the primary source of GM-CSF following the bitherapy (figure 4H). Additionally, as CD8 T cells mediate NK cell recruitment in response to the bitherapy (figure 3), we observed that both GM-CSF production and macrophage maturation were absent when CD8 T cells were depleted (figure 4, I and J and online supplementary figure S7F). Finally, in Ncr1cre Csf2flox mice in which NK cells cannot produce GM-CSF, bearing MC38 tumors and treated with the bitherapy, we observed that the absence of NK-cell-derived GM-CSF blocked the expansion of MHC II high macrophages (online supplementary figure 7G).

Collectively, these data demonstrate that the bitherapy promotes NK cells to produce GM-CSF, which is essential for macrophage maturation in the TME. This macrophage maturation is crucial for the overall efficacy of the bitherapy in controlling tumor progression.

CXCL11 recruits CD8 T cells to the tumor at later time points after treatment

We observed an increase in the CD8 T cell population at day nine following the bitherapy with anti-VEGFR-2 and anti-PD-1 (figure 1E). To better understand the role of anti-PD-1, we discriminated different subpopulations of CD8 T cells using PD1, Tim3 and SlamF6 to define non-exhausted (triple negative), progenitor exhausted (PD1+SlamF6+) and terminally exhausted CD8 T cells (PD1+Tim3+). At this time point, terminally exhausted CD8 T cells were preferentially increased in the combination treatment group, further supporting a role for anti-PD-1 in shaping the functional state of tumor-infiltrating CD8 T cells (Extended Data, online supplementary figure 8A). While anti-VEGFR-2 promoted CD8 T cell accumulation, anti-PD-1 is required for optimal acquisition of effector functions, including granzyme B and IFNγ production across multiple CD8 T cell subsets (Extended Data, online supplementary figure 8 A, B and C). To better understand the underlying mechanisms driving this increase, we assessed CD8 T cell proliferation by measuring Ki-67 expression. No significant difference in Ki-67 expression was observed, suggesting that the increase in CD8 T cells was not due to enhanced proliferation but rather to increased recruitment of these cells to the tumor (figure 5A). We next investigated the molecular mechanisms underlying CD8 T cell recruitment by evaluating the expression of CXCL9, CXCL10, and CXCL11, which are ligands of CXCR3 and are known to attract CD8 T cells.^[18] All three chemokines showed increased mRNA expression compared with controls in tumor bulk analyses (online supplementary figure S8 C to E). Interestingly, only CXCL11 protein levels were significantly increased in CT26 tumors following bitherapy treatment at day 9 (figure 5B, online supplementary figure S8, F and G). While our data support CXCL11 as the predominant functional CXCR3 ligand in this setting, we cannot exclude that CXCL9 and CXCL10 may also contribute to the overall CXCR3-dependent response, potentially through transient production and rapid local consumption within the TME. Although CCL5, a ligand of CCR5, is also capable of attracting CD8 T cells,^[19] it was not significantly increased 9 days after treatment (online supplementary figure S8H).

To determine when CXCL11 is increased in tumors, we performed a kinetic analysis of this protein and showed that the bitherapy significantly increases CXCL11 secretion in CT26 tumors from day 5 in vivo (figure 5C). To test the importance of CXCL11 in the efficacy of the bitherapy, we used a monoclonal antibody targeting CXCR3. A tumor growth assay demonstrated that blocking CXCR3 significantly impaired the efficacy of the bitherapy in CT26 tumors (figure 5D), highlighting the importance of the CXCR3 axis. Flow cytometry analysis showed that CXCR3 blockade inhibited the increase in CD8 T cells in tumors at day 9 (figure 5E), confirming that CXCR3 is crucial for CD8 T cell recruitment in response to the bitherapy.

Macrophages’ secretion of CXCL11 is GM-CSF dependent

To identify the source of CXCL11 in the TME after treatment, we sorted DCs and macrophages from CT26 tumors treated or not with the bitherapy at day 5 post-treatment initiation. This specific time point was chosen as it matches the observed increase in CXCL11 secretion in response to the bitherapy (figure 5C). Macrophages, which are among the known producers of CXCL11, were the only immune population that showed a significant increase in Cxcl11 mRNA expression after treatment (figure 5F; Online supplementary figure S8I-J).

Mature macrophages secrete CXCL11, and GM-CSF is known to promote macrophage maturation. In vivo, GM-CSF production triggered by the bitherapy was critical for the increase in CXCL11, as blockade of GM-CSF completely prevented the upregulation of CXCL11 in the TME 9 days after treatment (figure 5G). Finally, given that CXCL11 recruits CD8 T cells, we investigated whether GM-CSF plays a role in T cell recruitment after the bitherapy. We observed that GM-CSF blockade abolished the increase in CD8 T cells observed at day 9 post-treatment in vivo in CT26 tumors (figure 5H), confirming the role of GM-CSF in this process.

As GM-CSF is produced by NK cells, we examined the role of NK cells in CXCL11 production and CD8 T cells accumulation at day 9. We show that NK cell depletion using the anti-asialo-GM1 monoclonal antibody prevented the increase in both CXCL11 and CD8 T cells (figure 5, I and J), confirming that NK cells are essential for CXCL11 production and therefore for the recruitment of CD8 T cells to the tumor. Finally, given that CD8 T cells promote NK cell recruitment 1 day after treatment, we investigated whether CD8 T cell depletion could impact CXCL11 production in tumors. CD8 T cell depletion resulted in a loss of the CXCL11 increase in the tumor after treatment (figure 5K), indicating that CD8 T cells play a role in sustaining CXCL11 production.

In summary, our data demonstrate that the increase in CD8 T cells within the tumor at day 9 post-treatment is primarily driven by their recruitment via CXCL11, produced by macrophages. This recruitment is dependent on GM-CSF, NK cells, and CD8 T cells, and plays a crucial role in the antitumor efficacy of the bitherapy.

Treg recruitment by the bitherapy hinders its efficacy

CD4 Tregs are known to express high levels of CXCR3 and can be recruited by its ligands: CXCL9, CXCL10, and CXCL11.^[20] In our study, we observed increased levels of CXCL11 in tumors treated with the bitherapy (figure 5, B and C). To assess whether the Treg population in CT26 tumors was similarly affected by the treatment, we performed flow cytometry analyses. As anticipated, we found that the bitherapy led to a significant increase in Tregs in CT26 tumors 9 days after treatment but not at day 1 (figure 6A and B). Interestingly, Treg did not produce increased levels of IL-10 on anti-VEGFR-2 treatment (online supplementary figure S9A). Furthermore, we confirmed that Tregs express CXCR3, and we observed that the bitherapy reduced CXCR3 expression on the Treg surface, consistent with prior reports indicating that CXCR3 is internalized and degraded following ligand engagement^[21] (figure 6C).

The increase in Tregs was completely abolished when CXCR3 signaling was blocked using a monoclonal antibody against CXCR3 (figure 6D), confirming that Tregs are recruited to the tumor by the bitherapy via the CXCR3 pathway. Additionally, a NanoString analysis of CT26 tumors at day 9 following the treatment showed overexpression of genes associated with Tregs (figure 6E), which corroborated the findings from flow cytometry.

Tregs are known to foster an immunosuppressive environment that limits the antitumor immune response. To explore whether targeting Tregs could improve the efficacy of the bitherapy, we used a monoclonal antibody targeting CD25, which is the alpha subunit of the IL-2 receptor required for Treg survival, allowing for specific Treg inactivation.^[22] In a CT26 tumor growth assay, the combination of the bitherapy along with the anti-CD25 antibody, which targets Treg cells without affecting CD8 T cells or NK cells, resulted in a dramatic improvement in tumor control, with nearly 100% of mice achieving complete tumor regression (figure 6, F and G and online supplementary figure S9 B to F). Rechallenge of the cured mice on the contralateral leg with CT26 tumor cells revealed no tumor development, suggesting the establishment of robust, long-lasting antitumor immunity (figure 6H).

Our previous work demonstrated that CXCL11 is induced by GM-CSF production from NK cells (figure 4). In line with this, we found by flow cytometry that GM-CSF blockade prevented the bitherapy-induced increase in Tregs in CT26 tumors (figure 6I). Similarly, NK cell depletion also abrogated the increase in Tregs observed after bitherapy treatment (figure 6J). Given that NK cells are recruited by CD8 T cells (figure 3), we next investigated whether CD8 T cell depletion would impact Treg numbers in CT26 tumors post-treatment. As expected, depletion of CD8 T cells completely prevented the increase in Tregs in CT26 tumors observed 9 days after treatment (figure 6K), further confirming the sequential cascade of events that leads to CXCL11 production and Treg recruitment.

Together, these findings suggest that Tregs are recruited to the tumor by the bitherapy, involving CXCL11 and GM-CSF production as well as the activation of NK and CD8 T cells. Despite the observed beneficial role of IL-10 in NK cell recruitment, Tregs still exert pro-tumor functions, and their elimination allows for the full antitumor efficacy of the bitherapy.

Discussion

The combination of an AA and an immune checkpoint blockade (ICB) has received Food and Drug Administration and European Medicines Agency approval for the treatment of multiple cancers, demonstrating significant improvements in progression-free survival and overall survival for patients.^[23][25] There is a strong rationale for combining an ICB with an AA. Indeed, while AAs normalize tumor vessels and promote the infiltration of immune cells into tumors,[12 26] they also improve the response to anti-PD1 in mouse tumor models.^[4] Together, this therapeutic combination holds the promise to enhance the overall antitumor response. However, the use of ICB alone is not recommended for patients with CRC with microsatelitte stable (MSS) cancer and combination therapies are needed to reverse resistance and recent clinical trials suggest a potential efficacy of AA and ICB in MSS CRC.^[6] However, the interactions between antiangiogenic therapy and checkpoint inhibitors within the immune microenvironment are poorly understood.

The present study elucidates the immune mechanisms triggered by the combination of anti-VEGFR-2 and anti-PD-1 therapies in two preclinical CRC models. Unexpectedly, our findings provide compelling evidence that CD8 T cells act as the initial responders to this treatment and play a critical role in shaping both the innate and adaptive immune responses within the tumor in mice. We further characterized the sequence of immune events elicited by the bitherapy, involving CD8 T cells, NK cells, macrophages, and Tregs as well as the molecular cues linking these cell populations (see graphical abstract online supplementary figure S14).

CD8 T cells are well-established as key mediators of antitumor immunity. In our mouse models, we observed an increase in CD8 T cells within the tumor following treatment, consistent with findings in clinical studies where CD8 tumor-infiltrating lymphocytes (TILs) increased after ramucirumab (anti-VEGFR-2) therapy.^[27] Notably, depletion of CD8 T cells resulted in a marked reduction in the efficacy of the bitherapy. Depletion experiments at distinct time points revealed that CD8 T cells exert their most significant antitumor functions within 3 days of the initial anti-VEGFR-2 treatment, with limited effects at later stages. We observed that among the CD8 T cells expressing VEGFR-2, over 60% of them are specific for a tumor antigen, which aligns with the known fact that VEGFR-2 is upregulated on T cell activation.^[28] VEGF has been shown to suppress T cell proliferation.^[12] This is consistent with our observation of enhanced CD8 T cell proliferation following treatment with anti-VEGFR-2. These data suggest that blockade of the VEGFR-2/VEGF-A axis facilitates the rapid expansion of CD8 T cells in our two mouse tumor models of CRC.

Interestingly, depletion of CD8 T cells also led to a significant reduction in NK cell numbers within the tumor. To investigate the interaction between CD8 T cells and NK cells, we performed scRNA-seq and uncovered an unexpected relationship. Indeed, our analysis revealed that increased IL-10 production by CD8 T cells after anti-VEGFR-2 treatment was a key factor driving NK cell recruitment to the tumor.

The production of IL-10 by CD8 T cells has been well-documented and typically requires strong antigenic stimulation.^[29] While VEGF has not been previously linked to IL-10 production, it is known to contribute to T cell exhaustion,^[28] and we observed that anti-VEGFR-2 treatment not only enhanced CD8 T cell proliferation but also induced IL-10 production by these cells.

Despite its well-documented immunoregulatory properties,[13 30] IL-10 has also been shown to activate NK cells, enhancing their proliferation and cytotoxicity.^[31][33] In our study, unexpectedly, IL-10 plays a non-canonical role as a chemoattractant for NK cells toward the tumor, thereby increasing their population following treatment. The precise role of IL-10 in cell migration remains contentious, as it can both promote^[34][36] and inhibit migration.[37 38] Recent reports indicate that IL-10 activates the mTORC1 pathway in human NK cells and enhances their functions.^[31] Interestingly, the mTOR signaling pathway is known to mediate cell migration.[39 40] Mechanistically, we demonstrated that IL-10 is sufficient to induce NK cell migration to the tumor in mice, and this process requires mTOR signaling, as evidenced by the inhibition of IL-10-induced NK cell migration on treatment with rapamycin, an mTOR inhibitor.

ediThis study describes a novel CD8 T cell-NK cell interaction in mice, where NK cells are influenced by CD8 T cells. This finding suggests that CD8 T cells not only serve as effectors but also influence other innate immune populations, such as NK cells. It also describes how IL-10, typically associated with immune tolerance, can promote NK cell recruitment in the context of immune activation, marking an emerging concept in murine immunology.

Having established the early immune events triggered by the bitherapy, we next explored the anti-tumor functions of NK cells. Depletion of NK cells led to a significant loss of therapeutic efficacy. We found that NK cell-derived GM-CSF promoted macrophage maturation, a phenomenon previously observed in inflammatory arthritis.^[17] NK cells expressed both the CD2 and IL2rb genes, potentially explaining their production of GM-CSF as observed before.^[41] These macrophages, in turn, exhibited enhanced CXCL11 production, a chemokine known to attract CXCR3-expressing cells, including CD8 T cells,^[42] which likely explains the increase in CD8 T cells observed at later stages of therapy in our preclinical model. Moreover, we confirmed that depletion of both CD8 T cells and NK cells abolished the upregulation of CXCL11 in macrophages, further validating our immune cascade.

Nevertheless, the CXCL11/CXCR3 axis not only allows for the recruitment of antitumor CD8 T cells, but also of Tregs,^[43] which is consistent with our observations. Interestingly, anti-VEGFR-2 blockade alone leads to a decrease in Tregs^[44] when the bitherapy resulted in an increase of Tregs. As expected, depletion of Tregs using an anti-CD25 monoclonal antibody improved the efficacy of the bitherapy. Additionally, we validated the cascade of events leading to increased CXCL11 production and Treg recruitment. Interestingly, although IL-10 is typically associated with Treg-mediated immunosuppression,^[45] our findings suggest that IL-10 is crucial for fostering a potent anti-tumor immune response, reflecting a paradigm shift in our understanding of IL-10’s role in tumor immunity in mice.

In conclusion, this study reveals a layered immune choreography that explains why dual blockade of VEGFR-2 and PD-1 can synergize even in poorly immunogenic tumors in mice. This work identifies a novel immune cascade, with IL-10 production by CD8 T cells as a cornerstone of the bitherapy’s efficacy. The specific characteristics of the TME required for IL-10 to mediate antitumor effects remain to be delineated. Until then, caution must be exercised when considering the development of novel therapeutic strategies that use IL-10 to modulate antitumor immune responses. If successfully translated into a human setting, these findings could provide a useful framework for hypothesis generation regarding vascular–immune crosstalk and have important implications for the design of combination immunotherapies, particularly in cancers and diseases where CD8 T cell and NK cell mediated immunity is critical to therapeutic success.

Limitations of the study

This study is strictly limited to mouse immunity, using two mouse models of CRC to explore the link between CD8 T cells and NK cells, these findings may not translate to human immunity. Additionally, this study used rat monoclonal antibodies directed against PD-1 and VEGFR-2, that are different from the antibodies used against human cancers, so the conclusions drawn might not be translatable in a human setting. Further studies involving human-derived samples are required to ensure the cascade of events we unraveled in this study can potentially apply to humans. Finally, this study was not designed to evaluate therapeutic efficacy but rather to explore the immune cascade behind the association of anti-VEGFR-2 and anti-PD1 therapies in mouse models of CRC.

Limitations of the study

This study is strictly limited to mouse immunity, using two mouse models of CRC to explore the link between CD8 T cells and NK cells, these findings may not translate to human immunity. Additionally, this study used rat monoclonal antibodies directed against PD-1 and VEGFR-2, that are different from the antibodies used against human cancers, so the conclusions drawn might not be translatable in a human setting. Further studies involving human-derived samples are required to ensure the cascade of events we unraveled in this study can potentially apply to humans. Finally, this study was not designed to evaluate therapeutic efficacy but rather to explore the immune cascade behind the association of anti-VEGFR-2 and anti-PD1 therapies in mouse models of CRC.

Materials and methods

Study design

The aim of this study was to shed light on the immune response occurring after a combination treatment including an anti-VEGFR2 antibody and an anti-PD1 antibody. To achieve this, CT26 or MC38 colorectal cell lines were injected subcutaneously (s.c) in Balb/c or C576BL/6J mice respectively and received a total of four doses of anti-VEGFR-2 and/or anti-PD1 antibodies. At different time points during the treatment (1, 3 or 9 days after the first VEGFR2 treatment), tumors were harvested and tumor-infiltrating immune cells were analyzed by flow cytometry, ELISA, reverse transcription quantitative real-time PCR (RTqPCR), Luminex, NanoString. scRNA-seq was next performed to study putative interactions between CD8 T cells and NK cells. Ex vivo and in vivo experiments allowed us to confirm the CellChat analysis data. No data were excluded. Mice were allocated to groups randomly on the day of the first treatment, they were then tattooed on their toes for identification and an Excel sheet was established to link each mouse to their specific treatments and schedule.### Tumor cell lines culture

CT26 (CRL-2638, ATCC) and MC38 (ENH204-FP, Kerafast) cell lines were cultured at 37°C under 5% CO2 in RPMI 1640 or Dulbecco’s Modified Eagle’s Medium respectively supplemented with 10% fetal bovine serum (FBS) (Dutscher) and Penicillin-Streptomycin, (10,000 IU/mL Penicillin, 10 mg/mL Streptomycin, PAN Biotech). Cells were divided three times per week before the confluence exceeded 80%. Mycobacteria contamination was routinely tested using the MycoAlert kit (LT07, Lonza). For orthotopic transplantation, the CT26 cell line used was CT26-Luc (pAIPCMVpLuciferase_Puro), kindly provided by Dr Olivier Micheau and Abdelmnim Radoua.### Mice

All mouse protocols were carried out in accordance with Federation of European Laboratory Animal Science Associations (FELASA) guidelines and all experiments were approved by the University of Burgundy Ethics Committee. All mouse experiments were performed at the University of Burgundy in specific pathogen-free conditions, with a 12-hour day/night alternation in a room at 22°C and 55% humidity. Female C57BL/6J (reference 632C57BL/6J), Balb/c (reference 627BALBC/CBYJ) and Nude (reference 639NU/NUMRI) were purchased from Charles River Laboratories. IL10−/− mice were purchased from Taconic (reference 15660) and NCR1cre Csf2flox were gifted by Professor Wick. All mice were used in experiments between 8 and 12 weeks of age. The protocol numbers are #35480, #46231 and #55889.### Heterotopic subcutaneous injection of tumor cells and growth monitoring

To induce s.c tumor formation, 3.105 CT26 cancer cells or 1.106 MC38 cancer cells were injected into Balb/c or C57BL/6J mice respectively. 7 days after tumor cell injection, mice tumors were measured using a caliper, and divided into groups with equal mean tumor size. For tumor growth experiments, tumor size was determined three times per week. Tumor surface is expressed in mm² and was calculated by multiplying the width by the length of each tumor. All mice were killed before reaching the limit points in accordance with the FELASA guidelines.### Orthotopic cecal injection of tumorous cells and growth monitoring

To minimize the risk of postoperative infections, antibiotic prophylaxis with 22,000 IU/kg of Duplocillin (MSD, Animal Health) was administered s.c daily within 72 hours prior to the orthotopic cecal injection. Mice were shaved at the incision site the day before the procedure.

10 min before anesthesia, to avoid pain, a s.c injection of buprenorphine (0.1 mg/kg) was administered. Moreover, to prevent dehydration, 300 µL of 0.9% saline solution (B Braun) was injected s.c. To prevent corneal damage, an ocular gel (Ocry-gel, TVM) was applied. To enhance the local anesthesia, lidocaine (Emla, Apsen) was applied to the abdomen. Anesthesia induction was performed with an intraperitoneal (i.p) injection of ketamine (75 mg/kg) and xylazine (10 mg/kg).

Asepsis was achieved by double application of Vetadine at the surgical site, followed by the application of an adhesive film (Medline).

First, a midline skin and peritoneum incision was made to access the abdominal cavity. A quick exploration was conducted, and the cecum was exteriorized. Subsequently, a subserosal injection of 30 µL of a solution containing 1.5×10⁵ CT26 Luciferase cells diluted in Matrigel (Fisher Scientist) was performed using a sterile insulin syringe (B Braun). This step was performed under a trinocular stereo microscope (OZM 933, Kern). Finally, the peritoneum and the skin layer were sutured with respectively continuous stitches and interrupted stitches using Monocryl 6/0 (Laboderm, 46378). To reinforce skin closure, a biological glue (topical skin adhesive, derma+flex) was applied to the incision immediately after surgery. Between procedures, instruments were sterilized using a beads sterilizer (Germinator 500).

After the surgery, mice were placed in an incubator (Cimuka) at 37°C and monitored until fully awake. During monitoring, hydration was assessed and 100 µL of saline solution (B. Braun) was reinjected s.c if needed. Temperature and cyanosis were also checked.

Postoperative follow-up included daily monitoring of wound healing progress, weight recovery, grooming behavior, and pain assessment. To promote rehydration and nutrient intake, a gel containing water and nutrients was given to the mice (Safe Geldiet). Analgesia continued for 3 days with two times a day s.c injections of buprenorphine (0.1 mg/kg).

7 days after the surgery, an initial bioluminescence measurement was performed (IVIS Luminalll, Perkin Elmer) 15 min after i.p injection of luciferin (150 mg/kg). Once the tumors were established (bioluminescence value of at least 2×10⁶ p/sec/cm²/sr), mice were divided into groups with equal average bioluminescence at the start of treatment. Bioluminescence was measured every three to 4 days until the defined ethical criteria. Bioluminescence images were analyzed using Living Image software (Perkin Elmer), and results were represented as fold-change in bioluminescence.### Mice treatment

The mice were treated with an anti-PD-1 antibody (200 µg/mouse, clone RMP1-14, BE0146 BioXCell) at D2, D4, D6 and D8 or its isotype control (Rat IgG2a clone 2A3, BE0089 BioXCell), with an anti-VEGFR-2 antibody (200 µg/mouse, clone DC101, BP0060 BioXCell) at D0, D2, D4, D6 or its isotype control (rat IgG1 clone HRPN, BP0088 BioXCell), with an anti-CXCR3 antibody (200 µg/mouse, clone CXCR3-173, BE0249 BioXCell) at D0, D2, D4, D6 and D8 or its isotype control (polyclonal Armenian hamster IgG, BE0091 BioXCell) with an anti-GM-CSF antibody (200 µg/mouse, clone MP1-22E9, BE0259 BioXCell) at D0, D2, D4, D6 and D8 or its isotype control (Rat IgG2a clone 2A3, BE0089 BioXCell), with an anti-CD8β (100 µg/mouse, clone 53–5.8, BE0223 BioXCell) at D0 and D7 or its isotype control (rat IgG1 clone HRPN, BP0088 BioXCell), with an anti-IL10 antibody at D0, D2, D4, D6 and D8 (100 µg/mouse, clone JES5-2A5, BE0049 BioXCell) or its isotype control (rat IgG1 clone HRPN, BP0088 BioXCell), with an anti-IL10R antibody at day 7 and 14 (20 µg/mouse, clone 1B1.3A, BP0050 BioXCell) or its isotype control (rat IgG1 clone HRPN, BP0088 BioXCell), with an anti-Asialo-GM1 antibody (10 µl/mouse, Fujifilm Wako) at day D0, D2, D4, D6 and D8 or phosphate-buffered saline (PBS) 1X, and with an anti-CD25 (200 µg/mouse, mcd25c2-mab10-1, InvivoGen) at D0. The mice were also treated with the anti-VEGF-A antibody at D0, D2, D4 and D6 that was kindly provided by Genentech (100 µg/mouse). All the treatments were diluted in PBS 1X and injected i.p.

For mice rechallenge, cured mice received 3.105 CT26 cancer cells in the opposite side s.c. 51 days after the first tumor cells injection.### Tumor dissociation

Tumors were harvested, cut into small pieces using scissors and placed into a gentleMACS C Tube (130–093-237, Miltenyi Biotec) with 2,35 mL of RPMI 1640, 100 µl of Enzyme D, 50 µl of Enzyme R and 12,5 µl of Enzyme A from the tumor dissociation kit (130–096-730, Miltenyi Biotec). For CXCR3 analysis, the Enzyme R quantity was reduced to 10 µL.

The tubes were added to the gentleMACS Octo dissociator (130–096-427, Miltenyi Biotec) following the 37mTDK1 program. Tumor supernatants used for subsequent cytokine analysis were collected at this step. The tumor cells were filtered on a 70 µm filter (130–110-916, Miltenyi Biotec) followed by three washing steps using PBS 1X. Cells were then ready to use for subsequent analysis.### Immune cells isolation

When several immune cell populations were analyzed (CD8 T cells, NK cells, macrophages and DC), fluorescence-activated cell sorting (FACS) was used using a BD FACSAria III (BD Biosciences) following tumor dissociation. CD8 T cells were defined as CD45+ CD3+ CD8a+ cells, NK cells as CD45+ CD3− NKp46+ and CD49b+, macrophages as CD45+ CD11b+ F4/80+ and DC as CD45+ CD11c+ cells. The antibodies used were the following: viability dye (FVS780, BD Biosciences, 565388), anti-CD45 (VioGreen, REA737, Miltenyi Biotec, 130–110-803), anti-CD3 (FITC, 17A2, Miltenyi Biotec, 130–118-958), anti-CD49b (BB700, HMα2, BD Biosciences, 742140), anti-NKp46 (APC, REA815, Miltenyi Biotec, 130–112-202), CD8a (V450, 53–6.7, BD Biosciences, 560469), anti-CD11b (PE-Vio770, REA592, Miltenyi Biotec, 130–116-246), anti-F4/80 (BV605, T45-2342, BD Biosciences, 123133), anti-CD11c (PE, N418, BioLegend, 117308).

In case of single immune cell population isolation, magnetic sorting was performed (QuadroMACS Separator, 130–091-051, Miltenyi Biotec). For the purification of intratumor NK or CD8 T cells, the mouse NK Cell Isolation kit followed by the mouse CD45 (TIL) Microbeads kit (130–115-818 and 130–110-618, Miltenyi Biotec) or the CD8 (TIL) Microbeads kit (130–116-478, Miltenyi Biotec) were used respectively. For splenic immune cell isolation, spleen and lymph nodes from naïve mice were harvested and then dissociated before being filtered on a 70 µm filter (130–110-916, Miltenyi Biotec). Red blood cell lysis buffer (EDTA 0.1 mM, NH4Cl 0.83% et KHCO3 0.1%) was next added for 1 min. After centrifugation, NK cells or CD8 T cells were isolated using the mouse NK Cell Isolation Kit (130–115-818, Miltenyi Biotec) or the mouse CD8a (Ly-2) MicroBeads kit (130–117-044, Miltenyi Biotec), respectively, following the manufacturer’s instructions. Regarding peritoneal macrophage isolation, peritoneal cells were harvested through peritoneal lavage using cold PBS 1X containing 3% FBS. Cells were then washed and macrophages were sorted using the F4/80 MicroBeads UltraPure purification kit (130–110-443, Miltenyi Biotec) following the manufacturer’s instructions.### Ex vivo immune cell culture

After FACS sorting, the cells were kept in RPMI 1640 supplemented with 10% FBS (Dutscher) and Penicillin-Streptomycin (10,000 IU/mL Penicillin, 10 mg/mL Streptomycin, 25 µg/mL Amphotericin B, PAN Biotech, P06-07300) for 24 hours. Supernatants were collected then frozen while the cells were resuspended in TRIzol reagent and frozen for subsequent analysis.

Following magnetic sorting, intratumor CD8 T cells were kept 24 hours in RPMI 1640 containing 10% FBS, 1% Penicillin-Streptomycin, 1% MEM Non-Essential Amino Acids Solution (Gibco, 11140–035), 1% Sodium Pyruvate (Gibco, 11360–070) and 1% L-Glutamine (Gibco, 25030–081) (referred to after as complete RPMI). Supernatants were then collected and frozen for subsequent analysis.

Splenic NK cells were incubated for 24 hours with increasing concentrations of mouse IL-10 (Miltenyi Biotec, 130–094-068) in complete RPMI supplemented with 80 IU/mL of mouse IL-2 (Miltenyi Biotec, 130–120-662). Cells were then harvested and analyzed by flow cytometry.

Splenic CD8 T cells were incubated in activating (coated plate with 2 µg/mL of anti-CD3 (InVivoMab, BE0002) and anti-CD28 (InVivoMab, BE0015-1) antibody) or resting conditions with increasing concentrations of mouse VEGF-A (Miltenyi Biotec, 130–094-086). At 24 hours, 48 hours and 72 hours, supernatants were collected and then frozen for subsequent analysis.

Peritoneal macrophages were incubated with increasing concentrations of mouse GM-CSF (Miltenyi Biotec, 130–094-043) in complete RPMI. At 24 hours and 48 hours, supernatants were collected and then frozen for subsequent analysis.### Transwell assay

50.103 isolated mice spleen or intratumoral NK cells were added in the top chamber of the 5 µm HTS Transwell 96 well plate (3387, Corning) in RPMI 1640 containing 10% FBS supplemented with 80 IU/mL of murine IL-2 (130–120-662, Miltenyi Biotec) and in some cases, rapamycin (553210, R&D systems) at 20 nM as previously described (10.3389/fimmu.2021.619195). In the bottom chamber, RPMI 1640 containing 10% FBS supplemented with murine IL-2 and increasing murine IL-10 (Miltenyi Biotec, 130–094-068) concentrations were added. After 4 hours of incubation at 37°C, cells in the bottom chamber were harvested, stained for mouse NK cell surface markers and viability, and then acquired on a flow cytometer using counting beads (424902, BioLegend). Migration index was calculated by doing a ratio between migrated cell numbers of each condition compare to the control condition (without IL-10).### Flow cytometry analysis

To study cytokine production by immune cells, tumor cells were incubated for 4 hour in a 96-well U bottom plate in Cell Stimulation Cocktail plus protein transport inhibitors (eBioscience, 00–4975-93, Thermo Fisher Scientific) diluted in RPMI 1640 medium supplemented with 10% FBS (Dutscher) and Penicillin-Streptomycin, (10,000 IU/mL Penicillin, 10 mg/mL Streptomycin, PAN Biotech).

Viability staining was performed in PBS 1X for 10 min in the dark at 4°C. To avoid a specific antibody binding, cells were incubated with the FcR Blocking Reagent (130–092-575, Miltenyi Biotec) according to the manufacturer’s instructions. Then, cell surface markers labeling was performed 15 min in the dark at 4°C in Brilliant Stain Buffer (566349, BD Biosciences) using antibodies diluted at 1/50e. Intracellular staining was performed after fixation and permeabilization of the cells using the Foxp3 Staining Buffer Set (130–093-142, Miltenyi Biotec) for 30 min. Finally, intracellular markers labeling was performed 45 min in the dark at 4°C using antibodies diluted at 1/25e. For Foxp3 labeling, the fixation and permeabilization step as well as the labeling step were both realized for 1 hour. Stained cells were acquired using a BD LSR Fortessa (BD Biosciences) and analyzed with FlowJo software. To identify NK and CD8 T cells (online supplementary figure S10-11), cells were stained with a viability dye FVS510 (BD Biosciences, 564406), anti-CD45 (VioBlue, REA737, Miltenyi Biotec, 130–110-664), anti-CD8a (BUV395, 53–6.7, BD Biosciences, 563786), anti-CD3e (FITC, 145–2 C11, BioLegend, 100306), anti-Nkp46 (APC-Vio770, REA815, Miltenyi Biotec, 130–112-361), anti-CD49b (BB700, HMα2, BD Biosciences, 742140), anti-Granzyme B (PE, REA226, Miltenyi Biotec, 130–116-486), anti-CD69 (PE-Cy7, H1.2F3, BioLegend, 104512), anti-IFNγ (PE-Cy7, xmg1.2, BioLegend, 505825), anti-PD1 (BV605, J43, BD Biosciences, 563059), anti-Tim3 (Pe-Cy7, RMT3-23, BioLegend, 119716), anti-SlamF6 (FITC, 13G3, Miltenyi Biotec, 130–109-859) and anti-Ki-67 (BV605, 16A8, BioLegend, 652413). To identify macrophages (online supplementary figure S12), cells were stained with a viability dye (FVS575, BD Biosciences, 565694), anti-F4/80 (FITC, BM8, BioLegend, 123108), anti-CD11b (VioGreen, REA592, Miltenyi Biotec, 130–113-811), IA/ie, (APC-Vio770, REA983, Miltenyi Biotec, 130–112-233). To identify CD4 and Treg (online supplementary figure S13), cells were stained with a viability dye FVS510 (BD Biosciences, 564406), anti-CD4 (BUV805, BD Biosciences, 612900), anti-CD25 (BV421, BD Biosciences, 564370), anti-Foxp3 (PE, Invitrogen, 12–5173-82), anti-CXCR3 (BV510, BioLegend, 126528). To identify basophiles, cells were stained with a viability dye, anti-CD45 (BUV805, 30-F11, BD, 568336) anti-CD49b (BB700, HMα2, BD, 742140), anti-CD200R3 (VioBlue, REA128, Miltenyi, 130–103-387) and anti-CD11b (VioGreen, REA592, Miltenyi, 130–113-811) while being negative for CD3 (APC-Vio770, REA641, Miltenyi, 130–119-793). To identify eosinophiles, cells were stained with a viability dye anti-CD45 (BUV805), anti-CD11b (VioGreen), anti-CD11c (PeCy7, HL3, BD, 558079), anti-siglecF (BB515, E50-2440, BD, 564514), were negative for CD3 (APC-Vio770). To identify neutrophils, cells were stained with a viability dye, anti-CD45, anti-CD11b, anti-Ly6G (BV421, 1A8, BioLegend, 127641). To identify monocytes, cells were stained with a viability dye, anti-CD45, anti-Cd11b, anti-Ly6C (PE, 1G7.G10, Miltenyi, 130–117-522), negative for F4/80 (BV605, T45-2342, BD, 569237) and CD3 (APC-Vio770). Subpopulations of DCs were defined as cDC1: CD45+ (BUV805) CD3− (APC-Vio770), CD49b− (BB700), CD11c+ (PeCy7), XCR1+ (PE-Dazzle594, ZET, BioLegend, 148234), while cDC2 were defined as CD45+ (BUV805), CD3− (APC-Vio770), CD49b− (BB700), CD11c+ (PeCy7), CD11b+ (VioGreen).

Additional antibodies were used: anti-SPP1 (AF405, NBP3-20774AF405, Bio-Techne), anti-Tnfsf9 (APC, Miltenyi Biotec, 130–116-087), anti-VEGFR-2 (PE, BioLegend, 121906) and anti-CD210 (PE, BioLegend, 112705). The H-2Ld MuLV gp70 Tetramer- SPSYVYHQF (MBL International, TB-M521-2) staining was performed according to the manufacturer’s instructions.

In the case of the labeling of NK cells from mice blood, the blood was first collected from the mandibular vein of the mouse in tubes containing EDTA (Microtube SARSTED, 41.1395.105). The red blood cells were lysed using a homemade red blood cell lysis buffer (EDTA 0.1 mM, NH4Cl 0.83% and KHCO3 0.1%) prior to the staining procedure described above.

The number of intratumor immune cells was all normalized using the following formula:### Transfer of CD8 T cells into CT26 tumors

We gathered CD8 T cells coming from CT26 tumors bearing WT Balb/c mice or IL-10−/− mice induced as described earlier, 1 day after a treatment with an anti-VEGFR2 (200 µg/mouse). After dissociation using the tumor dissociation kit (130–096-730, Miltenyi Biotec), we magnetically sorted CD8 T cells from tumors using CD8 (TIL) Microbeads kit (130–116-478, Miltenyi Biotec). We then injected those CD8 T cells, WT or IL-10−/− within CT26 tumors on IL-10−/− mice and analyzed the NK populations 24 hours later. 80,000 CD8 T cells were transferred in the tumor of each mouse.

We also transferred CD8 T cells VEGFR2+ or VEGFR2− into CT26 tumors on Nude mice. We isolated CD8 T cells from CT26 tumors on WT mice 1 day after the first anti-VEGFR2 treatment by flow cytometry (Aria II, BD) and sorted CD8 T cells defined as: CD45+, CD3+, CD4−, CD8+ Viability and VEGFR2+ and VEGFR2−. Those CD8 T cells were transferred into CT26 tumors on Nude mice with 50,000 CD8 being transferred per tumor. Mice were then treated by the bitherapy as previously described and tumor growth was assessed three times per week using a caliper. All mice were killed before reaching the limit points in accordance with the FELASA guidelines.### Cytokine analysis

Cytokine analysis was performed on CT26 tumor supernatants at D1–D5 and D9 after the treatment. Tumor supernatants were centrifuged at 12,000 g for 5 min and then stored at −20 °C until being used. The LEGENDplex Custom Mouse Panel including TNF-α and the Luminex Assay Customization including GM-CSF and IFNγ were used following the manufacturer’s instructions.

ELISA to measure the concentration of murine IL-10 (DY417, R&D systems), murine GM-CSF (DY415, R&D systems), murine CXCL9 (DY492, R&D systems), murine CXCL10 (DY466, R&D systems), murine CXCL11 (DY572, R&D systems) and murine CCL5 (DY478, R&D systems) were also performed according to the manufacturer’s instructions.### Real-time quantitative PCR

After tumor cell preparation, 1.106 cells were resuspended in TRIzol reagent (15 596 026 Invitrogen) and RNA extraction was performed following the manufacturer’s instructions. The quantity of RNA extracted was determined using the Nanodrop 2000 (Thermo Fisher Scientific). 300 ng of RNA was reverse transcribed into complementary DNA (cDNA) using the PrimeScript RT Reagent Kit (Takara, RR037A). Then, the samples were diluted at 1/10e and cDNA was quantified using the PowerUp SYBR Green Master Mix (A25742, Applied Biosystems). The results were obtained using the ViiA 7 Real-Time PCR System and the QuantStudio Real-Time software. Relative expressions were determined relative to βactin with the 2−Δ ΔCt method. The primers sequence was: Cxcl9 forward 5’-TAGGCAGGTTTGATCTCCGT-3’ and reverse 5’-CGATCCACTACAAATCCCTCA-3’, Cxcl10 forward 5’-CCAAGTGCTGCCGTCATTTTC-3’ and reverse 5’-GGCTCGCAGGGATGATTTCAA-3’, Cxcl11 forward 5’-CCGAGTAACGGCTGCGACAAAG-3’ and 5’-CCTGCATTATGAGGCGAGCTTG-3’ reverse, Csf2 forward 5’-TTTTCCTGGGCATTGTGGTCTA-3’ and reverse 5‘-TCTCTCGTTTGTCTTCCGCT-3’.### NanoString analysis

For the nCounter analysis (NanoString Technologies), 100 ng of RNA were used with the mouse-specific nCounter PanCancer Immune Profiling Panel. The protocol was carried out according to the manufacturer’s instructions. Samples were prepared and mixed with the nCounter panel Master Mix (containing the Reporter CodeSet and Capture ProbeSet), and each sample was then loaded onto an nCounter cartridge for data collection. The data were analyzed and collected using a NanoString analyzer. The nCounter RNA counting data were normalized using the geometric mean of positive controls and housekeeping genes. Normalization was performed using z-score calculations.### Single-cell RNA sequencing

Cell preparation

CT26 tumors from mice treated or not with an anti-VEGFR-2 antibody (200 µg/mouse, clone DC101, BP0060 BioXCell) were harvested and a single-cell suspension was obtained following the protocol from 10x Genomics (CG000147). Then, a debris removal (130–109-398, Miltenyi Biotec), a dead cell removal (130–090-101, Miltenyi Biotec), anti-F4/80 microbeads Ultrapure magnetic sorting (130–110-443, Miltenyi Biotec) followed by a CD45 Microbeads (130–110-618, Miltenyi Biotec) magnetic sorting were performed.### Single-cell RNA-seq

15,000 cells were captured thanks to a single-cell 3′ Gel Bead kit V.3.1 and single-cell 3’GEM kit V.3.1 on each well of a Chip G (10X Genomics). Library preparation, from reverse transcription to final amplification, was performed with a single-cell 3′ library kit V.3.1 (10X Genomics) following manufacturer’s instructions. Each library was paired-end (28/90 bp) sequenced on a NextSeq 2000 device (Illumina) at a depth of 1 billion reads, resulting in a read depth of 20,000 reads per cell.### Raw data analysis

Cell Ranger mkfastq (V.7.1.0) (10x Genomics) was used to generate demultiplexed FASTQ files from the raw sequencing reads. The resulting reads were then aligned to the mouse mm10 reference genome and gene counts were quantified using Cell Ranger count (V.7.1.0) (10x Genomics).### Single-cell RNA-Seq analysis

Count matrices corresponding to D1Ig and D1αR2 were loaded into R (V.4.3.3) and filtered using the following parameters: cells having9,000 genes and presenting over 10% mitochondrial transcript were excluded. Filtered gene expression matrices were then merged into a single object, normalized and scaled using Seurat^[46] (V.5.0.3). The top 2,000 most variable genes were calculated prior to integration using cca method. Uniform manifold approximation and projection (UMAP) dimensional reduction was calculated using the top 35 principal components. Seurat clusters were identified using a resolution of 1.2. Clusters were annotated using canonical markers from the literature and the most differentially expressed genes (DEGs). UMAP plots of clusters and Ptprc and violin plots were generated using respectively Seurat DimPlot, FeaturePlot and VlnPlot. Dot plots were generated using DotPlotHeatmap from RightOmicsTools^[47] (V.2.2.0). The Joint Density plots were generated using PlotDensityJointOnly from scCustomize^[48] (V.2.0.1). The volcano plot was generated using a custom ggplot2 (V.3.5.1) code. The cell–cell communication analysis was performed using CellChat^[49] (V.2.1.2) on a subset containing four NK cells clusters and two T cells clusters. Bubble plots were generated using netVisualbubble and networks were generated using ccnetwork from CCPlotR^[50] (V.0.99.3). The gene set enrichment analysis was performed using fgsea^[51] (V.1.28.0) on the statistically significant (p