Accumulation of MDSC and Th17 Cells in Patients with Metastatic Colorectal Cancer Predicts the Efficacy of a FOLFOX–Bevacizumab Drug Treatment Regimen

Colorectal cancer is the fourth most commonly diagnosed cancer worldwide and is a major cause of cancer-related deaths. Numerous studies have demonstrated a link between chronic inflammation and many cancers, including colorectal cancer. For example, inflammatory bowel disease is associated with increased incidence of colorectal cancer (1). Colorectal cancer is controlled by the immune system: Accumulating evidence shows that T-cell infiltration of primary tumors is associated with better prognosis (2, 3). Similar findings were also observed in patients with metastatic colorectal cancer (mCRC; ref. 4). In particular, infiltration with memory CD8 T cells, follicular helper T cells, or regulatory T cells (Treg) is associated with better tumor prognosis in colorectal cancer (5–7). However, infiltration with Th17 cells is associated with poor prognosis (8) due to the secretion of proinflammatory cytokines like IL17A, which can promote angiogenesis or other protumor effects (9).

Myeloid-derived suppressor cells (MDSC) are a heterogeneous population composed of myeloid cells blocked at several stages of differentiation. These cells were observed to accumulate in the blood, lymph nodes, bone marrow, and tumor sites of tumor-bearing patients and in experimental animal models of cancer (10, 11). These cells are characterized by their ability to inhibit both innate and adaptive immune responses, thus having a negative effect on antitumor immunity (12). In humans, MDSCs are not well characterized, partially because no unified markers are currently available for these cells. However, these cells typically express the common myeloid markers, CD33 and CD11b, but lack markers of mature myeloid cells, such as HLA-DR. Among human MDSCs, the monocytic subset comprises CD14⁺ cells, and the granulocytic subset comprises CD15⁺ cells (13, 14). We previously reported that 5-fluorouracil (5-FU) could induce MDSC depletion in mouse models of cancer (15) but enhance Th17 cell accumulation (15, 16).

Currently, mCRC patients are frequently treated in front line with a combination of chemotherapies, which associates 5-FU with either oxaliplatin (FOLFOX) or irinotecan (FOLFIRI). In addition to the chemotherapeutic regimens, patients are often treated with biotherapies, such as antiangiogenic agents like the anti-VEGF mAb bevacizumab or anti-EGFR mAb. The effect of such combination chemotherapies on peripheral immune responses is currently poorly explored. In addition, the prognostic value of initial immune parameters and their putative alteration by chemotherapy is also unknown.

In this work, we have first compared blood repartition of memory T-helper cells and MDSC [granulocytic MDSC (gMDSC) and monocytiic MDSC (mMDSC)] subsets in 20 healthy volunteers versus 25 consecutive mCRCs treated with FOLFOX plus bevacizumab combination as first line and tested the association of immune parameters with patients' outcome.

Between January 2014 and October 2014, we collected whole blood from healthy volunteers (n = 20) and mCRC patients (n = 25) at the Centre Georges François Leclerc (Dijon, France). mCRC patients were diagnosed in our cancer center and were proposed to be treated with FOLFOX plus bevacizumab as a first-line regimen. FOLFOX regimen was given every 14 days as follows: oxaliplatin (85 mg/m² over 2 hours), leucovorin (400 mg/m² over 2 hours), and 5-FU (400 mg/m² bolus, then 2,400 mg/m² over 46 hours). Bevacizumab was given at a dose of 5 mg/kg once every 2 weeks.

After 12 cycles or sooner in case of toxicity, responder patients received a maintenance treatment with bevacizumab, leucovorin (400 mg/m² over 2 hours), and 5-FU (400 mg/m² bolus, then 2,400 mg/m² over 46 hours every 2 weeks). All patients must have received at least four cycles of chemotherapy to be evaluable. Tumor response was prospectively assessed every four cycles according to RECIST criteria by CT scan (17). Treatment was repeated until the occurrence of disease progression or unacceptable toxicity, whichever occurred first.

We added another study of immunomonitoring performed on 20 patients treated in Centre Georges François Leclerc for mCRC in second or third line by 5-FU–based chemotherapy plus bevacizumab combination.

All patients gave informed consent approved by the local Ethics Committee. The collection of blood sample is authorized by the French authorization (nr. AC2014-2460). No additional blood samples beyond those required for routine testing were taken. Whole blood of mCRC patients was sampled before (D0) and after chemotherapy (D15, D30, and D60) on EDTA-K2 tubes (BD Biosciences) for complete blood count, which was performed in our Clinical Biology Unit (Centre George François Leclerc) and on heparinized tube for leucocyte phenotyping. All analyses were performed following the first 6 hours after sampling. Review of pathology reports confirmed the diagnosis. Information regarding clinical, pathologic, and biological characters of patients and healthy volunteers is presented in Supplementary Table S1.

Anti-CXCR3-PE-Vio700 (REA232), anti-CCR6-PE (REA190), anti-CD25-APC (4E3), anti-CD45RA-APV-Vio770 (T6D11), anti-CD4-VioGreen (VIT4), anti-CD8-VioGreen (BW135/80), anti-CD33-APC-Vio770 (AC104.3E3), anti-HLA-DR-Vioblue (AC122), anti-CD15-VioGreen (VIMC6), anti-CD14-PerCP-Vio700 (TUK4), anti-CD3-FITC (BW264/56), anti-CD56-FITC (REA136), anti-CD19-FITC (LT19), anti-CD20-FITC (LT20), anti-TNF-α–PE (REA656), and anti-Foxp3-PE (3G3) were purchased from Miltenyi Biotec. Anti-CCR4-BV450 (1G1), anti-PD-L1-APC (MIH1), anti-Ki-67-eFluor450 (20Raj1), and anti-PD-1-PerCP-eFluor710 (J105) were purchased from BD Biosciences and eBioscience, respectively. Anti-CD4-Alexa Fluor 700 (RPA-T4), anti-CD39-PE (A1), anti-CD73-Brillant Violet 421 (AD2), and anti-IL17A-Pacific Blue (BL168) were purchased from BioLegend. All events were acquired by a BD LSR-II cytometer equipped with BD FACSDiva software (BD Biosciences), and data were analyzed using FlowJo software (Tree Star).

For leucocytes identification by flow cytometry, whole blood removed to heparinized tube (100 μL) was stained with different antibody cocktail for 45 minutes at room temperature. For MDSC identification, we used lineage cocktail (CD3, 56, 19, and 20), CD33, CD15, CD14, and HLA-DR antibody. For Treg analysis, we used CD4, CD45RA, CD25, and Foxp3 antibodies, and for other T-helper subsets, we used CD4, CD45RA, CD25, CCR6, CXCR3, and CCR4 antibodies. The gating strategy is described in Supplementary Fig. S1 and S2. After surface staining, 2 mL of red blood cells lysis solution (BD Biosciences) was added for 10 minutes, centrifuged (400 × g, 5 minutes), and then resuspended in flow cytometry buffer (eBioscience). Foxp3 staining was carried out according to the manufacturer's protocol using the fixation/permeabilization solution (eBioscience).

To test MDSC subset suppressive activity, total CD3⁺ lymphocytes and MDSC subsets were sorted from mCRC patient blood (around 15 mL). For gMDSC and granulocytes isolation (around 7.5 mL of blood), we used Whole Blood CD15 MicroBeads and column following the manufacturer's instructions (Miltenyi Biotec). For mMDSC and monocytes isolation (around 7.5 mL of blood), we performed peripheral blood mononuclear cell (PBMC) isolation on lymphocyte separation medium (Eurobio). CD15⁺ cells and PBMCs were then stained with anti-CD33-APC-Vio770 (AC104.3E3), anti-HLA-DR-Vioblue (AC122), anti-CD15-VioGreen (VIMC6), anti-CD14-PerCP-Vio700 (TUK4), and anti-CD3-FITC (BW264/56) in flow cytometry buffer (eBioscience) for 30 minutes. Myeloid subsets (gMDSC, mMDSC, monocytes, and granulocytes) and total T cells were cell sorted on ARIA-III (BD Biosciences). CD3⁺ cells were activated with 2 μg/mL anti-CD3 (Bio X Cell) and anti-CD28 (Bio X Cell) as effector cells and cocultured with or without gMDSC, mMDSC, granulocytes, or monocytes (T-cell/myeloid cell ratios are 10:1 or 25:1) for one week in culture medium (AIM V Medium, Fisher Scientific) in the presence of 2 μmol/L of ATP (Sigma). In some experiments, the CD39-neutralizing antibody (OREGA Biotech, clone BY40, 10 μg/mL) and/or anti-PD-1 antibody (Nivolumab, 2 μg/mL) were added. In some experiments, TNFα and Ki67 expression was assessed in CD3⁺ T cells by flow cytometry after 24 hours of coculture with or without MDSCs. This was performed in the presence of anti-CD3 (2 μg/mL, Bio X Cell) and anti-CD28 (2 μg/mL, Bio X Cell). Mouse IgG1 (11711, R&D Systems) and human IgG4 (ET904, eBioscience) antibodies were respectively used as a control for anti-CD39 and anti-PD-1 antibody efficacy.

After 5 days of culture, cell culture supernatants were assessed by ELISA for human TNFα (Biolegend) according to the manufacturer's protocol.

For intracellular cytokine staining, cells were stimulated for 4 hours at 37°C in culture medium containing PMA (50 ng/mL; Sigma-Aldrich), ionomycin (1 μg/mL; Sigma-Aldrich), and monensin (GolgiStop; 1 μL/mL; BD Biosciences). After staining for surface markers [anti-CD3-FITC (BW264/56)], cells were fixed and permeabilized according to the manufacturer's instructions (Fixation/Permeabilization Kit; eBiosciences), then stained for intracellular products. Antibodies used for intracellular staining were as follows: phycoerythrin (PE)-conjugated anti-TNFα or eFluor 450–conjugated anti-Ki67.

Total RNA from T cells was extracted with TRI Reagent (Ambion), reverse transcribed using M-MLV Reverse Transcriptase (Invitrogen), and was analyzed by qRT-PCR with the SYBR Green method according to the manufacturer's instructions using the 7500 Fast Real Time PCR System (Applied Biosystems). Expression was normalized to the expression of human ACTB. Primers designed to assess gene expression are as reported in Supplementary Table S2.

For the analysis of data, comparison of continuous data was achieved by the Mann–Whitney U test or Wilcoxon Test and comparison of categorical data by Fisher exact test, as appropriate. All P values are two tailed. P < 0.05 was considered significant. Data are represented as mean ± SEM. All patients were followed up until death or the end of data recording (May 30, 2015). Progression-free survival (PFS) was calculated from the date when therapy started to the date of disease progression, and overall survival was calculated from the date when therapy started to the date of death. Median follow-up with its 95% confidence interval (CI) was calculated using the reverse Kaplan–Meier method. Survival probabilities were estimated using the Kaplan–Meier method, and survival curves were compared using the log-rank test. Analyses were performed using MedCalc Software.

We proposed to separate patients using median as a cutoff to have comparable number of patients in the two groups. The rate of PFS under FOLFOX–bevacizumab is about 60% at 6 months (18). We decided to only foster on clinically relevant biomarker that could separate a group of patients with 6 months PFS rate of 40% versus 80%. With a risk α of 5% and a power of 80%, we needed a minimum of 11 patients per group to detect a difference. For this reason, we decided to include 25 patients.

We first analyzed the frequency of circulating immune cell populations from mCRC patients at baseline compared with healthy individuals. The clinical characteristics and blood parameters of patients and healthy volunteers are summarized in Supplementary Table S1 and Supplementary Fig. S3. To assess the frequency of memory T-cell subpopulations, we relied on their chemokine receptor expression using a gating strategy adapted from Mahnke and colleagues (19). The transcription factor Foxp3 occurred with a concomitant high expression of CD25, while Th1, Th2, Th17, and Th17/Th1 cells were analyzed on the basis of their expressions of CCR6, CXCR3, and CCR4 (Supplementary Fig. S1). To validate our gating strategy, we analyzed IFNγ and IL17A cytokine expressions in memory T-CD4 cells regarding chemokine receptor expression (Supplementary Fig. S4). The frequency of memory Th2 cells was comparable between healthy volunteers and mCRC patients while there was a small decrease in memory Th1 (P = 0.035) in mCRC (Fig. 1A). In contrast, we observed a significant increase in the number of Treg (P < 0.0001) and Th17 cells (P = 0.019) while inflammatory Th17 cells expressing the CXCR3 marker, and called Th17/Th1 in the article, did not accumulate in mCRC patients (P = 0.071) (Fig. 1A). We then studied the prognostic role of accumulation of Treg or Th17 cells. We separated patients into two groups using median as a cutoff, and we did not observe any association between Treg (P = 0.68) or Th17 (P = 0.42) accumulation and PFS during first-line chemotherapy and overall survival (Fig. 1B and C and not shown). Th1, Th2, and inflammatory Th17 cells were also not associated with prognosis (not shown). Together, these data show alterations in the frequency of memory T-helper populations in mCRC patients compared with healthy donors. However, these changes do not seem to affect tumor prognosis.

Chemotherapy using 5-FU was shown to affect both MDSC levels and T-cell polarization (15, 16). Oxaliplatin was also shown to improve antitumor T-cell functions via a mechanism called immunogenic cell death (20). We observed that one cycle of chemotherapy did not significantly alter the number of circulating lymphocytes (Supplementary Fig. S5). T-helper frequencies (Th1, Th2, Treg, Th1/Th17, and Th17) were monitored at days 0, 15, 30, and 60 after chemotherapy. In the 25 mCRC patients, we did not observe any change in the frequency of Th1, Th2, and Th17/Th1. For Treg and Th17 cells, we observed only a respective decrease (P = 0.013) and increase (P = 0.047) frequency at day 15 that did not persist during follow-up (Fig. 2A). As a control, we could not detect significant modification of Treg and Th17 frequencies in untreated patients (data not shown). Importantly, when we focused on the impact of modifications of memory CD4 T-helper cell frequency on PFS, we observed a subset of patients whose Treg number is decreased 15 days after the first cycle of chemotherapy. This modulation is maintained during the whole follow-up (D30, P = 0.021; D60, P = 0.013). Nevertheless, Treg frequency modulation at day 15 was not associated with PFS (P = 0.42; Fig. 2B and C). Concerning Th17 cells' frequency modulation, we also observed a subset of patients whose Th17 percentage increased 15 days after the first cycle of chemotherapy. This increase is only transitory. However, this increase in Th17 cell frequency at day 15 was associated with a poor PFS (P = 0.048; Fig. 2D and E). In the group of patients with increased Th17 frequency, we observed 46% of partial response after 3 months of treatment, while all patients presented a partial response upon RECIST criteria in the group of patients with a decrease of Th17 cell frequency (P = 0.03, Fisher exact test). We could observe similar prognostic role of Th17 frequency decrease after a cycle of 5-FU–based chemotherapy plus bevacizumab in a series of 20 metastatic colorectal cancer patients treated by this protocol in second or third line [median PFS of 7 months vs. 4 months for patients with decreased vs. increased Th17 frequency (P = 0.04; data not shown)]. Together, these results show that after the first cycle of chemotherapy, Treg frequency decreased to a normal level while Th17 frequency increased in some patients. Increased Th17 level is associated with poor prognosis and resistance to therapy.

We also investigated the frequency of MDSCs using the Lin⁻ CD33^high HLA-DR⁻ cell labeling. Using CD14 and CD15 labeling, we separated the two types of MDSCs: gMDSCs, which express CD15 and CD14 marker; and mMDSCs, which express the CD14 marker only (Supplementary Fig. S2). To validate our gating strategy, we cell sorted from mCRC whole-blood CD15⁺ cells using magnetic beads. These cells were then separated cells upon CD33 labeling and we observed that only CD15⁺ CD33^high cells exerted immunosuppressive function and could be called gMDSCs, while CD15⁺ CD33^low cells did not exert suppressive function and could be called granulocytes (Fig. 3A–C). For CD14⁺ cells isolated from PBMC, all cells expressed CD33 labeling but only cells with HLA-DR^low expression had immunosuppressive function and could be called mMDSCs (Fig. 3A–C). In addition, only CD15⁺ CD33^high gMDSCs but not CD15⁺ CD33^low granulocytes expressed CD124 (IL4Rα), a classical marker of MDSC (Supplementary Fig. S2D). The proportion and absolute number of total MDSCs in peripheral blood from untreated mCRC patients was significantly increased compared with the proportion found in healthy donors (P = 0.0027; Fig. 3D). We also observed that both the frequency (Fig. 3D) and the absolute number (Fig. 3E) of gMDSC (P = 0.0013 and P < 0.001), but not mMDSC (P = 0.15 and P = 0.485), significantly accumulate in the blood of mCRC patients compared with healthy donors. We then studied the prognostic role of MDSCs. As described above, we separated patients into two groups using median MDSC frequency as a cutoff. We observed that high levels of gMDSC at baseline are significantly associated with poor PFS [7 months vs. median not reached (NR), P = 0.04] and overall survival (12 months vs. 23 months, P = 0.04; data not shown), while mMDSC level is not associated with prognosis (Fig. 3F and G). We also confirmed that decrease in the frequency of gMDSC is significantly associated with longer median PFS after a cycle of 5-FU–based chemotherapy plus bevacizumab in a series of 20 colorectal patients treated by this protocol in second or third line [median PFS 6 months vs. 4 months for patients with decreased vs. increased gMDSC frequency (P = 0.05; data not shown)]. Together, these data show that mCRC patients have high levels of MDSCs, and initial high levels of gMDSCs are associated with poor prognosis.

We previously observed that 5-FU induced MDSC cell death in humans, so we tested the effect of the first cycle of FOLFOX–bevacizumab chemotherapy on the frequency of MDSCs. We tested the evolution of MDSC frequency during the first 2 months of treatment. We observed that FOLFOX–bevacizumab chemotherapy did not impact on total MDSC or mMDSC frequency. In contrast, we observed a significant decrease in the number of gMDSCs 2 months after the beginning of the treatment (P = 0.008; Fig. 4A). As a control, we could not detect significant modification of MDSC frequency in untreated patients (data not shown). However, when we studied gMDSC, we found that 15 of 25 patients have a decreased frequency of gMDSC after the first cycle of chemotherapy compared with baseline (Fig. 4B). Analysis of gMDSC frequency at days 30 and 60 after initiation of the treatment confirmed that decreased frequency of gMDSC is maintained during the whole follow-up. For the other 10 patients, we did not observe a decrease in gMDSC frequency. Importantly, we observed that decrease in the gMDSC frequency is significantly associated with better prognosis and longer PFS (P = 0.006; Fig. 4C). In contrast for mMDSC, we observed that patients with a decreased frequency of mMDSC at day 15 did not maintain this evolution during the whole follow-up (D30, P = 0.57; D60, P = 0.99). Patients with an increased frequency of mMDSC at day 15 maintained this evolution during the whole follow-up (D30, P = 0.0013; D60, P = 0.006; Fig. 4D). The increase of mMDSC frequency did not change PFS (P = 0.09; Fig. 4E). We also confirmed that decrease in the frequency of gMDSC is significantly associated with longer median PFS after a cycle of 5-FU–based chemotherapy plus bevacizumab in a series of 20 metastatic colorectal cancer patients treated by this protocol in second or third line [median PFS 6 months vs. 4 months for patients with decreased vs. increased gMDSC frequency (P = 0.05; data not shown)]. Together, these data show that FOLFOX–bevacizumab induces a decrease of gMDSC, and this decrease is associated with better PFS.

Previous reports suggested that MDSCs in human cancers could suppress immunity using PD-L1 and CD39 (14, 21). PD-L1 interacts with PD-1, which is located on the surface membrane of effector memory T cells, and transmits an inhibitory signal reducing the proliferation and activity of these cells. CD39 is an ectonucleotidase that converts extracellular ATP or ADP into AMP. CD39 works together with CD73, another ectonucleotidase that converts AMP into adenosine, which is the final immunosuppressive molecule (22). Thus, both enzymes are required to convert ATP into the immunosuppressive molecule adenosine. We first tested the expression of PD-L1, CD73, and CD39 in the peripheral different myeloid cells (gMDSC and mMDSC, granulocytes, and monocytes) of mCRC patients and healthy volunteers. We observed that gMDSC is the myeloid population that expressed the highest levels of PD-L1, CD73, and CD39 (Fig. 5A–C). Interestingly, a higher expression of PD-L1, CD39, and CD73 is observed in gMDSCs of mCRC patients compared with healthy volunteers, thus suggesting a more important immunosuppressive activity in mCRC patients (Fig. 5A–C). We also confirmed that gMDSC of mCRC patients expressed more Pdl1, Entpd1, and Nt5e (coding for CD39 and CD73, respectively) mRNA than mMDSC (Supplementary Fig. S6).

To exert its immunosuppressive effect, PD-L1 must encounter PD-1 on a T cell. We checked the level of expression of PD-1 in CD4 and CD8 T-cell subsets of mCRC patients and observed a high expression of PD-1 on memory CD8 T cells in both healthy volunteers and mCRC patients. In CD4 T cells, PD-1 is only found on Th1 cells and a higher level is found in mCRC patients compared with healthy volunteers (Fig. 5D–F).

Then, we tested the expression of CD39 and CD73 in effector CD4 and CD8 T-cell subsets and observed that while CD8 and CD4 T cells expressed reduced level of CD73 in mCRC patients, a similar level of CD39 was observed in mCRC patients compared with healthy volunteers (Fig. 5D–F).

Together, these data demonstrate that in mCRC patients, gMDSC expressed high levels of PD-L1, CD73, and CD39, while Th1 and CD8 memory cells expressed PD-1, thus suggesting that gMDSC could blunt T-cell response using both ectonucleotidases and PD-L1.

To test the immunosuppressive function of MDSC from mCRC patients, we sorted CD3⁺ T cells and gMDSC and mMDSC from 5 mCRC patients. In an autologous model, we stimulated T cells with CD3 plus CD28 mAb and added in the culture either gMDSC or mMDSC in the presence of ATP molecule. As a control, we used granulocytes in this experiment. We observed that both gMDSC and mMDSC subsets have an immunosuppressive effect on T cells, demonstrated by reduced TNFα and Ki67 proliferation marker expression, however, gMDSC exert a more important immunosuppressive effect (Fig. 6A and B). Similar results were obtained using TNFα ELISA assay (Fig. 6C). Importantly, the addition of nivolumab, a clinically available anti PD-1 mAb, and/or the antagonistic anti-CD39 mAb BY40 blunted the immunosuppressive effect of gMDSC on CD3⁺ T cells, while it did not affect the immunosuppressive effect of mMDSC (Fig. 6A–C).

Together, these data show that gMDSCs that accumulate in mCRC patients have a high immunosuppressive function that could be targeted with anti-PD-1 or anti-CD39 antibodies.

Previous reports have shown that a high number of MDSC is associated with a poor prognosis in different solid cancers and in hematologic malignancies like chronic lymphocytic leukemia or myeloma (23–25). However, the prognostic role of MDSCs has not been yet addressed in mCRC. We observed here that accumulation of gMDSC in mCRC is associated with a poor outcome. However, due to the low number of patients, further studies are required to confirm this observation and determine whether gMDSC accumulation is an independent prognostic factor in mCRC.

We observed that mMDSCs are the most frequent MDSC population in patient blood. Such data contrast with previous reports on MDSC in colorectal cancer, in which the authors did not observe expression of CD15 or CD14 on MDSC or only detected CD15⁺ gMDSC (14, 21). Definition of human MDSC phenotype is still controversial. Although CD14⁺ Lin⁻ HLA-DR⁻ cells were commonly admitted to be mMDSC, the definitive of gMDSC is not consensual. Most reports define gMDSCs as low-density granulocytes copurified with PBMC during Ficoll gradient. Here, we observed that blood CD15⁺ CD33^high Lin⁻ HLA-DR⁻ cells are immunosuppressive cells and could be called MDSC, while CD15⁺ CD33^low Lin⁻ HLA-DR⁻ cells are not immunosuppressive and are granulocytes. Moreover, we could compare the functions of both gMDSC and mMDSC. We observed that gMDSC have a higher immunosuppressive function than mMDSC on a per cell basis. Previous data suggested that MDSC from mCRC patients have a more important immunosuppressive function compared with MDSC from healthy donors, but the comparison of the function of gMDSC and mMDSC was not addressed (21). We found that gMDSCs have higher levels of PD-L1, CD39, and CD73 expression and exerted a stronger immunosuppressive activity than mMDSCs. Together, such data suggest that gMDSC is the major subset of MDSC involved in immunosuppression in mCRC and strengthen the rationale to combine PD-1/PD-L1 or ectonucleotidase-neutralizing antibodies with chemotherapy to blunt immunosuppression induced by gMDSC in mCRC.

In a previous report, we observed that 5-FU mediated MDSC depletion via an induction of cell death (15). In humans, a recent study on a few colorectal cancer patients treated with FOLFOX (6 patients) or FOLFIRI (4 patients) showed that while FOLFOX treatment of patients with colorectal cancer led to a decrease in MDSC levels, FOLFIRI had the opposite effect (26). In this report, we also observed that FOLFOX–bevacizumab induces a decrease of MDSC but only of the gMDSC subset. This reduction of gMDSC frequency is preferentially observed in patients with initial high level of gMDSC (≥1% in leucocytes), thus raising the hypothesis that accumulating gMDSC from mCRC patients may have a specific biology that renders them more sensitive to chemotherapy-induced cell death.

We had previously reported that MDSC cell death mediated by 5-FU induced the activation of NLRP3 inflammasome and was associated with IL1β secretion. This IL1β secretion promoted induction of Th17 response. We demonstrated that Th17 induction was associated with the accumulation of proangiogenic molecules in the tumor. Such phenomenon is deleterious, and depletion of IL17A or IL1β is associated with enhanced efficacy of 5-FU treatment in mice (16). In this report, we observed that FOLFOX–bevacizumab induced Th17 accumulation after one cycle. Concomitantly, we also noted a decrease of gMDSC in patients with high initial levels. Such data strongly support that this combination of chemotherapy also affects antitumor immune response through similar mechanisms to those observed with 5-FU monotherapy in mouse and human models. Importantly, increase in Th17 cells and absence of reduction of gMDSC frequency are both associated with a poor prognosis, thus suggesting, in a similar way to mouse observations, that FOLFOX–bevacizumab induced a deleterious effect on immune response via induction of Th17 polarization.

In conclusion, our study shows that FOLFOX–bevacizumab chemotherapy induced accumulation of Th17 cells in patients, and this parameter is associated with poor prognosis, while gMDSC depletion is associated with better outcome. However, the most important point in this study is that gMDSC is an MDSC subset that could be targeted in mCRC using molecules that target the PD-1/PD-L1 or CD39/CD73 ectonucleotidase pathways. Thus, our data give rationale to use ectonucleotidase inhibitors or anti-PD-1/PD-L1 in association with FOLFOX–bevacizumab regimen in further clinical development of chemoimmunotherapy in human colorectal cancer.

N. Bonnefoy is a consultant/advisory board member for OREGA Biotech. L. Apetoh is a consultant/advisory board member for Bristol-Myers Squibb. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E. Limagne, S. Ladoire, L. Apetoh, F. Ghiringhelli

Development of methodology: E. Limagne, V. Derangère, F. Végran, S. Ladoire

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Limagne, R. Euvrard, M. Thibaudin, N. Bonnefoy, J. Vincent, L. Bengrine-Lefevre

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Limagne, R. Euvrard, M. Thibaudin, V. Derangère, R. Boidot, L. Apetoh, F. Ghiringhelli

Writing, review, and/or revision of the manuscript: E. Limagne, R. Euvrard, M. Thibaudin, C. Rébé, V. Derangère, F. Végran, N. Bonnefoy, D. Delmas, L. Apetoh, F. Ghiringhelli

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Rébé, A. Chevriaux, R. Boidot, S. Ladoire

Study supervision: E. Limagne, L. Apetoh, F. Ghiringhelli

We thank Aurelie Bertaut (statistician) for calculation of the required number of patients.

This work was supported by the Ligue Nationale contre le Cancer, the Institut National du Cancer, the Association pour la recherche sur le cancer, the Conseil Régional Bourgogne/INSERM, Fondation pour la Recherche Médicale (F. Ghiringhelli) the Fondation de France, the Fondation Lilliane Betancourt, the French National Research Agency (ANR-13-JSV3-0001 and ANR-11-LABX-0021), the Ligue Régionale contre le Cancer Comité Grand-Est, the Canceropôle Grand-Est, and the European Community (Marie Curie Fellowship PCIG10-GA-2011-303719 to L. Apetoh). This work was also supported by institutional grants from INSERM and Université de Montpellier 1 and from the LabEx MabImprove (N. Bonnefoy).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.