Immunosuppressive Myeloid Cells Induced by Chemotherapy Attenuate Antitumor CD4⁺ T-Cell Responses through the PD-1–PD-L1 Axis

Recent advances in therapeutic antibodies, cancer vaccines, and adoptive T-cell therapy (ACT) have manifested the tremendous therapeutic potential of immunotherapy in treating patients with cancer (1). However, the efficacy of cancer immunotherapy is often restricted by various immune suppressive mechanisms preestablished in the tumor microenvironment. The prevalent immunoregulatory mechanisms include immune suppressor cells such as T regulatory cells (Tregs) and myeloid-derived suppressor cells (MDSC), tolerogenic enzyme indoleamine 2,3-dioxygenase (IDO), immunosuppressive soluble factors such as TGFβ, IL10, and prostaglandin E₂ (PGE₂), and checkpoint molecules such as CTLA-4 and PD-1 (2). To enhance the efficacy of cancer immunotherapy, chemotherapy is often included in treatment regimens to condition an immune milieu conducive to therapeutic immunologic modulations. The notion of achieving synergistic antitumor effects through combined chemoimmunotherapy has been substantiated by the finding that many widely used chemotherapeutic agents have immunomodulatory effects (3). One complicated issue associated with chemoimmunotherapy strategy is that chemotherapy causes substantial cell death and tissue injuries, which often result in dynamic recovery of lymphoid and myeloid cells, along with extensive inflammation in the postchemotherapy setting. How these events shape the ultimate outcome of immunotherapy is not fully understood.

The antineoplastic agent cyclophosphamide (CTX) is frequently used in the treatment of hematopoietic malignancies. More importantly, cyclophosphamide represents a well-studied example of chemotherapeutic agent with strong immunomodulatory effects. At high doses, cyclophosphamide is cytotoxic and lymphoablative, causing severe immune suppression. At lower doses, cyclophosphamide exhibits multifaceted immune-potentiating effects (4, 5). It has been shown that cyclophosphamide can induce immunogenic tumor cell death that enhances antigen presentation (6), and mitigate immunosuppression by reducing the number and activity of T-regulatory cells (7). In addition, cyclophosphamide induces transient lymphopenia, creating niche-“space” and providing growth factors for the expansion and survival of transfused tumor-reactive T lymphocytes (8). Thus, the postchemotherapy period represents a window of opportunity to exploit for immunotherapy, especially adoptive T-cell therapy (9, 10). Despite cyclophosphamide's well-characterized immunostimulatory effects, it has long been documented that cyclophosphamide can induce a population of cells exhibiting immunosuppressive activities (11, 12). Some early studies suggested that cyclophosphamide-induced suppressor cells were a heterogeneous population with CD11b⁺ myeloid lineage cells being most suppressive (13, 14). Recent studies reported that cyclophosphamide-induced suppressor cells were mainly CD11b⁺Gr1⁺ MDSCs (15, 16). However, because MDSCs themselves are a heterogeneous population of immature myeloid cells (17), the cellular identity of cyclophosphamide-induced suppressor cells has not been well defined. More importantly, the biologic relevance of these cells in the context of cyclophosphamide-based cancer therapy has not been addressed.

In this study, we report that certain cytotoxic chemotherapeutic agents, including cyclophosphamide, can drive the expansion of myeloid cells consisting of monocytic and granulocytic populations. In mice with advanced B-cell lymphoma or lung metastasis of colon cancer, adoptive transfer (AT) of tumor-specific CD4⁺ T cells following cyclophosphamide resulted in heightened inflammation and enhanced myeloid cell expansion. Paradoxically, the inflammatory monocytes among therapy-induced myeloid cells exhibited immunosuppressive function, and promoted tumor escape by tolerizing CD4⁺ effector cells through the PD-1–PD-L1 axis. With direct clinical relevance, our study demonstrates that targeting therapy-induced monocytes, either by blocking their suppressive activity or reducing their accumulation, can lead to durable antitumor immunity and curative outcome. These findings may aid the design of more efficacious chemoimmunotherapy strategies.

Female BALB/c mice of 4 to 6 weeks old were purchased from the National Cancer Institute (Frederick, MD). Thy1.1^+/+ HA-TCR Tg mice expressing an αβ TCR specific for amino acids 110 to 120 from influenza hemagglutinin (HA) presented by MHC class II molecule IA^d were generous gifts from Dr. H.I. Levitsky (The Johns Hopkins University School of Medicine, Baltimore, MD). The PD-1KO mice on a BALB/c background were purchased from RIKEN BioResource Center (Ibaraki, Japan). Thy1.1^+/+ PD-1KOHA-TCR Tg mice were obtained by crossing the two transgenic strains. All mice were housed under specific pathogen-free conditions by Laboratory Animal Services (LAS) of the Georgia Regents University (GRU). All animal experiments were approved by the Institutional Animal Care and Use Committee of GRU.

The detailed information of antibodies used in the study is provided in the Supplementary Data.

The detailed descriptions of procedures of flow cytometry analysis, cell quantification, and cell sorting are provided in the Supplementary Data.

The generation and maintenance of HA-expressing murine B-cell lymphoma cell line A20 (A20HA) and colon cancer cell line CT26 (CT26HA) were described previously (18, 19). A20HA tumor cells were subcutaneously inoculated to the right flank of mice (5 × 10⁶ per mouse). Tumor growth was monitored by caliper measurement of the tumor area every three days, and expressed as the product of two perpendicular diameters in mm². Mice were euthanized when tumor size reached 400 mm² or when tumor sites ulcerated. For adoptive T-cell transfer, spleens and lymph nodes from HA-TCR Tg mice were harvested to enrich for CD4⁺ T cells by MACS (Miltenyi Biotec). A total of 2.5 to 3 × 10⁶ CD4⁺ TCR⁺ T cells were injected intravenously into each recipient. Cyclophosphamide was dissolved in PBS and intraperitoneally injected to mice at the dose of 150 mg/kg unless otherwise specified. Gemcitabine and 5-fluorouracil was dissolved in PBS and intraperitoneally injected to mice at 75 or 40 mg/kg, respectively, following the specified schedule. All chemotherapy solutions were filtered through a 0.22-μm membrane each time before injection. To deplete CD11b⁺Ly6C^hiCCR2^hi monocytes or CD11b⁺Ly6G^hi granulocytes, αCCR2 mAb (MC21, 20 μg per injection) or αLy6G mAb (1A8, 100 μg per injection), respectively, was intraperitoneally injected to mice following the specified schedule. CCX872 was given to mice by daily subcutaneous injection for 14 days, starting four days after cyclophosphamide treatment. In vivo PD-1 and PD-L1 antibody blockade was conducted as described previously (19).

For non-antigen-specific suppression, spleen cells from HA-TCR Tg or normal Balb/c mice were labeled with 0.5 μmol/L carboxyfluorescein diacetate succinimidyl ester (CFSE) and seeded into a round-bottom 96-well plate (1 × 10⁵ cells/well in 200 μL medium), with or without the addition of 1 μg/mL anti-CD3 Ab (145-2C11) and 5 μg/mL anti-CD28 Ab (37.51). Varied numbers of sorted monocytic or granulocytic myeloid cells were added to the culture. When using CFSE dilution as the readout for suppression, cells were harvested on day 3 or day 4 after culture, and stained with CD4 for FACS analysis. When using ³H-thymidine incorporation as the readout, cells were cultured for three days, then pulsed with ³H-thymidine (1 μCi/well) for additional 8 hours before harvest. ³H-thymidine uptake was counted using a liquid scintillation counter and expressed as CPM. In some experiments, anti-PD-1 (10 μg/mL, RMP1-14, Bio X Cell) and anti-PD-L1 (10 μg/mL, 10F.9G2, Bio X Cell) mAbs were added. For antigen-specific suppression, CD4⁺ T cells (5 × 10⁴/well) purified from HA-TCR Tg mice were mixed with cognate peptide-pulsed CD11c⁺ dendritic cells (5 × 10³/well) purified from a Balb/c mouse by MACS beads (Miltenyi Biotec).

Data were analyzed using Prism 4.0 (GraphPad Software, Inc.). The statistical significance of the results was determined using the Student t test. Data for mouse survival were analyzed using a log-rank test. P values less than 0.05 were considered statistically significant.

Using a mouse model of B-cell lymphoma, we previously reported that AT of tumor-specific CD4⁺ T cells after cyclophosphamide treatment gave rise to polyfunctional CD4⁺ effector cells, which played a critical role in mounting a robust antitumor immune response (18). However, antitumor CD4⁺ effector cells were susceptible to functional tolerization in the face of persistent residual tumors. Consequently, the antitumor immunity elicited by chemoimmunotherapy was not durable, and most mice succumbed to tumor relapse (19). Because immunosuppression mediated by MDSCs is one of the major tumor-induced immunoregulatory mechanisms by which tumors escape immune attacks, we set out to examine the presence and phenotype of myeloid cells in our model system. As shown in Fig. 1A schema, mice with established subcutaneous A20 tumors expressing HA (A20HA) were either untreated (No Tx) or treated with cyclophosphamide followed by AT of HA-specific CD4⁺ T cells (CTX+CD4 AT). There was only marginal presence of myeloid cells in the spleens and tumors of untreated mice, as demonstrated by both FACS analysis (Fig. 1A, No Tx) and immunofluorescence (IF) staining (Supplementary Fig. S1) of CD11b⁺ cells. In contrast, the presence of CD11b⁺ myeloid cells was markedly increased in mice treated with the combination of cyclophosphamide and CD4 AT (Fig. 1A and Supplementary Fig. S1; CTX+CD4 AT). These CD11b⁺ cells mainly consisted of two subpopulations expressing varied levels of myeloid lineage markers Ly6C, Ly6G, and Gr1 (Fig. 1B, left). Giemsa stain showed that the CD11b⁺Ly6C^hiLy6G⁻Gr1^int subset had a mononuclear feature typical of monocytes, whereas the CD11b⁺Ly6C^loLy6G^hiGr1^hi subset exhibited a polymorphnuclear feature characteristic of granulocytes/neutrophils (Fig. 1B, right). Additional phenotypic analysis indicated that the two subsets of myeloid cells were distinct from conventional dendritic cells (cDC) and macrophages (MΦ), and that the monocytic myeloid subset expressed higher levels of CD14 and IL4Rα compared with the granulocytic myeloid subset (Fig. 1C).

The robust expansion of myeloid cells suggested that these cells were proliferative. Indeed, both subsets of myeloid cells expanded after CTX+CD4 AT were Ki67 positive (Fig. 1D). Interestingly, the monocytes in both spleen and tumor exhibited higher Ki67 levels than the granulocytes, suggesting that the former population had a proliferative advantage. Moreover, as we reported previously (19), the CTX+CD4 AT regimen resulted in heightened inflammation, as evidenced by markedly elevated levels of IL1β, CSF1, CXCL10, CXCL12, and CCL2 (Supplementary Fig. S2). Corresponding to this inflammatory immune milieu, the monocytes in both spleen and tumor expressed high levels of CCR2 (Fig. 1E), a key chemokine receptor involved in mediating the trafficking of inflammatory monocytes (20).

To determine which component of the treatment regimen (cyclophosphamide or CD4 AT) led to the induction and expansion of myeloid cells, we quantified the two myeloid cell subsets in mice receiving different treatment. We first compared the frequency of each myeloid subset in treated or untreated tumor-bearing mice. The frequencies of monocytes in untreated mice were low in the spleen and tumor, and were not increased in mice receiving tumor-specific CD4⁺ T cells (Fig. 2A and B, No Tx vs. CD4 AT). In contrast, the frequencies of monocytes in cyclophosphamide-treated mice were significantly increased both in the spleen and tumor, and were further boosted in the presence of tumor-specific CD4⁺ T cells (Fig. 2A and B, cyclophosphamide vs. CTX+CD4 AT). For granulocytes, their frequencies in the spleen followed similar pattern as monocytes (Fig. 2C, top). Interestingly, granulocytes were rare in tumor, and their presence in tumor became evident only after the combined treatment of CTX+CD4 AT (Fig. 2A and C, bottom). In a colon cancer model (CT26HA), we confirmed that cyclophosphamide can induce the expansion of myeloid cells, and addition of tumor-specific CD4⁺ T cells can intensify myeloid cell expansion (Supplementary Fig. S3).

Next we conducted time course experiments to determine the kinetics of myeloid cell expansion after treatment. The absolute numbers of the two subsets of myeloid cells were used to reflect the actual cell accumulation in specified tissues. In the spleen, there was an initial reduction in the numbers of monocytes after cyclophosphamide treatment (day 2) followed by a rebound thereafter (Fig. 2D, top, cyclophosphamide). In the tumor, the rebound of monocytes peaked on day 7, and by day 10 reached a stable level that was elevated than the starting point (Fig. 2D, bottom, cyclophosphamide). Notably, CD4 AT following cyclophosphamide intensified the magnitude of monocyte expansion both in the spleen and tumor (Fig. 2D, CTX+CD4 AT). Similarly, CD4 AT following cyclophosphamide also enhanced the expansion of granulocytes in the spleen and tumor (Fig. 2E). Our previous study showed that in mice with established tumors, the transferred tumor-specific CD4⁺ T cells were destined to become dysfunctional and suppressive, but cyclophosphamide allowed these cells to differentiate into activated effector cells (18). Thus, antigen-driven CD4⁺ T-cell effector differentiation might be a prerequisite for enhanced myeloid cell expansion. Consistent with this idea, AT of irrelevant DO11.10 CD4⁺ T cells (OVA-specific) after cyclophosphamide did not boost myeloid cell expansion compared with cyclophosphamide only (data not shown). Altogether, the data indicate that cyclophosphamide is necessary and sufficient to drive systemic myeloid cell expansion in tumor-bearing hosts, and that activated tumor-specific CD4⁺ effector cells can amplify this expansion.

To determine whether therapy-induced myeloid cells possessed immunosuppressive activities, we conducted in vitro suppression assays to test the ability of each myeloid cell subset to suppress CD4⁺ T-cell activation. Using a CFSE dilution assay as the readout for cell proliferation, Fig. 3A shows that CD4⁺ T-cell response to nonspecific stimuli (αCD3 and αCD28 mAbs) was inhibited, in a dose-dependent manner, by monocytes isolated from the spleens of mice receiving the CTX+CD4 AT regimen, whereas granulocytes from the same mice were not suppressive. Similarly, monocytes recovered from the tumors of treated mice were immunosuppressive, whereas tumor-infiltrating granulocytes were nonsuppressive (Fig. 3B). The immunosuppressive activities of monocytes were also confirmed by ³H-thymidine incorporation assays (Supplementary Fig. S4A). Furthermore, therapy-induced monocytes can also suppress CD4⁺ T-cell activation in response to antigen-specific stimulation (Supplementary Fig. S4B).

We asked whether the emergence of immunosuppressive monocytes after therapy was peculiar to the tumor microenvironment, or the presence of tumor-specific CD4⁺ T cells. To address this, naïve mice (tumor-free) were treated with a single dose of cyclophosphamide, and spleen cells were harvested at different time points to enumerate the numbers of monocytic and granulocytic myeloid cells. Figure 3C shows that the numbers of monocytic and granulocytic myeloid cells were both reduced shortly after cyclophosphamide, reaching the nadir by day 2 and day 4, respectively; and both populations rebounded thereafter, resulting in net increases in cell number by day 10. Induction of myeloid cell expansion was seen in mice receiving cyclophosphamide treatment in the range of 100 to 300 mg/kg (Supplementary Fig. S5). In vitro suppression assays showed that the monocytes, but not granulocytes, from cyclophosphamide-treated tumor-free mice can suppress CD4⁺ T-cell activation (Fig. 3D right). As controls, neither subset of myeloid cells from the spleens of untreated naïve mice was suppressive (Fig. 3D left), suggesting that immunosuppression is an acquired property of monocytic myeloid cells induced by cyclophosphamide.

We then extended the above assays to some other chemotherapeutic agents to determine whether induction of immunosuppressive monocytes was a unique feature of cyclophosphamide. We found that the frequencies of monocytic myeloid cells also increased in mice treated with melphalan (Mel) or doxorubicin (Dox; Supplementary Fig. S6A), and these cells were equally capable of suppressing CD4⁺ T-cell activation as cyclophosphamide-induced monocytes (Supplementary Fig. S6B). Altogether, the data indicate that certain cytotoxic anticancer drugs can induce the expansion of immunosuppressive monocytes, which phenotypically resemble the well-described monocytic myeloid-derived suppressor cells (mMDSC; ref. 17).

We then sought to dissect the mechanism(s) by which therapy-induced inflammatory monocytes mediated immune suppression. A number of molecules, including Arginase 1, iNOS, IDO, IFNγ, IL10, and TGFβ, have been implicated in suppression mediated by inflammatory monocytes or MDSCs (17, 21). We repeated in vitro suppression assays with the addition of inhibitors or neutralizing mAbs targeting these molecules. However, these inhibitors used either individually or in combination had no significant effects on monocyte-mediated suppression (data not shown). Because the co-inhibitory molecule PD-L1 was prominently expressed in inflammatory monocytes (Fig. 4A), we hypothesized that the PD-1 pathway might be involved in monocyte-mediated suppression. To test this, we used two complementary approaches to disrupt PD-1/PD-L1 interactions. First, therapy-induced monocytes were cocultured with PD-1–sufficient (WT) CD4⁺ T-cell responders in the presence of a PD-1/PD-L1 blocking mAb cocktail. PD-1/PD-L1 blockade largely restored proliferation of the responder cells (Fig. 4B, left). The second approach used PD-1–deficient (PD-1KO) CD4⁺ T cells as responders. Strikingly, therapy-induced monocytes, which efficiently suppressed PD-1–sufficient CD4⁺ T-cell responders, were unable to suppress the proliferation of PD-1–deficient CD4⁺ T cells (Fig. 4B, right). Together, the data support the notion that the PD-1/PD-L1 pathway is involved in monocyte-mediated suppression on CD4⁺ T cells.

Our in vitro suppression data (Fig. 4B) suggested that disruption of the PD-1/PD-L1 pathway may prevent CD4⁺ effector cell tolerization in vivo. To test this, tumor-bearing mice were treated with cyclophosphamide followed by AT of either PD-1–sufficient (WT) or PD-1–deficient (PD-1KO) tumor-specific CD4⁺ T cells. Figure 5A shows that there were similar levels of monocytic and granulocytic myeloid cells in mice that received PD-1KO CD4⁺ T cells compared with those that received WT CD4⁺ T cells after cyclophosphamide. Nonetheless, the transferred PD-1KO CD4⁺ T cells persisted in mice and maintained a polyfunctional effector phenotype, that is Foxp3^loCD40L^hiIL2⁺IFNγ⁺TNFα⁺; by contrast, WT donor CD4⁺ T cells eventually acquired a tolerized phenotype, characterized by elevated levels of PD-1 and Foxp3, but reduced CD40L and lack of production of pro-inflammatory cytokines (Fig. 5B). Correspondingly, AT of PD-1KO CD4⁺ T cells after cyclophosphamide resulted in prolonged survival in the majority of mice, whereas relapse was prevalent in mice receiving WT CD4⁺ T cells after cyclophosphamide (Fig. 5C, CTX+PD1KOCD4 AT vs. CTX+WTCD4 AT). As an alternative approach, adding αPD-1/PD-L1 antibody blockade after CTX+WTCD4 AT recapitulated the beneficial effect in mouse survival achieved by CTX+PD-1KOCD4 AT. These data support the notion that disrupting the PD-1–PD-L1 axis can prevent CD4⁺ effector cell tolerization in vivo.

A20 tumor cells constitutively express PD-L1 (22). Thus, either tumor cells or monocytic myeloid cells can provide PD-L1 to engage PD-1 on CD4⁺ effector cells and consequently render them tolerant. We reasoned that if monocytes were the relevant PD-L1–expressing cells, then reducing the presence of these cells should confer significant therapeutic benefits. Our finding that therapy-induced monocytes were highly proliferative (Fig. 1D) suggested that these cells may be more sensitive to low-dose chemotherapy. We chose to test this using gemicitabine (Gem) because it has been shown that gemicitabine can preferentially eliminate CD11b⁺Gr1⁺ MDSCs but largely spare T lymphocytes (23). As expected, gemicitabine given after CD4⁺ T-cell transfer reduced the frequency and number of therapy-induced monocytes in peripheral blood, spleen, and tumor (Fig. 6A and B). Interestingly, reduction of granulocytes was also evident in blood and spleen, but was insignificant in tumor. Notably, reduction of therapy-induced monocytes correlated with remarkable therapeutic benefits. As shown in Fig. 6C and D, gemicitabine treatment following CTX+CD4 AT (CTX+CD4 AT+Gem) led to complete tumor remission and long-term survival in the majority of mice. In contrast, the combination of cyclophosphamide and gemicitabine did not differ from cyclophosphamide alone in tumor growth and mouse survival (CTX+Gem vs. cyclophosphamide). Moreover, the combination of gemicitabine and CD4 AT had no therapeutic effects (data not shown). Altogether, our results indicate that the effectiveness of gemicitabine in potentiating CTX+CD4 AT therapy is not simply because of the added cytotoxicity on tumor cells by two anticancer drugs, but rather a synergistic effect of the tripartite regimen. To test the idea that the efficacy of the tripartite regimen is not restricted to lymphoma, mice with lung metastasis of CT26HA tumors were treated with different combinations of cyclophosphamide, CD4 AT, and gemicitabine (Supplementary Fig. S7). Again, only the tripartite regimen resulted in substantial improvement in long-term survival in this aggressive tumor model. In addition to gemicitabine, 5-fluorouracil (5-FU) has been shown to deplete MDSCs (24, 25). We demonstrated in the A20HA tumor model that 5-FU can effectively reduce the presence of therapy-induced monocytes (Supplementary Fig. S8), further supporting the notion that therapy-induced immunosuppressive monocytes are sensitive to certain anticancer drugs used at low dose.

We sought to more specifically deplete therapy-induced monocytes so as to determine their role in CD4⁺ effector cell tolerization and tumor relapse. Based on our finding that therapy-induced monocytes preferentially expressed high levels of CCR2 (Fig. 1E), we administered a CCR2-specific mAb (MC21) to mice after CTX+CD4 AT therapy. Anti-CCR2 mAb injection led to selective depletion of CD11b⁺Ly6C^hi monocytes in peripheral blood, spleen, and tumor, whereas granulocytes (CD11b⁺Ly6C^low) were largely unaffected (Fig. 7A, second row). For comparison, a Ly6G-specific mAb (clone 1A8) was used to selectively deplete granulocytes but spare monocytes (Fig. 7A, bottom row). The functional impact of selective depletion of monocytes or granulocytes on the efficacy of chemoimmunotherapy was assessed by tumor growth and mouse survival. The CTX+CD4 AT regimen led to initial tumor regression followed by relapse in most mice (Fig. 7B, third graph, 7/8 mice had relapse). Depletion of inflammatory monocytes, by administering αCCR2 mAb after CTX+CD4 AT therapy, resulted in uniform and complete tumor regression in all mice, and substantially reduced the occurrence of relapse (Fig. 7B, fourth graph, 3/11 mice had relapse). By contrast, depletion of granulocytes seemed to accelerate tumor regrowth (Fig. 7B, last graph). It is important to note that the combination of cyclophosphamide and αCCR2 mAb was not different from cyclophosphamide alone (Fig. 7B, first and second graphs), suggesting that αCCR2 mAb by itself was not toxic to tumor cells that were CCR2 negative (data not shown), and that the beneficial effects afforded by CCR2-specific mAb require the presence of tumor-specific CD4⁺ effector cells. In line with the tumor growth data, the survival results clearly demonstrated that selective depletion of therapy-induced monocytes promoted, whereas depletion of granulocytes impaired, long-term survival of tumor-bearing mice (Fig. 7C). Moreover, the transferred CD4⁺ T cells were readily detectable in cured mice, and maintained an effector phenotype (data not shown). As an alternative approach, we replaced CCR2-specific mAb with CCX872, a potent and selective antagonist of CCR2. As shown in Fig. 7D, administration of CCX872 following CTX+CD4 therapy significantly improved long-term survival compared with the control mice that received CTX+CCX872. Altogether, our data indicate that disrupting the CCL2–CCR2 axis can effectively relieve tumor-specific CD4⁺ effector cells from inflammatory monocyte-mediated tolerization and sustain a productive antitumor immunity.

Because many mice had complete tumor remission and survived long term after CTX+CD4 AT therapy in combination with either CCR2-specific mAb (MC21) or gemicitabine, we asked whether these long-term survivors (LTS) had developed immune memory. To this end, mice that became LTS after combinatorial therapy were rechallenged with A20HA. Figure 7E shows that these mice were completely protected from tumor rechallenge, whereas all naïve control mice succumbed to rapid tumor growth.

Induction and expansion of suppressive inflammatory monocytes or MDSCs in the tumor setting has been amply documented. Using tumor cell lines engineered to express proinflammatory cytokines, previous studies have reported that tumor-induced inflammation recruited and expanded inflammatory monocytes/MDSCs capable of suppressing antitumor immune responses (26–28). In addition to mediating immune suppression, tumor-induced inflammatory monocytes were also found to facilitate tumor metastasis in breast cancer (29). Different from these published studies that focus on cancer-induced inflammation, this study addresses the role of therapy-induced inflammation in regulating ongoing immune responses. Although the tumor-promoting effect of chronic inflammation has been well established (30, 31), the impact of therapy-induced inflammation (often acute) on eventual treatment outcomes has been a subject of debate (32). In our study, it was chemotherapy with cyclophosphamide that drove the expansion of immunosuppressive monocytes, regardless of the presence or absence of tumor. The fact that cyclophosphamide-induced inflammatory monocytes acquired suppressive activities in the absence of tumor growth (Fig. 3D) suggests that this feature may serve as a negative feedback loop to control excessive inflammation. Surprisingly, this counter-regulation mechanism was reinforced by antitumor CD4⁺ effector cells in a therapeutic setting (Fig. 2). It has been shown that cyclophosphamide induces an inflammatory immune milieu, in which myeloid growth factors and chemotactic factors such as GM-CSF, G-CSF, and CCL2 are abundant (8, 33, 34). We previously showed that this inflammatory milieu was markedly intensified in the presence of CD4⁺ effector cells (19). It is intriguing that cyclophosphamide-induced monocytes shared some similarities with alternatively activated macrophages (M2), such as IL4Rα expression and suppressive activities, raising the question whether M2 cells were included in therapy-induced monocytes. However, IL4 and IL13, which are essential for inducing M2 cells (35), were not detected in the immune milieu after CTX+CD4 AT therapy (data not shown), arguing against the emergence of M2 cells in our mouse model. We postulate that CD4⁺ effector cell–mediated enhancement of myeloid cell expansion is because of increased myeloid cell homeostasis under heightened inflammation. However, singly neutralizing GM-CSF, IFNγ, TNFα, or IL6 after CTX+CD4 AT did not have significant impact on the numbers of myeloid cells (data not shown), suggesting that multiple, overlapping inflammatory cytokines may contribute to myeloid cell expansion in this setting.

In addition to cyclophosphamide, we showed that at least two other anticancer drugs, doxorubicin and melphalan, can induce the expansion of monocytic myeloid suppressor cells (Supplementary Fig. S6). This is consistent with the report by Nakasone and colleagues that doxorubicin treatment led to a specific, acute recruitment of CCR2⁺ monocytic myeloid cells that contributed to tumor regrowth (36). Intriguingly, Alizadeh and colleagues recently reported that low-dose doxorubicin transiently reduced the number and diminished the suppressive function of MDSCs in the 4T1 mammary cancer model (37). It is important to note that in this study doxorubicin was given to tumor-bearing mice to examine its acute effect on existing tumor-induced MDSCs, whereas in our study the drug was injected to tumor-free naïve mice to investigate therapy-induced myeloid cells, It is possible that doxorubicin can exert cytotoxic effect on tumor-induced MDSCs, meanwhile its myeloid depletion effect may result in homeostatic myeloid reconstitution, which gives rise to therapy-induced MDSCs. In future studies it will be of interest to assess the relation between preexisting tumor-induced MDSCs and therapy-induced MDSCs. Chemotherapy-driven MDSC expansion has also been observed in patients with cancer. In a study conducted in patients with breast cancer, circulating MDSC numbers were significantly increased in patients receiving doxorubicin–cyclophosphamide chemotherapy, and correlated with clinical cancer stage and metastatic tumor burden (38). It is also important to note that induction and expansion of inflammatory myeloid cells with immunoregulatory function is not restricted to chemotherapy. Other cancer treatment modalities, including radiation therapy and surgery, may have similar effects. It was reported in a mouse melanoma model that total body irradiation (TBI) induced rapid reconstitution of MDSCs with enhanced suppressive activity (39). Moreover, a correlative study in patients with pancreatic cancer found that the prevalence of CCR2⁺CD14⁺ inflammatory monocytes in the peripheral blood following tumor resection correlated inversely with survival (40). There is accumulating evidence that cancer immunotherapy can also induce immunosuppressive myeloid cells. Mitchell and colleagues reported that therapeutic vaccination induced inflammatory monocytes that counter-regulated vaccine-induced immunity (41). Furthermore, Hosoi and colleagues found that AT of pmel-specific CD8⁺ T cells to mice with B16 melanoma led to accumulation of monocytic MDSCs, which acted to temper CTL antitumor activity (42). Altogether, these studies indicate that a broad spectrum of cancer therapies may give rise to inflammatory myeloid suppressor cells that counteract therapy efficacy.

Although this study focused on monocyte-mediated suppression on antitumor CD4⁺ T cells, it is worth noting that CD8⁺ T-cell activation was also inhibited by cyclophosphamide-induced inflammatory monocytes (data not shown). Therefore, it is reasonable to speculate that targeting therapy-induced myeloid suppressor cells may augment the long-term efficacy of some current cancer immunotherapy strategies, such as cancer vaccines and adoptive T-cell therapy, in which cyclophosphamide is often a component of the treatment regimen. From a therapeutic standpoint, our study outlines multiple targeted approaches that can effectively reduce the number or abolish the suppressive function of inflammatory monocytes, thereby tipping the balance toward unrestrained antitumor immunity. Here we showed that the proliferative nature of inflammatory monocytes rendered them susceptible to low-dose gemcitabine and 5-FU. Alternatively, disrupting the relevant CCL2/CCR2 chemotactic pathway represents an attractive approach to reduce the recruitment and accumulation of inflammatory monocytes (Fig. 7). Along this line, the use of CCL2-specific neutralizing mAb or small molecule inhibitors for CCR2 has also shown beneficial effects in several preclinical models (29, 40, 41, 43). A novel finding of our study is that therapy-induced inflammatory monocytes suppress CD4⁺ T-cell responses through the PD-1–PD-L1 axis. This provides a mechanistic explanation for our previous observation that the PD-1 pathway is critically involved in CD4⁺ effector cell tolerization (19). Although A20 tumor cells constitutively express PD-L1, it has been shown that direct encounter of A20 tumor cells in vivo was stimulatory, rather than tolerogenic, to the activation of tumor-specific CD4⁺ T cells (44). Here we showed that targeted depletion of inflammatory monocytes was quite effective in preventing PD-1–dependent CD4⁺ T-cell tolerization (Fig. 7), supporting the notion that the relevant ligands for PD-1 primarily come from inflammatory monocytes as oppose to tumor cells.

In summary, our study reveals that inflammation engendered after chemoimmunotherapy may facilitate tumor escape by engaging myeloid suppressor cells that attenuate antitumor immunity, and that mitigating this counter-regulation mechanism can augment the efficacy of chemoimmunotherapy. We have identified multiple clinically applicable approaches that can effectively target chemotherapy-induced myeloid suppressor cells, leading to durable antitumor immunity. These findings imply that modulation of therapy-induced inflammation represents an attractive strategy to augment the efficacy of chemoimmunotherapy and achieve a long-lasting curative effect.

M. Walters has ownership interest (including patents) from Chemocentryx. A.L. Mellor has other commercial research support from NewLink Genetics Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Mack, B.R. Blazar, G. Zhou

Development of methodology: Z.-C. Ding, P. Chandler, A. Krasinski, M. Mack, G. Zhou

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Z.-C. Ding, X. Lu, M. Yu, H. Lemos, P. Chandler, K. Liu, G. Zhou

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.R. Blazar, G. Zhou

Writing, review, and/or revision of the manuscript: Z.-C. Ding, P. Chandler, M. Mack, B.R. Blazar, A.L. Mellor, D.H. Munn, G. Zhou

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Huang, M. Walters, M. Mack, A.L. Mellor, G. Zhou

Study supervision: G. Zhou

The authors thank W. King for assisting cell sorting, J. Oakley and Y. Kang for technical assistance, and Dr. E. Celis for critical reading of the article.

This work is supported by NIH grant R01CA158202 and the American Cancer Society Research Scholar Grant RSG-12-169-01-LIB to G. Zhou; R01HL56067, R01CA72669, and P01AI056299 to B.R. Blazar; AI083005 and AI103347 to A.L. Mellor; R01CA112431 to D.H. Munn.

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.