Myeloid-Derived Suppressor Cells: A Propitious Road to Clinic

In recent years, myeloid-derived suppressor cells (MDSC) emerged as important regulators of immune responses in many pathologic conditions, including cancer. There is now ample evidence demonstrating the crucial role of these cells in the formation of premetastatic niche and promotion of tumor metastasis (1). MDSC research has gone a long way from phenomenological observations 20 years ago to a more defined biology at the current time. MDSCs represent a relatively stable form of pathologic activation of neutrophils (PMN) and monocytes (MON). As a reflection of their lineage association, two major groups of MDSCs are currently recognized: PMN-MDSCs and M-MDSCs. Although these two groups of cells have distinct phenotypes and gene expression profiles, they share many common features in gene expression, as well as biochemical and functional activities. The pathologic activation of MDSCs is different from the classic activation of PMNs and MONs associated with the response of these cells to microbial and viral pathogens. These differences have been described in previous reviews (1, 2). Although the concept of MDSCs is now widely accepted, many aspects of their biology remain not well defined and require further investigation. This includes better characterization of surface molecules specific to these cells, a better understanding of molecular mechanisms regulating the function of these cells, and better delineation of MDSCs from classic PMNs and MONs (3). However, because MDSCs do not represent actual subsets of myeloid cells but rather a state of activation (albeit lasting for the duration of their life span), it is likely that their functional activity is not controlled by one or even several defined regulatory mechanisms as observed for different subsets of T cells. The complex nature of MDSC biology presents a substantial challenge for therapeutic targeting. The wealth of information that has accumulated around the biology of these cells nevertheless provides a strong foundation for this effort, which will be discussed in this review.

Generation of MDSCs may be considered a two-phased, partially overlapping process: (i) expansion of myeloid cells and (ii) acquisition of MDSC features by these cells (refs. 1, 4; Fig. 1). MDSCs develop mainly in the bone marrow (BM), where tumor-derived factors influence myeloid cell differentiation. MDSC generation may also take place in other organs, such as the spleen or the liver of tumor-bearing (TB) mice or patients with cancer (5, 6). Enhanced myelopoiesis in cancer is somewhat similar to the emergency myelopoiesis observed in acute infection or trauma. However, it persists much longer and is less potent (if assessed by the number of cells produced), which causes substantial changes in the biology of myeloid progenitors. In the early stages of cancer, PMN-MDSCs are present in small numbers in blood and peripheral lymphoid organs of both mice and patients with cancer. They can be mostly identified by changes in the gene expression profile rather than functional activity. These cells have potent migratory activity, which apparently helps in the promotion of tumor growth and metastasis after settling down in tissues (7, 8). These cells may represent an early stage of MDSC differentiation and are sometimes called MDSC-like cells. Interestingly, in tumors, functionally potent MDSCs accumulate at the same time as tumors form. Recent data suggest that PMN-MDSCs migrate from BM to tumors and peripheral lymphoid organs independently and are influenced by the microenvironment (9, 10). In the tumor site, M-MDSCs differentiate into tumor-associated macrophages (TAM; refs. 1, 11). A recent study demonstrated that the immunosuppressive activity of macrophages might depend on the nature of their precursors. Although MONs differentiated largely into nonsuppressive macrophages, M-MDSCs developed into immunosuppressive macrophages. S100A8/A9 proteins were shown to contribute to the latter process (12), having been retained in immunosuppressive macrophages derived from M-MDSCs but lost in nonsuppressive macrophages derived from MONs.

Myeloid cells, including MDSCs, differentiate from granulocyte–macrophage progenitors (GMP), lineage-restricted progenitors. GMPs are generated from the hematopoietic stem cells through a sequential differentiation process involving several progenitors, the most notable of which are common myeloid progenitors (CMP). In cancer, BM is chronically exposed to a variety of growth factors, cytokines, and other mediators, deregulating the expression and function of transcription factors involved in myeloid differentiation, skewing the differentiation program, and generating immunosuppressive MDSCs instead of classic MONs or PMNs (Fig. 1). This process is controlled by a plethora of different and redundant tumor-derived factors, which complicate the effort for therapeutic regulation of MDSC generation. It is unlikely that therapeutic targeting of individual factors would be successful.

The activity of multiple transcription factors has been implicated in MDSC generation in mice. Downregulation of IRF8 in hematopoietic progenitors plays a key role in the expansion of PMN-MDSCs. During normal myelopoiesis, IRF8 is upregulated during the transition from CMPs to GMPs and is then downregulated during the differentiation of PMNs. Monocyte precursors retain high expression of IRF8. In TB mice, IRF8 is downregulated in GMP and in the more differentiated granulocyte precursors (GP). Importantly, IRF8^lo GPs were shown to give rise to a large number of PMN-MDSCs, and IRF8 downregulation was necessary for PMN-MDSC differentiation in tumors (4, 13, 14). Within GMP, IRF8 interaction with GFI1 has been recently identified as a critical point in emergency myelopoiesis for neutrophil generation at the level of proneutrophils, a discrete subset of unipotent precursors (15, 16). These data were corroborated by the identification of a single cluster of unipotent neutrophil precursors expanded in TB mice. These cells had low expression of IRF8 and were identified as the main source of PMN-MDSCs in TB mice derived from GMPs (17). Recently, in mice, monocyte-like precursors of granulocytes (MLPG) were implicated in the generation of a proportion (up to 50% in some cases) of PMN-MDSCs. These cells were regulated not by IRF8 but by downregulation of the retinoblastoma protein (Rb; ref. 11). Similar precursors, albeit less characterized, were also identified in patients with cancer (11). It is important to point out that IRF8 or Rb downregulation, although key for the expansion of cells with PMN-MDSC features, is not sufficient for the conversion of PMNs to PMN-MDSCs.

Although C-EBPα is the main regulator of myelopoiesis under steady-state conditions, C-EBPβ is one of the master regulators of emergency myelopoiesis. The two transcription factors have the opposite effect on MDSC generation. Although C-EBPα restrains MDSC expansion and suppressive function, C-EBPβ promotes MDSC generation. In line with this, C-EBPβ deficiency strongly reduced the generation of MDSCs in TB mice, and myeloid cells were unable to suppress T-cell response. This suggests a major role for this transcription factor in MDSC generation (18). Retinoic acid–related orphan receptor 1 (RORC1) has also been shown to profoundly affect cancer-induced emergency myelopoiesis and regulate the generation of both MDSC subsets. RORC1 suppressed the negative regulators of myelopoiesis SOCS3 and BCL3 and promoted C-EBPβ. Furthermore, RORC1 also controlled the generation and suppressive function of M-MDSCs and their differentiation to immunosuppressive TAMs (19). STAT3 was one of the first transcription factors implicated in MDSC accumulation in mice (20). STAT3 phosphorylation, dimerization, and nuclear translocation can be induced by several cytokines that are increased in TB, including G-CSF, GM-CSF, and IL6 (reviewed in ref. 21).

Currently, it is not clear whether persistent exposure of BM to growth factors such as GM-CSF is sufficient to skew the differentiation program of neutrophils and monocytes to MDSCs. The data indicating rather weak suppressive activity of BM MDSCs compared with cells in other tissues (7) suggested that other factors may be required for the development of fully potent MDSCs. We recently demonstrated that GM-CSF contributes to the development of highly immunosuppressive CD14^hi PMN-MDSCs in tumors. However, blocking GM-CSF with a neutralizing antibody only partially reversed the expression of CD14, suggesting that other tumor-derived factors may contribute to the acquisition of CD14 and the immunosuppressive activity of PMN-MDSCs (9). GM-CSF was also responsible for the upregulation of the fatty acid transport protein 2 (FATP2) by PMN-MDSCs, which has been implicated in the accumulation of suppressive PMN-MDSCs in mice (22).

Exosomes (23), Toll-like receptor (TLR) ligands (24), proinflammatory cytokines such as TNFα (25), IL1β (26), and alarmins (HMGB1 and S100A8/A9; refs. 27, 28) produced during tumor progression all contribute to the acquisition of immunosuppressive activity by MDSCs. Most of these molecules signal via activation of the transcription factor NFκB (29). IFNγ has been shown to stimulate the suppressive activity of M-MDSCs through activation of the transcription factor STAT1, which in turn promotes transcription of NOS2, one of the main enzymes responsible for M-MDSC–suppressive activity (refs. 30–32; Fig. 1).

MDSCs are characterized by the accumulation of unfolded proteins in the endoplasmic reticulum (ER), which causes ER stress. This phenomenon is considered a major driver for the pathologic activation of the MDSC immunosuppressive phenotype (1). It can be activated in tumor tissues as a result of hypoxia, low pH, the effect of proinflammatory cytokines, or deprivation of nutrients (33, 34). It is also possible, but not yet proven, that ER stress can be a consequence of persistent stimulation of myeloid progenitors by growth factors, which stimulate their proliferation, requiring increased protein synthesis. Induction of ER stress in PMNs by the chemical inducer thapsigargin converted these cells to immunosuppressive PMN-MDSCs and upregulated the expression of lectin-type oxidized LDL receptor 1 (LOX1) in humans (35), as well as TRAIL receptor 2 (DR5) in both humans and mice (36), both of which can be used for the identification and targeting of MDSCs (as discussed below; Fig. 2). Interestingly, a recent study demonstrated different involvement of ER stress in the suppressive activity of M-MDSCs and PMN-MDSCs. In mouse tumor models, a critical role of IRE1α and ATF6 pathways of ER stress was demonstrated for PMN-MDSC suppressive activity, whereas these were largely dispensable for M-MDSC–suppressive activity (32).

PMN-MDSCs are very short-lived cells. Their life span may be affected by the specific tissue microenvironment. In line with this, parabiosis experiments showed distinct gene signatures and function for neutrophils infiltrating different organs in mice (10). This phenomenon is most likely applicable to PMN-MDSCs as well. Although limited by the use of in vitro–generated MDSCs, a study by Sceneay and colleagues (37) showed that the gene expression profile of adoptively transferred MDSCs in TB mice was different in various tissues. Specifically, MDSCs that migrate to the lungs had the highest expression of Arg2, Nos3, and S100a9; MDSCs in tumors upregulated Arg1 the most, whereas cells migrating to the spleen had an intermediate expression of all four genes (37). We recently found that PMN-MDSCs within murine tumors were enriched for PMNs that expressed CD14 and had potent suppressive activity. These CD14^hi PMN-MDSCs were rare in the spleens and virtually absent in the BM of TB mice (9), suggesting tissue-specific activation of the PMN-MDSC immunosuppressive program, which reflected the suppressive abilities of the total pool of PMN-MDSCs derived from the BM, spleen, or tumor (7). In mice, MDSCs isolated from liver metastasis had upregulation of pSTAT5, whereas those from lung metastasis had pSTAT3 upregulation (38). Furthermore, adoptive transfer of liver-derived MDSCs into lung metastasis shifted their dependency from pSTAT5 to pSTAT3 and promoted the differentiation of M-MDSCs (apparently containing MLPGs) to PMN-MDSCs (ref. 38; Fig. 2).

Under homeostatic conditions, MONs can differentiate into macrophages to support the immune response. In cancer, M-MDSCs differentiate into highly suppressive TAMs (12). M-MDSC differentiation to macrophages is tissue-dependent. After adoptive transfer, M-MDSCs slowly differentiate into macrophages in the spleen but rapidly differentiate into TAMs in tumors (39). This phenomenon was found to be regulated by hypoxia and mediated by HIF1α (39).

PMN-MDSCs may undergo different types of cell death. However, at this moment, apoptosis is considered the main mechanism (40). PMN-MDSCs require MCL1 to control apoptosis induction (41). ER stress that results in upregulation of DR5 TRAIL receptor induces apoptosis mediated by this mechanism (36). Importantly, DR5 upregulation may render PMN-MDSCs more susceptible to TRAIL-induced apoptosis in organs containing TRAIL-expressing cells, such as macrophages, activated natural killer (NK) cells, and T cells. This mechanism is thus being exploited for selective therapeutic targeting of MDSCs (42). MDSC apoptosis is also controlled by the death receptor Fas and may be induced by FasL-expressing activated T cells (43). M-MDSCs require continuous expression of c-FLIP to prevent apoptosis but are less susceptible to MCL1 ablation (41). Activation of PMN with strong stimuli, including tumor-derived factors, leads to neutrophil extracellular trap (NET) formation, which may promote tumor progression and metastasis (44). Our current understanding of the biological role of NETs is gained from studies of PMNs. It is likely that a similar mechanism exists for PMN-MDSCs, but the biological role of NETs in PMN-MDSC function needs to be elucidated.

Identification of the MDSC phenotype in humans (reviewed in ref. 1) allowed for evaluation of a link between the presence of MDSC in patients with cancer and their response to therapy. A growing body of evidence demonstrates that the presence of these cells is associated with poor clinical outcome in a variety of solid tumors, such as melanoma (45, 46), prostate cancer (47), breast cancer (48), pancreatic adenocarcinoma (49), glioma (50), colorectal cancer (51), and non–small cell lung cancer (NSCLC; ref. 52). Furthermore, low frequencies of PMN-MDSCs and M-MDSCs in peripheral blood are associated with longer progression-free survival and overall survival in patients with NSCLC treated with anti–PD-1 immunotherapy (53). In a pilot clinical trial in patients with NSCLC, peripheral MDSC ratio was used as a biomarker for the prediction of response to nivolumab therapy (NCT03486119). A subset of circulating MDSCs, Tie2^hi M-MDSCs that suppress antitumor-associated antigen T-cell response, has also been linked to the poor outcome in patients with NSCLC (54). Furthermore, a higher expression level of LOX1 can mark PMN-MDSCs in different human cancer tissues (35). In a longitudinal study in patients with NSCLC, the ratio of regulatory T cells (Tregs) to LOX1⁺ PMN-MDSCs (TMR) was chosen as a predictor of response to nivolumab. A TMR ≥0.39 after the first dose of nivolumab treatment was associated with the likelihood of being a responder to the treatment. Notably, the level of chemokines and soluble factors associated with the recruitment of MDSCs, such as CXCL2, CXCL23, and HMGB1, was significantly higher in nonresponders than in responders (55). Similarly, the NK cell to LOX1⁺ PMN-MDSC ratio was applied to predict the response to anti–PD-1 therapy in patients with NSCLC (56).

In patients with metastatic melanoma, the clinical responders to ipilimumab therapy showed significantly lower frequencies of M-MDSCs in peripheral blood compared with nonresponders (45, 57), suggesting that the patients with lower percentages of M-MDSCs could benefit from ipilimumab. These observations are consistent with the report that high frequencies of M-MDSCs are correlated with the reduced expansion and activation of tumor-specific T cells (58). Moreover, in patients with prostate cancer treated with a combination of a cancer vaccine with ipilimumab, low pretreatment frequencies of circulating M-MDSCs were a predictor of better clinical outcome (47). Thus, the circulating level of both PMN-MDSCs and M-MDSCs can be used as a predictive marker for immune checkpoint blockade (ICB)–based treatments for different tumors. The increased abundance of these cells in the tumor microenvironment (TME) is associated with poor clinical response to ICB and may be a key contributing factor to the resistance to these therapies. The challenge for the broader use of MDSCs as a predictive marker is the standardization of protocols for their evaluation. However, studies in recent years have offered a harmonized approach to MDSC identification in patients with cancer, which may help to overcome this challenge (59).

These results clearly demonstrate the clinical significance of MDSCs in cancer. However, can MDSCs be used as a true biomarker of responses? Currently, approximately 90 clinical trials are using MDSCs as biomarkers of response to cancer therapy. Although an appealing concept, there is not enough information at this moment to definitely answer this question, and the results of ongoing clinical trials are needed to assess whether MDSCs may be used as a true biomarker of response. A potential issue of using MDSCs as a biomarker is the fact that MDSCs are induced not only in cancer but also in other conditions such as infection, autoimmunity, and pregnancy (1, 2).

Despite the great success of ICB in the clinic, most patients either do not respond to these treatments or develop resistance after a period of response. Therefore, targeting the immunosuppressive MDSCs in the TME in combination with ICB or other immunotherapies has become an attractive approach to extend the benefits of these treatments. To effectively target these cells, three different approaches have been tested in both preclinical and clinical settings: (i) control of the expansion and survival of MDSCs, (ii) block of MDSC recruitment, and (iii) alteration of the immunosuppressive function. Although many issues remain to be elucidated, first lessons have become apparent. We will discuss them after a brief outline of different targeting mechanisms.

Various chemotherapeutics showed prominent effects on controlling the frequencies of MDSCs. Gemcitabine showed efficacy in a murine model of mesothelioma in overcoming the resistance to ICB (60). However, gemcitabine and 5-fluorouracil activate the NOD-like receptor family pyrin domain containing-3 protein–dependent inflammasome complex in MDSCs, leading to the production of IL1β. MDSC-derived IL1β further induces the secretion of IL17 in CD4⁺ T cells and blunts the antitumor efficacy of these chemotherapeutics (61). The oral CHK1 inhibitor SRA737 and a low dose of gemcitabine improved the outcome of PD-L1 blockade in small-cell lung cancer. These effects are mainly attributed to a significant decrease in immunosuppressive macrophages and MDSCs, with a concomitant increase in the expression of IFNβ, and chemokines CCL5 and CXCL10 (62). Phenformin, which causes activation of AMPK and blocks the mTOR regulatory complex, selectively reduced the numbers of PMN-MDSCs in the BRAF^{V600E/PTEN-null} mouse model and enhanced the efficacy of anti–PD-1 checkpoint blockade. The combination of phenformin and anti–PD-1 significantly downregulated the expression level of key immunosuppressive genes, such as Arg1, S100a8, and S100a9, with a concomitant increase in the CD8⁺ T-cell infiltration in tumors (63).

More specific depletion of MDSCs could be achieved using a more innovative approach—nanoparticles conjugated to chemotherapeutic agents. A study by De La Fuente and colleagues (64) identified four novel RNA aptamers specific to mouse and human tumor-infiltrating myeloid cells, which facilitated improved delivery of doxorubicin to tumor sites, increasing its therapeutic index and promoting tumor regression.

An intriguing link is also starting to emerge between MDSC accumulation and microbiome. MDSC frequencies have been reported to correlate with abundance of certain bacteria, such as Bacteroidales (65), and one study showed that gram-negative gut commensal bacteria translocation induced by chronic gastrointestinal disorders can drive accumulation of MDSCs in the liver and promote carcinogenesis through the TLR4/CXCL1/CXCR2 pathway (66). These findings might have therapeutic implications, as they open up an opportunity to regulate the frequencies of MDSCs through manipulation of gut microbiome composition.

Therapeutics intervening in the differentiation of MDSCs is another approach to reduce the frequencies of these cells in the TME. Earlier studies showed that the blockade of retinoic acid signaling by all-trans retinoic acid (ATRA) promotes the differentiation of MDSCs into macrophages and dendritic cells (DC) in both mice and humans (67, 68). Mechanistically, ATRA-induced MDSC differentiation involves ERK1/2 activation and glutathione generation in these cells (69). The patients with metastatic renal cell carcinoma (68) and lung cancer (70) treated with ATRA showed the reduction of MDSC frequencies and an improved response. Furthermore, ATRA in combination with ipilimumab showed strong pharmacodynamic results in a phase II trial in patients with advanced melanoma via the reduction in the frequencies of MDSCs and an improvement in CD8⁺ T-cell function (71). A similar phase I/II clinical trial applying a combination of pembrolizumab with ATRA in patients with advanced melanoma is under recruitment (NCT03200847; Table 1). Moreover, a recent study in an LKB1-deficient murine model of NSCLC demonstrated that ATRA treatment can overcome the accumulation of MDSCs and sensitize tumors to anti–PD-1, which provides a further rationale for combining ATRA with ICB in patients with NSCLC who have LKB1 mutation refractory to immunotherapy (72).

Treatment with casein kinase 2 (CK2) inhibitor reduced the frequencies of PMN-MDSCs in the BM and spleen of TB mice, owing to the induction of the differentiation block in these myeloid cells. CK2 inhibitor also synergized with the antitumor effect of anti-CTLA4 antibody in different syngeneic mouse models (73).

As described above, TRAIL receptor DR5 is upregulated on MDSCs but not on PMNs and MONs. Induction of MDSC apoptosis with agonistic DR5 antibody substantially reduced the presence of MDSCs in TB mice and improved the response to ICB (CTLA4; ref. 36). Selective elimination of MDSCs by agonistic DR5 antibody was demonstrated in patients with cancer in a phase I/II clinical trial (42). Qin and colleagues (74) described a peptidobody that targeted murine MDSCs and demonstrated a potent antitumor effect. It would be interesting to see the follow-up study from this observation. Another example of MDSC targeting is the liver X nuclear receptor (LXR). Activation of LXR in TB mice induced apoptosis in MDSCs, which was associated with decreased immune suppression and activation of the antitumor T-cell response (75). One of the LXR agonists, RGX-104, is currently being investigated in a phase I clinical trial (75).

It is worth noting that, given the homeostatic regulation of MDSCs, their depletion strategies may result in a rebound effect, leading to compensatory myelopoiesis in the BM, as has been demonstrated by several groups (36, 76). The presence of depletion-resistant MDSCs has also been reported in the liver and tumor (76). Combination of MDSC depletion with potent immunotherapeutic strategies would be critical to achieve an antitumor effect. In this case, decreased tumor burden would dampen MDSC generation and provide for sustained antitumor effect.

MDSCs have been shown to migrate to tumors due to the chemokine gradient generated by the growing tumor. Migration of PMN-MDSCs mainly relies on the expression of chemokine receptor CXCR2 and its cognate ligands CXCL1, CXCL2, and CXCL5 in mice and humans (77) and CXCL8 in humans (78, 79). A CXCR1/2 inhibitor, SX-682, blocked MDSC migration and enhanced the effect of NK-cell immunotherapy in a mouse model of head and neck cancer (80). CXCR2 blockade also improved the efficacy of anti–PD-1 immunotherapy in models of rhabdomyosarcoma (81) and pancreatic ductal adenocarcinoma (82). Moreover, pharmacologic inhibition of CXCR2 significantly boosted the outcome of a combination of anti–PD-1 and CSFR1-targeted therapy in a syngeneic mouse model (77). Another study highlighted the increased recruitment of PMN-MDSCs as a mechanism of resistance to immunotherapy (83). SX-682 is being tested in patients with melanoma (NCT03161431) and in patients with RAS-mutated, highly metastatic, microsatellite stable (MSS) colorectal cancer in combination with nivolumab (NCT04599140; Table 1).

Epigenetic therapies are usually aimed at killing tumor cells, but earlier studies indicate that such treatments could also target tumor-promoting myeloid cells, including MDSCs. A recent study showed that low-dose adjuvant epigenetic therapy could disrupt premetastatic niche formation by inhibiting the recruitment of MDSCs and their differentiation into macrophages (84). In a lung premetastatic mouse model, low doses of 5-azacytidine and entinostat reduced MON and MDSC recruitment through downregulation of CCR2 and CXCR2, respectively, and prolonged the overall survival of mice after surgical resection of the primary tumor (84). Furthermore, the PARP inhibitor olaparib reduced the expression of SDF1α released from cancer-associated fibroblasts, thereby dampening CXCR4-mediated migration of MDSCs. A combination of olaparib with EGFRvIII-targeted chimeric antigen receptor (806–28Z CAR) T cells also significantly enhanced the efficacy of CAR T cells in a model of breast cancer (85).

Due to their nature as pathologically activated PMNs and MONs, MDSCs employ a wide range of mechanisms to mediate immunosuppression (2). Many of these mechanisms are now well characterized and are targetable pharmacologically. During the past 15 years, the field has accumulated a wealth of information on preclinical activity of MDSC-targeting agents (Fig. 3), and the first evidence of clinical effect is starting to emerge.

One of the first effector mechanisms described for MDSC is upregulation of arginase 1 (ARG1)—an enzyme that can degrade the nonessential amino acid arginine into ornithine and urea. ARG1 activity can induce localized immune tolerance and exert profound immunosuppressive effects on T-cell responses in vitro, driving T-cell anergy and apoptosis through interference with TCRζ chain expression and cell-cycle regulation (86). Specific and potent inhibitors of ARG1 were developed. However, they demonstrated a modest effect in combination with chemotherapy or PD-1 blockade.

Another strategy exploited by MDSCs is upregulation of the tryptophan metabolizing enzyme indoleamine 2,3-dioxygenase (IDO). Effector T cells can be particularly sensitive to tryptophan depletion by IDO, leading to their proliferative arrest as a result of a GCN2-dependent block in the G₁- to S-phase transition. Foxp3⁺ Treg populations, on the other hand, can thrive under low-tryptophan conditions. IDO-positive tumors have been associated with MDSC expansion and more aggressive phenotypes in both humans and mice (87). Pharmacologic inhibition of IDO had yielded promising results preclinically (86). Such preclinical findings generated excitement about testing IDO inhibitors such as epacadostat in clinical trials. These agents had an encouraging start in the clinic in combination with checkpoint blockade. However, the failure to demonstrate an improvement in progression-free survival or overall survival for such combinations in a later phase III trial (88) led to discouragement. Nevertheless, alternative approaches of targeting this pathway are still being considered, including blockade of downstream effector pathways such as the aryl hydrocarbon receptor, as well as dual IDO/TDO inhibition, which may address potential redundancy issues.

In addition to degrading arginine and tryptophan, MDSCs can also suppress T-cell function by limiting the availability of cysteine in TME. This amino acid is important for T-cell activation and function, as well as for protection from free radicals typically produced by MDSCs. Cysteine is considered an essential amino acid for T cells due to the lack of relevant enzymes (cystathionine γ-lyase) and transporters (SLC7A11) that could supply cysteine in these cells. Thus, T cells rely on other cells to import cystine through the SLC7A11 transporter, convert it to cysteine, and excrete it extracellularly through alanine–serine–cysteine (ASC) transporter activity. MDSCs have devised a strategy to exploit this vulnerability of T cells by upregulating the SLC7A11 but not the ASC transporter, which allows them to sequester cystine without returning cysteine to the microenvironment. MDSC-mediated T-cell immunosuppression can be partially reversed by supplementation with exogenous N-acetylcysteine (NAC) in vitro (89), and NAC-supplied diet is able to delay tumor growth in vivo (90). Despite these tantalizing findings, it is not yet clear whether the effects observed with NAC in vivo can be attributed to inhibition of MDSC-mediated T-cell suppression.

Nitric oxide (NO) is an important mechanism by which MDSCs (particularly the M-MDSCs) can mediate their immunosuppressive effects. Inducible nitric oxide synthase (iNOS) and pan-NOS can be pharmacologically inhibited using L-NMMA and L-NAME, respectively. These agents have demonstrated some inhibitory effects on tumor growth in animal models. These modest antitumor effects could be further improved when combined with chemotherapy (91, 92). Other agents targeting reactive nitrogen species (RNS) have also been developed and include NO scavengers, which have shown some benefit in preclinical models of adoptive T-cell therapy (93). However, attempts to test the effect of RNS-targeting compound in the clinic produced very limited results (94).

PMN-MDSCs, on the other hand, are capable of producing abundant levels of reactive oxygen species (ROS), primarily superoxide and hydroxyl peroxide, which can be triggered by various factors, including IL10, TGFβ, IL6, and GM-CSF. ROS production by MDSCs is driven by the upregulation of NADPH oxidase isoforms, primarily NOX2. Scavenging ROS protected T cells from ROS-induced cell death and loss of function in vitro, demonstrating the potential of targeting this axis therapeutically. Several anticancer therapies currently under investigation may be able to reduce the levels of ROS in either a direct or indirect manner. For some of these agents, the antitumor effect, at least in part, has already been attributed to MDSCs (95, 96).

Interaction of superoxide and NO generates highly reactogenic peroxynitrite (PNT; ONOO⁻). MDSCs have been shown to generate PNT in large quantities (97), which has been linked to their immunosuppressive function. PNT is able to nitrate several amino acids, including tyrosine and methionine. Protein nitration has been implicated in the MDSC-mediated induction of antigen-specific T-cell tolerance. Specifically, PNT has been shown to nitrate the TCR–CD8 complex, interfering with the conformational flexibility of TCR chains and pMHC binding to CD8⁺ T cells, rendering them unable to respond to antigen-specific stimulation (97). In addition to TCR, MDSCs can hamper antitumor immunity through nitration of other molecules, including CCL2 (98), STAT1 (99), and LCK (100).

Targeting MDSC-mediated protein nitration in cancer could thus represent a promising therapeutic strategy. Encouraging preclinical data have been generated demonstrating that scavenging PNT with uric acid in vivo can sensitize refractory tumors to ICB (100) as well as significantly delay tumor growth in combination with DC vaccination (97). Furthermore, in vivo administration of a small-molecule blocker of PNT AT38 could improve tumor-infiltrating lymphocyte migration into the tumor and synergize with adoptive cell therapy (98).

Myeloperoxidase (MPO) is an enzyme that catalyzes the production of a potent oxidant, the hypochlorous acid. MPO is upregulated by PMN-MDSCs from TB mice, as well as in plasma from patients with cancer. MPO catalytic activity can have a negative impact on cross-presentation capacity of CD103⁺ DCs in vitro, whereas cross-presentation of tumor-associated antigens is improved in MPO-deficient mice. Furthermore, pharmacologic blockade of MPO with 4-aminobenzoic hydrazide showed synergy with checkpoint blockade in tumor growth inhibition in vivo, although it was ineffective as monotherapy (101).

There is an accumulating body of evidence that oxidized lipids, especially prostaglandin E2 (PGE2), play a major role in MDSC-mediated suppression. FATP2 was shown to be critically involved in the accumulation of arachidonic acid (AA)–containing lipids. Polyunsaturated AA is highly susceptible to oxidation via oxidation machinery present in PMN-MDSCs. Encouraging antitumor effects have been observed with the antagonists of EP4 receptor for PGE2 preclinically, including synergy with checkpoint blockade (102), and these agents are currently being investigated in clinical trials (Table 1).

The blockade of VEGF receptor using a tyrosine kinase inhibitor, axitinib, in combination with anti-CTLA4 antibody increased survival in melanoma models of subcutaneous and intracranial brain metastasis. The reduced tumor growth effects were due to an enhanced antigen presentation function of DCs, reduced suppressive capacity of M-MDSCs, and improved T-cell function (103). In a head and neck cancer preclinical model, selective inhibition of PI3Kδ and PI3Kγ isoforms with low-dose IPI-145 enhanced CD8-dependent responses to anti–PD-L1 therapy. This was due to abrogated production of arginase and iNOS from MDSCs induced by IPI-145 in a dose-dependent fashion (104). This was consistent with previous reports implicating PI3Kγ in immunosuppressive activity of myeloid cells (105). PI3Kγ inhibitors are currently being tested in clinical trials (Table 1). The PI3Kγ inhibitor IPI-549, for example, has demonstrated early clinical activity and a significant reduction in MDSC numbers in some patients (106). Furthermore, Li and colleagues (107) identified c-Rel, a member of the NFκB family, as a myeloid checkpoint. In TB mice, genetic ablation of c-Rel abrogated generation of MDSCs and also reprogrammed their metabolism toward a proinflammatory state through downregulation of protumoral genes (Arg1, Cebpb, and Nos2). Moreover, pharmacologic inhibition of c-Rel alone could reduce tumor growth and also synergized with anti–PD-1 therapy in a melanoma model (107). Recently, a new target on myeloid cells emerged: C-EBPα transcription factor, which acts as a negative regulator of MDSC differentiation. A recent phase I clinical trial of MTL-CEBPA, a first-in-class small activating RNA therapeutic designed to upregulate the transcription of C-EBPα, has demonstrated encouraging safety and efficacy in hepatocellular carcinoma. Of note, MTL-CEBPA was able to downregulate CXCR4 expression on immune cells in this study (108). We found that MTL-CEBPA could block the suppressive activity of M-MDSCs, but not PMN-MDSCs, in preclinical mouse models of cancer. In patients with cancer, treatment with MTL-CEBPA resulted in reduced presence of M-MDSCs and repolarization of macrophages toward the M1-like phenotype (109).

Although many mechanisms of MDSC targeting described above have not yet been formally tested in the clinic, key messages are already starting to emerge, which could be important for the development of effective therapeutic strategies.

Overreliance on only transplantable tumor models for the assessment of MDSC-targeting compound activity may often produce misleading results. Transplantable subcutaneous tumor models are very good starting points in studying MDSCs. They usually have a strong inflammatory component and express tumor-associated antigens. They often generate a large number of MDSCs and have a short kinetic of tumor growth, which helps in the evaluation of the antitumor activity of therapies. However, these benefits come with drawbacks. Biological characteristics of MDSCs are often skewed toward one or two of the most prominent factors that mediate their activity, depending on the nature of the tumor. Targeting these molecules produces strong antitumor effects that may not be easily reproduced in the clinic. For example, high production of MDSCs in these models helped to demonstrate impressive antitumor effects of MDSC migration blockade. However, translation of these findings to a much more settled process in patients with cancer may present a challenge. Transplantable orthotopic models could alleviate some of these problems, since a much lower number of tumor cells is usually injected, and the inflammatory component is much weaker. This may affect the biological response of MDSCs (7). Transgenic tumor models may provide additional benefits due to much slower tumor development. However, the nonimmunogenic nature of most transgenic models and the challenges associated with the evaluation of antitumor responses make the use of these difficult for an extensive evaluation of therapeutic compounds. It seems that lesson one in the development of MDSC targeting is the need for validation of potential therapeutic compounds in at least two different types of models: transplantable (preferably subcutaneous and orthotopic) and transgenic (Fig. 4).

Because MDSCs are pathologically activated PMNs and MONs that are easily isolated from peripheral blood, it is tempting to test therapeutic compounds directly in these cells. The challenge is that PMN-MDSCs are terminally differentiated cells, and their targeting in vitro (except direct elimination strategies) would not be very effective. M-MDSCs cultured on plastic get activated quickly. It appears from the therapeutic targeting standpoint that the most effective way to generate MDSCs in vitro would be from progenitors. In this case, therapeutic intervention can be applied during MDSC development, which would reflect the natural process in patients with cancer more closely. The challenge is to determine the appropriate way to develop these cells in vitro, which would reflect the natural conditions. The use of conditioned medium from preselected tumor cell lines or tumor explants could provide the most adequate condition for this purpose.

It has now become evident that many immunosuppressive pathways used by MDSCs are redundant. Effective targeting of one pathway may result in a rapid upregulation of the other. Typical examples could be ARG, iNOS, and IDO. Thus, targeting one pathway alone may not be sufficient. Rapid turnover of MDSCs would result in the replacement of existing cells by newly differentiated cells that could use different suppressive mechanisms. Moreover, certain components of the immunosuppressive pathways may play critical roles in the regulation of normal metabolism and cell function, which could undermine such therapeutic efforts. For instance, superoxide is an important mechanism of the PMN-MDSC–mediated suppression. However, it is also critical for antigen processing by DCs. Blockade of superoxide can thus have a negative effect on the development of the immune response. Another example is the inhibition of PGE2, which is beneficial for the control of MDSC suppressive activity. However, systemic inhibition of PGE2-mediated signaling is associated with substantial toxicity due to its key role in cell metabolism. Targeting individual MDSC suppressive pathways thus poses many challenges. Developing novel strategies to reprogram these cells away from the immunosuppressive phenotype or to block their differentiation could be alternative MDSC-targeting approaches to be considered. Thus, lesson two is that intelligent combinations of therapeutics targeting distinct biological pathways could be a more effective approach to MDSC targeting.

PMN-MDSCs and M-MDSCs/macrophages exploit different mechanisms of suppressive activity and distinct differentiation pathways. Selective targeting of one group of cells may not be sufficient to control tumor progression. There are a number of examples of this phenomenon. Targeting of macrophages with inhibitors of CSF1R resulted in the increased infiltration of tumors by PMN-MDSCs, which neutralized the antitumor effect of therapy (77). Furthermore, inhibition of histone deacetylase 1 (HDAC1) with entinostat resulted in the blockade of suppressive activity of PMN-MDSCs only and showed a limited antitumor effect in mouse models, as well as a very modest clinical effect (110, 111). In a follow-up study, it was noticed that M-MDSCs had low HDAC1 activity. Instead, they had high histone deacetylase 6 (HDAC6) activity. A combination of HDAC1 and HDAC6 inhibitors blocked the suppressive effect of both PMN-MDSCs and M-MDSCs and produced a strong antitumor effect without further addition of checkpoint inhibitors (112). Thus, lesson three is the need to target both arms of myeloid cells: PMN-MDSCs and M-MDSCs/macrophages.

A large number of immunosuppressive mechanisms described for MDSCs suggested that they cannot all be employed at the same time. In different types of cancer and different stages of disease, the prevailing mechanisms of MDSC activity may be different. For instance, PMN-MDSCs are a predominant group of cells that are expanded in many types of cancer. However, in melanoma, prostate cancer, and multiple myeloma, it is representation of M-MDSCs that is quite substantial, strongly suggesting that distinct mechanisms of regulation would be required. Furthermore, although it is widely accepted that a cohort of patients with cancer exhibits substantial accumulation of MDSCs, the analysis of individual patients showed that a proportion of patients had in fact no increase in MDSCs. Therefore, it is likely that MDSC targeting in these patients would be ineffective. Thus, lesson four is that the critical component of rational MDSC targeting is the selection of patients with a clear increase in a defined subset of MDSCs. Because MDSCs use different mechanisms to mediate their suppressive activity, which may depend on the type of cancer, it would be critical to identify the prevailing mechanisms employed by MDSCs in patients with specific pathologies, with the goal to tailor therapeutic strategies accordingly.

No disclosures were reported.