Understanding and Overcoming Immunosuppression Shaped by Cancer Stem Cells

Ever since Richard Nixon declared a “War on Cancer” in 1971 (National Cancer Act of 1971) and Barack Obama and Joe Biden began the Cancer Moonshot initiative in 2016 (Obama-Biden-cancer moonshot), understanding of the cellular and molecular mechanisms responsible for tumorigenesis and cancer development has made impressive advances (1). This improved understanding of cancer biology has paved the way to improved treatments. Conventional chemoradiotherapy (CRT) plus targeted (at mutant oncoprotein) therapy can reduce tumor load, but therapeutic refractoriness and cancer relapse remain a major challenge (2, 3). The heterogeneous nature of cancer helps account for therapeutic resistance (1, 2, 4–6). Clinical evidence increasingly supports a correlation between the biological characteristics of cancer stem cells (CSC) and cancer relapse-associated mortality, and the therapy refractory tumor fraction is often enriched with CSCs (Fig. 1A; refs. 7, 8).

CSCs are defined as the cell population within a tumor that has abilities of self-renewal and tumor initiation, and more rigorously the ability to survive stress as well as support tumor regrowth after therapy (9–13). CSCs can originate from normal stem cells (9–11, 14); however, CSCs can also be derived from differentiated cells, from transdifferentiation, or even from reprograming to different developmental stages, accounting for the common phenotypic plasticity of tumors (7, 15, 16). On the other hand, CSCs can contribute to generating phenotypic plasticity by giving rise to many progenies, with different proliferative capacities, lineage impairments, and epigenetic landscapes (Fig. 1B; refs. 7, 17). In addition, the number of CSCs varies from relatively rare to abundant in different types of cancers and is modulated by therapy, which generally induces CSC expansion (18, 19). Furthermore, the complicated tumor microenvironment (TME) can also play a role in positively or negatively influencing CSC survival, proliferation, and metastasis (Fig. 1A; refs. 13, 20–25). Therefore, understanding the properties of CSCs and their cross-talk with the TME is of great significance as researchers seek to improve cancer treatments, especially immune therapy.

Over the past decade, the use of immune checkpoint blockade (ICB) immunotherapies has remarkably improved patient outcomes across a range of cancer types (26, 27). ICB therapy targets key negative regulators of immune activation using antibodies blocking the CTLA4 and PD-1 pathways. ICB antibodies have been shown to induce durable tumor regressions in some patients across a range of cancer types (26, 27). As with CRT, not all cancer types or patients respond effectively to ICB-based immunotherapy (28, 29), and CSCs play a role in immunotherapy resistance (7, 24). In this review, we discuss the impact of the immune privilege of CSCs and the CSC-associated immune suppressive microenvironment in regulating the response to ICB-based therapy.

A large body of research has focused on characterizing the roles the TME plays in supporting tumorigenesis and cancer development (30, 31). The TME is comprised of tumor-associated stroma, including cancer-associated fibroblasts (CAF), endothelial cells, and immune cells. Functionally, immune cells are the main component of the TME. Among these immune cells are innate immune cells, including tumor-associated macrophages (TAM), myeloid-derived suppressor cells (MDSC), neutrophils, and dendritic cells (DC), and adaptive immune cells, including T, B, and natural killer (NK) cells (24, 32, 33). NK cells are a prototypical subset of innate lymphoid cells (ILC; Figs. 1A and 2A; ref. 34). There are two categories of TAMs. The first category is antitumor M1-like, which is induced by IFNγ, produces nitric oxide, and has the potential to target tumor cells either directly or indirectly via NK activation. The other category is protumor M2-like that is activated by IL4, IL10, or IL13 and produces polyamines and prostaglandin E2 that can promote tumor cell growth (35–37). It is well known that MDSCs suppress T cells’ ability to access and attack tumor cells (38–41). Developmentally, MDSCs are stress-activated progenitor cells that normally give rise to macrophages, neutrophils, and DCs. T-cell subsets include cytotoxic CD8⁺ T cells that, upon activation, are cytolytic against tumor cells (CTL; defined by expression of GNLY, GZMB, and PRF1) and CD4⁺ T cells that consist of Treg cells (FOXP3 positive), Th1 cells (BET, IRF1, IL12RB2, and STAT4 positive), Th2 cells (IL4, IL5, and IL13 positive), and Th17 cells (RORC and IL17a positive). Although Treg cells inhibit CD8⁺ T-cell activity, Th1 and Th2 cells have opposing functions on antitumor immunity akin that of M1-like and M2-like macrophages, respectively (33, 40, 42–49). Th1 cells support priming of CD8⁺ T cells that can target cancer cells, whereas Th2 cells suppress CD8⁺ T cells to exert anti-inflammatory and protumorigenic functions. DCs play a critical role in presenting tumor neoantigen to CD8⁺ T cells via MHC-I as facilitated by CD86/CD80 (50). NK cells have a killing power similar to that of CD8⁺ CTL cells (51). However, DCs must pre-educate CTL cells if they are to be capable of recognizing tumor cells. In contrast, NK cells must sense a difference in MHC-I expression to become activated to kill tumor cells (52). B cells can target tumor cells through the production of specific antitumor antibodies, but select B-cell populations (Breg cells) can also provide immunosuppression signals to T cells via IL10, IL35, and TGFβ (Fig. 2A; ref. 53).

Clinically, ICB-based therapy has been shown to have high efficacy against certain types of cancers, such as melanoma, non–small cell lung cancer, and renal cancers (54, 55). However, as mentioned above, not all cancer types respond to immunotherapy. For example, in colorectal cancer, immunotherapy is efficacious in only 5% to 8% of patients who exhibit high mismatch-repair/microsatellite instability and in approximately 20% of patients with colorectal cancer who have high expression of programmed cell death 1 ligand 1 (PD-L1; refs. 56, 57). In recent years, the use of immunotherapy to treat triple-negative breast cancer (TNBC), the most aggressive breast cancer subtype, has provided a new approach for its management. Unfortunately, monotherapies targeting PD-1, PD-L1, or CTLA4 have shown poor efficacy in TNBC (58, 59). There are a variety of treatment combinations that incorporate immunotherapy to address a broad range of cancers, but treatment outcomes vary significantly (60, 61). It is apparent that there is an urgent need to understand why some cancer types respond to immunotherapy whereas other types do not.

One of the main challenges prohibiting the efficacy of ICB-based immunotherapy is the fact that T cells have a limited ability to infiltrate tumors (62). Analysis of the qualitative relationship between immunophenotype and tumor mutational burden across cancer types has revealed three categories of tumor microenvironment that correspond to response to immunotherapy (63–65). These immunophenotypes are based on the spatial distribution of CD8⁺ T cells in the TME: immune-inflamed, immune-excluded, and immune-desert (Fig. 2B). The immune-inflamed category is characterized by strong IFNγ signaling, which stimulates MHC-I gene expression and facilitates MHC-I–mediated tumor neoantigen presentation. Non–small cell lung cancer, melanoma, and renal cancer, all of which respond well to immunotherapy, belong to the immune-inflamed category. The immune-excluded category is characterized by the presence of MDSCs, which inhibit CD8⁺ T-cell infiltration to the tumor site via TGFβ and IL10 signaling. Colorectal cancer, gastric cancer, pancreatic cancer, and glioblastoma fall into this category. Finally, the immune-desert category is characterized by dominant fatty acid metabolism and abnormally active Wnt/β-catenin signaling, both of which prevent CD8⁺ T cells from infiltrating the tumor (62, 63). Small cell lung cancer, HR⁺ breast cancer, and prostate cancer are examples of cancers that fall into this category. Of note, these cancer immunophenotypes are not absolute regarding a response or nonresponse, but they are predictive of the relative likelihood of response. Tumors that fall into either the immune-excluded or immune-desert category are considered “cold” tumors, which means they are essentially unresponsive to immunotherapy (66). Therefore, a major focus of current basic and clinical research is determining how to convert cold tumors into immune-inflamed (hot) tumors. Such a discovery would make effective immunotherapy treatment an option for a broader range of cancer types (67–72).

As discussed, CSCs can be derived from, and contribute to, the phenotypical plasticity of cancer cells, thus adding complexity to effectively treating tumors (73). Although many CSCs are resistant to CRT, not all are. It has been well documented that actively cycling CSCs are sensitive to CRT, whereas slow-cycling CSCs are relatively tolerant to CRT (74–77). In addition to the slow-cycling characteristic, a key feature of CSCs is the upregulation of stress response genes that occurs in response to treatment (13, 78). Another distinct feature of CSCs that sets them apart from differentiated tumor cells is their immune privilege (79–81). This quality has been established by observations that CSCs show overexpression of surface markers that make them refractory to the action of T cells, NK cells, and macrophages (82, 83). For example, overexpression of SIRPα can prevent macrophages from attacking via CD47 (84), and CD24 interacts with its receptor Siglec-10 that contains an immune inhibitory–motif in both T cells and macrophages (19, 35, 84). Upregulation of TIM3 and PD-L1 reduces T-cell activity (19, 85, 86). All these features functionally define therapy-refractory CSCs (13, 19, 87, 88).

It is well recognized that several developmentally regulatory pathways, including Wnt, Notch, Hedgehog, TGFβ/BMP, FGF, and PI3K/Akt, play major roles in a variety of biological processes in CSCs, including self-renewal, proliferation, metastasis, and treatment resistance (17, 89, 90). However, immune-escape regulation in CSCs has only recently emerged as a critical feature (19, 91). For example, TGFβ upregulation protects CSCs by suppressing cytolytic immune cells (92). Wnt/β-catenin signaling has also been shown to be increased in CSCs and to antagonize with antitumor immunity (93). Mechanistically, β-catenin can directly bind to the loci of numerous immune checkpoint genes, including PD-L1, TIM3, and CD24. Increased PI3K/AKT signaling in CSCs can upregulate expression of these immune regulators via phosphorylation at Ser552/Ser765 of β-catenin (19, 94). Intriguingly, this AKT-dependent upregulation of immune checkpoint genes has been predominantly detected in leukemia stem cells (LSC) but not in leukemia blasts, which confers LSCs with immune privilege (19). The interaction between MTDH and SND1 has been identified as a stress response program that enhances the metastatic capabilities of CSCs by providing protection against immune cells. The MTDH–SND1 complex reduces the stability of mRNA encoding key components of the antigen presentation machinery to suppress antitumor T-cell response (95–99). NF-κB is also well known to be widely involved in the regulation of immune cells in cancer (90, 100, 101), and NF-κB has been reported to be critical for many biological CSC properties (102–104). Furthermore, Hedgehog signaling supports stem-like phenotypes while also promoting tumor immune tolerance (105). The signaling module CD200-CD200R was shown to involve in the immune evasion of CSCs (106). Importantly, the signaling pathways involved in immunosuppression vary in different tissues and in different cancer types (Fig. 3).

Although the role of the TME in tumorigenesis has been well documented (30, 31), CSCs are the critical cell population underlying treatment refractoriness and are the primary driver of tumor regrowth (13, 78, 107). The direct interaction between CSCs and their associated niche is therefore critical in terms of therapy resistance, which includes immunosuppression and subsequent cancer relapse (24, 33, 87, 88, 108–110). For example, an IL33/TGFβ regulatory loop between macrophages and CSCs has been reported (111). In addition, a systematic analysis revealed that different TME components dynamically interact with CSCs via dynamic molecular signaling modules (13). A key finding from these studies is that, under stress, CSCs can actively recruit tumor-associated monocytes and macrophages, which in turn support CSC survival and expansion posttreatment (13, 30, 112–114). Furthermore, bidirectional cross-talk between CSCs and other TME cells, including T cells, MDSCs, B cells, and CAF via various pathways (Fig. 4A; refs. 13, 115). For example, MIF1/CD74 signaling between CSCs and MDSCs or Breg cells was shown to mediate the immunosuppression of cytotoxic CD8⁺ T cells in melanoma and glioblastoma (via TGFβ, IL10, and PD-L1). In addition, cancer cells can secrete RPS19 to activate C5AR1 signaling in MDSCs, which suppresses immune response by promoting generation of Tregs and reducing infiltration of CD8⁺ T cells into tumors (13, 81, 116, 117). NF-κB was shown to be a major player in regulating immune response via impacting the TME, including mesenchymal stem cells, innate immune cells, and adaptive immune cells (118, 119). As an essential regulator of innate and adaptive immunity, Notch plays a critical role in the immune regulation of TME cellular components, including polarization of macrophages and helper T cells, as well as T-cell development and maturation (120–122). Another known CSC marker, ALDH, was shown to be involved in the immune regulation of MDSC in the TME to promote tumor growth (123). In addition, LILRB signaling, which can be activated by APOE, mediates immunosuppression (124, 125), which is supported by the fact that lack of APOE was shown to reduce growth and increase CD8⁺ T-cell infiltration (126). Furthermore, alterations to the extracellular matrix in the TME can impact immune infiltration. DDR1 signaling in cancer cells promotes collagen alignment (127), and CAFs, which can be activated by CSCs, secrete hyaluronan (128), which helps form a physical barrier for T cells. Overall, regulation of various components of the TME by CSCs plays an integral role in promoting immunosuppression and therapy resistance.

Because Wnt/β-catenin signaling endows CSCs with immune privilege, one plausible strategy to overcome this immune privilege is to inhibit the Wnt/β-catenin signaling module (19, 92, 111, 129). For example, the widely used anthracycline chemotherapeutic drug doxorubicin was shown to inhibit Akt-mediated phosphorylation of β-catenin at S552 at a low dosage (DXR-low). Indeed, DXR-low can reduce pS552 β-catenin levels but not β-catenin protein level per se, and it has been shown to suppress LSCs by inhibiting expression of multiple immune checkpoint genes in a manner that depends on the presence of CD8⁺ T cells in a T-cell acute lymphoid leukemia mouse model (19). This observation has been verified in a small-scale clinical trial that evaluated treatment of residual acute myeloid leukemia (AML) with another anthracycline daunorubicin (19). Inhibition of the interaction between MTDH and SND1 reduces breast cancer metastasis by suppressing the immune protection of CSCs, thereby enhancing efficacy of anti-PD-L1–based immunotherapy (95). Hedgehog signaling was shown to be involved in the immune tolerance of cancer (130), and inhibiting the Hedgehog pathway has been shown to remodel the immune microenvironment (105). Expression of CD200 was reported to be associated with Hedgehog expression and CSC features, and the CD200-CD200R interaction was reported to increase CSC immune evasion (106, 131), suggesting this pathway as another potential target to overcome CSC immune privilege.

In addition to the large body of research that focuses on the role of the TME as it relates to whole tumors, there are numerous reports on overcoming the immunosuppression imparted by the TME (132–134). These reports include the use of anti-PD-L1, anti-IL1β, and anti-TGFβ1 antibodies that target T cells (127, 135), NK cells (136, 137), MDSCs, B cells (138–141), and DCs (142). In this subsection, we discuss the cross-talk between CSCs and the niche.

Bidirectional cross-talk between CSCs and the TME actively occurs during tumor progression, especially when CSCs are under stress (13, 115). For example, in response to CRT, CSCs are challenged by CD8⁺ T-cell-mediated FASL/TNFαR signaling that promotes tumor cell death (13). In turn, CSCs respond by shaping the TME into an immunosuppressive barrier that involves MDSCs, Treg, and Breg cells as dominant players (143–145). Because TGFβ and IL10 are the primary mediators of immunosuppressive signaling (146), clinical trials are now evaluating an approach that inhibits TGFβ signaling (147–149). Of note, directly inhibiting TGFβ and IL10 reduces immunosuppression but comes with potential toxicity (111). Alternatively, targeting TME cellular components that produce immunosuppressive signals may be a better therapeutic scenario. For example, human anti-DDR1 antibodies were reported to significantly increase T-cell infiltration in breast cancer treatments (127, 140). Anti-LILRB4 antibody was shown to reduce immunosuppression and is now being used in a clinical trial for the treatment of monocytic AML (124, 150). Furthermore, suppression of MIF1-CD74 and/or RPS19-C5AR1 signaling between CSCs and MDSCs, Tregs, or Bregs has a potentially broader effect to facilitate the conversion of immune-excluded tumors into immune-inflamed tumors in a variety of cancers (13, 81, 116, 117, 151). Given the essential role of NOTCH and NF-κB signaling in TME immune system regulation, targeting each of these pathways to overcome immunosuppression is a promising approach (100). However, because of the broad influence of these two pathways, it is imperative that this strategy be carefully designed to reduce cytotoxicity and negative effects on normal immune function (Fig. 4A).

Metabolic reprogramming is a key feature of cancer and can impact the TME, making modulating metabolic pathways a potential strategy to stimulate antitumor immunity. For example, fatty acid metabolism is known to mediate immunosuppression (152), and it can contribute to innate immune and adaptive immune cell proliferation, differentiation, and functionality. Wnt signaling regulates lipogenic gene expression, de novo lipogenesis, lipid desaturation, and lipid metabolism; this is, in part, mediated by control of Mlxipl and Srebf1 transcription by Wnt signaling (153–158). Fatty acid metabolism can contribute to innate immune and adaptive immune cell regulation, suggesting it could be an immunotherapeutic target. In the TME, macrophage polarization from the antitumor M1-like to protumor M2-like state primarily relies on oxidative phosphorylation (159), and metabolic mediators, such as AMPK, mTORC2, PPARs (key lipid sensors), and CD36 (transporter for free fatty acids), impact macrophage phenotypes. The observation that MDSCs gathered from patients with cancer or tumor-bearing mice have higher lipid accumulation compared with controls indicates a correlation between fatty acid metabolism and potential MDSC immunosuppression activity (160, 161). The presence of both M2-like macrophages and MDSCs is associated with decreased T-cell infiltration and activity, leading to immune suppression. Indeed, CD8⁺ T-cell activity was drastically suppressed by MDSCs with high lipid overload compared with MDSCs with normal lipid content (162). All these factors involved in metabolic regulation represent plausible targets that could be used to reduce the immune suppression of CSCs and the associated TME.

In summary, phenotypic plasticity endows CSCs with intrinsic immune privilege and facilitates the formation of an extrinsically immunosuppressive barrier. Understanding the cellular and molecular mechanisms underlying the phenotypic plasticity in CSCs provides valuable insight into how to overcome immune privilege and immunosuppression. As described above, targeting the cellular components and molecular signaling involved in generating immune-excluded and immune-desert phenotypes opens the possibility of converting these two immunologically “cold” tumor types into “hot” immune-inflamed tumors; in theory, tumors converted into immune-inflamed phenotype should be more responsive to ICB treatment. Recent studies have uncovered how the balance between activation and exhaustion of T cells is tightly regulated (163–166), which points to the critical role of ICB in keeping active T cells from becoming exhausted. However, overcoming CSC-mediated immunosuppression and ICB resistance will require additional approaches. It has been shown that inhibition of Wnt signaling plus anti-CTLA4 treatment reduces melanoma progression (167), so it is very likely that combinatorial treatment involving immune checkpoint inhibition and approaches that overcome CSC immune privilege and suppression will be necessary for achieving optimal treatment outcomes for a variety of cancers (Fig. 4B).

No author disclosures were reported.

The authors appreciate Xi He for the scientific discussion and Michael Epp for manuscript editing. The authors are grateful to M. Miller for the artistic illustration. L. Li was supported by SIMR funding (SIMR-1004), NIH grant (U01DK085507), as a member of the Intestinal Stem Cell Consortium funded by NIDDK and NIAID, and by NCI grant (P30 CA168524). R.A. Jensen was supported by an NCI Cancer Center Support grant (P30 CA168524).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.