The Cross Talk between Cancer Stem Cells/Cancer Initiating Cells and Tumor Microenvironment: The Missing Piece of the Puzzle for the Efficient Targeting of these Cells with Immunotherapy

The expression of HLA class I and class II molecules and APM has been investigated in CSCs/CICs isolated from colorectal cancer (CRC) and glioblastoma multiforme (GBM) showing an overall aberrant expression of these molecules, with, in some cases, failure in their modulation by the pre-treatment with IFNs (both alpha and gamma) or DNA demethylating agent (5-Aza CdR) [64, 65]. This impairment in antigen processing and presentation by CSCs/CICs lead, upon co-culture of these cells with autologous T cells, to a preferential selection and differentiation of TH2 type T cells and failure in eliciting effector functions [64, 65]. The suboptimal expression of HLA class I molecules and APM components was also reported in CSCs/CICs isolated from different type of solid tumors [64‐69]. These peculiar observations suggested that the defective expression of HLA molecules could represent a tool for the identification of CSCs/CICs [70]. On the contrary, the side population (SP) cells derived from CRC and endowed with stemness properties, showed detectable level of HLA class I molecules as well susceptibility to antigen-specific cytotoxic T lymphocytes (CTLs) [71], however this study has been performed using long-term in vitro established cell lines, that could have lost phenotypic properties of primary CSCs/CICs. Indeed, stem-like cells isolated by sphere-forming assay displayed aberrant expression of HLA class I and APM components [16, 65]. Contradictory results were obtained also in glioblastoma multiforme (GBM); CSCs/CICs isolated as sphere forming cells from this tumor have been shown to exhibit the expression of HLA class I molecules [72] while, when applying these analyses to primary GBM-derived sphere forming cells, defective expression of HLA class I and APM molecules was detected [64]. Stem-like cells expressing ABCB5 and isolated from melanoma were found to express suboptimal levels of HLA class I molecules while they were positive for HLA class II (45). APM components (e.g., LMP2, LMP7 MECL-1, TAP1 and TAP2) detected through mRNA analyses were found to be expressed in tumor sphere-models from different solid malignancies, representing a tool for in vitro enrichment of stem-like cells and for the investigation of micro-metastasis [73]. However, HLA class I and class II molecules were down-modulated in these cells as compared to differentiated tumor cells, also following their pre-treatment in vitro with IFN-γ, highlighting an impairment of antigen presentation by these cells [64]. It needs to be considered that this study did not analyze the expression at protein levels of APM molecules, therefore post-transcriptional mechanism could affect their expression. Moreover, long term in vitro established cell lines were used to isolate tumor cell spheres while in other studies reporting defective expression of APM components, primary CSCs/CICs have been investigated. These examples highlight that an overall suboptimal immunogenic potency by CSCs/CICs resulting in low or impaired susceptibility to T cell mediated immune responses (Fig. 1). This represents a mechanisms of evasion by immune responses that is shared with normal stem cells, that could represent a typical feature of cells with stemness properties [74]. Importantly, the failure in the expression of HLA molecules by tumor cells was found as one of the mechanisms of failure of the clinical activity of immune checkpoint blockade agents in cancer patients [75, 76], indicating that either CSCs/CICs can display immune evasion mechanisms shared by differentiated tumor cells or that indeed the suboptimal expression of HLA molecules of these cells and their resistance to T cell recognition can protect these cells from immunotherapy interventions, leading to tumor recurrence or progression. However, the lack of standardization in methods for both, the isolation of cells with stemness properties and to analyze HLA and APM molecules, represents a limitation in providing conclusive results. Nevertheless, detailed analysis to identify the molecular mechanisms that lead to aberrant expression of HLA molecules and APM components are warranted.

The suboptimal expression of HLA class I molecules if associated with detectable NKG2D ligands, can drive the increased susceptibility of CSCs/CICs to Natural Killer (NK) cells. This phenomenon has been observed in CSCs/CICs from glioma, melanoma, and CRC [68, 77‐79]. However, down-modulation of NKG2D ligands on CSCs/CICs has been documented, e.g., in GBM patients [64] suggesting that the expression of low levels of NK cell activating ligands can result in the impairment of anti-CSCs/CICs innate immune responses (Fig. 1). The expression profile of molecules activating innate immune responses on CSCs/CICs can be affected by their crosstalk with TME, and thus, by their plasticity that can influence the fate in vivo of these cells.

Tumor associated antigens (TAAs) can be recognized by T lymphocytes when exposed on the surface of tumor cells in the form of peptide/HLA complexes [80, 81]. They are categorized into three groups; (i) the overexpressed/self-antigens that are expressed at high levels by tumor cells and detectable, although at lower levels, on normal tissues (e.g., MART-1/Melan-A, hTERT, EGFR, survivin). (ii) Cancer testis (CT) antigens that are detectable on tumor cells and not on normal cells, except for testis and trophoblast (e.g., NY-ESO1, MAGE A3-A4, PRAME, CT83, SSX2). (iii) Neoantigens or mutated antigens derived by non-synonymous mutations in cancer cells (e.g., MUM-1, CDK4, ME1, ACTN4, HLA-A2) [82]. The neoantigens are higher immunogenic compared to differentiation/self TAAs since are tumor specific and do not induce tolerogenic mechanisms in immune cells [83‐85]. Neoantigens have been shown to drive immune responses and to mediate efficient T cell recognition of tumor cells, leading to cancer eradication in patients treated with either mutanome based vaccines or adoptive cell therapy (ACT) with tumor infiltrating lymphocytes [84‐86]. Notably, CSCs/CICs bearing a somatic mutation in the CRC-associated “driver” gene SMAD4, could elicit antigen-specific T cell responses directed to both stemness and differentiated components of tumor [87].

A transcriptome analysis of the SP cells and main population (MP) derived from CRC, breast and lung cancer revealed a preferential expression of 18 CT antigens (MAGEA2, MAGEA3, MAGEA4, MAGEA6, MAGEA12, MAGEB2, GAGE1, GAGE8, SPANXA1, SPANXB1, SPANXC, XAGE2, SPA17, BORIS, PLU-1, SGY-1, TEX15 and CT45A1) in CSCs/CICs [32]. The TAA DNAJB8, that is a member of the heat shock protein (HSP) 40 family, was found to be preferentially expressed in renal cell carcinoma (RCC); interestingly this protein played an important role in the maintenance of CSCs/CICs. DNAJB8-specific immune responses could be detected in a mice model study of DNA vaccination for RCC, rendering this molecule appealing for targeting CSCs/CICs by the immune system [32, 33] (Table 1). Recently a new antigen, Ankyrin repeat and SOCS box protein 4 (ASB4), was described as target molecule of CTLs recognizing CSCs/CICs and not the differentiated cellular components of the tumor [88]. Suboptimal expression of TAAs (MART-1, ML-IAP, NY-ESO-1, and MAGE-A) was reported in melanoma-derived CSCs/CICs (Fig. 1)[89]. Similar results were obtained in CSCs/CICs isolated from GBM and CRC (Fig. 1 and Table 1) [64, 65]. On the other hand, CD133⁺ CSCs/CICs isolated from melanoma were shown to express either NY-ESO-1 or DEAD/H (Asp-Glu-Ala-Asp/His) box polypeptide 3, X-linked (DDX3X) representing target of tumor-specific T cells [90, 91]. Other studies have described the isolation of T lymphocytes recognizing TAAs expressed by CSCs/CICs such as IL-13Rα2, SOX2 and CD133 in GBM, CEP55 and COA-1 in CRC and EpCAM in retinoblastoma (Table 1) [28, 65, 71, 92].

Natural killer (NK) cells are the first line of defense against cancer development and metastasis. NK cells have been described to efficiently recognize and kill in vitro CSCs/CICs isolated from CRC, melanoma and glioblastoma [68, 78, 93, 94]. The efficiency of NK cell-mediated lysis of CSCs/CICs was dependent on the expression of NCR ligands (NKp30 and NKp44), NKG2D ligands and when suboptimal or negative expression of HLA class I molecules were found on the surface of CSCs/CICs (Fig 1) [68, 78, 93‐95]. Tallerico et al. found that CSC/CIC but not their differentiated counterpart of CRC is susceptible to NK cells [77]. Similar results have been reported in GBM and melanoma, highlighting that the amount of ligands of activatory NK receptors on CSCs/CICs was determinant for efficient innate immune responses [68, 77, 78]. In patients with acute myeloid leukemia (AML), the suboptimal expression of NKG2D ligands has been described as a mecahnisms of escape by tumor cells from NK cell recognition [96], confirming that these molecules can affect the susceptibility of cancer cells to innate responses. The observations that NKG2D ligands could represent as biomarkers for prediction of clinical responses to immune checkpoint blockade in melanoma highlight that the pattern of NKG2D ligands expression by tumor cells can affect the type and efficiency of elicited anti-tumor immune responses [97]. Therefore, the levels and pattern of expression of NKG2D ligands by tumor cells, including CSCs/CICs could be a predictive marker for the choice of the type of immunotherapy interventions.

Dendritic cells (DCs) are antigen presenting cells (APCs) that can activate either innate or adaptive immune responses [98]. In addition, they play an important role in the formation of anti-tumor T- and B cell immunologic memories [99]. Immature DCs can capture the tumor-derived antigens by phagocytosis or pinocytosis and then migrate to lymphoid organs where they present these TAAs in the form of HLA/peptide complexes to T cells, resulting in antigen-specific immune responses [100‐102]. However, DCs depending on their morphological and phenotypic subtypes can either induce anti-tumor immune responses or promote tumor growth and progression [103]. The crosstalk of tumor with their TME is a crucial factor which results in the development of cancer [104]. Along this line, it has been described that high extent of expression of the chemokine (C-X-C motif) ligand 1 (CXCL1) by tumor and stromal cells can promote CSCs/CICs survival and proliferation and attract at tumor site DCs with suppressive functions, that could correlate tumor progression and poor survival of patients [104].

Macrophages represent important players for innate immune responses and can act as APCs similarly to DCs [105]. Based on their phenotype and functions they can be distinguished in two subpopulations: 1. The M1 subtype that are characterized by elevated pro-inflammatory cytokines, such as IL-12, IL-1β, IL-6, and tumor necrosis factor α (TNF-α), increased expression of HLA class II molecules, generation of reactive oxygen and nitrogen intermediates and ability to induce TH1-type T cell responses [106]. 2. In the presence IL-4, IL-10, and IL-13, macrophages can polarize towards M2 phenotype. These cells express scavenging, mannose and galactose receptors, IL-10, vascular endothelial growth factor (VEGF), matrix metalloproteinases (MMPs) and activation of the arginase pathway, leading to pro-tumoral effects [105‐107, 109]. The cross-talk between CSCs/CICs and TAM is orchestrated by STAT3 signaling [102, 103]. Upon iper-modulation of STAT3 in TAM, they can promote stemness, survival and proliferation in cancer cells while the latest cells can induce the immunosuppressive properties of TAM, leading to the impairment of cancer immune-surveillance [110].

Myeloid derived suppressor cells (MDSCs) are immune cells endowed with suppressive functions that can inhibit the effector functions of immune responses [111]. The frequency of these cells either at tumor site or in the circulation has been described as a prognostic factor for patients’ survival as well as of responsiveness to immunotherapy [112]. Interestingly, STAT3 can lead the differentiation of monocytes towards MDSCs in pancreatic tumors [113] regulating also the development of CSCs/CICs [113]. The secretion of pro-inflammatory cytokines and chemokines by tumor cells can induce the differentiation and recruitment of immunosuppressive cells that can also contribute to sustain the inflammatory TME and to the interaction and reciprocal influence of CSCs/CICs and their niche [110, 113‐115]. Another key regulator of the cross-talk between CSCs/CICs and TAM and DCs is represented by CD47 [116]. This molecule is over-expressed by CSCs/CICs of B cell malignancies. The binding of this molecule to the signal regulatory protein alpha (SIRPα), that mediates phagocytic functions in DCs and macrophages, has been shown to mediate the impairment of innate responses [116]. The cross-talk between CSCs/CICs and myeloid cells can affect both the fate and immunological profile of these cells, with implications for their susceptibilities to immune responses. Further studies should be designed to dissect the interactions of CSCs/CICs with different immune cells, although the major limitation is represented by the lack of in vivo models to monitor the interaction of these different immune cell population in the context of TME.

The observations that CSCs/CICs isolated from different tumor types can secrete soluble factors, such as Galectin-3, GDF-15, IL-10, IL-13, PGE2 and TGFb, or express immune checkpoint molecules with immunosuppressive functions, have suggested that these cells can regulate the impairment of immune responses as well as regulating a pro-tumoral TME (Table 2) [4, 51, 18‐130]. These immunosuppressive factors have been described to induce the differentiation of regulatory T cells (Tregs) or MDSCs and M2 macrophages, resulting in the impairment of effector functions of innate and adaptive responses [4, 51, 130, 131]. Moreover, pro-inflammatory cytokines such as IL-6, IL-8, IL-10 and IL-13, released by CSCs/CICs can contribute to maintain an inflammatory and suppressive TME representing the “niche” sustaining cellular stemness (Table 2) [132, 133]. Indoleamine 2,3-dioxigenase (IDO), that mediates the catabolism of tryptophan, has been shown to be expressed by CSCs/CICs, contributing to the differentiation of Tregs, skewing the cytokine profiling of T cells toward TH2 -type and inhibiting the survival and proliferation of CTL (Table 2) [133, 134].

In addition, the CSC/CIC-associated expression of IL-4 and CD200, through cell-to-cell interaction, lead to the inhibition of T cell effector functions (Table 2) [65, 135]. It has been demonstrated that the over-expression of IL-4 by CRC-derived CSCs/CICs has led to inefficient TCR-mediated proliferation and antigen recognition of CTLs [65]. Interestingly, the neutralization of this cytokine by mAb could overcome the T cell mediated anti-tumor impairment and induce antigen-specific recognition of both CSCs/CICs and differentiated tumor cells [65]. This study showed that, upon up-regulation of HLA class I and APM expression through IFN-γ treatment of CSCs/CICs, T cells could specifically recognize a neoantigen, SMAD4, generated by a non-synonymous mutation bearing stem-like cells and bulk tumor cells [65]. Thus, CTL reactivity against CSCs/CICs and the TAA-specific immunosurveillance could be improved by the usage of strategies to correct the low immunogenic profile of these cells.

The immune suppressive profile of CSCs/CICs has been also confirmed by evidences describing the expression by these cells of immune checkpoint molecules (e.g. CTLA-4, PD-L1, B7-H3 or B7-H4) (Table 2) [4, 34, 64, 65, 136]. These observations highlighted the similarities between CSCs/CICs and normal stem cells in terms of the immune profile [34, 137, 138]. Moreover, altered expression of STAT3 pathway in CSCs/CICs can also affect their immune suppressive activity through inhibiting T cell proliferation and activation, inducing the differentiation of Tregs and triggering T cell apoptosis [121]. The observations reported above show that multiple mechanisms and molecular pathways are either up-regulated or aberrantly activated in CSCs/CICs resulting in their immune suppressive properties, therefore the blockade of these signaling through the combination of inhibitory agents should be considered in order to rescue the tumor-specific immune responses.

miRNAs are non-coding RNAs regulating at post-transcriptional levels, through complementary binding to target mRNA, the expression of genes [139]. The altered regulation of gene expression in tumor cells can occur by both up- or down regulation of miRNA [139]. The most common activity of miRNAs in CSCs/CICs is represented by the control of the expression of either oncogenes (e.g., MiR34a, MiR31 or MiR205) or tumor suppressor genes [139]. The aberrant expression of few miRNAs, such as miRNAs 451and 199b-5p, has been shown to affect stem-like cell properties isolated from different type of tumors (e.g., GBM, breast cancer and medulloblastoma) [139‐143]. Of note, miRNAs displaying regulatory activity on immune-related genes (e.g, miRNA-199a that can regulate the IFN-mediated responses) can play a role in the differentiation of mammalian CSCs/CICs [144]. The level of miRNA-124, through regulating the expression of STAT3, can affect the efficiency of anti-CSC/CIC T cell responses in GBM [145]. Along this line, miR203 and miR92 can control the stemness and immunological profiles of melanoma cells [146, 147].

The recognition of TAAs expressed by CSCs/CICs by T cells have been documented (see Table 1). These in vitro or in vivo models were based on the usage of TAAs that represented sources of antigens for the therapeutic administration of cancer vaccines in cancer patients [81]. However, the principle limiting factor of the clinical efficacy of this strategy is represented by the usage of “self”/tolerogenic TAAs, shared with normal tissues [81]. The low or negative expression of these categories of antigens and of CT-TAAs by CSCs/CICs can represent an additional reason of failure of high rate and long duration of clinical responses observed in cancer patients treated with cancer vaccines [4, 34]. In addition, the sub-optimal levels of HLA class I molecules and APM by stem-like cells can drive the failure in targeting CSCs/CICs by cancer vaccines leading to the development of tumor dormancy and tumor recurrence, although the observance of initial clinical efficacy of these therapeutic interventions [4, 34].

DC-based vaccines, exploiting these cells as APC to present TAAs to T cell-mediated responses, represent also a therapeutic strategy for cancer patients showing encouraging clinical activity [102, 156‐162]. DCs loaded with either CSC/CIC-lysates or mRNA isolated from these cells represented source of antigens for vaccination in the context of Phase I/II clinical trials of GBM patients [163, 164]. These studies provided proof of principle of improved overall survival of cancer patients treated with CSC/CIC targeted immunotherapy [163‐165]. Immune responses, with, in some cases increased frequency of circulating NK cells, were detected in patients showing clinical benefit from these treatments [163, 165]. Of note, these therapeutic interventions could overcome the failure of CSCs/CICs in expressing efficient levels of HLA class I molecules and in presenting TAAs to T cells, documenting for the first time, that cancer vaccine, if eliciting NK cell-mediated responses, could target stem-like tumor cells [158, 163‐165]. Tumor cell clones expressing immunogenic neoantigens can undergo immune selection due to the recognition and elimination by T lymphocytes, leading to the survival of tumor cell clones not expressing strong immunogenic antigens and maintaining the expression of low immunogenic TAAs [166] (and see https://​www.​biorxiv.​org/​content/​10.​1101/​536433v1). This process is also associated with immune evasions mechanisms developed by both tumor cells and TME [166].

In some tumors, the decrease in antigen presentation is a result of epigenetic silencing of the genes involved in antigen presentation machinery. The usage of demethylating agents such as 5-Aza-2′-deoxycytidine to reduce the methylation of genes involved in antigen presentation, is a potential strategy to increase the antigen presentation in these CSC/CICs. The effect of demethylation has been shown in CSCs/CICs from breast cancer, where it resulted in high expression levels of TAP1, which is involved in antigen presentation [167].

In addition, increased antigen presentation also improves the potential of discovering novel antigens, which can then be helpful in development of new anti-cancer vaccines (Fig. 2).

Immune checkpoints, including CTLA-4, PD-1 and PD-L1, are important physiological regulators of innate and adaptive immune responses [168]. Biological inhibitory agents have been clinically developed, revealing striking therapeutically success [169‐175]. However, a significant proportion of cancer patients failed to benefit from these therapies.

The effectiveness of immune checkpoint blockade (ICB) is largely dependent on the tumor microenvironment [176]. Tumors such as melanoma, bladder cancer and non-small cell lung cancer (NSCLC) can be characterized as “hot” tumors due to their inflamed TME, high levels of and neo-antigen expression and of T cell infiltration and detection of PD-L1. These tumors have been reported to be associated with higher frequency of susceptibility to immune checkpoint treatments. On the other hand, prostate cancer is considered to be a “cold” tumor, due to minimal level of T cell infiltration, and limited response to single agent checkpoint inhibition [176].

Expression of immune checkpoint molecules has been observed in CSCs/CICs from different histological origins [51]. PD-L1 expression was detected at high levels in CSCs/CICs isolated from primary human head and neck squamous cell carcinoma (HNSCC), gastric and breast cancer, CRC and GBM [34, 64, 65, 177‐179]. ,CSCs/CICs could theoretically be targeted in vivo by immune checkpoint blockade agents, enhancing the clinical efficacy of cancer vaccines, as demonstrated in a mouse model [180]. However, recent reports describing that clinical failure of these therapies was associated with defective expression of HLA class I molecules by tumor cells [75, 76], suggest that these cells might evade from the ICB-mediated unleash of immune responses (Fig. 2).

ACT is represented by the isolation of T lymphocytes from cancer patients, their ex vivo expansion, and the infusion back into patients [181‐183]. In addition, T lymphocytes engineered to express TCR with high affinity for a cognate TAA could be exploited for ACT studies [183]. Highly encouraging and sustained responses mediated by adoptively transferred TCR targeting the TAA NY-ESO-1 have recently been reported in different tumor types, such as breast cancer and myeloma [86, 184]. Nevertheless, the antigen choice is highly relevant to prevent severe toxicities due to “off-target” cross-reaction with normal tissues sharing the same antigens or expressing molecules mimicking the TAAs [182].

Neoantigens have been described as candidate TAAs efficiently recognized by T cells that can be exploited for ACT of cancer patients and, interestingly, CTL targeting these antigens could be isolated from tumor infiltrating lymphocytes (TILs) of melanoma and other type of malignant lesions [183, 185, 186]. Nevertheless, ACT to target neoantigens can represent a promising approach for treatment of cancer patients upon assessment of HLA expression by both CSCs/CICs and differentiated tumor cells and, in case of suboptimal levels of expression, the achievement of their up-regulation by pre-treatment with immunomodulating agents [4, 34, 51].

T cell can be genetically modified to express a chimeric antigen receptor (CAR) that is composed of epitope-specific domains isolated from mAbs linked to T cell-derived activatory/costimulatory molecules [182, 187‐189]. CAR-T cells can recognize TAAs independently on the expression of HLA molecules and APM components [189]. CAR-T cell therapy for some subgroups of hematological malignancies represent the salvage intervention leading to stable clinical responses and improved overall survival of patients refractory to standard therapies or with recurrences [187, 189‐192]. The usage of CAR-T cell therapy for solid tumors is currently under investigation, showing encouraging results in cancer patients with aggressive tumor types, including malignant mesothelioma, pancreatic cancer and GBM [193‐197].

CAR-T cells targeting TAAs, such as CD133, EGFRvIII, EpCAM, CSPG4 and B7-H3, expressed by different type of solid tumors, including CSC/CIC components, have been developed in pre-clinical studies [196‐201]. These studies have shown that the targeting of TAAs that are expressed only by tumor cells, including CSCs/CICs, and not by normal cells, could provide the rational for safe and efficient clinical development of CAR-T cells therapy for these tumors(Fig. 2) [200, 202]. Moreover, the combination of CAR-T cells targeting dual TAAs, EGFRvIII and CD133, has been used for the successful therapeutic treatment of a patient with advanced cholangiocarcinoma [203].

CAR-T cells targeting NKG2D ligands on CSCs/CICs have been investigated and tested both in vitro and in vivo [204, 205]. CSCs/CICs from glioblastoma expressing detectable NKG2D ligands could be efficiently targeted by CAR- T cells. These tools showed to efficintly eliminate also xenograft tumors [205]. Nevertheless, a major limitation associated with the use of these CAR-T cells is represented by the variability of the levels of NKG2D ligands on the surface of CSCs/CICs, depending on their origin and methodology used for their ex vivo isolation.

Nevertheless, further investigations aimed at a comprehensive genomic and immunological characterization CSCs/CICs are warranted to implement the efficiency and safety of ACT strategies.