Cancer stem cells in gliomas: Evolving concepts and therapeutic implications

Abstract

Cancer stem cells have been implicated in the hierarchical heterogeneity and treatment resistance of hematologic and solid tumor malignancies, including gliomas, for several decades now, but their therapeutic targeting has not been fully realized. Recent studies have uncovered deeper layers of cancer stem cell complexity, related to developmental origins, plasticity, cellular states, and interface with the microenvironment. To harness the potential of cancer stem cells for clinical application, there is urgent need to continue to investigate their complex nature and myriad interactions, to better understand the contribution of these self-renewing, stem-like cancer cell populations in the pathogenesis and therapy resistance of malignant brain tumors.

Cancer is a heterogeneous disease caused by a variety of genetic mutations in susceptible cells with varying disease manifestations in individual patients. Despite numerous advances in cancer treatment, many patients succumb to tumor progression, relapse and reduced mortality. In the last several decades, it has been suggested that a population of cancer cells that behave like stem cells, much like their normal counterparts, may drive tumor heterogeneity and growth, thus contributing to treatment resistance and recurrence¹.

The cancer stem cell (CSC) theory posits that subsets of self-renewing cancer cells propagate and maintain long-term tumor growth². In contrast to the stochastic model whereby each cancer cell is equally potent in driving tumor formation, the CSC model implies a tumor hierarchy where CSCs sit at the apex and non-CSCs constitute the majority of the tumor bulk (Figure 1A–1B). CSCs, first identified in leukemias, have since been identified in solid tumors, and numerous strategies have been devised to investigate these cells in the hopes of eradicating them to treat cancer^2,3.

Figure 1. Models of cellular hierarchy in cancer and the cancer stem cell hypothesis.

A. Stochastic model. Tumor cells are equipotent and the majority of tumor cells have the ability to maintain tumor growth. B. Hierarchical cancer stem cell model. Self-renewing cancer stem cells sit at the top of the hierarchy and are mainly responsible for tumor growth and maintenance. C. Cancer stem cell evolution model. Cancer stem cells are the self-renewing cancer cells that sustain tumor growth, but due to extrinsic (e.g. tumor niche, exposure to therapy) and intrinsic (e.g. genomic instability) factors, evolve to produce new cancer stem cell clones that are able to propagate the tumor.

Cancer stem cells in gliomas

Gliomas are the most common primary malignancies in the central nervous system (CNS), and the most aggressive form, glioblastoma (GBM), is also the most prevalent. These tumors are refractory to standard therapies including surgical resection, radiation and chemotherapy; hence, prognosis of these patients remain dismal, with a median survival of about 15 months from time of diagnosis^4,5.

In GBM, functionally defined CSCs were identified in human brain tumors by serial xenotransplantation assays, where CSCs exhibit increased tumorigenic potential and shorter latency compared to non-CSCs⁶. More recent studies use genetic ablation techniques to eradicate putative CSCs, showing inhibition of tumor growth, and increasing efficacy of chemotherapeutic agents such as temozolomide against glioblastoma⁷. Glioma CSCs have been shown to play important roles in mediating resistance to radiation and chemotherapy, which preferentially target non-CSCs, angiogenesis and recurrence^7–9. Furthermore, lineage mapping studies demonstrate the hierarchical nature of CSCs, showing long-term generation of dominant clones and recapitulating heterogeneity². In a human glioblastoma model, barcoded slow-cycling stem-like cancer cells were found to give rise to rapidly dividing progenitor cells that generate differentiated tumor cells¹⁰.

Cancer stem cell origins and brain tumor development

Most cancer cells express cell lineage markers that recapitulate the developmental programs of stem cells and their lineage differentiation¹¹. As such, CSCs have been purported to arise from transformed stem or progenitor cells. Mouse models have demonstrated that neural stem cells in the subventricular zone (SVZ), the largest neurogenic niche in the adult mammalian brain, are cells of origin for glioblastoma¹². This has since been verified in human GBM, where whole exome and single-cell sequencing showed the presence of low-level GBM driver mutations in normal SVZ away from matched tumors of GBM patients¹³. SVZ neural stem cells that carry driver mutations were also found to migrate from the SVZ and establish tumors in distant brain regions. Lineage-restricted CNS progenitors and oligodendrocyte precursor cells (OPCs) have also been shown to be GBM-initiating cells^14,15, whereas more differentiated CNS cell types were shown to be least susceptible to gliomagenesis¹⁶.

Recent studies suggest a relatedness of radial glia, the evolutionarily conserved embryonic neural stem cells in vertebrate nervous system, with CSCs in glioblastoma. Single-cell RNA sequencing (scRNA-seq) of glioblastoma mapped onto a reference framework of the developing and adult human brain showed multiple CSC populations, including an outer radial glia-like invasive cluster¹⁷. During development, these cells expand the stem cell niche in the normal human cortex; in glioblastoma, these radial glia-like CSCs undergo characteristic mitotic somal translocation via PTPRZ1 (Protein tyrosine phosphatase receptor type Z1), suggesting that subsets of CSCs undergo reactivation of developmental programs, which contributes to the invasive phenotype. These reports, taken together, support a model whereby transformed stem and progenitor populations in specialized niches such as the SVZ give rise to CSCs that maintain tumor growth and heterogeneity in gliomas (Figure 2).

Figure 2. Cancer stem cell origins in glioblastoma.

Transformed adult neural stem cells and lineage-restricted progenitors in specialized niches give rise to cancer stem cells that propagate glioblastoma growth.

Plasticity and cellular states of cancer stem cells

Single cell technologies have revolutionized our understanding of normal tissue and tumor heterogeneity¹¹. Recent single cell sequencing efforts have demonstrated the plasticity of tumor cells in gliomas. scRNA-seq of isocitrate dehydrogenase (IDH) wild type glioblastoma revealed four main transcriptional or “cellular” states: astrocyte-like, OPC-like, neural progenitor cell (NPC)-like, and mesenchymal-like, and suggesting interconversion between the four states¹⁸. On the other hand, IDH-mutant gliomas, which include chromosome 1p/19q-co-deleted oligodendrogliomas and ATRX-TP53 mutant astrocytomas, show the majority of tumor cells as differentiated, non-proliferating cells, transcriptionally resembling either oligodendrocytes or astrocytes. The minority of proliferating cells were restricted to the more NPC stem/progenitor-like cells^19,20, reflecting the slow growing nature of IDH-mutant gliomas. In pediatric CNS malignancies, the highly aggressive midline H3K27M (histone H3 lysine-27-to-methionine) mutant gliomas exhibit a high fraction of undifferentiated and proliferating progenitors which are more OPC-like, giving rise to non-proliferating astrocyte-like and oligodendrocyte-like tumor cells²¹. These data suggest that different pathologic classes of glioma are propagated by different CSC populations, with more highly aggressive tumors showing higher proportion of putative CSCs compared to less aggressive tumors, and that these CSC populations exist in different cellular states. The dearth of neuronal signatures in glioma single cell studies is also notable, reflecting the paucity of neuronal differentiation in gliomas compared to glial cells and the neuronal pathway not being conducive to gliomagenesis¹⁶.

The transcriptional distinctions captured by scRNA-seq studies reflect the diversity of tumor cells in gliomas and are not inconsistent with the hierarchical organization of tumor cells, with more stem-like cells giving rise to more differentiated cells. These data also suggest the “plasticity” of CSCs, leading to the possibility that primary CSCs may evolve to produce more malignant and therapy-resistant versions over time or in response to therapies. Hence, new CSC clones may emerge upon internal stresses such as genomic instability, and external factors, such as the tumor microenvironment and exposure to therapeutic agents such as chemotherapy and radiation (Figure 1C). Some epithelial cancers such as colorectal cancer show evidence for hierarchical plasticity, whereby phenotypic transitions from non-CSCs to CSCs have been shown to occur, similar to how adult stem cell hierarchies in intestinal niches appear to be non-rigid^2,22. In contrast, the glioblastoma hierarchy appear to be unidirectional and consistent with the traditional stem->progenitor->differentiated cell adult neural stem cell hierarchy⁷. This unidirectional hierarchical model was implicated in a recent study identifying core transcription factors (POU3F2, SOX2, SALL2, OLIG2) that contribute to the CSC phenotype in glioblastoma²³.

Cancer stem cells and GBM molecular subtypes

In a single cell transcriptome study of IDH-wild type glioblastoma, cells with proliferation signatures were found in multiple transcriptional states, and the relative frequencies of each state within a tumor varies¹⁸. Predominant states appear to be associated with specific GBM drivers, similar to what was reported with computationally-derived TCGA (The Cancer Genome Atlas) molecular subtypes^24,25. Amplifications in EGFR were associated with astrocyte-like transcriptional states (similar to the TCGA Classical subtype), PDGFRA for OPC-like (similar to Proneural subtype), CDK4 for neural progenitor cell (NPC)-like, while mutations in NF1 were similarly associated with mesenchymal-like transcriptional profiles (similar to Mesenchymal subtype)¹⁸. Cells of any given state can propagate tumors and recapitulate the diversity of “cellular” states. On the other hand, functionally conserved glioblastoma transcriptional subtypes based on the cell-of-origin lineage have also been demonstrated in mouse and human models^14,26. These cell-of-origin-based subtypes exhibit unique tumor phenotypes and drug sensitivities. Type I GBM are SVZ neural stem cell-derived tumors that exhibit EGFR/SOX9 activation, whereas Type II GBM are OPC-derived tumors with prominent ERBB3/SOX10 dependency that lends to selective pharmacologic ErbB3 inhibition²⁶. These subtypes also appear to contain subtype-specific CSCs in mouse and human models and appear to be stable within tumors and across organisms (Xie et al., unpublished data). Thus, the above cited transcriptional states that are susceptible to transitions may represent intrinsic (e.g. driver mutations) or adaptive mechanisms due to extrinsic factors, and hence are distinct from the more stable cell lineage-based profiles. These studies underscore the intertumoral heterogeneity of IDH-wildtype glioblastoma and are consistent with distinct CSC populations in genetic driver-and cell of origin-based molecular subtypes (Figure 3). These imply that CSCs from different glioma subtypes may exhibit both unique and common susceptibilities that need to be further explored.

Figure 3. Cancer stem cells in molecular subtypes of glioblastoma.

GBM molecular subtypes, based on genetic driver, cell of origin and other determinants, exhibit unique cancer stem cell populations, but share common, dynamic tumor cell states, which can evolve as a result of extrinsic and intrinsic factors.

Treatment resistance of cancer stem cells

The diversity, plasticity, dynamic cell states, redundant evasive mechanisms and enhanced adaptability to harsh microenvironments prime CSCs for resistance to current cancer therapies. The relative quiescence of CSCs also contributes to therapeutic resistance, as conventional therapies mostly target dividing cells. In mouse models of glioblastoma, quiescent CSCs are resistant to treatment with temozolomide; however, upon diphtheria toxin-mediated genetic ablation of CSCs, these tumors develop increased susceptibility to chemotherapy⁷. In human glioblastoma models, it was shown that CSCs can reversibly transition to a slow-cycling, persistent state in response to targeted therapy using receptor tyrosine kinase inhibitors²⁷. This was mediated by Notch-dependent upregulation of neurodevelopmental programs and reorganization of repressive chromatin via histone demethylases KDM6A/B. Such mechanisms potentially allow CSCs to transition to alternate pathways and less susceptible states in response to therapy.

On the other hand, GBM CSCs have been shown to exhibit radio-resistance by preferential activation of the DNA damage and NF-κB response^8,28. The same pathways that facilitate DNA damage repair are also required for maintaining stemness of CSCs. Other resistance mechanisms include upregulation of drug efflux pumps and enhanced protection against reactive oxygen species^2,29. Hence, exposure to anti-proliferative treatment regimens leads to eradication of mostly dividing tumor cells, leaving behind the treatment-resistant, quiescent CSCs. These become activated to produce rapidly-dividing transit amplifying cells and reconstituting the tumor bulk, leading to tumor relapse (Figure 4). This is akin to the re-initiation of neurogenesis in the SVZ upon anti-mitotic treatment, causing transition of quiescent neural stem cells into activated neural stem cells, which can then actively proliferate^30,31.

Figure 4. Therapeutic resistance of cancer stem cells.

Cancer stem cells are resistant to chemotherapy and radiation, which target non-cancer stem cells. This leads to activation of quiescent CSCs that in turn produce rapidly dividing transit amplifying cells that reconstitute the tumor, leading to tumor recurrence.

Avenues for therapeutic targeting of cancer stem cells

Malignant gliomas present a unique challenge as current treatment options are limited. The central role of CSCs in tumor maintenance and therapeutic resistance makes these cells promising targets in the search for novel therapies for gliomas. Low-hanging fruit type of targets include the numerous regulatory pathways that have been reported in CSCs. Developmental signaling pathways that regulate stem cell self-renewal and differentiation have been implicated in CSCs, including Notch, Wnt and Sonic Hedgehog³². Drugs targeting these pathways are in various stages of clinical trials for solid tumors in general, though its efficacy in gliomas is not clear. However, a Phase II clinical trial of monotherapy with Vismodegib, a SMO inhibitor, in recurrent resectable GBM did not yield significant survival difference³. Another key challenge will be the implementation of multi-drug therapy regimens into experimental pre-clinical and clinical trials.

Epigenetic regulation of CSCs, which can contribute to plasticity and phenotypic transitions, can also represent attractive therapeutic targets. Protein arginine methyltransferase 5 (PRMT5) catalyzes arginine demethylation in histone and non-histone proteins, and its inhibition was shown to disrupt splicing and stemness of CSCs in glioblastoma³³. Meanwhile, CSCs have been shown to exhibit unique metabolic requirements, and oxidative phosphorylation has emerged as a CSC dependency in gliomas, which can be selectively inhibited by small molecule inhibitors such as Gboxin that target mitochondrial oxidative phosphorylation complexes³⁴.

Unlike their normal counterparts, CSCs can survive and thrive in harsh tumor microenvironments. Like other tumor cells, CSCs are able to evade immune surveillance and eradication, and promote a pro-tumorigenic niche. A recent study showed that CSCs from glioblastoma elicit immune evasion by establishing an immune suppressive microenvironment³⁵. This is acquired via epigenetic immunoediting, rather than subclonal selection, that launches myeloid-affiliated transcriptional programs in CSCs following an immune attack, leading to increased recruitment of tumor-associated macrophages. Another study suggested that CSCs from glioblastoma promote macrophage polarization toward the immunosuppressive and pro-tumorigenic M2 phenotype³⁶. Additional studies that look into specific interactions between CSCs and immune cells vis-à-vis non-CSCs will shed greater light on these immune evasive mechanisms.

To date, unlike for other cancer types, the efficacy of immunotherapy in gliomas remains to be established. Immunotherapy strategies that aim to promote immune surveillance and clearance of glioma CSCs involve using chimeric antigen receptor (CAR) T cells and immune checkpoint inhibitors³⁷. Intracranial infusions of CAR T cells that target the GBM-associated antigen interleukin-13 receptor alpha 2 (IL13Rα2) led to tumor regression without major toxicities in a recurrent glioblastoma patient; such approaches await confirmation in clinical trials³⁸. On the other hand, immune checkpoint blockade against programmed cell death protein 1 and its ligand PD-1/PD-L1 has shown limited efficacy in glioblastoma clinical trials except in case reports of patients with mismatch repair deficiency^39,40. These immunomodulatory molecules are expressed in CSCs and non-CSCs alike and it is possible that such strategies are undermined by the plasticity of CSCs and the presence of other niche cells in GBM⁴¹. For example, a unique population of CD73^hi macrophages was found to persist following anti-PD-1 treatment in GBM patients, suggesting that combinatorial therapies may improve immunotherapy response⁴². Moreover, a recent early phase trial of neoadjuvant anti-PD-1 therapy showed a survival benefit compared to adjuvant alone, hence it is possible that a therapeutic window exists⁴⁰. Modulation of the lymphatic vasculature (e.g. via VEGF-C) is another potential strategy to promote checkpoint inhibitor therapy in immune-privileged sites such as the CNS⁴³.

Concluding Remarks and Future Directions

Despite advances in our understanding of cancer biology and emerging treatment strategies in other cancers, to date, no new therapies have significantly altered the course of malignant glioma patients. The central role played by CSCs in tumor progression and therapeutic resistance provides avenues for finding new targets. However, the simple notion that eradication of CSCs, if they can be comprehensively identified and eradicated at all, will eliminate cancer has not borne out. What is clear though is that there are multiple fluid and redundant mechanisms that overlie CSC capabilities in gliomas. It is complicated by the plasticity and adaptability of these cells, cross-talk between key signaling pathways and contribution of the tumor microenvironment. This requires a multi-pronged approach that not only targets both quiescent and activated CSCs, but also addresses their genetic, epigenetic and metabolic vulnerabilities and disrupts interactions with the tumor-promoting and immune evasive niche. Only with a more wide-ranging understanding of the complex nature of these self-renewing cells and its environs will pave the way for novel therapeutic approaches particularly for these intractable CNS malignancies.