Glioblastoma Stem Cells: Driving Resiliency through Chaos

Abstract

Glioblastoma is an aggressive and heterogeneous tumor, with glioblastoma stem cells (GSCs) at the apex of an entropic hierarchy, imparting devastating therapeutic resistance. The high entropy of GSCs is driven by a permissive epigenetic landscape and a mutational landscape that revokes critical cellular checkpoints. The GSC population encompasses a complex array of diverse microstates, defined and maintained by a wide variety of attractors including the complex tumor ecosystem and therapeutic intervention. Constant dynamic transcriptional fluctuations result in a highly adaptable and heterogeneous entity, primed for therapeutic evasion and survival. Analyzing the transcriptional, epigenetic and metabolic landscapes of GSC dynamics in the context of a stochastically fluctuating tumor network will provide novel strategies to target resistant populations of GSCs in glioblastoma.

Glioblastoma

Glioblastoma (World Health Organization grade IV glioma) is a universally lethal disease for which there is no effective therapy. Current standard-of-care includes maximal surgical resection, concurrent radiotherapy and treatment with the orally available alkylating agent temozolomide, followed by adjuvant temozolomide, a treatment regimen which extends survival to a median of only 14.6 months (1). Glioblastoma is a heterogeneous tumor, as reflected by its previous designation, glioblastoma multiforme, with multiple subclonal driver mutations creating a highly adaptable entity that is resistant to all therapeutic approaches (2,3). Glioblastomas are complex ecosystems, which rapidly evolve in response to harsh environmental conditions. As tumors have been characterized as “wounds that do not heal”, tumor cells can coopt stem-like features to survive and thrive (4). Further, tumors actively remodel their microenvironments through modulation of the immune system, stroma, and vasculature (5). Thus, numerous drugs showing promising results in preclinical studies have failed to demonstrate efficacy in clinical trials.

Intratumoral heterogeneity and therapeutic resistance that characterize glioblastomas are thought to be promoted by glioblastoma stem cells (GSCs), which demonstrate two principal features of stem cells: self-renewal and differentiation (6–8). GSCs recapitulate the heterogeneity of the parental tumor in vivo, and their biological relevance is demonstrated by their functional role in tumor growth and recurrence (8–10). GSCs drive pharmacologic, radiation and surgical resistance, and are thus a critical therapeutic target (9–13). GSCs thrive in harsh, complex microenvironmental niches, unencumbered by stringent checkpoints on proliferation and survival that constrain their normal counterpart (14–18). Several markers, including CD133 (PROMININ1), CD15 (stage-specific embryonic antigen-1, SSEA1), L1CAM and SOX2 enrich for GSCs, although, similar to normal stem cells, no marker or set of markers (i.e. immunophenotype) has been identified that exclusively and comprehensively mark GSCs (8,19,20). Intertwined with this question of classification is the ongoing controversy regarding the structure, immutability and linearity of the cellular hierarchy in glioblastoma (21–23). While specific pathways that contribute to the augmented aggressiveness and resiliency of GSCs have been described, effective therapies remain elusive. Successfully targeting glioblastoma heterogeneity, driven by subclonal variation, regional features (e.g. vasculature, hypoxia, inflammation, etc.) and the repopulating ability of GSCs, will require a shift away from the binary characterizations of cell state and linear gene interactions and towards an understanding of the molecular landscape of cellular potential and its relationship with microenvironmental, epigenetic and metabolic characteristics of GSCs.

Characterization of intratumoral heterogeneity, tumor evolution and single cell transcriptomes suggests that the glioblastoma hierarchy represents a dynamic network, with GSCs occupying a spectrum of multipotent microstates at positions of highest entropy (3,24–28). In this context, chaotic oscillations of cell states create heterogeneity, adaptability and therapeutic resistance (29–34). As cells differentiate, they are constrained to a limited transcriptional program (31). This model predicts greater therapeutic resistance for cells at higher entropy due to a greater diversity of available escape routes (31,34). By applying current knowledge and the growing library of –omics data to this scaffold, we can derive a more nuanced understanding of tumor evolution and therapeutic resistance as emergent properties of cellular potential.

GSCs – The Apex of a Dynamic Network

Hierarchical models of cellular differentiation, such as that proposed for glioblastoma, are characterized by predominantly unidirectional progression from founder stem populations to more differentiated progeny, a phenomenon eponymously visualized as Waddington’s landscape (22,23,35,36). Research into the apparent thermodynamic favorability proceeding towards differentiation has drawn from principles of dynamic network theory and statistical mechanics (33,37). GSCs, like normal stem cells, appear to be poised at a critical state of maximal entropy relative to their more differentiated counterparts (29,38). This state is maintained by a permissive epigenetic landscape and low, oscillating expression of a large number of genes (29,38). As a result, individual stem cells shift stochastically within a potential landscape through transcriptional fluctuations, enhancing the diversity of the overall population (33,38). Multipotency may therefore be defined as an emergent property of a cell population in constant dynamic flux (30). While entropically primed for adaptation and differentiation, the stem cell population remains relatively resilient to large, random state changes. Instead, state changes are driven by non-stochastic forces or events — known as attractors — generated by the microenvironment, therapeutic intervention or cell-cell interactions (33,39). Thus, cancers cells removed from their environment may inhabit a potential landscape that is distinct from that of the original tumor, and treatment may not only select for, but also generate, new subpopulations. This hypothesis is supported by observations that genetic discovery efforts performed in parallel in vivo and in vitro yield largely non-overlapping results, with a greater number of molecular dependencies in vivo, suggesting additional attractor states in vivo (40). As stem cells differentiate, they are drawn down into an energetic valley, becoming locked into a transcriptional profile, expressing fewer genes (31). Thus, as a whole, the stem cell population is poised to respond to a wide variety of stimuli, while differentiated cells exhibit fewer degrees of freedom and more stringently regulated genetic programs. Greater therapeutic resistance of the cancer stem cell population derives both intrinsically from chaotic fluctuations in gene expression and extrinsically from the complexity of interactions with a variety of attractor states.

The network dynamics that govern tumor cell fate are shared with normal stem cells, but without the constraints that direct normal multicellular development. Somatic mutations and genetic instability appear to re-define the potential space occupied by stem and non-stem tumor cells, making cancer networks distinct from their normal counterparts (27,29,41). While conventional tumor genetics categorize mutations into oncogenes and tumor suppressor genes, recent genetic analyses of gliomas and other malignancies have detected dysregulation of chromatin regulators, which may lead to plasticity in the epigenetic cell state (42–45). When analyzed in this context, cancer networks reside at a net higher entropy than corresponding normal tissues, as well as distinct energetic relationships between stem and non-stem populations (29). Differences in entropy between stem and non-stem cancer cells are smaller than in those of a normal tissue, particularly in GBM (29). Thus, prospective delineation of clear, binary transcriptional distinctions between stem and non-stem glioblastoma cells on a single-cell level have proven challenging (25). GSCs have also proven resistant to differentiation strategies, suggesting that terminal differentiation may have different connotations in the context of a perturbed cancer network (46).

The conceptual framework of network dynamics underlying the glioblastoma hierarchy has several concrete implications for the modeling and treatment of glioblastoma. GSCs sit atop a hierarchy of entropy and are defined by noisy transcriptional fluctuations. Probabilistically, the population will, therefore, expand to dynamically occupy all available microstates, reconstituting lost populations and regenerating heterogeneity (38,47). In essence, the significance of GSCs lies not solely in the capacity of a single cell, but in the chaotic resiliency of the network (31). GSCs move dynamically through this space in response to perturbations, such as therapeutic intervention. One challenge for the neuro-oncology research community, indeed the entire cancer stem cell community, is that measuring cell states is limited by the lack of strong immunophenotypes for cancer stem cells and limited functional assays to measure tumor biology. Therapeutic development targeted at resistant populations must therefore address not only cell state or defined cell fate, but rather cellular potential. This potential space of GSCs is defined intrinsically by genetic and epigenetic landscapes, and externally by complex tumor microenvironments acting as attractors.

The variable metabolic, inflammatory and cell-cell cues within these environmental niches serve to maintain heterogeneity, which is further reinforced by feedback as GSCs generate and remodel their environments (5,28). Interactions between GSCs and their environmental niches represent important attractor states and are critical in generating tumor heterogeneity, thereby promoting development of treatment-resistant populations.

GSCs and the Tumor Microenvironment – A Tangled Hierarchy

Three major microenvironments have been described in glioblastoma — the hypoxic-necrotic core, the perivascular niche and the invasive edge (48). Each biome serves as a unique attractor, activating a variety of cellular programs in GSCs, which, in turn, serve as architects to actively remodel the microenvironmental architecture. Niche interactions may, therefore, be critical in maintaining the breadth of states that the GSC population can occupy, promoting both heterogeneity and robust maintenance of stem properties.

The perivascular niche provides critical cues for maintenance of stemness, while inducing pathways that enrich for GSCs capable of migration and DNA repair (49–51). Endothelial cells (ECs) promote a stemness phenotype through NOTCH, sonic hedgehog and nitric oxide signaling pathways, among many others, while other perivascular cell populations, such as tumor associated macrophages, secrete chemokines that promote GSC growth and expansion (51–54). Signaling within the perivascular niche also generates GSCs that may be adapted to certain types of therapeutic resistance. TGF-β is highly expressed around tumor vasculature and promotes stem cell maintenance as well as activation of DNA repair pathways and expression of matrix metalloproteinase 9 (MMP9), an important mediator of invasion (49,53). CXCL12, a ligand expressed by ECs, provides a chemotactic signal and positive regulator of MMP expression, priming a population of GSCs in the perivascular niche for invasion (50,55). GSCs remodel and maintain the perivascular niche, producing high levels of proangiogenic factors, such as VEGF, driving EC proliferation, survival, migration and blood vessel permeability (56). GSCs can give rise to pericyte-like cells, critical regulators of vascular remodeling and stabilization, while differentiation of GSCs into tumor ECs remains controversial (57–60). Conflicting results may reflect the challenges of applying immunophenotypic and functional definitions derived from normal cellular hierarchies to the cancer cell hierarchy, which may not achieve classical differentiation. Despite the importance of the perivascular niche in tumor growth, antiangiogenic factors have not performed well in clinical trials (61). While some resistant tumors retain high levels of vascularity, presumably through angiogenic pathways that circumvent VEGF targeting, in others, the hypoxic niche expands and becomes predominant (62,63).

Hypoxic, necrotic regions are a hallmark of glioblastoma, and support GSC maintenance, proliferation, and therapeutic resistance (15,64). Hypoxic stress generates a subpopulation of cells adapted to survive in nutrient-restricted conditions, promoting shifts towards aerobic glycolysis and glutamine-mediated fatty acid production (65). Furthermore, as in normal neural stem cell niches, hypoxia is hypothesized to promote quiescence, a phenotype that could significantly contribute to the enrichment of chemo and radio-resistance populations (17). The effects of hypoxia are mediated in large part through hypoxia-inducible factor-1 (HIF-1) and HIF-2 (48). HIF-2α remains elevated under chronic hypoxia and is involved in the activation of signaling pathways regulating stem cell maintenance, including KLF4, SOX2 and OCT4 (14,66). HIF-1α, a key player in the acute hypoxic response, regulates metabolic adaptation to nutrient deprivation and promotes a mesenchymal shift in hypoxia-treated GBM cells and expression of pro-survival factors such as ERK (65,67,68). HIF-1α also promotes VEGF expression, inducing angiogenesis in hypoxic regions (69,70). Thus, the existence of the hypoxic niche primes cells to regenerate the perivascular niche, a prime example of how the multiplicity of subpopulations and signaling pathways induced by different attractor states promotes dynamic heterogeneity within the tumor.

The third major glioblastoma microenvironment is the invasive niche. GSCs are enriched for their invasive potential, a finding consistent with ability of the leading edge cells to drive tumor recurrence following surgical resection (12). These ‘surgically resistant’ populations migrate along the vasculature and white matter tracts utilizing cadherins and integrins, cleaving their way through extracellular matrix using matrix metalloproteinases, such as MMP2, MMP9 and ADAMT2 (11,71). Invasion is facilitated by several signaling pathways that are upregulated in GSCs, including L1CAM and CXCR4 (55,72). GSCs also express multiple mediators of the epithelial-mesenchymal transition, during which cancer cells convert to a more invasive, metastatic phenotype, including the TWIST1-SOX2 signaling axis, N-cadherin, STAT3, NF-κB and periostin (12,73–76). The invasive and migratory phenotype is promoted by signaling in the hypoxic and perivascular niches and is modulated by the differential tissue mechanics and matrix stiffness of blood vessels and white matter tracts (48,77). Therefore, normal brain tissue could be considered an attractor in the development of the invasive niche.

These three tumor microenvironments serve as attractors that texturize and stretch the fabric of the GBM landscape, generating a spectrum of GSC subpopulations and increasing the probability that any one population will survive a therapeutic challenge to reconstitute the others (47). Efforts to comprehensively target the heterogeneous GSC population must therefore incorporate the diversity of the tumor ecosystem to fully appreciate the complex landscape of cellular potential. We recently showed that GSCs residing in separate niches express distinct GSC markers, transcriptional profiles, and reciprocal dependencies on core epigenetic regulators, the polycomb repressive complexes (43). Differences in epigenetic regulation were reflected in differential sensitivity to BMI1 and EZH2 antagonists. Unresolved is the interconversion between GSCs in different niches, but recent evidence supports multiple stem/progenitor populations in normal tissues (e.g. bone marrow and gut) that can repopulate other depleted populations. In cancer, recapitulating cancer stem cell niches in homogenous, nutrient-rich in vitro cultures and in animal model is an ongoing challenge. To address this issue, several culture systems have attempted to mimic the heterogeneity of the tumor microenvironment in vitro through 3-dimensional culture or microfluidic approaches (78–80). Replacing convenient, but simplistic models with more accurate, complex ones will be critical in developing effective therapies. Microenvironments both rely upon and maintain the inherent heterogeneity in the GSC population that allows for dynamic transitions and flexible adaptation.

The interaction between the microenvironment and tumor genetics in shaping GSC cellular states remains an open area of investigation. Mouse modeling studies have demonstrated that specific mutational events such as NF1 loss or PDGFB overexpression shift the tumor ecosystem towards macrophage infiltration or vascular dysfunction respectively (81). Consistent with human studies demonstrating NF1 loss following temozolomide treatment in recurrent tumors (5), NF1 silencing correlated with temozolomide resistance (81). Single cell RNA-sequencing studies have begun to characterize the changing profile of GBM cells in the context of different genetic and microenvironmental attractors (24). Additional multiregional and single cell studies may further elucidate the dynamic interplay between attractors and their role in shaping the landscape of tumor heterogeneity, adaptation and resistance.

Intratumoral Heterogeneity and Tumor Evolution

Glioblastoma exhibits significant intratumoral molecular and phenotypic heterogeneity, and targeting any one component has proved minimally effective. IDH1 or IDH2 mutant glioblastomas are fundamentally distinct from IDH wildtype tumors, with relatively better prognosis and response to therapy. Classification of IDH wildtype glioblastoma has resulted in transcriptional profiles divided into three major subtypes — proneural, classical (or proliferative) and mesenchymal — that are distinguished by distinct prognostic significance, molecular signatures, biologic phenotype and stemness signatures (82). However, multiregional sampling and single cell RNA sequencing have revealed the presence of multiple subtypes within a single tumor (25). Intratumoral heterogeneity not only increases the likelihood of the emergence of resistant subclones, a phenomenon that characterizes the inevitable tumor recurrence of glioblastoma, but has also been shown to facilitate tumor growth (3,26). For example, simultaneous implantation of cells with high or inhibited HIF-1α expression led to more rapid tumor growth while co-implantation of GSCs with senescent, differentiated glioblastoma cells promotes tumorigenesis (83,84).

Subtype conversions (e.g. proneural-to-mesenchymal transition) occur frequently during tumor recurrence, presumably driven by both cell autonomous and shifting tumor microenvironments to add new attractor states secondary to a therapeutic intervention (5). The pervasive heterogeneity at the apex of the tumor hierarchy has implications for overall tumor architecture, as emergence of resistant subclones likely originates in part from the diversity of the GSC population, which then propagates changes to the whole tumor (25,26). As noted above, a single tumor can harbor different subtypes of GSCs, while GSC-derived subclones from a single patient tumor exhibit significantly different growth patterns (85–87). Ongoing efforts to characterize an exclusive, comprehensive population of GSCs have revealed the difficulty of finding a binary marker of stemness. CD133 (Prominin) is a commonly used marker with a high specificity, but low sensitivity, for GSCs in that subsets of CD133⁻ cells demonstrate stemness characteristics and are tumorigenic (8,88). Several other markers, including CD15 (SSEA-1) (20), integrin α6 (89), ALDH (90), NESTIN (91), SOX2 (92), OLIG2 (25,93) and NANOG (94), also enrich for populations with stem cell-properties. However, a single, comprehensive GSC marker has yet to be identified, and thus the nature of the cellular hierarchy (immutable vs. adaptable) in GBM remains controversial (21,23,24). It is possible that a factor or combination of markers exists that can definitively mark stem vs. non-stem cells, but which has not yet been determined. A more likely scenario is the existence of a hierarchy in which the path to differentiation is better defined not as a transition between discrete states, but as a series of reversible transitions through many microstates, with a heterogeneous population of stem cells exhibiting reversible, random oscillations until an attractor drives them towards a committed, differentiated state (28,31,95). Accordingly, single cell sequencing has revealed a gradient of stem cell markers expressed by individual cells within a tumor (25). In this context, dedifferentiation is more likely to occur on a small scale, as a series of microstate transitions, while transitions from the nadir to the apex of the hierarchy are generally improbable (and, thus, unfavorable) in naturally occurring biologic states. However, reprogramming can be externally induced through delivery of transcription factors, both in normal and tumor hierarchies (96,97).

With the capacity to regenerate a tumor and elevated therapeutic resistance, GSCs are thought to drive recurrence and evolution. Modeling of clonal evolution suggests that recurrence is driven by subclones that diverged early from the dominant clone of the primary tumor (3). However, the details of resistance in tumor evolution remain obscure.

Multiple mechanisms related to temozolomide resistance have been reported including induction of stem cell markers and de-differentiation towards a stem-like phenotype (98) or differentiation of some GSCs towards endothelial-like cells which reciprocally support the GSC population (60). Heterogeneous genetic changes may be selected for or induced in recurrent tumors (5,26,99). Treatment is often characterized as a Darwinian process, selecting for existing, adapted subclones within a tumor. Indeed, subtype conversion in tumor recurrence may represent simply an expansion an existing population. Many studies have shown that therapeutic intervention induces adaptive changes in a more Lamarckian process that may even promote elevated aggression and resistance (39). Accordingly, mutational signatures consistent with treatment paradigms such as alkylating agents are frequently found in recurrent tumors, likely originating from the GSCs that survived treatment (100). Although not yet explicitly investigated in glioblastoma, this process of evolution depends not on selection of existing clones, but on the role of treatment as an attractor state, inducing a compensatory shift in tumor cell populations. GSCs are highly enriched for this adaptive potential, exhibiting upregulation of DNA damage response, an ability to preferentially utilize nutrients in response to harsh environmental conditions and a general phenotype of therapeutic resilience (9,10,101,102).

GSC-Targeted Therapeutics and Mechanisms of Resistance

Treatment options for glioblastoma remain limited and prognosis is dismal (2). GSCs inhabit the entropic peaks within the tumor, predicting that they will have the greatest diversity of escape routes in the face of therapeutic intervention, due to less restricted genetic programs and constant flux within the population. Three characteristics of the GSC population exemplify and drive this model of resistance: genetic instability, signaling promiscuity and population heterogeneity.

As previously described, while following the same entropic pattern, cancer networks are distinct from normal cell networks (27,41,47). In GSCs, genetic instability is a significant driver of this phenotype and promotes DNA repair, aberrant tumor cell survival and mutation tolerance. Critical replicative checkpoints are commonly mutated in glioblastoma, including p53, TERT, ATRX, NF1, CDKN2A and RB1, (103,104). Building upon these founder mutations, and seemingly paradoxically, GSCs excel at DNA damage repair, upregulating key players in recognition and repair, such as damage detection and checkpoint kinases Chk1, Chk2, ATR, ATM and RAD17 and repair enzymes, PARP1 and TIE2 (9,105,106). Many of the pathways that are critical for maintaining stemness in GSCs also facilitate DNA damage repair. For example, NOTCH signaling, which is important for GSC survival, mediates radioresistance in GSCs through upregulation of pro-survival pathways PI3K/AKT and Bcl-2 (107). This balance of checkpoint regulators and repair enzymes in mutation tolerance and repair efficiency in GSCs represents a likely mechanism driving rapid evolution and plasticity in the context of environmental stress, and thus a potential target for radio- and chemo-sensitization of glioblastoma (106,108).

Signaling promiscuity characterizes GSCs both on a global epigenetic level and in specific pathways. Compared to differentiated glioblastoma cells, chromatin profiling reveals that GSCs demonstrate a widespread loss of repressive histone marks compared to normal human astrocytes and broad activation of multiple transcription factors networks that do not normally coincide (109). This pattern would allow for greater noise in gene expression, generating a dynamically fluctuating population with greater access to alternative pathways and state transitions in response to therapy. GSCs are also more metabolically flexible than their differentiated counterparts. While differentiated glioblastoma cells rely on the well-known Warburg effect for glucose metabolism, GSCs can more adeptly switch between aerobic glycolysis and oxidative phosphorylation (110). Differentiated tumor cells and most other cancer cells predominantly express pyruvate kinase isozyme 2 (PKM2), which promotes aerobic glycolysis and is primarily found in proliferating cells. However, GSCs also express the PKM isozyme PKM1, which facilitates oxidative phosphorylation, providing a potential mechanism for the higher mitochondrial utilization of GSC and the greater flexibility of GSC metabolic regulation (110). Rather than eliminating the tumor, targeting any single aspect of metabolic regulation may instead just induce a metabolic switch in GSCs.

Finally, the potential space of cellular states available to glioblastoma is diverse. The noise inherent in the genetic and epigenetic landscape of GSCs, enhanced by the heterogeneity afforded by multiple complex attractor states, generates a redundant system able to tolerate failure of any one component, such as arises from targeting by a particular treatment regimen (24,28,111). Current therapeutic modalities target specific cell states, simply selecting for or generating adaptive subclones rather than eliminating the tumor. For example, radiation induces DNA damage, triggering GSC-dependent, NF-κB-driven interconversion between subtypes towards the mesenchymal signature (112). Radiotherapy may therefore select for or induce GSCs adapted to rapid repair with a high apoptotic threshold. Both radiation and temozolomide also preferentially target proliferating cells. Single cells sequencing studies have revealed high overlap between stem cell signatures and proliferative markers, suggesting that GSCs are more proliferative than previously thought (24,25). Quiescent GSCs may exist as a subpopulation, maintained by signaling from the microenvironmental niche (113), giving rise to, or existing in parallel with, proliferative GSCs (23). While the relationship between proliferative and quiescent GSCs remains an open area of investigation, multiple studies have demonstrated that GSCs maintained in a quiescent state are more treatment resistant (110,113,114). As previously noted, anti-angiogenic agents such as bevacizumab have not been effective in clinical trials, despite demonstrating initial promise in pre-clinical models (111,115,116). Research into mechanisms of resistance to anti-angiogenic therapy suggest that it can be circumvented either as an intrinsic property of cellular heterogeneity in the perivascular population, which precludes reliance on a single pathway for angiogenesis (VEGF), or by a global shift in tumor constitution to rely on another major tumor microenvironment, the hypoxic niche (62,63,117). In addition to underscoring the importance of accurate tumor modeling, this result may represent the futility of targeting a single component of the tumor ecosystem.

When tumors are modeled as dynamic networks, higher entropy correlates with greater drug resistance and cellular potential becomes a critical mediator of resistance (34). This framework favors several therapeutic strategies. Studies targeting markers of the cell state with highest potential are already underway, utilizing immunotherapeutic, peptide- and nanoparticle-targeting strategies. While unlikely to eliminate the entire population of potential GSCs, these approaches may be able to check the recurrent or adaptive potential of the tumor, making it more susceptible to other therapeutic interventions. Another potential avenue is directing GSCs cells into particular valleys using drugs or other interventions as attractors (41). Perhaps the most obvious application of this strategy occurs in the context of differentiation therapy. However, the challenge of terminally differentiating cancer stem cells in vitro suggests that the genetic profile of these cells is resistant to complete differentiation (46). Attempts to bottleneck a tumor by driving GSCs into a predictable valley, not necessarily of terminal differentiation, but of restricted transcriptional options will require a more integrated understanding of network potential in the context of both epigenetic and genetic driving forces.

Concluding Remarks

Glioblastoma is a complex and diverse entity that has largely thwarted attempts at therapeutic intervention (see Outstanding Questions). The resiliency of glioblastoma is founded in heterogeneity and adaptability, characteristics that are enriched in GSCs (6). The GSC population is defined by chaotic state fluctuations in a perturbed genetic landscape, stripped of the natural checkpoints that constrain cellular differentiation and sculpted by a permissive epigenetic profile and a complex array of attractor states (27,29,41). Conceptualizing GSCs in the context of their thermodynamic potential suggests an alternative interpretation of the complexity of the observed tumor hierarchy, one in which the transition between apparently irreversible states occurs through many stochastic shifts and reversions. In modeling these systems as analogue states with linear dynamics, potentially critical nuances may be lost. This theory also further emphasizes the importance of contextualizing GSC function and accounting for cellular potential. Drug discovery in neuro-oncology has been characterized by a narrowing funnel of efficacy, from in vitro systems to pre-clinical animal models and finally therapeutic trials. By removing the complexity of attractors from the equation in preclinical studies, the cellular landscape becomes constrained and the tumor’s adaptive potential becomes handicapped. Furthermore, high-throughput strategies to identify nodes for precision therapy must account for the fact that they are not targeting a static population, but a dynamically adaptive one. Defining the multifaceted dependencies of the GSC population by mapping the complex intratumoral interactions that facilitate resilience may inform novel therapeutic strategies targeted not just at a snapshot of cellular state, but at a topographic landscape of cellular potential.

OUTSTANDING QUESTIONS.

• Tumor cells in different niches express specific transcriptional signatures. Do microenvironmental interactions drive distinct dependencies in GSCs?

• Epigenetic dynamics regulate expression of important oncogenic pathways in GBM. How does the epigenetic landscape differ in GSCs versus their differentiated progeny? What role do histone modifications play in maintaining the stem state?

• GSCs adapt rapidly and effectively to therapeutic intervention. Can integration of GSC epigenetic and genomic networks predict likely avenues of cellular adaptation?

Figure 1. Attractor state model of glioblastoma.

GSCs at the center of the tumor hierarchy have the highest entropy and capacity for adaptation. Attractor states (e.g. microenvironmental niches, genetic mutations, therapeutic intervention) drive the development of different tumor cell populations. Each colored petal depicts different attractor states which drives the proportion of each cellular state. Arrow on the different cellular state represent the directionality allowed on presence of attractor state. Abbreviations: GSC, glioblastoma stem cell

Figure 2. Glioblastoma stem cells across tumor niches.

GSCs are found in each tumor microenvironments and maintain heterogeneity through unique cell-cell interactions and niche properties throughout the tumor. These niches are not stable and independent, but instead are dynamic drivers of cellular adaptation and resistance, communicating and interconverting as the tumor grows and adapts. Abbreviations: GSC, glioblastoma stem cell

Figure 3. Therapeutic approach to glioblastoma stem cell adaptation and heterogeneity.

i) Classical therapeutic approaches often spare the GSC population or target individual components of the tumor landscape – e.g. tumor vasculature or rapidly dividing cell populations. This generates new attractor states along with the older untreated states and allows for the tumor to evolve and repopulate. ii) The attractor state model implies that effective therapy will require a combinatorial approach. The first treatment bottlenecks tumor adaptation by applying an initial stimulus that drives cells towards one state, and the second intervention is targeted at the resulting specific cellular state. Abbreviations: GSC, glioblastoma stem cell

HIGHLIGHTS.

• Glioblastoma contains a dynamic cellular hierarchy in which stem cell-like tumor cells (glioblastoma stem cells) occupy positions of highest entropy.

• Glioblastoma stem cells are critical drivers of treatment resistance and recurrence in glioblastoma.

• Heterogeneity and chaotic fluctuations in glioblastoma stem cell populations prime tumors for adaptation and evolution.

• Therapeutic strategies targeting cellular potential will be essential for development of effective treatments.

GLOSSARY

Cancer stem cell (CSC)

a cell that is capable of recapitulating a tumor and exhibits the two defining properties of stem cells: self-renewal and differentiation

Glioblastoma (GBM)

grade IV glioma; the most common malignant brain tumor

Glioblastoma cancer stem cell (GSC)

tumorgenic cancer stem cells in glioblastoma

Neural stem cell (NSC)

a multipotent progenitor cell that gives rise to multiple cell types in the central nervous system