Lineage Plasticity: the New Cancer Hallmark on the Block

Abstract

Plasticity refers to the ability of cells to adopt a spectrum of states or phenotypes. In cancer, it is a critical contributor to tumor initiation, progression, invasiveness, and therapy resistance, and it has recently been recognized as an emerging cancer hallmark. Plasticity can occur as a result of cell-intrinsic factors (e.g. genetic, transcriptional or epigenetic fluctuations), or through cell-extrinsic cues (e.g. signaling from components of the tumor microenvironment (TME) or selective pressure from therapy). Over the past decade, technological advances, analysis of patient samples, and studies in mouse model systems have led to a deeper understanding of how such plastic states come about. In this review, we discuss: (i) the definition of plasticity; (ii) methods to measure and quantify plasticity; (iii) the clinical relevance of plasticity; and (iv) therapeutic hypotheses to modulate plasticity in the clinic.

Towards a consensus definition of plasticity

Lineage plasticity is a key feature of embryonic development, where it underlies normal cellular differentiation. In this setting, cell states and tissue-organizing principles are well-defined; plasticity is tightly controlled to enable the regulated formation of mature cell types. Plasticity also operates in adult organs under conditions of cellular or physiologic stress, where it enables the evolution, adaptation and regeneration of tissue, sometimes predisposing to malignant transformation (1,2). In tumors, cancer cells can co-opt these developmental plasticity programs to adopt cell states that enable invasiveness, metastatic dissemination, and therapy resistance (3). Epithelial-to-mesenchymal transition (EMT) is a classic example of this (2–5), but many other distinct forms of plasticity have emerged across tumor types (6). While plasticity is generally viewed through the lens of cellular differentiation, other cell states may also undergo phenotypic adaptations, including metabolic plasticity or other changes in cellular function. For the purposes of this review, we consider lineage plasticity to encompass changes in cell state that may be recognized histologically or morphologically, and are associated with broad epigenetic, transcriptional or proteomic preprogramming, as these manifestations of plasticity are readily observed in the clinic in association with therapy resistance. Of note, we may consider plasticity as one component of tumor heterogeneity as it may result in multiple cell states co-existing in a single patient’s tumor.

Broadly, cancer plasticity resulting in altered cell identity may be categorized as either dedifferentiation or trans-differentiation (1,6). Dedifferentiation refers to the adoption of a more primitive state or phenotype, while trans-differentiation refers to the conversion from one mature state or phenotype to another, which may or may not involve a dedifferentiation step. Importantly, the cell states adopted by tumor cells undergoing plasticity need not recapitulate cell states that exist in the normal adult or even in the embryo; instead, cancer cells engaging in plasticity may explore a spectrum of non-physiological states. In the next sections, we will provide concrete examples of how these modes of plasticity occur in different tumor types and how we might target them.

The transcriptional, epigenetic, and post-translational adaptations underlying plasticity are heavily influenced by the genetic background of the cell in which they occur (7–10). Stated differently, the genetic landscape of a cell constrains the cell state and phenotypic spectrum that a cell may adopt (9–12). In certain cases a genetic event may facilitate plasticity (11,13), while in other cases cells may exhibit plasticity as a result of transcriptional or epigenetic fluctuations independent of genetic changes (14,15). Cell-extrinsic factors, including hypoxia, nutrient availability, growth factor signaling, inflammation, cell-cell contact, and treatment with chemotherapy, immunotherapy or targeted therapies also influence cell state (15–19). The ability to generate and test hypotheses to modulate cellular plasticity for therapeutic benefit will depend on our ability to fully delineate these cell-intrinsic and -extrinsic factors.

Measuring plasticity in model systems and the clinic

Much of our knowledge regarding cellular plasticity comes from model systems, which may elucidate the spectrum of cell states or phenotypes that a cell can adopt. Historically, measurements of plasticity in both human patients and mouse models have involved one of three principles: (i) observing a redistribution of cell states or phenotypes in tissue across timepoints (e.g. under a perturbation condition); (ii) identifying cells that have features of more than one state, or, in other words finding ‘hybrid states’, using genomic, epigenetic, transcriptomic or proteomic profiling; and (iii) performing single-cell cloning of cells and observing the cell states or phenotypes adopted by clonal progeny. While these approaches have been observationally consistent with the notion of plasticity, they either fail to definitively prove the existence of plasticity or are restricted in measurements of plasticity outside of native tissues, particularly human tumors.

Advances in single cell and lineage tracing technologies using genetic barcodes or imaging now enable more direct measurements of plasticity both in vitro and in vivo(20–22). While previous approaches will continue to provide valuable information, future efforts to understand cellular plasticity will likely involve: (i) deep single cell or spatial profiling of cells within the native tissue architecture over time, thereby demonstrating the emergence of new cell states; and (ii) multicellular model systems that allow the lineage of a single cell to be traced, thereby demonstrating that its progeny exhibit multiple cell states. This type of advanced lineage tracing has now been employed in several studies(20,22–24). These methods entail DNA barcoding approaches (including expressed barcodes) or fluorescent reporters; the pros and cons of different lineage tracing approaches have recently been reviewed and must be tailored to the phenotype of interest(25).

With the generation of high-dimensional data to model plasticity, a need for better models of quantitative features of plasticity has also emerged, including reconstruction of clonal relationships between cells and inference of transition rates. Towards this end, conventional methods that model state transitions as Markov processes have proven useful(26,27), and provide an intuitive framework for quantitating transition rates. In this framework, a Markov transition matrix is used to define the transition rate from one state onto itself, or any other state. Non-zero rates between two distinct states is suggestive that plasticity exists, and different transition rates between tumor types, clonal populations within a tumor, or between individual cells, can suggest heterogeneous plasticity properties.

Of course, the approaches described above are most easily applied in model systems; studying plasticity in human tumors is more difficult as it must rely on inference rather than direct measurement of cell state changes in real time. While barcoding methods and similar lineage tracing techniques cannot be applied in humans, other techniques are available. For example, single cell approaches relying on patterns of passenger mutations or mitochondrial DNA mutations allow for the assembly of phylogenetic trees, permitting an overlay of cell lineage and cellular phenotype(28). Together, by leveraging experimental and computational technological advances and established, well-defined model systems, we will deepen our understanding of the molecular underpinnings of lineage plasticity.

Plasticity and the cancer stem cell hypothesis

Lineage plasticity may also have bearing on the cancer stem cell hypothesis (29), which posits that a stem cell-like state may give rise to cells with diverse states or phenotypes in the context of tumor initiation or therapy resistance. In the setting of chemoresistance, several studies have now described rare cancer persister cells and one putative stem-like state that are either dormant or have unique cycling properties in the setting of therapy selective pressure(23,30–32). The ability of such persister cells to survive, proliferate and give rise to diverse cell states raises the possibility that plasticity is an inherent property of stem-like cancer cells. Of note, this hypothesis implies that treatment resistance emerges from a ‘vertical’ (i.e. progenitor-to-progeny) differentiation path from stem-like cells. An alternative hypothesis is that plasticity is an inherent property of most or all tumor cells, and that adoption of a stem-like state is not needed for cell survival in the face of therapy. In this setting tumor cells may adopt diverse treatment-resistant states through ‘horizontal’ differentiation paths (i.e. “trans-differentiation”) that do not first require adoption of stem-like phenotypes. The latter phenomenon has been studied extensively in vitro(33,34) and new lineage tracing techniques are now enabling cellular trajectories to be studied in vivo(20,22). More data needs to be generated to better understand the contribution of vertical and horizontal differentiation paths in tumor initiation and treatment resistance. It is likely that the relative contribution of each may depend on the particular tumor type and treatment context.

A key (unanswered) question is whether cells within a given tumor have different plasticity properties. A quantitative framework for this would involve measuring the transition rates between cells harboring distinct cell states. Clinical observations suggest that a cell’s genetic background, or treatment-related selective pressure, can alter the rate of cell state transitions(11,35,36). As experimental methodologies for single cell analysis and lineage tracing continue to evolve, researchers will achieve a more precise understanding of the heterogeneous plasticity phenotypes within individual tumors. Such insights, together with a more detailed molecular understanding, are likely to reveal viable therapeutic targets.

Molecular drivers of plasticity

Decades of effort have elucidated some of the key transcriptional, epigenetic, metabolomic and microenvironmental features of plasticity, which have been discussed in several recent reviews(7,8,37–39). Our understanding of how these features interact has evolved with the application of emerging single-cell approaches, functional tools, and in vitro and in vivo models of plasticity. These new platforms have revealed several tumor cell intrinsic and extrinsic features that are either shared or unique to various tumor types or model systems(15).

Each cell is born with a particular genetic and epigenetic makeup. As a result, change in cell state – the result of coordinated activity of various transcription factors (TFs) and epigenetic regulators – are constrained by this inherited genomic landscape (Figure 1). Within this context, it is reasonable to think of cell state transitions as reactions whose rates are governed by the activity of such transcriptional regulators, many of which are lineage-biasing TFs. Two well-studied examples are epithelial-to-mesenchymal plasticity (EMP) and neuroendocrine transformation. In EMP (discussed in greater detail below), several TFs (e.g. SNAIL, TWIST and ZEB proteins), repress epithelial gene expression and activate mesenchymal gene expression(2,3). In neuroendocrine transformation, a different set of TFs (e.g. SOX2, ASCL1 and FOXA2 proteins), govern cellular transitions to a neuroendocrine state by mechanisms only partially elucidated(40–42). But as noted above, the epigenetic landscape of a cell is rate limiting, as the activity of a given TF depends on the permissiveness of chromatin state. In the case of EMP, for example, cell state is heavily influenced by the activity of the epigenetic regulator EZH2, which represses the expression of the epithelial gene CDH1 through the deposition of an H3K27me3 repressive chromatin mark(37).

Figure 1.

Molecular features that influence plasticity. The genetic architecture of a tumor cell may constrain the spectrum of cell states a cell can adopt (blue outline). Within this spectrum, cells may exist in dense neighborhoods defined in a multidimensional feature space (cell states). Tumor cell intrinsic (e.g. transcription factors, epigenetic state, among others) and extrinsic factors (e.g. growth factors, immune response, metabolomic environment, mechanical forces, among others) influence the rate of transition between cell states. The rates of transition between two states are one possible metric for the level of plasticity exhibited by a cell. Both intrinsic and extrinsic factors may influence each other, and both may also be influenced by other tumor extrinsic perturbations such as treatment.

Cancer cell-extrinsic features present in the tumor microenvironment, such as the metabolic milieu, inflammatory or growth factor signaling, and cell-cell communication may also influence plasticity phenotypes, as described in several recent reviews (38,43–49). Again using EMP as an example, growth factor signaling via TGFb (among others) causes cells to shed their epithelial features and gain mesenchymal features (EMT) through the activation of EMT-promoting transcriptional and epigenetic regulators(18,38). Deeper characterization of model systems, and in vivo models alongside analysis of patient samples is further elucidating the role of extrinsic features in influencing plasticity phenotypes(15,20). An emerging field is the interplay of the immune system and tumor cell states, with recent studies suggesting a role of interferon signaling in influencing plasticity phenotypes in tumor cells(11,46,47,49). Conversely, plasticity may also drive immune evasion, with several studies suggesting epigenetic modification of cancer cells can alter the tumor immune microenvironment and/or alter the susceptibility of cancer cells to immune recognition and killing(45,48,50–53).

The clinical relevance of plasticity

Despite the limitations of measuring plasticity in patients, there is abundant evidence that cellular plasticity contributes to cancer progression and treatment resistance. Below, we highlight several examples, noting that the list is by no means comprehensive.

Plasticity and the development of cancer.

Cellular plasticity plays a role at the very earliest stages of carcinogenesis, where changes in cell identity in normal tissues – a phenomenon known as metaplasia – can, under some circumstances, predispose to cancer. Metaplasia is a widespread response to tissue injury in organs ranging from the esophagus to the liver to the cervix (54). In the esophagus, for example, the squamous epithelium that normally lines the tube is converted to a cuboidal epithelium reminiscent of the intestine (“Barrett’s metaplasia”), which is thought to protect the tissue from the damaging effects of chronic acid exposure. However, the presence of Barrett’s increases the risk of developing esophageal cancer by 50- to 100-fold (55). A similar form of metaplasia occurs in the stomach, where the replacement of the normal foveolar and oxyntic epithelium with an intestine-like epithelium (spasmolytic polypeptide-expressing metaplasia, or “SPEM”) predisposes to gastric cancer (56). The precise mechanism by which metaplasia primes cells for malignant transformation is unknown, but it is thought that the epigenetic remodeling which accompanies altered cell identity creates permissive conditions for oncogene activity. Evidence for this model has recently emerged in the setting of pancreatic cancer, where injury-induced “acinar-to-ductal” metaplasia leads to an epigenetic state that facilitates the oncogenic functions of mutant Kras (57). Plasticity is thus a (non-genetic) mechanism by which environmental factors influence the risk of some cancers.

Epithelial to mesenchymal plasticity.

Amongst the most well described and pervasive forms of plasticity is epithelial to mesenchymal plasticity (EMP), which includes epithelial to mesenchymal transition (EMT) and the reverse process known as mesenchymal to epithelial transition (MET). For years, there was doubt as to whether EMP occurs in cancer; however, there is now abundant evidence of EMP in human patients across many tumor types including breast, lung, pancreatic, colorectal and head and neck cancers among others (3,15,18,58–65). Mesenchymal cell states, in particular, have been implicated in chemotherapy resistance (15,64,66,67) and there is growing evidence of the role of EMP in resistance to targeted therapies and immunotherapy as well (50,68,69). A number of ongoing efforts are trying to reach a consensus on the definition of EMP and the best ways to measure it (4). Of note, several efforts have now described tumor cells as lying on a spectrum of epithelial to mesenchymal states, including the identification of hybrid states that share epithelial and mesenchymal features(14,15,22,38,70,71). Cells that share features of epithelial and mesenchymal states have been implicated as having increased metastatic potential and stem-like features(72,73). While these cells share features of epithelial and mesenchymal states to different extents, it remains to better characterize how cells on this continuum may underlie different plasticity phenotypes and transition rates towards more extreme epithelial and mesenchymal phenotypes.

The molecular mechanisms underlying EMP have been studied extensively and are an area of intense active investigation. EMT may be induced by several pleiotropic factors, including TGFb, EGF, FGF, HGF, NOTCH and Wnt ligands(38). Signaling through receptors of these factors leads to downstream expression of several transcription factor families implicated in EMT, including ZEB, SNAIL, TWIST, YAP/TAZ and PRRX(2,3). Several other processes including post-transcriptional regulation by microRNAs, alternative splicing, epigenetic factors, and post-translational modifications of key EMT factors have been implicated in the regulation of EMP (2). Despite the elucidation of these molecular mechanisms underlying EMP, no effective therapeutic targeting of this form of plasticity has yet made its way to the clinic. Future efforts focused around identifying novel, targetable pathways in EMP and understanding the pharmacodynamics of EMP modulation will be essential for clinical translation of targeting EMP.

Dedifferentiation.

A canonical example of dedifferentiation occurs in melanoma cells treated with BRAF and MEK inhibitor combinations, and even immunotherapy (50,74–77). With such combinations, tumor cells downregulate MITF, a key regulator of melanocytic proteins, and upregulated AXL and neural crest markers, including neural growth factor receptor (NGFR), leading to cells that resembled a dedifferentiated neural-crest state (33,75,78). The same phenomenon of dedifferentiation in melanoma cells has also been observed as a mechanism of immune evasion and may be reversibly induced by treatment with TNFa and IFNg in vitro (50,74).

Trans-differentiation.

Trans-differentiation, also known as “histological transformation,” is now recognized as a common feature of non-genetic resistance across many tumor types. The most prevalent manifestation involves a transformation from adenocarcinoma to neuroendocrine histology, which has been observed in both non-small cell lung cancer (NSCLC) (79,80) and prostate cancer (11). In NSCLC, where driver mutations in the EGF receptor (EGFR) are common, 14% of patients exhibit small cell lung cancer transformation in the setting of EGFR directed therapies (81). A similar phenomenon occurs in prostate cancer in the setting of androgen receptor directed therapy, where adenocarcinoma to neuroendocrine transformation occurs in association with RB1 and TP53 loss (42,82). In both scenarios, tumors are highly resistant to therapy as a result of these histological makeovers. The mechanisms for these transformations are poorly understood; in some cases, plasticity is associated with various molecular drivers (e.g. RB1, TP53, SOX2) whereas in other cases the mechanism remains to be determined. More recently, adenocarcinoma to squamous transdifferentiation has been observed in a subset of lung cancer patients treated with KRAS G12C inhibitors (83). In this subset of patients for whom there were paired biopsies, no acquired mutations were observed in either KRAS or in genes implicated in bypass pathways, suggesting that histological transformation is the likely mechanism of resistance. Several modes of plasticity have also been observed in primary brain cancers, including in neuroblastoma where reprogramming between mesenchymal and adrenergic states has been linked to therapy resistance and immune escape (48,84). In this setting, inflammatory signaling may facilitate mesenchymal transition(48); the importance of inflammatory signaling, particularly the interferon pathway, has been observed in adenocarcinoma to neuroendocrine transformation as well (11).

Drug-tolerant states.

Another potential mode of plasticity has emerged from our understanding of drug-tolerant states (85,86). Recent efforts suggest that drug tolerant states may emerge from transcriptional variation within a population (33). Recent studies using novel methods for lineage tracing have suggested the presence of cycling persister cells in lung cancer in the setting of targeted therapy (23,87) and cells in a diapause-like state in colorectal cancer in the setting of chemotherapy (32); however, these experiments were not designed to study if cells may reversibly adopt drug-tolerant states. In glioblastoma, there is evidence in fact that cells may reversibly switch to slow cycling persister states in the setting of treatment, and this is influenced by chromatin remodeling (88). Further investigation is needed in this area and, in our belief, the behavior of tumor cells to adopt drug-tolerant states may vary by tumor type and depend on adaptation to the spatial and metabolic environment in which they exist (89).

Therapeutic hypotheses for targeting plasticity

The therapeutic approach to plasticity will depend on the underlying mechanism of plasticity that leads to cancer progression or therapy resistance. One may categorize these into three therapeutic hypotheses, outlined below (Figure 2): (i) preventing dedifferentiation, transdifferentiation or the formation of other drug tolerant states, (ii) targeting unique vulnerabilities in drug tolerant states, and (iii) promoting differentiation in the setting of blocked differentiation.

Figure 2.

Therapeutic hypotheses for addressing plasticity for patient benefit. To overcome plasticity and treatment-resistant states, we must identify combinations of current therapies with novel drugs that either: 1) prevent emergence of treatment-resistant states; 2) promote plasticity towards treatment-sensitive states; or 3) address dependencies or synthetic lethalities in treatment-resistant states. Such novel drugs need to be assessed for their pharmacodynamic activity in modulating cell states, and if effective, they have the potential in combination with current therapies to increase tumor cell killing, address tumor cell-intrinsic heterogeneity and prevent the emergence of treatment resistance to improve progression free survival, and possibly overall survival, of patients.

Preventing plasticity.

Several approaches have been proposed to prevent plasticity that leads to treatment-resistant states. In the context of EMP – where across several tumors mesenchymal states have been implicated in chemotherapy resistance, invasiveness and metastasis – one proposed strategy is to prevent EMT and maintain cells in treatment-sensitive epithelial states. TGFb has been identified as a key inducer of EMT, often from tumor cell-intrinsic and microenvironmental sources, and attempts have been made in the clinic in recent years to inhibit TGFb signaling using both biologics and small molecule inhibitors (90). This strategy has been employed in pancreatic cancer in the neoadjuvant setting using Losartan, an angiotensin receptor blocker (ARB) with weak TGFb inhibiting activity, in combination with chemoradiation for which clinical readouts are still pending (NCT03563248). Galunisertib, a small molecule inhibitor of the TGFb type 1 receptor, has been tested in combination with chemoradiation in rectal cancer recently (91) with promising results demonstrating a 32% complete response (CR) rate compared to historical control rates of around 10%. Several ongoing studies are testing biologic agents known as TGFb traps, which prevent TGFb binding to its receptors, in early-stage trials (90).

In prostate cancer, JAK/STAT and fibroblast growth factor receptor (FGFR) signaling is thought to be important for the neuroendocrine histological transition that confers resistance to AR-directed therapy (11,92). Knowledge of state-specific resistance can be leveraged to risk stratify patients for therapy and inhibition of JAK/STAT signaling may have potential to prevent plasticity or even reverse the emergence of resistant phenotypes in this setting (11). In melanoma, strategies to prevent the emergence of dedifferentiated neural crest-like states that are resistant to BRAF/MEK inhibitor therapy have been proposed, including inhibition of NGFR (CD271) or RXR signaling, or targeting cells that express high levels of AXL (76,93). These latter approaches have yet to be tested clinically.

One potentially intriguing emerging approach to modulating plasticity is targeting epigenetic regulators of plasticity phenotypes. An example of this may be found in recent studies that have demonstrated reversal of therapy resistant neuroendocrine phenotypes seen in prostate cancer using EZH2 inhibitors, which has led to resensitization of androgen-receptor (AR) directed therapies(82,94). LSD1 inhibitors have similarly shown neuroendocrine reversal in both prostate and small cell lung cancer (SCLC)(95,96). Pharmacologically, one may consider combinations with EZH2 or LSD1 inhibitors to potentially prevent plasticity towards treatment resistant neuroendocrine states. Several inhibitors for each of these targets are currently in clinical development, and novel epigenetic targets continue to emerge. A key question moving forward relates to the therapeutic window that these targets afford and the pharmacodynamics of cell state modulation that may confer clinical benefit.

Identifying vulnerabilities conferred by plasticity.

In cases where the emergence of treatment-resistant states has occurred as a result of plasticity, reversing the phenotype of all cancer cells to a treatment-sensitive state may be difficult to achieve pharmacologically. In this situation, an optimal strategy may be to target newly-arising vulnerabilities associated with such treatment-resistant states. Several efforts including the Cancer Dependency Map (97) have yielded tumor-specific and cell state-specific dependencies. In the context of EMP, early studies had suggested that gemcitabine may selectively kill mesenchymal cells (67), however these findings appear model-dependent. Recent efforts have demonstrated dependencies on ferroptosis and ER stress pathways as dependencies in mesenchymal cells (98,99), and these findings have led to ongoing drug-development efforts. More generally, the identification of synthetic lethal dependencies can open new avenues for targeting cell state-dependent lethalities (100), and future synthetic lethal screens anchored plasticity may afford opportunities for novel targets.

Differentiation therapies.

Although not a formal “hallmark,” the absence of specialized characteristics consistent with a loss of cellular differentiation is a near-universal feature of cancer. Across cancer, poor tumor differentiation portends a poor clinical outcome. But in some cases, the block to cellular differentiation presents a therapeutic opportunity. The best example of this paradigm is acute promyelocytic leukemia (APL), which is characterized by a balanced chromosomal translocation between chromosome 15 and 17 which fuses the PML locus with the RARa locus in myeloid progenitor cells. While this locks cells in a proliferative progenitor state, treatment with all-trans retinoic acid (ATRA) and arsenic trioxide (ATO) can be curative(101). The mechanism for pro-differentiation is partially understood; ATRA binds RARa and promotes degradation of the PML-RARa fusion protein while ATO blocks the function of PML-RARa as a transcriptional repressor. Together both drugs promote differentiation of promyelocytes into mature myeloid cells. Other examples of blocked differentiation may be seen in aberrant expression of SOX10 in melanoma, which prevents differentiation of progenitor cells to melanocytes(102). In some sense, promoting differentiation may be thought of as enabling plasticity.

Concluding remarks

Cellular plasticity is an important feature of cancer initiation, progression, and therapy resistance across many tumor types. To translate our understanding of plasticity from models to patients, we must employ the right tools to measure, quantify, and dissect the mechanisms by which cells shift their identities or states in clinically relevant settings. Such information may allow us to anticipate and counteract the dangerous consequences of plasticity with new interventions that act in a cell state-specific manner.