Impact of lineage plasticity to and from a neuroendocrine phenotype on disease progression and response in prostate and lung cancers

Summary:

Intratumoral heterogeneity can occur via phenotype transitions, often after chronic exposure to targeted anticancer agents. This process, termed lineage plasticity, is associated with acquired independence to an initial oncogenic driver, resulting in treatment failure. In non small cell lung (NSCLC) and prostate cancers, lineage plasticity manifests when the adenocarcinoma phenotype transforms into neuroendocrine (NE) disease. The exact molecular mechanisms involved in this NE transdifferentiation remain elusive. In small cell lung cancer (SCLC), plasticity from NE to nonNE phenotypes is driven by NOTCH signaling. Herein we review current understanding of NE lineage plasticity dynamics, exemplified by prostate, NSCLC and SCLC.

Overview.

Studies of cancer progression and therapy resistance largely focus on “driver” genetic mutations that perturb growth circuits. However, cancers can also become aggressive and treatment-resistant through radical shifts in cell identity, yielding relapses that may bear little transcriptional or morphological resemblance to early-stage pre-treatment tumors. This phenomenon, known as lineage plasticity, is a less well understood mechanism of tumor progression and therapy resistance. In this perspective, we address lineage plasticity in two exemplar epithelial cancers, prostate (PCa) and lung adenocarcinoma (LUAD), where a therapy indifferent disease state is observed in ~10% of cases following targeted therapy (anti-androgens and tyrosine kinase inhibition respectively) that manifests as transdifferentiation to a small cell, neuroendocrine (NE) phenotype. Conversely, in small cell lung cancer (SCLC) a NE tumor, NOTCH signaling directs a reverse transition to a more chemoresistant non-neuroendocrine (non-NE) phenotype in a cell subpopulation that provides paracrine support for NE cell metastasis. We discuss the role of specific genetic (TP53, RB1, and PI3K pathway), transcriptional (ASCL1, NEUROD1, POU2F3, YAP1), epigenetic (perturbations of EZH2, SETD2, LSD1, and the SWI/SNF complex) alterations and their generalizability beyond specific contexts (e.g. SCLC) to a broader molecular definition of the NE phenotype in lung and prostate cancer. We propose several important research directions to systematically characterize the role of noncoding alterations, chromatin remodeling, and microenvironmental factors in creating and maintaining the NE state. Without an understanding of the consequences of increased and earlier use of potent targeted therapies, the frequency of “plastic” and untreatable small cell cancers will continue to rise. We argue that an improved understanding of the dynamics of lineage plasticity around the NE cellular identity may be necessary to devise better therapeutic approaches for both de novo and targeted therapy-associated small cell lung and prostate cancers.

In this perspective, we propose a working concept for lineage plasticity in three solid tumors, exploring the interplay between complex genomic structural variations, epigenetic events, and alterations in the tumor microenvironment on epithelial, NE and non-NE cell phenotype transitions.

1. Hidden Lineage Plasticity Revealed by Precision Cancer Therapy

The classic model of metazoan tissue organization is built along an axis of differentiation, from pluripotent progenitors that differentiate to more stable cell lineages along a well-defined tree topology. A more modern understanding (developed over the past three decades) acknowledges deviations from this hierarchy in both physiologic conditions (e.g., stress, wound healing, inflammation) and disease (e.g. cancer) (see comprehensive review on this topic by Le Magnen et al. (2018)). Such lineage plasticity makes it particularly challenging to link cancer histologies (e.g. adenocarcinoma, small cell cancer (SCC)) to their cell-of-origin simply on the basis of their cellular morphology or transcriptional state. Addressing this cell-of-origin question has thus required the development of lineage tracing systems in genetically engineered mouse models (GEMMs), where combinations of oncogenic driver alterations can be activated in specific cell types. While these reverse genetics approaches have validated certain long-held intuitions, (e.g. that lung adenocarcinomas arise from type 2 alveolar cells, small cell lung cancers arise from neuroendocrine cells) they have also identified alternate cell identities (e.g. basal cells, club cells) that readily give rise to adenocarcinomas and SCCs, given the right driver combinations (Ferone et al., 2020). In the prostate, while both multipotent basal cells and terminally differentiated luminal cells can give rise to adenocarcinomas following PTEN deletion, basal cells must transsit through a luminal cell state prior to transformation (Blanpain, 2013). Lineage plasticity is also thought to affect later cancer stages, where it drives invasion and metastasis through epithelial-to-mesenchymal transition (EMT). EMT is a transcriptional program activated in normal wound healing, where epithelial cells take on features of fibroblasts and multipotent progenitors (see review on EMT by Kalluri and Weinberg (2009)).

Over the past ten years, a surprising example of lineage plasticity has emerged in lung adenocarcinoma (LUAD) and prostate (PCa) adenocarcinoma, where tyrosine kinase inhibitors (TKI) and Androgen Receptor signaling inhibitors (ARSi), respectively, have transformed the care of patients with advanced disease (Konieczkowski et al., 2018). However, most LUAD and PCa patients treated with these agents acquire resistance within 1–2 years, usually through “on target” genetic lesions that overcome (EGFR T790M) or bypass (MET amplification) targeted blockade, restoring the function of these cell-identity specific growth circuits (Konieczkowski et al., 2018). Remarkably, a small fraction (5–20%) of drug-resistant cases recur with a neuroendocrine (NE) phenotype (Boumahdi and de Sauvage, 2020; Quintanal-Villalonga et al., 2020; Rickman et al., 2017). The resulting cancers acquire the histomorphology of small cell lung cancer (SCLC), an aggressive and highly metastatic cancer thought to arise from NE cells. Like SCLC, these small cell prostate cancers (SCPCa) and LUAD-SCLC transitions rapidly progress, and a subset (LUAD-SCLC) are also strikingly sensitive to SCLC cytotoxic drugs (platinum, etoposide). SCLC, SCPCa, and LUAD-SCLC also demonstrate striking transcriptional similarity relative to adenocarcinomas from their tissue of origin (Balanis et al., 2019).

In this Perspective, we argue that these three SCC types represent a convergent cell state, resulting from an inherent lineage plasticity in epithelial tissues towards the NE phenotype. This plasticity allows adenocarcinoma cells to escape targeted inhibition (TKI’s in LUAD, ARSi in PCa) of cell identity-specific growth circuits by adopting a “therapy indifferent” phenotype (Rickman et al., 2017). The joint study of shared and divergent mechanisms among these distinct tumor types may provide opportunity for transferring biological and clinically translatable insights. This includes key observations regarding specific transcriptional and epigenetic mechanisms, some of which have been deeply characterized in one tumor type (e.g. SCPCa) but not evaluated in another (e.g. LUAD-SCLC). Can some of the approaches proposed to reverse PCa and LUAD lineage plasticity be used to treat SCLC? Do recent insights into the origins, genomic features, and drug resistance patterns of SCLC have relevance for prevention and treatment of drug resistant SCPCa and LUAD-SCLC transitions? Do a subset of de novo SCLCs arise through an epithelial-NE transition prior to diagnosis? As we discuss, many of these questions can be addressed through synthesis of existing research directions. Looking into the future, we also discuss the prospects of SWI/SNF, structural genomic, and microenvironmental perturbations as potential novel classes of lineage plasticity drivers.

As more potent third and fourth generation inhibitors are developed to overcome on-target resistance, we foresee that the prevalence of SCC will rise precipitously. Though LUAD and PCa are the current exemplar cancers, the tendency of epithelial tissues towards NE lineage plasticity may be much more widespread and even occur outside of the targeted therapy context. Recent application of a NE gene expression signature to 7000 (predominantly pre-treatment) primary tumors from the TCGA, uncovered a subset of high-grade epithelial tumors with poor survival across many epithelial cancers, including bladder, endometrial, and gastric cancer (Balanis et al., 2019). These signatures may represent a latent NE lineage plasticity across many additional tissue contexts, which may result in many additional cases of SCC as the targeted therapy paradigm is extended more broadly and applied earlier in the disease course.

2. Prostate Cancer and Small Cell Lineage Plasticity (Figure 1)

Figure 1. Lineage plasticity as a mechanism of disease progression in Castration Resistant Prostate Cancer (CRPC).

Schematic representation of the current state of knowledge. In this perspective, we explore how around 10% of CRPC transition to a lethal form of prostate cancer by becoming indifferent to ARSi. We posit that understanding the mechanism of how this occurs could help prevent this deadly turn to the most aggressive form of prostate cancer. Modified from Cyrta et al. (in press).

Emerging evidence suggests that lineage plasticity plays an important role in the progression of advanced PCa during the course of treatment with ARSi, like enzalutamide or abiraterone acetate (Abida et al., 2019; Aggarwal et al., 2018; Beltran et al., 2016). The mainstay of treatment for metastatic PCa is androgen deprivation therapy until progression occurs and then ARSi are included. Although this potent AR blockade is initially effective, this therapy ultimately fails and progression to castration-resistant PCa (CRPC) occurs. One form of lineage plasticity observed is characterized by androgen receptor (AR) indifference and progression to SCC, which shows a distinct histomorphology and expresses neural-like markers (Rickman et al., 2017). Unlike more commonly recognized mechanisms of ARSi resistance due to AR mutations or amplification, SCPCa no longer responds to AR-targeting therapy and has a mean survival of 12 months (Metzger et al., 2019). Together, there is mounting evidence supporting the role of epigenetic events as a mechanism for transdifferentiation of PCa to an AR-indifferent state under specific genomic conditions, involving, but not limited, to TP53, RB1, and PTEN loss (Park et al., 2018) (Table 1). Based on a recent molecular pathology review of 430 CRPC (Abida et al., 2019), the incidence of SCPCa is 11%, which is consistent with a second published report that found SCPCa cases in 17% of patients with advanced PCa after AR targeted therapy (Aggarwal et al., 2018). In contrast, the incidence of SCC in treatment-naïve PCa patients is extremely rare (~0.1%) (E.J. Small, 2014). What is currently unknown is whether aggressive localized PCa tumors that contain TP53 or PTEN loss have an innate programming to acquire a small cell phenotype or whether SCPCa evolves following sequential treatments and modification of the tumor environment as an iatrogenic-driven pathology (Mateo et al., 2020; Zou et al., 2017). These concepts are not necessarily mutually exclusive.

Table 1.

Similarities in key molecular events characteristic of SCLC, LUAD-SCLC and SCPCa

Key molecular events	SCLC	LUAD-SCLC	SCPCa
TP53 loss	universal	universal	universal
RB1 loss	frequent	universal	frequent
MYC family gene upregulation	frequent	unknown	frequent
Altered TF expression (ASCL1, NEUROD1, POU2F3)	variable	unknown	frequent
PIK3CA mutation or PTEN loss	frequent	frequent	frequent
EZH2 upregulation	frequent	unknown	frequent

3. Non Small Cell Lung Cancer (NSCLC) lineage plasticity

Lung cancer spans three major histological subtypes – lung adenocarcinoma (LUAD), lung squamous cell cancer (LUSC), and SCC of the lung, in order of prevalence (Travis et al., 2015). Classically, NSCLCs (LUAD, LUSC) behave more indolently and are treated surgically at early stages, while lung SCCs are diagnosed as high-grade metastases and treated with systemic chemotherapy. Canonically, LUAD, LUSC, and SCLC histologies are linked to specific anatomical and cellular origins in the lung: LUAD arising distally from type 2 alveolar (AT2) cells, LUSC from basal cells in the proximal airway, and SCLC from pulmonary NE cells (Blanpain, 2013). However, accumulated evidence from autochthonous GEMMs suggests considerable infidelity in these canonical mappings (reviewed in Blanpain (2013)). For example, KRAS G12V mutations in club cells or TP53 / Rb1 mutation in basal cells can give rise to LUAD or SCLC histology, respectively. LUSC can be induced through the overexpression of SOX2 across a multitude of cellular contexts (AT2, club cells, goblet cells). These results indicate that cellular plasticity may drive the natural life history of lung cancer even prior to the onset of treatment. Much of this plasticity may be inherited from the normal lung epithelium, where the network of cell identities and state transitions and their impact on lung cancer development is being actively revised using single cell RNA-sequencing (scRNA) (Lambrechts et al., 2018; Laughney et al., 2020; Marjanovic et al., 2020; Maynard et al., 2020; Vieira Braga et al., 2019).

In 2004, the discovery of oncogenic EGFR lesions predicting erlotinib response in LUAD ushered in a new era of genomics-driven targeted therapies across cancer (Lynch et al., 2004; Paez et al., 2004; Pao et al., 2004). LUAD has remained at the vanguard of this paradigm, with the subsequent discovery of oncogenic and drug-sensitizing lesions in other Receptor Tyrosine Kinase (RTK) / Ras / Raf pathway members including ALK, ROS, RET, and NTRK family proteins (Brown et al., 2018). These render 30–50% of LUAD patients as eligible for targeted therapy with a tyrosine kinase inhibitor (TKI). However, as in other cancer types, the inevitable consequence of TKI therapy in LUAD is resistance, which usually occurs within 18 months of treatment (Camidge et al., 2014). 5–15% of EGFR mutant LUAD patients with TKI resistance demonstrate small cell histology in their relapse tumors (Oser et al., 2015). As with SCLC, these tumors demonstrate TP53 mutations, loss of RB1 expression, rapid progression, and remarkable sensitivity to SCLC chemotherapy.

Interestingly, loss of function RB1 and TP53 mutations may be clonally present in the pre-treatment tumor and may provide a useful biomarker to assess for relapse risk (Quintanal-Villalonga et al., 2020). These phenotypically distinct SCC relapses invariably harbor the identical TKI-sensitizing EGFR lesion, demonstrating that they are descendants of the original adenocarcinoma rather than an independent primary tumor. Remarkably, certain cases demonstrate a complete response to SCC chemotherapy (etoposide, cisplatin), followed by a return of the original EGFR-mutant adenocarcinoma, which regresses after TKI inhibition and is followed by a second SCC relapse (Niederst et al., 2015; Sequist et al., 2011). In cases where SCC and adenocarcinoma relapses coexist, the adenocarcinoma-associated EGFR resistance alleles (e.g. EGFR T790M) are not found in the SCC relapse. These observations and others (George et al., 2015; Lee et al., 2017; Niederst et al., 2015) indicate that conversion to SCC provides a prevalent alternative resistance mechanism to alterations that directly evade or overcome TKI blockade.

4. Small Cell Lung Cancer (SCLC) Lineage Plasticity

SCLC is the prototype NE cancer and is undergoing a second ‘golden age’ of research where discoveries about SCLC biology made in elegant studies of GEMMs are being validated in increasingly developed and highly heterogeneous patient derived models (Gazdar et al., 2017). The emerging SCLC ‘blueprint’ of de novo NE cancer biology may serve to inform studies of those less well characterized NE tumors that arise via lineage plasticity in NSCLC and PCa (described above). Whereas NSCLC and PCa epithelial tumor cells transit to aggressive small cell NE cancers, in SCLC, plasticity is observed from a NE to non-NE phenotype.

The aggressive behavior of SCLC lies both within its capacity for early widespread metastases and rapid acquisition to cisplatin-based chemotherapy resistance. Up to a fifth of all lung cancers are SCLCs, which are the 6^th major cause of cancer related deaths (~250,000 deaths worldwide) with a sobering 5-year survival of only ~5% (Bunn et al., 2016). SCLC is strongly associated with tobacco smoking, harboring a high tumor mutation burden (TMB) and prevalent circulating tumor cells (CTCs) even in early stage disease (Tay et al., 2019). Until recently, the Standard of Care (SoC) for SCLC, a platinum agent plus etoposide, remained largely unchanged for 30 years. Acquired chemoresistance commonly occurs within 1-year whereafter patients live on average only 30 weeks (Bunn et al., 2016). The high TMB in SCLC raised expectations that immunotherapies would be effective. Immune checkpoint blockade (ICB) was approved as first-line SoC in the USA after studies of CTLA4 and PD1 ICB in patients with metastatic SCLC with progression after platinum-based chemotherapy demonstrated modest 3-month overall survival benefit. In June 2020, the FDA approved the polymerase II inhibitor Lurbinectedin for second line treatment that extends Overall Survival (OS) from 5.0 to 11.9 months (Trigo et al., 2020). Although a number of targeted therapies (e.g., to DNA damage repair, oncogene-induced replicative stress, and cell cycle control) are entering early clinical trials in SCLC, none so far are approved for routine use and prospective patient stratification has not yet been implemented (Bunn et al., 2016). Recent reports suggest increased efficacy of ICB in additional combinations including gemcitabine, temozolomide, CHK1 inhibitors and DNA damaging agents (Drapkin and Farago, 2019; Sen et al., 2019).

The genomic landscape of SCLC is characterized by almost universal loss of TP53 and frequent loss of RB1 similar to what is observed in SCPCa. MYC family genes (MYC, MYCL, and MYCN) are recurrently amplified alongside mutations in NOTCH family genes), also similar to SCPCa (Beltran et al., 2011; Lee et al., 2016; Rubin and Demichelis, 2019). PTEN deficiencies are reported in 7.4% of SCLC patients (Sundaresan et al., 2017) and loss of PTEN in the context of RB1 and P53 loss accelerates SCLC progression and metastasis in a mouse model (Cui et al., 2014). These findings are also highly consistent with mouse models of advanced PCa (Rickman et al., 2017).

Although SCLC is currently treated as a homogeneous disease, four SCLC subtypes were recently defined based on TF expression: master NE TFs Achaete-Scute Complex Homolog-Like 1 (ASCL1) and Neurogenic Differentiation Factor 1 (NEUROD1), the tuft cell TF POU class 2 homeobox 3 (POU2F3) and YAP1, a component of the HIPPO signaling pathway (Rudin et al., 2019). A fifth subtype based on predominant expression of another NE TF Atonal (ATOH-1) was also recently reported (Simpson et al., 2020). SCLC phenotypic diversity is evidenced by a spectrum of cell morphologies (classic to variant) and a NOTCH signaled NE to non-NE cell lineage transition (Schulze et al., 2019).

As more comprehensive molecular details of the biology and plasticity of SCLC emerges, there is optimism for future biomarker driven personalized treatments.

SCLC GEMMs have enabled detailed characterization of cell-of-origin (comprehensively reviewed in Ferone et al. (2020)), disease progression, and lineage plasticity (Quintanal-Villalonga et al., 2020). An impressive collection of studies modelling SCLC in GEMMs suggests that both driver lesions and cell-of-origin impact upon SCLC subtypes. It was long thought that NE lung epithelial cells were the only cell type that give rise to SCLC and this contention was refined by Oudah et al. (2019), who suggested a subset of NE cells, termed NE^STEM, were the primary cell-of-origin. However, GEMM studies have also shown that, with varying efficiency and particular genetic drivers, AT2 cells and tracheobronchial basal cells can also be SCLC cells-of-origin (Ferone et al., 2020). GEMM studies have recently shown that cell-of-origin impacts SCLC mechanisms that underlie metastatic progression and specifically, that a dependence on the TF NFIB is dispensable when tumor arise from mature NE cells (Yang et al., 2018a).

These recent insights into SCLC subtypes and cell-of-origin have critical relevance for our emerging understanding of SCPCa and LUAD-SCLC. Which of these SCLC subtypes to these treatment-associated SCCs most closely resemble? Conversely, do a significant portion of SCLC arise from epithelial rather than NE cells, mirroring the epithelial-neuroendocrine transition (though in the absence of treatment) that advanced PCa and LUAD tumors undergo. The increasingly robust corpus of molecular knowledge in SCLC provides an important axis on which the emerging entity of treatment-associated SCCs can be projected, and potentially better understood.

5. Emergent Directions in the Search for Lineage Plasticity Drivers

The search for molecular drivers of NE lineage plasticity have focused on transcriptional circuits that promote the small cell and/or abolish the adenocarcinoma phenotype, largely in the context of in vitro and animal models. A radically altered cellular phenotype, such that associated with therapy associated lineage plasticity, may be the result of combined perturbations of the epigenome, genome, and microenvironment. A systematic study of known and potential mechanisms driving entry to and exit from the NE phenotype across model systems and human samples across these various instantiations of SCC may be necessary to comprehensively identify lineage plasticity drivers, and leverage these insights for diagnosis and therapy. We outline several existing and future research directions below.

Master regulators of the adenocarcinoma to small cell transition.

A sophisticated understanding of transcriptional circuits driving the adenocarcinoma to SCC transition has emerged over the past several years through the study of advanced PCa. In a GEMM of PCa with combined Trp53 and Pten loss, lineage tracing provided evidence that NE tumor cells can directly arise from pre-existing luminal adenocarcinoma cells and do not emerge from a second, independent population of NE or intermediate cells (Zou et al., 2017). Patient-derived PCa xenografts that develop NE features following castration display genomic relatedness to pre-existing adenocarcinoma (Lin et al., 2014). Moreover, mouse models with Trp53 and Rb1 genomic loss show lineage plasticity but epigenetic therapy can re-sensitize those tumors towards ARSi treatment (Ku et al., 2017).

Overall, these data have converged to the notion that PCa to SCPCa lineage plasticity is driven by epigenetic changes that arise in a specific genomic context (Mu et al., 2017; Zehir et al., 2017). In patient cohorts, SCPCa are characterized by an overexpression of several epigenetic regulators (e.g., enhancer of zeste homolog 2 (EZH2)) and a specific DNA methylation profile (Beltran et al., 2016; Davies et al., 2018; Puca et al., 2018). Recently, Nyquist et al. (2020), using a large panel of CRPC tumor samples exhibiting different TP53 and RB1 status, found that biallelic loss of TP53 and RB1 does not necessarily promote outgrowth of androgen-independent PCa, such as SCPCa. Rather, only 46% of tumors manifested a small cell phenotype and 35% were classified as androgen receptor active PCa. These findings show that biallelic loss of TP53 and RB1 can co-occur with small cell-transdifferentiation and AR-independence but does not obligate either of these characteristics.

Insights into what additional epigenetic factors could provoke multipotent PCa cells to transdifferentiate into SCPCa are still emerging. Extensive studies have shown that DNA methylation plays a significant role in mediating mechanisms in PCa but also other common cancers (Zhao et al., 2020). Yet, most studies have focused on global effects and not specific mediators of transdifferentiation. The dysregulation of EZH2 is particularly intriguing and relevant to PCa progression independent of ARSi. EZH2 is the catalytic core subunit of polycomb repressive complex 2 (PRC2). The activity of EZH2 heavily relies on the functionality of its cystine-rich region and its SET domain. In cancer, EZH2 induces anchorage-independent colony formation and cell invasion and it has been linked to self-renewal in poorly-differentiated cancers and cancer stem cells (Kim and Roberts, 2016). Gain-of-function mutations in the EZH2 SET domain have been reported in a variety of cancers, including Non-Hodgkin lymphoma and melanoma, while in other cancers, such as PCa, EZH2 levels are elevated in CRPC without genomic alterations (Abida et al., 2019; Armenia et al., 2018; Varambally et al., 2002). As noted above, attempts to inhibit EZH2 activity in advanced CRPC can re-sensitize the tumor cells to ARSi, suggesting that EZH2 is suppressing lineage differentiation. However, the regulation of EZH2 has been elusive in PCa.

Recently, Yuan et al. (2020) demonstrated an inverse correlation with EZH2-mediated H3K27me3 and SETD2-mediated H3K36me3 levels. Loss-of-function mutations or depletion of SETD2 were previously reported to be enriched in SCPCa when compared to CRPC (Beltran et al., 2016). Yuan et al. assessed the tumor suppressive function of SETD2 in PCa and discovered EZH2 as one of its direct non-histone substrates. In the study, the authors demonstrated that SETD2 directly monomethylates EZH2 at its K735 lysine residue, which ultimately triggers a Smurf2 E3 ligase-dependent degradation, and were able to confirm their observations using GEMM. They propose a model for PCa progression, in which SET2D loss-of-function triggers EZH2 overexpression. The effect of SETD2 loss with EZH2 over expression is further enhanced by the loss of PTEN. Both results are consistent with decreased ARS and NE differentiation.

In PCa, it has also been shown that EZH2-mediated transcriptional programs can be induced by overexpression of neuronal transcription factor N-Myc (Dardenne et al., 2016). Interestingly, other epigenetic factors have been linked to influence N-Myc-induced oncogenic behavior in cancer. Shi et al. (2020) exhibited that loss of ARID1A, a component of the canonical BAF complexes within the mammalian SWI/SNF chromatin remodeling family, triggers increased invasive potential and adrenergic-to-mesenchymal transition in N-Myc-driven neuroblastoma (25% of all neuroblastomas). The authors discovered that ARID1A loss modifies enhancer-mediated gene expression in these neuroblastoma cells, by altering the binding capacity of BAF or PBAF complexes. Thus, ARID1A loss contributes to epigenetic reprogramming and, in concert with N-Myc amplification, potentiates the induction of phenotype switching in neuroblastoma with increased invasive potential. Albeit, N- Myc is not a direct target of ARID1A, the ARID1A gene is almost always deleted in neuroblastomas with MYCN gene amplification. Clinically, combination of those genotypic traits manifest as the high-risk specimen of neuroblastoma, emphasizing a synergistic role between N- Myc and ARID1A deletion in cancer progression (Shi et al., 2020). These findings support a mechanism in which altered SWI/SNF complexes can modify the epigenetic landscape favoring transdifferentiation, especially in cancer types that harbor neuronal lineage traits.

In line with these findings in neuroblastoma, recent work from the Rubin laboratory has identified the chromatin remodeling mSWI/SNF complexes as a potential regulator of SCPCa lineage plasticity (Cyrta et al., in press). Specialized assemblies of the SWI/SNF complex with distinct functions are observed at different stages of embryonic development and tissue maturation (Ho et al., 2009; Ho et al., 2011). The most notable changes in SWI/SNF composition described to date occur during neuronal differentiation. Cells committed to the neural lineage initially express a neural progenitor form of the complex (termed npBAF), which incorporates among others the BAF53A, BAF45A/D and SS18 subunits (Lessard et al., 2007; Wu et al., 2007; Yoo et al., 2009). However, upon differentiation to post-mitotic neurons, the complex undergoes a dramatic switch to the neural variant (nBAF) and incorporates the respective paralogs of these subunits (i.e., BAF53B, BAF45B/C and SS18L1). This switch is mediated by repression of BAF53A by micro-RNAs in response to a downregulation of REST (Yoo et al., 2009). We recently reported the presence of “neuronal” SWI/SNF subunits outside of the nervous system, characterized by the expression of BAF53B and BAF45B in SCPCa (Cyrta et al., in press). Although their expression appears to be specific to SCPCa, it remains unclear whether they play a role in activating neural-like gene programs or are simply expressed as a consequence of this process.

A pleotropic role for the SWI/SNF chromatin remodeling complex may depend on the genomic and/or the epigenetic context - a paradigm which has been gaining support both in regards to SWI/SNF and to other epigenetic regulators (Roy et al., 2015; Sun et al., 2017; Velcheti et al., 2019). Although the complex is often described as a tumor suppressor in multiple cancer types (Jones et al., 2010; Kadoch et al., 2013; Versteege et al., 1998; Witkowski et al., 2014), there is also increasing evidence for tumor-promoting functions of SWI/SNF in other malignancies, including leukemia, breast, liver and pancreas cancer, melanoma, glioblastoma, neuroblastoma and synovial sarcoma (Buscarlet et al., 2014; Clark et al., 1994; Hiramatsu et al., 2017; Jubierre et al., 2016; Kadoch and Crabtree, 2013; Laurette et al., 2020; Wu et al., 2016). Mutations in SWI/SNF genes are very rare in PCa (Abida et al., 2019; Armenia et al., 2018; Baca and Garraway, 2012; Barbieri et al., 2012; Beltran et al., 2016; Cancer Genome Atlas Research, 2015; Muthuswami et al., 2019), in contrast to several other cancer types (Kadoch et al., 2013; Shain and Pollack, 2013). From the functional perspective, inhibition of the SWI/SNF subunits BAF57 (SMARCE1) or BAF53A (ACTL6A) in PCa cells has shown to abrogate androgen-dependent cell proliferation (Jin et al., 2018; Link et al., 2008). Similarly, Sandoval et al. (2018) reported that SWI/SNF interacts with ERG in PCa cells harboring the TMPRSS2:ERG gene fusion and is required to activate specific gene programs and maintain cell growth.

The SWI/SNF composition in PCa is not a hard-set feature; instead, specialized forms of SWI/SNF complexes may assemble in cancer cells depending on their phenotype. One such candidate may be the embryonic stem cell form of the BAF complex (esBAF), which is characterized by the presence of SMARCA4, SMARCC1 and SMARCD1, and the absence of their somatic cell homologs: SMARCA2, SMARCC2 and SMARCD3 (Ho et al., 2009; Ho et al., 2011; Yan et al., 2008). Interestingly, overexpression of Smarcc1 and Smarca4, in addition to an ectopic expression of the “four factors” (Oct4, Sox2, Klf4 and c-Myc), was shown to increase the efficacy of reprogramming of mouse somatic cells into induced pluripotent stem cells (iPS) (Singhal et al., 2010). This result was also observed in the absence of c-Myc. In this setting, overexpression of Smarcc1 alone had a more potent effect than overexpression of Smarca4 alone, and combined overexpression of Smarcc1 and Smarca4 was shown to be synergistic rather than additive. Likewise, in another study, knock-down of their paralogs, i.e. Smarca2 and Smarcc2, was also shown to improve reprograming efficacy (Jiang et al., 2015). Such changes in SWI/SNF subunit levels may favor the assembly of esBAF-like complexes in lieu of somatic-type BAF complexes, which in turn enhances cell reprogramming. Therefore, the SWI/SNF complex may represent another an example of epigenetic regulation of tumor plasticity in advanced PCa.

Other modes for epigenetic regulation of lethal CRPC have been proposed. For example, Sehrawat et al. (2018) identified Lysine-specific demethylase 1 (LSD1), a histone demethylase to promote PCa cell survival functioning independently of its demethylase function and AR signaling. They proposed that LSD1 acts via LSD1’s binding protein, ZNF217. Moreover, LSD1 is commonly overexpressed in CRPC, while enzalutamide-resistant CRPC models showed increased sensitivity to LSD1 depletion, emphasizing its role as an important epigenetic regulator in PCa. To this end, they demonstrated that a small-molecule LSD1 inhibitor – SP-2509 – blocks important demethylase-independent functions and suppresses CRPC viability. Although they do not demonstrate lineage plasticity by siRNA-mediated knock-down of LSD1, they did identify the binding of LSD1 to CoREST, a mediator of neural cell identity (Qureshi et al., 2010), using rapid immunoprecipitation (IP) mass spectrometry of endogenous proteins (RIME) in ARSi resistant PCa. Taken together, LSD1 modulation may indirectly enable a neural identity.

The best integration of clinical and experimental data suggests that epigenetic regulation most probably occurs in a specific genomic background (Beltran et al., 2019; Quintanal-Villalonga et al., 2020). The genomic background or landscape enabling epigenetic regulation towards tumor plasticity has been predominately suggested to be TP53, RB1, and PTEN mutations. However, as shown above, in PCa genomic alterations are probably necessary but not sufficient and require additional epigenetic events to enable lineage plasticity leading to not only to a small cell phenotype but also to other states, including a double negative state (AR and SCPCa negative). There are intriguing similarities between alterations seen in SCPCa and in SCLC (Table 1). As in PCa, altered epigenetic regulation of SCLC gene expression is also common and notably gene expression of the epigenetic modifier EZH2 gene is stronger in SCLC compared to any other tumor type in the cancer genome atlas (Schulze et al., 2019). One of its functions in SCLC is to repress expression of SLFN11 supporting the emergence of acquired resistance to chemotherapy (Gardner et al., 2017). The relevance of these alterations in LUAD-SCLC have yet to be defined.

Drivers of (epi)genetic escape from the NE state.

SCLC has provided unique opportunities to study the NE cell identity, through both model systems (GEMMs) and patient tissue (CDX models). Recent studies have shed light into unexpected examples of transition from the NE state to more chemoresistant, non-NE identities. Such examples of “reverse” lineage plasticity in SCLC evolution may provide important lessons for the treatment of advanced PCa and LUAD, where the optimal algorithms for patient stratification and treatment are still evolving.

In a disease where tumor biopsy is challenging but CTCs are prevalent, a complementary approach to studying the biology of human SCLC, is the generation of CDX models in immunocompromised mice that recapitulate the donor patients’ tumor morphology, diagnostic small cell marker expression and chemotherapy responses (Hodgkinson et al., 2014; Lallo et al., 2019). CDX, implanted subcutaneously, undergo metastasis in mice, notably to those sites commonly seen clinically, including brain, liver and lungs. A bank of more than 40 CDX models encompassing all of the previously reported TF based subtypes, including 8 CDX pairs generated pre-treatment and at disease progression are illuminating the complex intra- and inter-tumoral heterogeneity that exists in human SCLC. CDX tumors can express more than one of the subtype TFs in adjacent cell subpopulations, the consequences of which are currently not completely clear (Simpson et al., 2020). Some CDX express more than one of the MYC family proteins in separate subpopulations, contrary to prior dogma that MYC family members were mutually exclusive (Simpson et al., 2020).

The Sage laboratory showed that endogenous NOTCH activity activates a NE to non-NE transition in the RPP GEMM (triple KO of Trp53, Rb1 and Rbl2, (Lim et al., 2017). CDX models contain both NE and non-NE cells (Simpson et al, 2020). Studies from the Berns laboratory in GEMM models remind us that NE and non-NE co-operate in the metastatic cascade (Calbo et al., 2011). CDX studies document evolution during disease progression exemplified by altered MYC and NOTCH gene expression and a switch ‘classic’ to ‘variant’ cell morphology reminiscent of the RPM GEMM of SCLC based on knock-out of RB1, TP53 with activation of c-MYC (Mollaoglu et al., 2017). Most recently, Ireland et al (2020) showed in the RPM GEMM that MYC activates Notch signaling to drive a trajectory of subtypes from ASCL-1 (the most NE) thru NEUROD1 (less NE) to YAP1 a non-NE subtype. The Dive group subsequently confirmed that within highly heterogeneous human CDX tumors, YAP1 is predominantly expressed in small clusters of non-NE cells where it co-localizes with REST, a known repressor of NE differentiation in SCLC (Pearsall et al., 2020). SCLC is a rapidly growing and highly hypoxic tumor, whilst hypoxia is implicated in tumor plasticity (e.g. in EMT), its impact on the NE to non-NE transition in SCLC is not yet clear, but certainly warrants further attention.

NE cells grow as floating aggregates in culture whilst non-NE cells (the minority subpopulation observed after harvest and disaggregation of CDX models) grow as loosely attached monolayers allowing easy separation and studies of distinct functional capabilities (Figure 2). When non-NE and NE cells from the same CDX tumors were separated and treated with cisplatin, the non-NE were 5–7.5-fold more resistant (Pearsall et al., 2020).

Figure 2. Studying plasticity in Small Cell Lung Cancer in CDX models.

Schematic showing how the NE and non-NE (NNE) cells within a SCLC CDX tumor, derived from a patient’s circulating tumor cells, can be disaggregated and placed into short term culture. The NE cells grow as floating cell spheres whilst the non-NE cells adhere to plastic (far right bright image). These differential growth behaviors allow easy separation and separate or combined study of the two phenotypes.

While SCLC NE to non-NE transition shares impacts with the EMT (metastasis and drug resistance), and here are some overlapping gene expression changes, there is to date no convincing evidence the NE to non-NE transition is reversible. SCLC NE cells are the more tumorigenic and Bern’s hypothesis of NE and non-NE cells co-operativity for metastasis is more plausible than reversible plasticity. The recent single cell analysis of CDX tumors and of patients’ CTCs showed how intratumoral heterogeneity increases during disease progression after therapy with multiple concurrent treatment resistance mechanisms (Stewart et al., 2020). Although the field of SCLC research is moving apace, precise relationships between cell-of-origin, driver mutations, tumor subtypes and lineage plasticity and treatment resistance, though potentially important for patient stratification and treatment, as yet, remain unclear. The recent single cell analysis of CDX tumors and of patients’ CTCs showed how intratumoral heterogeneity increases during disease progression after therapy with multiple concurrent treatment resistance mechanisms (Stewart et al., 2020).

Systematic search for noncoding and structural variant drivers of lineage plasticity (Figure 3).

Figure 3. Framework for systematic discovery of genomic and epigenomic lineage plasticity drivers.

(Left) pre and post-treatment LUAD and PCa samples undergo whole genome sequencing ± (middle) long-range and epigenomic profiling. (right) Statistical analysis of recurrent and chromatin perturbing SNVs, indels, and structural variants nominate loci that are under positive selection and/or recurrent chromatin perturbations. Such loci (rightmost panel) may represent enhancer-promoter interactions that bring complex combinations of distant loci together to drive cell identity changes through the creation (or disruption) of transcriptional condensates.

The conventional wisdom understanding of lineage plasticity in prostate and lung cancer is as a treatment-triggered shift in transcriptional and epigenetic programs. As discussed before, genetic alterations like TP53 and RB1 co-mutation are strongly associated with the small cell state, but are not thought to be sufficient to drive it (Boumahdi and de Sauvage, 2020; Quintanal-Villalonga et al., 2020; Rickman et al., 2017). A role for other (alternate or cooperating) genetic factors has not yet been defined, leading many to propose that somatic genetic selection plays a minimal role beyond the known genetic factors.

While the existence of a pre-existing small cell component in a pre-treatment adenocarcinoma cannot be ruled out in both SCPCa and LUAD-SCLC transition, it indeed seems unlikely given the natural history of these high-grade relapses, which frequently progress rapidly after 12 month or longer latency. These observations, however, do not rule out the accumulation of additional mutations during the course of treatment which trigger the small cell conversion. The existence of genetic as opposed to epigenetic drivers of small cell conversion may be critical for patient stratification and therapeutics, particularly for agents (e.g. EZH2 inhibitors) that attempt to reverse the small cell state.

The conventional wisdom that non-genetic mechanisms drive lineage plasticity in lung and prostate cancer has not been rigorously tested in patient data. Somatic genome landscapes of confirmed SCPCa and LUAD-SCLC transition cases have been largely limited to small cohorts (≤ 20 cases), targeted sequencing, and genomic case reports. The largest study in SCPCa comprised a whole exome sequencing (WES) survey of 20 distinct cases with confirmed NE phenotype, a minority of which had pre-treatment adenocarcinoma biopsies available. More extensive WES and WGS surveys of CRPC, including a recent WGBS / WGS study (Zhao et al., 2020), have included fewer than ten additional SCPCa cases, though these and other cohorts likely harbor additional SCPCa cases that weren’t annotated for NE histology. In LUAD-SCLC transition, the largest genome-wide study spanned five NE histology WGS cases and their matched pre-treatment adenocarcinoma profiles (Lee et al., 2017). Among other notable profiling efforts, Niederst et al. (2015) reported WES of multiple biopsies from a single LUAD-SCLC case and a more recent study reported WES profiles in four LUAD-SCLC cases (Marcoux et al., 2019). The remaining studies comprise case reports and/or have been limited to targeted sequencing (Gu et al., 2019; Offin et al., 2019; Pros et al., 2020; Sequist et al., 2011).

Such cohort sizes are grossly underpowered to nominate novel lineage plasticity drivers. Data-driven detection of positive somatic selection may require joint analysis of hundreds of samples to attain statistical power, particularly to detect variants or variant combinations that are less frequent (<10%) in the population (Lawrence et al., 2014; Martincorena et al., 2018). To maximize statistical power to associate genetic alterations with relapse, such studies would benefit from the maximal inclusion of pre-treatment adenocarcinoma samples to exclude variants acquired during early tumor development.

Among potential genomic targets, noncoding somatic variants provide an attractive source of potential lineage plasticity drivers. Noncoding variants can perturb regulatory elements, histone marks, and chromatin architecture, and theoretically promote or trigger cell state transitions (Khurana et al., 2016). Indeed, some noncoding variants have been implicated in dramatic developmental phenotypes, including high-penetrance Mendelian conditions (French and Edwards, 2020). More frequently, noncoding constitutional variants drive GWAS associations in complex traits through modulatory and tissue-specific effects on gene expression as expression quantitative trait loci (eQTL) (Albert and Kruglyak, 2015).

However, outside of TERT promoter mutations, statistically recurrent noncoding driver alterations in primary human cancers have been difficult to find, notwithstanding large-scale pan-cancer efforts scanning thousands of WGS samples (Khurana et al., 2016; Rheinbay et al., 2020). Though similarly-powered analyses are currently underway in WGS of later stage cancers, including thousands of metastases (Priestley et al., 2019), the analytic challenges may be more fundamental. Unlike coding drivers, where variants recur in single genes and have reasonably predictable and often highly penetrant effects on protein structure or function, the statistical nomination of noncoding drivers may require consideration of the cumulative or synergistic effect of many modulatory low-impact variants (Khurana et al., 2016; Yi et al., 2017). Furthermore, these mutations may be dispersed across a set of distant regulatory elements that control the expression of multiple promoters in a complex transcriptional program (Yi et al., 2017).

Notwithstanding these limitations, recurrent noncoding alterations have already been shown to impact targeted therapy resistance in CRPC. Several recent WGS and chromatin profiling studies implicated duplications of the AR enhancer in CRPC ARSi resistance (Quigley et al., 2018; Takeda et al., 2018; Viswanathan et al., 2018). Nomination of additional noncoding drivers of relapse and drug resistance may require joint analysis of large-scale compendia of WGS and chromatin profiles (e.g. open chromatin, methylation profiling, histone mark profiling) or 3D genome formation (e.g. Hi-C) in large compendia of tumor samples. Combined WGS and epigenome landscapes of patient cohorts have only been generated in a limited fashion in epithelial cancers, most notably in a pan-cancer ATAC-seq study from the TCGA, where a subset of cases had WGS (Corces et al., 2018). Recent studies employing genome-wide histone (Pomerantz et al., 2020) and methylation (Zhao et al., 2020) profiling ± WGS across many CRPC cases have begun to emerge, though only a subset of the published cases (~5%) are annotated as SCPCa.

Structural variants (SVs) may potentially provide a particularly powerful source of genetic lineage plasticity drivers, beyond those that mediate the loss of TP53 and RB1 in SCPCa and SCLC. Complex SV types like chromothripsis and templated insertion chains can bring distant regulatory elements in cis. SVs have been implicated enhancer hijacking across cancer (Li et al., 2020; Weischenfeldt et al., 2017), including neuroblastoma, a pediatric NE tumor (Zimmerman et al., 2018). Additional SV types, such double minutes (also known as eccDNA) (Verhaak et al., 2019) or breakage fusion bridge cycles, as well as recently characterized complex SV phenomena like pyrgo and tyfonas (Hadi et al., 2020), coalesce multiple copies of the same DNA loci into complex amplicons, many of which are associated with uniquely open chromatin states (Morton et al., 2019; Wu et al., 2019). Indeed, SCLC are among the most rearranged cancers (Li et al., 2020), where Hadi et al. (2020) have recently demonstrated tyfonas, a class of chromosomally integrated complex amplifications that incorporate tens of fold back inversion and other amplified junctions, as a mechanism of bringing many copies of the SCLC driver oncogenes NMYC and MYB in cis on a single complex chromosomally integrated locus. Given that 3D chromatin architecture is tightly regulated across human cell types and even across species, the impact of these dramatic perturbations on rewiring transcriptional programs is likely to be profound (Spielmann et al., 2018).

Recent research in gene regulation has focused on the role of transcriptional condensates, which are membraneless 3D compartments that arise through liquid-liquid phase separation, inside which multiple regulatory elements may interact to drive cell identity circuits (Sabari et al., 2020). We speculate that complex SVs could both help create cooperative 3D interactions that trigger the small cell state or, conversely disrupt, 3D condensates that are responsible for maintaining luminal / adenocarcinoma cell identity. Could some of these alterations provide the necessary perturbation to trigger a stable cell identity switch? The absence of WGS data in targeted therapy associated cancers, particularly those from patient-matched pre- and post-treatment samples, is the key obstacle to addressing this question. Such future studies may benefit from the use of long-range whole genome profiling technologies (e.g. linked-reads, long-reads, optical mapping (Sedlazeck et al., 2018) and the integration of primary DNA structure alterations with pairwise and multi-way 3D chromatin interactions (Kempfer and Pombo, 2020; Ulahannan et al., 2019). The integration of these data may help better elucidate the interplay of these two aspects (1D and 3D) of genome structure in lineage plasticity, and identify new directions for diagnosis and therapy.

Lineage plasticity triggers in the tumor microenvironment (TME).

The 2019 Nobel Prize in Medicine was awarded to scientists who unraveled the effect of oxygen delivery and metabolism in normal and malignant tissues. All solid tumors, including lung and prostate cancers and their metastases, contain sub- regions of abnormal cell metabolism, that can result from dynamic and differential gradients of oxygen consumption within the TME (Ashton and Bristow, 2020; Bhandari et al., 2019; Bharti et al., 2019). Tumor adaptation to imbalanced oxygen supply and demand is associated with poor prognosis and elevated genomic instability, resistance to chemotherapy and radiotherapy, immune dampening, altered autophagy, development of tumor stem cell protective niches, and increased proclivity for distant metastasis, such as bone metastases (Bristow and Hill, 2008; Johnson et al., 2017; Luoto et al., 2013; Nobre et al., 2018).

In PCa, independent observations in pre-clinical PCa models have linked tumor hypoxia to EMT (Bery et al., 2020; Tang et al., 2019), castrate resistance (Skov et al., 2004; Tran et al., 2020; Yapp et al., 2007) and acquisition of small cell phenotype (Guo et al., 2019; Qi et al., 2010). In one study, formation of SCPCa in the TRAMP mouse model was suppressed in mice lacking the ubiquitin ligase Siah2, which regulates HIF-1α availability. Cooperation between HIF-1α and FoxA2 transcription factors promoted expression of selected HIF-regulated genes (Hes6, Sox9, and Jmjd1a) required for the formation of SCPCa tumors. In more recent work (using a bioinformatic discovery approach for commonalities amongst SCC tumors), the gene ONECUT2 was identified as a candidate master transcriptional regulator of poorly differentiated SCPCa and SCLC tumors (Guo et al., 2019; Rotinen et al., 2018). In PCa, ONECUT2 synergizes with hypoxia by activating SMAD3, which regulates hypoxia signaling through modulating HIF-1α chromatin-binding, leading NE PCa to exhibit higher degrees of hypoxia compared to prostate adenocarcinomas

Once potential integrated model for the TME effect on lineage plasticity involves evolution and clonal selection of hypoxia stress-resistant cancer cells on the background of TP53 and other genomic losses within the primary SCC tumor and/or metastatic microenvironments. Pan-cancer approaches in 1,188 tumors spanning 27 cancer types, including PCa, NSCLC and SCLC, showed that elevated hypoxia is associated with increased mutational load across cancers, including TP53, MYC and PTEN (Bhandari et al., 2020), and this directs the evolutionary trajectory of these tumors. Importantly, mechanistic studies have shown that mutant TP53 and novel TP53 isoforms lead to selective growth advantage and increased metastatic capability due to an acquired tolerance to low oxygen conditions (Graeber et al., 1996) within an environment that is also immunosuppressed with higher levels of expression of PD1 and PDL1 (Kazantseva et al., 2019). In fact, in PCa (Glaser et al., 1989; Yang et al., 2018b) and in other cancers, the co-occurrence of hypoxia and genetic instability is synergistic in driving adverse clinical prognosis (Lalonde et al., 2014). Using bioinformatic approaches, the Bristow group quantified hypoxia in 8,006 tumors across 19 tumor types using the TCGA and ICGC whole exome and genome datasets. In ten tumor types, hypoxia was associated with elevated genomic instability. Of interest, in all 19 tumor types tested, hypoxic tumors exhibited characteristic driver mutation signatures for that histopathology (Bhandari et al., 2019). In aggressive localized PCa, hypoxia was associated with elevated rates of chromothripsis and genetic instability and an allelic loss of PTEN particularly in aggressive polyclonal tumors. Their work has shown that a constellation of aggressive features can occur in prostate glands resembling a tumor “nimbosus”– an aggressive cellular phenotype in which co-incident hypoxia, genetic instability and aggressive sub-pathologies co-occur (Chua et al., 2017).

More work is needed to understand further changes in genomics in hypoxic metastases in samples collected from metastatic treatment-sensitive and treatment-resistant cancers. We speculate that hypoxia-induced drug resistance observed following both non-targeted and targeted drug treatments (Jing et al., 2019; Ye et al., 2019) leads to further clonal adaption and increasing chromosomal instability (CIN) resulting in acquired resistance and SCC plasticity changes (Ashton and Bristow, 2020; Luoto et al., 2013). This is based on observations that hypoxia can modify the DNA damage response (DDR), and, in some cases, hypoxic cells are rendered DNA repair-deficient (Chan et al., 2010). Hypoxic cells deficient in homologous recombination (HR) or mismatch repair (MMR) can be sensitized in vitro by inhibitors of poly(ADP-ribose)polymerase (PARP) proteins, which function in DNA double- and single-strand break repair and base-excision repair (Borst et al., 2017; Chan et al., 2014; Chan et al., 2010). As a result, increased unrepaired double-strand breaks (DSBs), replication errors, and decreased centrosome function may accelerate genetic instability and lead to an aggressive, mutator phenotype. There may be a relationship between hypoxia related DDR alterations and SWI/SNF chromatin remodeling (Batie et al., 2018; Batie et al., 2019). SWI/SNF complexes have been implicated in facilitating repair of DSBs, by non-homologous end-joining (NHEJ), HR and nucleotide excision repair (NER). SMARCA2-BAF and SMARCA4-PBAF are recruited to DSBs and modify chromatin to enable spreading of ATM-induced gH2AX signaling during the DDR response and subsequently alter RAD51 and Ku70 activity (reviewed in (Ribeiro-Silva et al., 2019). This has yet to be studied in any comprehensive manner.

Collectively, these results highlight the synergy between genetic instability, lineage plasticity, TFs and hypoxia in driving SCC phenotypes and are the basis for in depth studies of in situ TME changes in relation to genetic and epigenetic changes during treatment resistance. How do these links between hypoxia and lineage plasticity impact the biology of the adenocarcinoma to SCC transition in lung? Does hypoxia-related NE lineage plasticity contribute to the early evolution of a subset of SCLC? How might the joint effect of hypoxia on transcriptional circuits (ONECUT2), chromatin remodeling (SWI/SNF), and genome instability conspire to promote the evolution of cooperating lineage plasticity driver alterations?

6. Preventing lineage plasticity

There are currently no standard paradigms for the treatment of SCPCa and LUAD-SCLC (Quintanal-Villalonga et al., 2020). Though the reversal of the NE phenotype provides an attractive target, a more realistic approach may be to prevent SCC conversion by identifying biomarkers for at-risk patient populations and developing co-targeting approaches to avoid transdifferentiation.

While TP53 and RB1 mutations are being actively investigated as pre-treatment biomarkers placing to predict SCC conversion in PCa and LUAD cancer, additional genomic characterization efforts are needed to identify at-risk patients. For such patients, dual inhibition of transcriptional circuits promoting the NE phenotype and growth circuits driving the growth of adenocarcinoma may help prevent transition to SCC. Such approaches are currently being investigated in phase 2 trials applying EZH2 inhibitors in combination with ARSi (see recent review Quintanal-Villalonga et al. (2020)). SWI/SNF complexes may also represent a novel therapeutic targeted for SCC prevention. Recent work by Ding et al. (2019) specifically proposed a synthetic lethal association between PTEN and SMARCA4 in PCa, identified through a CRISPR-Cas9 screen. This could have highly relevant translational implications, as 30% of clinically localized PCa cases and as many as 80% of CRPC demonstrate homozygous PTEN deletion (Abida et al., 2019; Aggarwal et al., 2018). They showed that in vitro, SMARCA4 knock-down leads to decreased cell proliferation in PTEN-negative cell lines (LNCaP, C4–2 and PC3), but not in PTEN-competent cells (22Rv1, BPH-1, and LAPC4). They extended these findings to a mouse model of early PCa by conditionally inactivating Pten and Smarca4 in Pten^PC–/–; Brg1^PC–/– mice and compared tumor growth and mouse-derived organoid growth with and without Pten loss in the context of Smarca4 loss. Their results were consistent with their in vitro cell line experiments. Therefore, targeting SMARCA4 or other components of the SWI/SNF complex may add another tool for epigenomic modulation of advanced PCa or other epithelial cancers.

A recent study by Li et al. (2019) found that the CXCR2, a chemokine surface protein, is more highly expressed in high-grade PCa in a subpopulation of AR negative cells. The CXCR2 positive cell population becomes enriched when exposing LNCaP, AR sensitive cells, to ARSi. The authors demonstrated using publicly available ChIPseq data that the over expression of CXCR2 shuts down the global expression program for luminal cell expression, helping to explain their lack of response to ARSi. They further demonstrated, in the ARSi resistant cell line C4–2B, that knocking out CXCR2 leads to the re-expression of AR signaling genes. Finally, in xenograft models they demonstrated that navarixin, a CXCR2i, could reduce tumor volumes of ARSi resistant tumors. Their work proposes a strategy to combine ARSi to attack the luminal component with CXCR2i to address the neuronal component. Intriguingly, they also discovered that AR-, CXCR2-positive cells secrete chemokines and cytokines that promote angiogenesis. Therefore, targeting CXCR2 would both have the potential effect of killing NE cells and act as an anti-angiogenic factor.

Our model suggests the additional potential of genetic instability-directed or hypoxia-directed therapy to offset lineage changes and resistance in prostate and lung SCCs (Ashton and Bristow, 2020; Borst et al., 2017; DiGiacomo and Gilkes, 2019; Guo et al., 2019; Lin et al., 2017; Salem et al., 2018). Our preclinical work supported the use of PARP inhibition in the DDR-deficient hypoxic tumor cells based on an acquired BRCAness which led to increased sensitivity to Olaparib, both in vitro and in vivo (reviewed in Ashton and Bristow (2020)). The clinical success of PARP inhibitors in PCa with germline and acquired mutations in BRCA2 and ATM lays the groundwork for their use in the hypoxic setting to prevent lineage plasticity (Gillessen and Bristow, 2020). As stated above, ONECUT2 ectopic expression in prostate adenocarcinoma synergized with hypoxia to suppress androgen signaling and induced SCPCa. Treatment with hypoxia-activated prodrug TH-302 potently reduced SCPCa tumor growth and emphasizes the potential of hypoxia-directed therapy for SCPCa patients (Guo et al., 2019; Rotinen et al., 2018). More recent data suggests that oncogenic and aberrant AR signaling, hypoxia and HIF1a pathways support PCa development through independent signaling pathways based on differential transcriptomic profiles. If AR and hypoxia/HIF1a signaling pathways independently promote treatment resistance and SCC plasticity, simultaneous therapeutic targeting of both pathways may be needed (Tran et al., 2020).

Summary:

We discuss the established and potential roles of epigenetic, genomic, and microenvironmental perturbations as lineage plasticity drivers, particularly in the evolution of SCC. We discuss two exemplar adenocarcinomas, PCa and LUAD, and compare and contrast their transition to SCC, as well as recent evidence of SCLC plasticity to a non-NE identity. Remarkably, each of these transitions are accompanied by a dramatic and clinically relevant phenotype, impacting drug sensitivity and survival. We posit that a more granular definition of the universal and specific manifestations of this aggressive and drug resistant NE state, and its adjacent cell identities, are become essential to guide the prevention and treatment of the increasing number and types of SCC cancers. A systematic dissection of the role of chromatin, genomic, and microenvironmental factors, through the large-scale profiling of patient tissues, but also through experimental in vitro and in vivo modeling, will be needed to define the forces shaping the lineage plasticity landscape around the NE state (Figure 4).

Figure 4. Lineage Plasticity as a therapy induced state.

A. In proposed models, there may be pre-existing differentiated cells that are simply selected for during the course of therapy. B. However, emerging data presented in this Perspective explores the lineage plasticity model as a series of events that can transition differentiated cells into a stem-like state and then under the right conditions into another differentiated state either reversible (middle) or irreversible (lower). In this perspective, we address the features that contribute to the formation of lethal resistant clones. We propose that key features include simple and complex genomic alterations (e.g. TP53, RB1, PTEN loss), epigenetic reprogramming (e.g., EZH2, SETD2, SWI/SNF) and genomic instability secondary to changes in the tumor microenvironment (e.g., hypoxia). These alterations occur over a time course in the context of selection based on distinct oncogenic therapies. We present examples of luminal tumors transdifferentiating to SCC (i.e., PCa and NSCLC) and an example of SCLC transdifferentiating to another mesenchymal state. Figure based Waddington’s original figure (Waddington, 1957) modified by Le Magnen et al. (2018).