Epithelial cell plasticity: breaking boundaries and changing landscapes

Abstract

Epithelial tissues respond to a wide variety of environmental and genotoxic stresses. As an adaptive mechanism, cells can deviate from their natural paths to acquire new identities, both within and across lineages. Under extreme conditions, epithelial tissues can utilize “shape‐shifting” mechanisms whereby they alter their form and function at a tissue‐wide scale. Mounting evidence suggests that in order to acquire these alternate tissue identities, cells follow a core set of “tissue logic” principles based on developmental paradigms. Here, we review the terminology and the concepts that have been put forward to describe cell plasticity. We also provide insights into various cell intrinsic and extrinsic factors, including genetic mutations, inflammation, microbiota, and therapeutic agents that contribute to cell plasticity. Additionally, we discuss recent studies that have sought to decode the “syntax” of plasticity—i.e., the cellular and molecular principles through which cells acquire new identities in both homeostatic and malignant epithelial tissues—and how these processes can be manipulated for developing novel cancer therapeutics.

In order to alter tissue identity, cells follow a set of “tissue logic” principles based on developmental paradigms. This review discusses cellular and molecular principles through which cells acquire new identities both in homeostatic and malignant epithelia.

Glossary

ADM

acinar‐to‐ductal metaplasia

AT1

type 1 alveolar epithelial cell

AT2

type 2 alveolar epithelial cell

CDX1

caudal type homeobox 1

CDX2

caudal type homeobox 2

EGFR

epidermal growth factor receptor

EMT

epithelial‐to‐mesenchymal transition

EZH2

enhancer of zeste homolog 2

FoxA1

forkhead Box A1

FoxA2

forkhead Box A2

GATA4

GATA‐binding protein 4

KRAS

Kirsten rat sarcoma

Krt5

keratin 5

Krt7

keratin 7

MAML2

mastermind‐like 2

MAPK

mitogen‐activated protein kinase

MECT1

mucoepidermoid carcinoma translocated‐1

MET

mesenchymal‐to‐epithelial transition

mTOR

mammalian target of rapamycin

NAPSA

novel aspartic proteinase of the pepsin family

Nkx2‐1

NK2 homeobox 1

NSCLC

Non‐small cell lung cancer

PKA

protein kinase A

PROTAC

proteolysis targeting chimera

RB1

transcriptional corepressor 1

SCLC

small‐cell lung cancer

SMG

submucosal glands

SOX2

(sex‐determining region Y)‐box 2

SOX9

(sex‐determining region Y)‐box 9

STK11

serine/threonine kinase 11

TCGA

The Cancer Genome Atlas

TF

transcription factor

TGFβ

transforming growth factor beta

Trp53

transformation‐related protein 53

Trp63

transformation‐related protein 63

TZ

transition zone

Introduction

Tissues differ dramatically in their cellular composition, architecture, and turnover. Collectively, these features dictate the unique characteristics and functions of a tissue. Within a given tissue, different cell types are located in anatomically distinct locations, facilitating cell‐autonomous and non‐cell‐autonomous regulatory mechanisms. In many cases, the constituent cell types of a tissue are organized in a hierarchical manner, where cells with higher potency, often referred to as stem/progenitor cells, are positioned at the top and more mature cell types (less potent) are placed at the bottom of the hierarchy (Hogan et al, 2014; Scadden, 2014; Visvader & Clevers, 2016; Clevers & Watt, 2018) (Fig 1). It has long been thought that cells differentiate irreversibly (Hsu et al, 2011; Rajagopal & Stanger, 2016; Tata & Rajagopal, 2016a). However, there is growing evidence that lineage hierarchies can be reversible within the same lineage, and, in some cases, cells can deviate from normal hierarchies to acquire other cell or tissue fates in response to environmental and genotoxic stresses. This phenomenon is termed “cellular plasticity”. The cellular plasticity has been extensively studied in invertebrates, and one of the best examples is dedifferentiation of germ stem cells of Drosophila melanogaster (Brawley & Matunis, 2004; Kai & Spradling, 2004; Fuller & Spradling, 2007; Sheng et al, 2009; de Cuevas & Matunis, 2011). Similarly, cardiomyocytes in zebrafish and muscle cells in newt limbs undergo dedifferentiation into progenitor states, which aid in tissue regeneration after injury (Jopling et al, 2010, 2011; Sandoval‐Guzmán et al, 2014; Tanaka et al, 2016). Recent studies using elegant cell lineage tracing models have demonstrated that plasticity is not restricted to lower organisms, but rather a common feature of tissues in higher vertebrates, including mammals (Rajagopal & Stanger, 2016; Tata & Rajagopal, 2016a; Tetteh et al, 2015, 2016; van Es et al, 2012, 1; Tata et al, 2013; Tata & Rajagopal, 2017; Rompolas et al, 2013; Tata et al, 2018a; Lynch et al, 2018). Here, we review different mechanisms of cellular plasticity and how they drive tissue‐wide changes that occur in the context of metaplasia and tumorigenesis in endoderm‐derived epithelial tissues. We discuss emerging models that explain the predictable ways in which cells acquire these alternate lineages, with implications for cancer prevention and treatment.

Figure 1. Plasticity within normal lineage hierarchies.

(A) During development and homeostasis, self‐renewing stem cells divide and generate progenitors, which are destined to form mature differentiated cells. (B) Mature cells may dedifferentiate and give rise to stem cells following injury or cell ablation. (C) Tissue injury may induce mature differentiated cells to change their cell fate, either directly or indirectly through reversion to a progenitor cell in a process called transdifferentiation. (D) In response to injury, a reserve stem cell population can be activated and contribute to generation of active stem cells, which in turn generate committed progenitors and mature differentiated cells. (E) In transdetermination, a self‐renewing progenitor cell transforms into another type of progenitor cell. (F) In response to acute injury, mature differentiated cells undergo activation of auto degradative responses and re‐enter the cell cycle to become proliferative in a process called paligenosis. Dashed lines indicate lineage reversion.

Cell plasticity: concepts and terminology

As discussed above, the cell types within a tissue can generally be organized into a lineage hierarchy. In most cases, tissue turnover is largely fueled by stem/progenitor cell proliferation and differentiation (Blanpain et al, 2004; Ito et al, 2005; Barker et al, 2007; Barker, 2014; Hogan et al, 2014). However, recent studies have demonstrated that under extreme conditions, such as severe damage, more differentiated cell types can regenerate lost cells within the normal lineage hierarchy. Such examples of cellular plasticity have been the subject of many excellent reviews (Slack, 2007; Jopling et al, 2011; Stange et al, 2013; Tata et al, 2013; Blanpain & Fuchs, 2014; Chua et al, 2014; Friedmann‐Morvinski & Verma, 2014; Hogan et al, 2014; Sánchez Alvarado & Yamanaka, 2014; Mills & Sansom, 2015; Tata & Rajagopal, 2016a; Tata & Rajagopal, 2016b; Tetteh et al, 2016; Visvader & Clevers, 2016; Clevers & Watt, 2018; Centonze et al, 2020). As illustrated in Fig 1, the relationships between stem/progenitor cells and their differentiated progeny have proven to be far more complex than previously appreciated. Through various processes such as dedifferentiation, transdifferentiation, reserve stem cell activation, and transdetermination, cells can undergo dramatic transformations that seemingly violate the rules of cellular hierarchies (Fig 1A–E). Some of these processes require extreme changes in cellular mechanisms that enable them to shed their machinery to acquire a progenitor cell state that facilitates cell division or conversion of one cell type or cell state to another. Recent studies identified a stepwise mechanism in which differentiated cells acquire a regenerative capacity to repair lost cells after injury. This process is termed “paligenosis” (Fig 1F). This process is best studied in the gastric epithelial tissues in which mature cells dynamically regulate mTOR signaling to control cellular metabolism. In the first step, downregulation of mTOR signaling is accompanied by activation of autophagy pathways, which in turn facilitates degradation of cellular machinery characteristic to the mature cells. In the second step, mTOR signaling is upregulated, which in turn downregulates lysosomal pathways followed by upregulation of gene expression programs relevant to progenitor cells and cell proliferation (Messal et al, 2018; Willet et al, 2018). Remarkably, paligenosis has been observed in multiple tissues and uses an evolutionarily conserved molecular network that facilitates the plasticity of cells in response to injury (Weis et al, 2017; Miao et al, 2020).

Cell proliferation is often associated with cell plasticity within normal lineage hierarchies (Banito et al, 2009; Ruiz et al, 2011). However, there are some exceptions. For example, it has been observed that transdifferentiation of pancreatic alpha cells to beta cells does not involve cell proliferation (Thorel et al, 2010). In light of these exceptions, there is some debate regarding the proper terminology to describe such processes in aggregate, given the unique idiosyncrasies of cellular plasticity in certain tissues. Here, we suggest the term “reprogramming” to collectively describe all such processes. For example, we can state that Cell‐A reprograms into Cell‐B, irrespective of whether that process requires proliferation or whether the starting cell and the resulting cell are related or not.

Cellular mechanisms driving cell plasticity

One striking feature about metazoan development is that a small number of key signaling pathways drive the generation of an enormous diversity of cell types, all arranged in highly ordered spatial patterns (Gerhart, 1999; Pires‐daSilva & Sommer, 2003; Salazar‐Ciudad, 2010; Martyn et al, 2018; Tan & Barker, 2018; Arnold et al, 2019; Newman, 2019). A prime example is the organization of epithelial tissues, which can exist in a variety of patterns, including squamous, columnar, and acinar structures, each containing hierarchically organized cells of varying potency. Irrespective of their germ layer of origin, epithelia in different organs can contain similar cell types, share histo‐morphological characteristics, and perform roughly similar functions (Osterfield et al, 2017). Even more strikingly, such tissues also utilize similar molecular circuits for their development and maintenance. For instance, squamous tissues that line different organs, such as the skin epidermis, esophagus, and reproductive tract, arise from ectoderm, endoderm, and mesoderm, respectively. Despite their embryologic differences, these three types of squamous epithelia share similar cell types, morphology, and molecular characteristics. For example, the transcription factor Trp63, the Notch pathway, and SMAD signaling are key molecular regulators of squamous tissues across different organs, helping to maintain the cellular composition and proportions of a tissue (Collier et al, 1996; Pellegrini et al, 2001; Blanpain et al, 2004; Senoo et al, 2007; Bass et al, 2009; Doupé et al, 2012; Alcolea et al, 2014; Guo & Ohlstein, 2015; Mou et al, 2016; Martincorena et al, 2018). Notably, these epithelial patterns are conserved not only within an organism but also across all metazoans. This implies that evolution utilized a common set of molecular circuits to mold epithelial sheets into specific architectures that satisfy the needs of a given tissue site (Hannezo et al, 2014; Osterfield et al, 2017). Studies have revealed that cells utilize these conserved programs of tissue patterning to drive histological substitutions observed during tumor initiation (Fig 2).

Figure 2. Types of metaplasia based on the resulting tissue.

(A) External stimuli can trigger conversion of columnar cells to metaplastic squamous cells. (B) Squamous epithelial cells may undergo metaplastic transformation into columnar and Goblet cells. (C) Glandular epithelia can differentiate into columnar epithelia. (D) The pancreas is composed of acinar cells with characteristic cytoplasmic zymogen granules and ductal cells. Following injury, pancreatic acinar cells undergo acinar‐to‐ductal metaplasia.

Metaplasia: plasticity at the tissue level

“Metaplasia” is derived from the Greek words μɛτα (“after”, “changed”) and πλασμα (“something formed”). The term metaplasia was first used in 1863 by Rudolf Virchow to describe the distinction between physiological tissues from pathological tissues resulting from a histological substitution—the conversion of one type of tissue into another morphologically distinct tissue (Virchow, 1865; Wagner, 1999; Kumar et al, 2010). As described in detail in the following section, such histological substitutions can originate de novo by complete reprogramming of resident cells or encroachment by neighboring cells/tissues (trans‐metaplasia). In a broad sense, metaplastic events can be classified based on the cell of origin and the type of substitution (i.e., the resulting tissue). For example, Barrett’s metaplasia (also known as intestinal metaplasia) involves the appearance of columnar epithelium resembling the neighboring stomach/intestine, in place of normal squamous esophageal epithelium (Slack, 2007; Burke & Tosh, 2012; Giroux & Rustgi, 2017).

Types of metaplasia based on tissue architecture

Squamous metaplasia of the airway is defined by the replacement of normal columnar cells with flat stratified cells (Puchelle et al, 2006; Slack, 2007; Rock et al, 2010; Giroux & Rustgi, 2017) (Fig 2A). Environmental pollutants, cigarette smoke (in airways), and vitamin A deficiency are the most commonly associated etiological factors associated with squamous metaplasia (Marchok et al, 1975; Puchelle et al, 2006). Importantly, replacement of normal ciliated columnar epithelia by squamous tissue can dysregulate mucociliary clearance, contributing to airway obstructive diseases. Recent studies have revealed that metaplastic squamous cells in the lung lack expression of the key lung transcription factor Nkx2‐1, demonstrating their considerable divergence from normal lung transcriptional programs (Tata et al, 2018b; Zewdu et al, 2021; Mollaoglu et al, 2018; Maeda et al, 2012; Camolotto et al, 2018, 1). In contrast, lung squamous cell carcinomas are uniformly negative for canonical lung markers and can be identified by expression of KRT5 and TRP63 (Mukhopadhyay & Katzenstein, 2011; Rekhtman et al, 2011). Although the lung does not normally contain squamous epithelium, squamous cell carcinomas nevertheless comprise around one‐third of all lung cancers. Similar to the lung, columnar tissues of the endocervix can also undergo squamous metaplasia. This histological substitution is associated with the onset of puberty, suggesting that hormonal changes play a central role in this process. Other sites of squamous metaplasia include the epithelium of the mammary, urinary bladder, prostate, pancreas, and sebaceous glands; however, these forms of metaplasia are less common (Slack, 2007; Giroux & Rustgi, 2017).

Columnar metaplasia is another commonly occurring type of histological substitution, in which columnar cells are present in a normally non‐columnar tissue (Fig 2B). As briefly discussed earlier, the classic example of columnar metaplasia is Barrett’s esophagus. In patients with Barrett’s esophagus, the squamous epithelium at the squamo‐columnar junction of the esophagus is replaced by columnar cells. From the perspective of tissue patterning and morphology, Barrett’s metaplasia is precisely the opposite of squamous metaplasia. Various factors, including genetic predisposition, inflammation, chronic gastro‐esophageal reflux, acids, and bile salts, are known to be associated with the development of columnar metaplasia. These metaplastic columnar cells seem to resemble the epithelia of the neighboring stomach glands and intestinal epithelial cells (Paneth cells, secretory cells, and enteroendocrine cells) implicating either de novo reprogramming of squamous cells or encroachment of neighboring cells in the emergence of Barrett’s metaplasia (see later section) (Colleypriest et al, 2009; Burke & Tosh, 2012; Garman, 2017).

Additionally, columnar metaplasia can also originate from glandular tissues. For example, it has been hypothesized that esophageal submucosal glands (SMGs) contribute to Barrett’s metaplasia (Leedham et al, 2008; Que et al, 2019) (Fig 2C). Recent single‐cell transcriptome studies from human Barrett’s metaplasia biopsies and porcine esophageal submucosal gland organoid cultures have further strengthened this model (Garman, 2017; Owen et al, 2018). However, mice do not have esophageal SMGs and yet can still develop Barrett’s metaplasia in the context of experimental perturbations that mimic chronic inflammation, namely overexpression of interleukin‐1β (Quante et al, 2012). This suggest that perhaps SMGs are not the predominant or sole source of cells for the development of these pathological lesions.

Another major example of metaplasia is acinar‐to‐ductal metaplasia (ADM), where an acinar/saccular epithelium acquires the features of the ducts (Fig 2D). Again, similar to squamous and columnar metaplasia, the identity of these metaplastic cells appears to resemble that of the neighboring ductal cells or the primitive embryonic‐like progenitor that shares the histological and molecular features common to both cell states (Slack, 2007; Prévot et al, 2012; Pan et al, 2013; Chuvin et al, 2017; Giroux & Rustgi, 2017; Vercauteren Drubbel et al, 2021). Common sites of ADM include the acini of the pancreas, mammary, and salivary glands. Chronic inflammation and/or stress caused by severe damage seems to play a major role in the induction of (or at least associated with) ADM in diverse glandular tissues (Batsakis, 1980; Giroux & Rustgi, 2017).

Thus, one common feature across all types of metaplasia is the emergence of ectopic cells that resemble cells from neighboring tissues (in many cases, tissues that directly connect with the site of metaplasia) or developmentally related primitive progenitors. Collectively, these phenomena highlight a remarkable adherence to conserved principles of embryonic tissue patterning as cells undergo extensive molecular and morphological remodeling in metaplasia (Slack, 1986, 2007). Intriguingly, there is evidence from epidemiological and experimental studies that metaplasia can be reversible (Dixon, 2001; Walker, 2003; Rigden et al, 2016). Because of these characteristics, metaplastic processes provide a valuable experimental window to uncover new insights into the molecular controls governing normal tissue maintenance and pathogenesis. The metaplasia‐to‐neoplasia sequence is a common mechanism for the emergence of tumor plasticity in many organs. Given the link between metaplasia and cancer, the reversibility of metaplasia suggests tangible benefits to furthering our understanding of this complex process.

Types of metaplasia based on cell of origin

Most metaplastic tissues consist of heterogeneous populations composed of normal cells intermingled with completely ectopic cell types (Fig 3A and B). This makes it ambiguous and sometimes controversial to discern the cell of origin. Prevailing model for the cell of origin of Barrett’s metaplasia is mainly based on the location of the abnormal columnar looking‐like cells. However, in vivo and ex vivo experimental models have indicated that at least 4 different types of cells contribute to Barrett’s metaplasia (Que et al, 2019). It is hypothesized that basal cells of the esophageal squamous epithelium undergo de novo reprogramming and convert into columnar cells in response to damage induced by bile reflex and inflammation (Fig 3A). This is referred to as de novo metaplasia. However, to date, there is no experimental evidence to support this model (Barbera & Fitzgerald, 2010; Xian et al, 2012). Yet, another model focuses on a small population of cells that are present at the squamo‐columnar junction of the esophagus and stomach. Historically, this model was based on electron microscopy studies that revealed cells with features of both squamous and columnar cells. Of note, two recent studies using mouse models have provided evidence to support this theory. In one study, using a Trp63 loss‐of‐function mouse model, it was observed that columnar epithelium at the squamo‐columnar junction expands into squamous epithelium of the esophagus. Additional evidence for this model came from another recent study, which used a transitional zone (TZ) epithelium—lineage tracing mouse model based on the Krt7 gene promoter (Jiang et al, 2017). In particular, the authors demonstrated that Krt7+ transitional basal cells can self‐renew and maintain the squamo‐columnar junction, and overexpression of CDX2 in these cells leads to the development of intestinal metaplasia. As described above, Barrett’s metaplasia is predominantly composed of cells that resemble the gastric and intestinal epithelium. Therefore, it is natural to speculate that the gastric and intestinal epithelial cells are the cell of origin for Barrett’s metaplasia (Barbera & Fitzgerald, 2010; Xian et al, 2012; Giroux & Rustgi, 2017). Indeed, it has been proposed that gastric epithelial cells expand and migrate to the esophagus when resident squamous cells are lost due to acid or bile reflux (Nakagawa et al, 2015). Here, we referred to this as “trans‐metaplasia” (Fig 3A). Additionally, submucosal gland cells can migrate to the surface epithelium to contribute to Barrett’s metaplasia (Fig 3B). Overall, it is our speculation that multiple cell sources, for example, cells from submucosal glands migrating toward the luminal surface or reprograming of a progenitor cell from transitional zone, may contribute to Barrett’s metaplasia in humans and the degree of contribution may vary according to the context‐specific milieu (Fig 3B).

Figure 3. Types of metaplasia based on the cell of origin.

(A) Normal squamous epithelium (orange), transition zone (TZ) (pink), and columnar cells (green). De novo metaplasia (a) originates from squamous epithelium. In specialized TZ progenitor (b), metaplasia emerges from transition zone cells. In trans‐metaplasia (c), metaplastic cells arise from neighboring columnar cells. (B) Epithelial metaplasia (shown in blue) may arise from a) de novo reprogramming of resident cells (e.g., squamous epithelium of esophagus), b) reprograming of transition zone cells, or c) migrating cells originating from different compartments (e.g., submucosal glands of esophagus or columnar epithelium of the stomach).

EMT in tumor plasticity

In many types of cancer, tumor cells have been shown to acquire developmental processes in response to environmental changes. Epithelial‐to‐mesenchymal transition (EMT) is a process that was originally described in developing tissues in which tightly packed epithelial cells transiently acquire mesenchymal cell characteristics (Nieto et al, 2016). These changes include loss of cell–cell adhesion and loss of epithelial but gain of mesenchymal cell molecular signatures, accompanied by cell migration through its dynamic interactions with the extracellular matrix (Revenu & Gilmour, 2009; Micalizzi et al, 2010; Zhang et al, 2014b; Jolly et al, 2016). Since its original inception, many studies have implicated EMT in many processes such as tissue morphogenesis and injury repair and more importantly in tumor cell plasticity. Classically, EMT has been seen as a hybrid cell state in which cells exhibit a mixture of both epithelial and mesenchymal phenotypes. With the advent of technologies such as clonal cell lineage tracing and single‐cell transcriptome have further stratified EMT into many subtypes (Pastushenko et al, 2018; Pastushenko & Blanpain, 2019). Interestingly, these subtypes can co‐exist in the same tissue, suggesting that they either co‐evolve or transitional states that represent EMT progression. Studies using lineage tracing coupled with oncogene expression in pancreas, mammary gland, prostate, skin, and intestine have led to the systematic identification of sequential events during EMT. These studies revealed that EMT precedes tumor establishment and metastasis initiation and cell dispersion during metastasis (Mani et al, 2008; Rhim et al, 2012; Chanrion et al, 2014; Koren et al, 2015; Zhao et al, 2016; Aiello et al, 2018; Pastushenko et al, 2018; Bakir et al, 2020). Further systematic analysis stratified cell transitions starting with epithelial, early hybrid, hybrid, late hybrid, and finally lead to a mesenchymal‐like state. Some studies have referred the hybrid states as partial‐EMT or incomplete EMT. Nevertheless, all studies point at EMT as a gradual process with many transitional states each can be characterized by their cell shape, markers, transcriptional states (Pastushenko et al, 2018). Interestingly, cells can also take an reverse process of EMT, in which a mesenchymal cell exhibits epithelial cell characteristics, a process often referred to as mesenchymal‐to‐epithelial transition (MET) (Bakir et al, 2020). It appears that there is a direct correlation between the degree of EMT with local inflammation. Significantly, induction of local inflammation augmented EMT in pancreatic tumors, suggesting that exogenous pressures influence EMT‐mediated cell plasticity (Rhim et al, 2012). Taken together, cells use yet another EMT as a plasticity mechanism to initiate, disseminate, and evade therapy.

Tumor cell plasticity: an evolving and adaptive mechanism

Given that cancers arise from normal cells, the resulting tumors would naturally be anticipated to retain the histological and molecular features of the tissue of origin. However, just as adult tissues can undergo metaplasia and exhibit plasticity, cancers frequently contain cell types that are foreign to the original site of tumorigenesis. Importantly, plasticity can arise at different stages in the disease course: during tumor initiation from a metaplastic state; subsequent tumor evolution, including hyperplasia, dysplasia, and carcinoma in situ; metastasis; evasion from therapy‐induced resistance; or microenvironmental selective pressures such as inflammation. In all cases, cells appear to follow the core principles of cell plasticity for survival and propagation.

Acquired plasticity during tumorigenesis and tumor evolution

Though the metaplasia‐to‐neoplasia sequence is a shared mechanism by which tumor plasticity can arise in diverse organs, it is not sufficient to explain the immense intra‐ and inter‐tumoral heterogeneity observed in patients. In particular, the existence of tumors with mixed histology indicates that plasticity can be acquired in the context of an established tumor, rather than merely passed on as a feature of the tumor‐initiating tissue at the time of oncogenic transformation. Indeed, recent studies have indicated that invasive mucinous adenocarcinomas of the lung are composed of a variety of tissues that resemble mid‐ and hindgut tissues, including gastric, duodenal, and small intestinal epithelia (Fig 4). Experimental models have revealed that such tumors predominantly begin with a reprogramming event into gastric type epithelium and subsequently develop into an array of tissues resembling the other gut tissues (Snyder et al, 2013; Camolotto et al, 2018; Tata et al, 2018b).

Figure 4. Understanding tumor plasticity through the lens of development.

(A) During embryonic development, the undifferentiated endoderm is prepatterned into multiple distinct organs along the anterior–posterior axis: the anterior foregut, posterior foregut, midgut, and hindgut. In lung tumors, plastic cancer cells adopt cell fates that are characteristic of gut organs. Lung squamous carcinomas (orange cells) manifest elements of the esophagus, whereas mucinous adenocarcinomas (yellow, blue, or green cells) are defined by cellular patterns similar to those found in stomach, duodenum, and hind gut epithelia. (B) Two possible pathways for mucinous lung adenocarcinoma development. As shown in Path‐1, lung cells first acquire stomach‐like cell fate (yellow), which then transition to multiple cell fates that resemble duodenum (blue) and/or hindgut‐like cells (turquois) in a stepwise manner. Alternatively, as shown in Path‐2, lung cells independently acquire different cell fates during tumorigenesis. AT2: type 2 alveolar cell and AT1: type 1 alveolar cell.

Lung carcinomas are broadly classified into small‐cell lung cancer (SCLC) and non‐small cell lung cancer (NSCLC). SCLC accounts for about 15% of all lung cancers and is heavily associated with smoking (Gazdar et al, 2017). SCLC tumors are histologically defined by their neuroendocrine differentiation features and characteristically exhibit mutations in Trp53 and RB1 (George et al, 2015). Loss of Trp53 and Rb1 in pulmonary neuroendocrine cells is sufficient to drive SCLC in mice (Sutherland et al, 2011). On the other hand, NSCLC is a highly heterogeneous disease, with numerous distinct histological subtypes defined by distinct cell types. Lung adenocarcinomas are comprised of cells that resemble the normal lung (i.e., alveolar epithelium) and are accordingly defined by their histochemical expression of classical lung markers such as NKX2‐1 and NAPSIN A (Mukhopadhyay & Katzenstein, 2011; Rekhtman et al, 2011). Among patients with primary SCLC diagnoses, 10‐24% of the tumors also possess features of NSCLCs on histopathology (Adelstein et al, 1986; Mangum et al, 1989). Collectively referred to as combined SCLC, these mixed tumors are associated with poorer prognosis and pose a diagnostic challenge (Hirsch et al, 1983; Marchevsky & Wick, 2015). As the clinical management of SCLC is different from that of NSCLC, understanding the plasticity intrinsic to combined SCLCs may help inform better treatment strategies (Oser et al, 2015). As another example of lung intratumoral heterogeneity, lung adenosquamous carcinomas possess the histological features of both lung adenocarcinomas and squamous carcinomas in a single tumor (Travis, 2011). Such tumors comprise 2‐3% of lung cancer diagnoses and are associated with poorer prognosis (Mordant et al, 2013). It has been demonstrated that the adenocarcinomatous regions of these mixed tumors retain expression of classical lung markers, while neighboring squamous regions are negative for these markers (Tata et al, 2018b). Though it is impossible to definitively rule out the co‐existence of two independent lung cancers in such patients, some sequencing analyses have indicated shared mutations between the adenocarcinomatous and squamous regions of the tumors, consistent with a clonal origin (Niho et al, 1999; Vassella et al, 2015). Interestingly, these histologically distinct tumor tissues are populated by different immune cells (Xu et al, 2014; Zhang et al, 2017; Mollaoglu et al, 2018). Similarly, tumors from different genetic mutations seem to recruit distinct stromal cells, suggesting that the cross‐talk between tumor cells and the microenvironment likely influences tumor heterogeneity as well as cell plasticity (Hill et al, 2020). In addition, genetic mouse models of lung cancer have identified that Stk11 mutations diversify the histology of Kras ^G12D‐driven tumors from purely adenocarcinomas to a mix of adenocarcinomas and squamous carcinomas (Ji et al, 2007). With that said, other studies utilizing tumor genomic sequencing have also provided evidence for the distinct origins of heterogeneous tumor cell populations within mature tumors (Zhang et al, 2014a; Raynaud et al, 2018; Wang et al, 2019).

In the salivary gland, mucoepidermoid carcinomas are the most common type of malignant tumor (Chenevert et al, 2011). First described in 1945 (Stewart et al, 1945), mucoepidermoid carcinomas are defined by the co‐existence of squamous epidermoid and mucinous cells on histology. Though it has historically been proposed that these tumors arise from a rare multipotent progenitor in the salivary gland (Batsakis, 1980), thus far no experiments have demonstrated the functional relevance or existence of such a cell. Remarkably, despite the immense cellular heterogeneity of these tumors, over half of all mucoepidermoid carcinomas possess a MECT1‐MAML2 gene fusion (Bell & El‐Naggar, 2013). This particular translocation leads to aberrant gene activation (Coxon et al, 2005; Wu et al, 2005) and is highly specific to mucoepidermoid carcinomas (Seethala et al, 2010). Given the strong connection between MECT1‐MAML2 translocations and mucoepidermoid carcinoma, it is likely that the MECT1‐MAML2 fusion protein somehow promotes cellular plasticity, thereby producing highly heterogeneous tumors.

Tumor plasticity arising from exogenous pressures

While cellular plasticity can be an intrinsic property of tumor cells, plasticity can also arise as an adaptation against environmental pressures. Analogous to how metaplasia can occur in response to repeated stress or injury, chemotherapy has been implicated in the reprogramming of tumors into alternative histological subtypes. In a cohort of chemotherapy‐treated patients with a final diagnosis of combined SCLC at surgery or autopsy, 7/17 (41%) of the patients had been originally diagnosed with NSCLC prior to initiation of chemotherapy (Adelstein et al, 1986). Furthermore, 2 out of 40 (5%) patients with an original diagnosis of SCLC were subsequently found to exclusively have adenocarcinoma at autopsy. While one cannot rule out these other tumor types were simply missed at diagnosis due to technical limitations, it is notable that an independent case series also revealed a similar phenomenon (Mangum et al, 1989). Furthermore, contemporary studies using targeted EGFR inhibitors in patients with EGFR‐mutant lung adenocarcinomas have indicated that drug resistance may arise through reprogramming from adenocarcinoma to SCLC (Sequist et al, 2011; Popat et al, 2013; Watanabe et al, 2013). Additionally, MAPK pathway activity was shown to dictate the target tissue identity adopted by NKX2‐1‐deficient lung tumors via a Wnt pathway‐dependent mechanism (Zewdu et al, 2021). Interestingly, a similar scenario has been observed in basal cell carcinomas of the skin, in which treatment of these tumors with vismodegib, an inhibitor of hedgehog pathway, led to activation of a transcriptional program accompanied by rapid Wnt signal transduction that drives cell plasticity (Biehs et al, 2018; Sánchez‐Danés et al, 2018). In this case, tumors shift from a hair follicle‐like identity to interfollicular and isthmus epidermal stem cells. These studies offer potential opportunities to target a shared Wnt pathway to block chemotherapy‐induced cell plasticity in these tumors (Biehs et al, 2018; Sánchez‐Danés et al, 2018; Zewdu et al, 2021). Consistent with this, a subset of the reprogrammed tumors were responsive to standard SCLC regimens (Sequist et al, 2011). In a similar vein, among prostate cancers that developed resistance to androgen receptor blockade, a sizeable fraction were found to exhibit small‐cell neuroendocrine differentiation (Beltran et al, 2014; Hu et al, 2015).

Additionally, a direct comparison of single‐cell transcriptome data from primary and metastatic NSCLCs and endoderm development revealed the existence of subpopulations of cells within these different tumors showed a convergence toward common developmental programs (Laughney et al, 2020). These subpopulations are marked by differential expression of SOX2 and SOX9. Interestingly, these seemingly distinct subpopulations that reflect developmental continuum exhibit differential sensitivities to immune cell surveillance and metastasis progression. Accordingly, loss and gain of function of these developmental transcription factors significantly affect tumor cell survival by either escape from immune cell‐mediated cytolysis or metastasis. These studies suggest that immune surveillance mechanisms normally control cell plasticity and that tumor cells subvert these extrinsic pressures by adapting developmental programs for survival and metastasis (Laughney et al, 2020).

Recent studies highlight microbiota as another extrinsic inducer of tumor cell plasticity. It is natural to presume that microbiota indirectly influence tumor cell characteristics by altering local inflammation. However, many studies have provided evidence for a direct role for microbes in tumor cell plasticity. For example, microbes produce genotoxins or free radicals, which induces DNA damage (Nougayrède et al, 2006; Goodwin et al, 2011). Additionally, microbes can promote cancer cell proliferation and induce the release of growth factors that increase tumor growth (Watanabe et al, 1998; Cougnoux et al, 2014; Dalmasso et al, 2014; Amieva & Peek, 2016). Studies have also shown that toxins produced by many different microbes may directly influence not only the tumor initiation but also disrupt adhesion between cells, which in turn leads to epithelial‐to‐mesenchymal transition (Wu et al, 1998; Rubinstein et al, 2013).

Tumor plasticity reflects a convergence toward common developmental programs

Large‐scale tumor profiling efforts such as The Cancer Genome Atlas (TCGA) have revealed the molecular portraits of cancers from diverse tissues of origin, providing a valuable resource to study mechanisms involved in tumorigenesis. While pan‐cancer analyses examining TCGA samples across 33 cancer types have found that tumors generally cluster together in a tissue‐ and subtype‐specific manner, squamous cancers are a notable exception (Hoadley et al, 2014, 2018; Campbell et al, 2018). In terms of their transcriptomic, genomic, and epigenomic features, squamous cancers arising in the lung, esophagus, head and neck, bladder, and cervix all appear to utilize a common oncogenic program, which may partly reflect the similarities in the molecular networks of their corresponding normal tissues. In addition, as previously discussed, small‐cell neuroendocrine cancers of the lung and prostate can both result from a reprogramming event following targeted therapy. Just as squamous cancers from different organs utilize similar molecular programs to drive tumorigenesis, the acquisition of RB1 mutations is a shared feature among both reprogrammed lung (NSCLC to SCLC) (Niederst et al, 2015) and prostate tumors (castration‐sensitive to resistant) (Sharma et al, 2010). Indeed, a recent study comparing lung and prostate neuroendocrine tumors revealed the molecular convergence of these two tumor types away from their distinct tissues of origin (Park et al, 2018). These studies collectively suggest that common programs underlie (tumor) cell plasticity and that these programs can operate in a tissue‐agnostic manner (Mollaoglu et al, 2018; Tata et al, 2018b). Although tumor plasticity can drive intra‐ and inter‐tumoral heterogeneity, it paradoxically offers an explanation for the phenotypic and molecular convergence of cancers from distinct organs toward common histological entities (Tata et al, 2018b).

An important challenge is to understand the molecular mechanisms by which these oncogenic programs engender tumor plasticity. As the prototypical example of an organ that gives rise to a vast array of different tumor types, the lung is a valuable system for dissecting the logic of cellular plasticity in cancer. Additionally, through analysis of human NSCLCs and genetic mouse models of lung cancer, a unifying framework has recently been proposed to explain how adenocarcinomas, squamous cell carcinomas, and mucinous adenocarcinomas can all arise from normal lung epithelium (Tata et al, 2018b; Laughney et al, 2020). Unlike adenocarcinomas, which retain their lung identity and are characterized by the expression of canonical lung markers, mucinous adenocarcinomas possess key transcriptomic and morphological characteristics of gastrointestinal tissues (Maeda et al, 2012; Snyder et al, 2013), while lung squamous carcinomas instead resemble the esophagus (Fig 4A). These results suggest that squamous carcinomas and mucinous adenocarcinomas are the result of a lineage conversion away from the lung toward neighboring endodermal organs. In light of its role as a key lung lineage‐specifying transcription factor during embryonic development, it has been proposed that Nkx2‐1 might be the critical “gatekeeper” restricting the lung epithelium from transforming into esophagus‐like squamous carcinomas or intestine‐like mucinous adenocarcinomas (Snyder et al, 2013; Camolotto et al, 2018; Tata et al, 2018b). Accordingly, several studies have shown that loss of Nkx2‐1 leads to mucinous adenocarcinomas when coupled with mutant Kras activation (Maeda et al, 2012; Snyder et al, 2013; Tata et al, 2018b), while instead driving squamous carcinoma when paired with ectopic Sox2 expression (Chen et al, 2014; Mollaoglu et al, 2018; Tata et al, 2018b). When Nkx2‐1 expression is maintained, mutant Kras or Sox2 overexpression led to adenocarcinomas, thus indicating the central importance of Nkx2‐1 in lung cancer as a guardian of lung identity. Notably, low expression of lineage‐specifying transcription factors is consistently associated with lineage plasticity in esophageal, intestinal, breast, and pancreatic cancers (Tata et al, 2018b), indicating that this phenomenon may indeed be a general mechanism by which plasticity arises in tumors from diverse organs.

The “nearest neighbor” model: reimagining tumor plasticity through the lens of development

As described above, studies suggest that most epithelial tumors exhibit remarkable similarity to normal tissues that exist within the same tissue or other tissues/organs. A question that naturally emerges is why tumor tissues would opt for predictable paths that obey developmental principles, rather than acquiring the characteristics of entirely unrelated tissues. For instance, why do lung tumor epithelial cells not acquire the characteristics of neural or skeletal tissues? In a related vein, why are lung squamous cell carcinomas composed of cells that resemble the endoderm‐derived esophagus, but not the ectoderm‐derived skin epidermis or oral epithelium?

The answers to these questions may be embedded within the epigenetic coding of cells. This concept has been described by Jonathan Slack as the “second anatomy”, referring to the internal memory by which an organism maintains the molecular code that was originally established as the organism developed from a single‐celled zygote (Slack, 1986). Based on this concept, we propose that the molecular principles by which different organs are patterned during embryonic development similarly drive tumor plasticity. Akin to how Nkx2‐1 is essential for proper specification and development of the lungs from the primitive gut during embryonic development (Kimura et al, 1999; Minoo et al, 1999), loss of Nkx2‐1 leads to remarkable lineage infidelity in lung cells when coupled with a tumorigenic stimulus. To resolve this crisis of cellular identity, lung cells traverse down alternative paths on the embryonic differentiation landscape, presumably guided by their “second anatomy”, producing tumors composed of non‐lung cell types. In this manner, lung cancers that have lost their lineage‐specifying program will adopt the characteristics of developmentally related “nearest neighbor” organs—for instance, the esophagus, duodenum, or intestine. Importantly, the nearest neighbor model applies not only to lung tissues, but also to several other tissues as well (Fig 4A and B). For instance, loss of CDX1 and CDX2, two key lineage‐specifying transcription factors of mid‐ and hindgut tissues, has been shown to induce changes that lead to loss of intestinal identity and acquisition of esophageal features (Hryniuk et al, 2012; Stringer et al, 2012; Simmini et al, 2014; Tata et al, 2018b). Alternatively, oncogene‐activated cells may first dedifferentiate into a primitive multipotent embryonic progenitor, which in turn give rise to cells that reflect the “nearest neighbor” characteristics. For example, upon activation of hedgehog signaling, esophageal squamous cells dedifferentiate into esophageal embryonic‐like progenitor that ultimately leads to squamous to columnar metaplasia (Vercauteren Drubbel et al, 2021). Similarly, under severe stress or oncogenic events pancreatic acinar cells regain multipotency and undergo acinar‐to‐ductal metaplasia, a key step in the process of pancreatic adenocarcinomas (Pan et al, 2013; Chuvin et al, 2017). Thus, the developmental history of an organ provides a roadmap for the emergence of tumor plasticity, restricting its possible paths within its lineage history (Tata et al, 2018b). In line with this conceptual framework, recent studies have revealed that the endoderm‐specifying factors FoxA1 and FoxA2 cooperatively restrict squamous transformation in established lung cancers that have lost Nkx2‐1 (Camolotto et al, 2018). Similarly, ectopic expression of GATA4 induces columnar cell characteristics in esophageal squamous epithelial cells by directly repressing the transcriptional networks needed for the maintenance of the later cells (Stavniichuk et al, 2021). This implies that developmental memory and cell plasticity principles co‐operate in tissue metaplasia and in tumorigenesis. While we provide some examples to support these models, more work is needed to further scrutinize such models in other tissues across different germ layers.

Conclusions

It appears that nature has evolved a set of common engineering principles to assemble tissues that ultimately meet the three fundamental needs: integrity, communication, and engagement, all of which are necessary to sustain homeostasis and regeneration. Remarkably, distinct tissues have also repurposed these mechanisms to convergently generate similar tissue architectures in response to stimuli.

In regard to cancer therapeutics, the conserved nature of tissue plasticity has both advantages and disadvantages. Presumably, one could target the key molecular nodes underlying a particular metaplastic change and effectively prevent this form of transformation in a wide variety of tissues. In patients with EGFR‐mutant NSCLC receiving targeted EGFR inhibitors, we speculate that it may be beneficial to preemptively inhibit signaling pathways that are critical for SCLC tumors, including EZH2 (Gardner et al, 2017), Notch (Lim et al, 2017), and PKA (Coles et al, 2020). Moreover, inhibitors that target signaling pathways including Wnt, Notch, or TGFβ have been reported as promising candidates for suppressing tumor cell plasticity (Qin et al, 2020). On the other hand, the conserved molecular logic underlying metaplasia necessitates the existence of “side‐effects” to normal tissues elsewhere (for instance, normal squamous tissues present in epidermal or esophageal epithelium). It is likely that such off‐target effects are at play with conventional chemotherapies that do not discriminate pathological versus physiological tissues. Therefore, moving forward it will be important to focus on understanding the differences between normal and pathological tissue patterns, as these differences may reveal new opportunities to target the molecular circuits that exclusively operate in pathological states. Targeting transcription factors (TFs) (alone or in complexes) offers an excellent avenue to modulate cell states. Although some TFs are expressed in multiple cell types or tissues, they act in concert with other nuclear protein complexes that regulate a specific cell‐ or tissue‐specific program. Such combinatorial TF modules offer a potential avenue for targeting specific cell states. For instance, as described above, NKX2‐1 controls lung cancer heterogeneity via its interactions with SOX2 or FOXA1 and FOXA2 and can serve as therapeutic module for targeting the growth and plasticity of specific cancer types. Similarly, identifying the upstream regulators, such as kinases or phosphatases that orchestrate specific combinations of TF protein complexes, can offer new therapeutic targets. For example, studies from mouse models have suggested that loss of LKB1 generated histologically different tumor phenotypes including squamous cell carcinoma, adenocarcinoma, and mucinous adenocarcinoma depending on loss of Stk11 alone or in conjunction with other tumor suppressor genes including PTEN or oncogenic mutations (Kras^G12D) or transcription factors (Nkx2‐1) (Zhang et al, 2017; Mollaoglu et al, 2018; Tata et al, 2018b). Such phenotypic correlations between these upstream and downstream regulators suggest the presence of distinct molecular nodes controlling tumor plasticity. In addition to the classical small molecule‐based approaches, recent advances in PROteolysis TArgeting Chimera (PROTAC)‐based protein degradation approaches offer unique niche to discriminate normal and pathological cell states. Since PROTAC‐based target protein degradation does not lead to a permanent change, they allow for reversible modulation of specific proteins (Neklesa & Crews, 2012; Burslem & Crews, 2020). Additionally, as discussed above, tumor plasticity likely represents a general strategy by which cancers can evolve resistance to therapy. To the extent that cancers arising from distinct organs exhibit differential responses to therapy, it is plausible that different tumor subtypes may interconvert toward alternative cellular fates that impart heightened resistance to a given treatment. Future work examining the acquisition of lineage plasticity in already established tumors may ultimately reveal new approaches to block lineage conversion, potentially eliminating an important avenue for the development of drug resistance. With the advent of next‐generation sequencing based high‐throughput genetic and pharmacological perturbations coupled with cell phenotyping approaches have the potential to chart out genetic circuits and molecular nodes associated with cell plasticity in normal and pathological contexts. More generally, we must further our understanding of the “second anatomy” in order to identify next‐generation molecular targets that specifically regulate pathological cellular plasticity.

Though much has been discovered in recent years, many important unknowns remain (see also Box 1). Nevertheless, the field of cellular plasticity is now well poised to discover fundamental insights into tissue regulation, potentially opening entirely new therapeutic avenues for both malignant and non‐malignant diseases.

Box 1: In need of answers.

1. What are the molecular and cellular mechanisms that distinguish pathological and physiological plasticity? Can these differences be therapeutically targeted in a tissue‐specific manner?

2. To what extent are histologically similar tissues or tumors driven by the same molecular programs? Advanced technologies such as single‐cell transcriptome and proteome profiling may uncover mechanistic basis for cell plasticity and the underlying cellular principles that control “second anatomy”.

3. How are mechanisms that control “second anatomy” intersect with germ layer specifying signaling networks?

4. How can our understanding of the “second anatomy” be harnessed to predict tumor evolution and drug resistance?

5. Does cellular plasticity have a role in transgenerational inheritance of adaptation to certain stress responses? If so, what are the mechanisms that contribute to transgenerational inheritance?

Box 1: In need of answers.

1. What are the molecular and cellular mechanisms that distinguish pathological and physiological plasticity? Can these differences be therapeutically targeted in a tissue‐specific manner?

2. To what extent are histologically similar tissues or tumors driven by the same molecular programs? Advanced technologies such as single‐cell transcriptome and proteome profiling may uncover mechanistic basis for cell plasticity and the underlying cellular principles that control “second anatomy”.

3. How are mechanisms that control “second anatomy” intersect with germ layer specifying signaling networks?

4. How can our understanding of the “second anatomy” be harnessed to predict tumor evolution and drug resistance?

5. Does cellular plasticity have a role in transgenerational inheritance of adaptation to certain stress responses? If so, what are the mechanisms that contribute to transgenerational inheritance?

Conflict of interest

P.R.T serves as a consultant for Cellarity Inc. and Surrozen Inc. on work not related to the contents of this manuscript. The other authors declare no conflict of interest.

EMBO reports (2021) 22: e51921.