Cellular Origins and Lineage Plasticity in Cancer

Abstract

All cancers arise from normal cells whose progeny acquire the cancer-initiating mutations and epigenetic modifications leading to frank tumorigenesis. The identity of those “cells-of-origin” has historically been a source of controversy across tumor types, as it has not been possible to witness the dynamic events giving rise to human tumors. Genetically engineered mouse models (GEMMs) of cancer provide an invaluable substitute, enabling researchers to interrogate the competence of various naive cellular compartments to initiate tumors in vivo. Researchers using these models have relied on lineage-specific promoters, knowledge of preneoplastic disease states in humans, and technical advances allowing more precise manipulations of the mouse germline. These approaches have given rise to the emerging view that multiple lineages within a given organ may generate tumors with similar histopathology. Here, we review some of the key studies leading to this conclusion in solid tumors and highlight the biological and clinical ramifications.

LINEAGE TRACING IN MURINE CANCER MODELS

Genetically engineered mouse models (GEMMs) are the workhorses for studying cancer origins. Most rely on promoters with some degree of tissue- or cell-type-specificity, which is used to define the target population at risk of carcinogenesis. Two early examples include the RIP-Tag model, in which the SV40 large T-antigen was expressed downstream of the rat insulin promoter (Hanahan 1985), and the MMTV/H-Ras transgenic mouse, in which the H-Ras oncogene was expressed downstream of the mammary-specific MMTV promoter (Sinn et al. 1987). These transgenic strains developed pancreatic islet tumors and mammary carcinomas, respectively, reflecting the ability of medium-length (kb) upstream regions to confer tissue-specific oncogene expression (Fig. 1).

Figure 1.

Genetic tools to model cancer in mice. Early genetically engineered models of cancer used transgenes to express oncogenes downstream of a known tissue-specific promoter. More recently, conditional knockout (or knockin) mice achieved cell lineage–specific expression by using Cre recombinase to delete loxP-flanked tumor suppressor genes (TSGs) or to excise loxP-STOP-loxP cassettes upstream of an oncogene. Multiple TSGs and oncogenes can be efficiently targeted with this approach. Tamoxifen-regulated CreER^TM alleles provide both spatial and temporal control to yield inducible knockout or knockin models. Lineage tracing is achieved through similar means with tissue-specific Cre alleles that delete a loxP-STOP-loxP cassette upstream of a reporter construct, often introduced into the Rosa26 (R26) locus. (Figure prepared with BioRender.com.)

Subsequent generations of GEMMs have used the Cre-loxP system, or variants thereof, to achieve spatial and temporal control over gene editing. Cre recombinase is a bacteriophage protein that recognizes specific DNA fragments (loxP sites). Researchers found that if loxP sites were engineered upstream and downstream of critical regions of a gene or regulatory sequence (in a specific configuration), Cre-mediated excision conditionally deleted the intervening DNA (Fig. 1). In this regard, tissue- and cell-type-specific Cre alleles allowed the expression of the recombinase (and hence DNA editing) in targeted cellular compartments of interest. One of the earliest in vivo examples of this approach placed Cre under the control of the interferon-responsive promoter, which in turn deleted the loxP-flanked DNA polymerase B gene (Kühn et al. 1995). Researchers found that Cre recombinase could also be delivered virally, such as the use of a Cre-containing adenovirus to delete the Apc tumor suppressor in the colon, resulting in rapid adenoma development (Shibata et al. 1997). Finally, fusion of the Cre protein to tamoxifen-inducible elements, resulting in “Cre-ER^TM” alleles, permitted temporal control over gene editing (Feil et al. 1996), as mice expressing Cre-ER^TM (or the optimized ER^T2) excise loxP sites only in the presence of exogenously administered tamoxifen (Fig. 1). Complementary systems were subsequently developed, including the related Flp-frt system, which pairs a yeast Flp recombinase with Frt sites, and the Dre-rox system, which pairs a bacteriophage Dre recombinase with rox sites. Such systems can now be combined to edit genes in multiple lineages simultaneously or sequentially (Schönhuber et al. 2014).

In the context of cancer, these inducible and conditional models afforded several opportunities. First, they could be used to misexpress an oncogene (by deleting a transcriptional stop sequence upstream of a mutant oncogenic allele) while simultaneously inactivating a tumor suppressor gene (TSG). Second, many of these systems bypassed the embryonic lethality associated with certain transgenic or gene knockout constructs. Third, these systems gave researchers the ability to select the cellular target of oncogene/TSG misexpression by selecting the Cre strain to be used. A prototypical example of this last point, discussed in greater detail in the sections below, has been the simultaneous activation of the Kras oncogene and deletion of the Tp53 TSG to give rise to diverse cancer models depending on the specific cell type targeted.

In parallel with the application of Cre-loxP technology to cancer, developmental biologists began using these tools to perform “lineage tracing”—a technique for studying cellular ancestries—in mouse embryos. Lineage tracing involves the conditional and heritable marking of discrete cellular populations whose resulting pedigrees could then be observed. Using so-called “reporter” strains, in which Cre recombination induces the irreversible expression of a fluorescent protein or β-galactosidase, Cre-loxP technology became a genetic means of determining lineage (Fig. 1; Lobe et al. 1999; Novak et al. 2000). Such approaches, combined with an ever-growing collection of mice bearing tailored loxP or Frt insertions in cancer-related genes, now allow investigators to refine the cell types and timing with which oncogenic mutations are introduced, while simultaneously providing an ability to track what becomes of those cells.

Plasticity as a Confounder in Cell-of-Origin Studies

Historically, tumors have been branded and treated based on two characteristics: the tissue in which they arise (i.e., breast, lung, liver, etc.) and their histopathological appearance. Such histological distinctions are valuable for distinguishing tumors that differ significantly from an embryological or clinical standpoint (e.g., carcinomas vs. sarcomas). However, the microscopic appearance of malignant cells is somewhat less useful when it comes to defining, with precision, the normal cell within a tissue from which a tumor arises—its “cell-of-origin.” The reason for this is “cellular plasticity.”

Plasticity refers to a feature of cells whereby they swap one molecular and functional identity for another. It is common during development, when cells must proceed through successive rounds of differentiation before attaining their eventual specialized identities. There is now an increasing awareness that cellular plasticity also occurs frequently in adult organs, where it is an adaptive response to tissue injury also known as “metaplasia” (Merrell and Stanger 2016; Bakir et al. 2020). Importantly, the molecular alterations accompanying metaplasia can result in dramatically different histological appearances; consequently, the microscopic and even anatomic designation of a tumor is not a reliable barometer of cellular history. This fact has important implications for cancer, as naming conventions are often interpreted as indicating that a tumor's cellular source is known. For example, one might reasonably assume that “pancreatic ductal adenocarcinoma” (PDAC) arises from the duct cells of the pancreas, “cholangiocarcinoma” (CCA) arises from the cholangiocytes of the liver, and “ovarian cancer” (OC) arises from some cell type within the ovary. But as described below, such seemingly straightforward assumptions can be misleading (Fig. 2).

Figure 2.

Plasticity as a confounder in cell-of-origin studies. The most straightforward or predicted cell-of-origin for a tumor may not be the only one (or even the most prevalent one). In pancreatic ductal adenocarcinoma (PDAC), mouse studies show that many if not most tumors arise from acinar cells, not ductal cells, because of acinar-to-ductal metaplasia (ADM). In cholangiocarcinoma (CCA), often referred to as “bile duct cancer,” the cholangiocytes that line the bile ducts were presumed to be the sole source of these tumors. But several studies suggest that a subset of CCAs are hepatocyte-derived following hepatobiliary metaplasia. And emerging evidence suggests that many if not most ovarian cancers are not derived from the ovary at all, but rather from the Müllerian cells of the fallopian tube. (Figure prepared with BioRender.com.)

CELL-OF-ORIGIN STUDIES IN SPECIFIC CANCER TYPES

Pancreas Cancer

The cellular origin(s) of pancreas cancer has been the focus of intense investigation. The pancreas consists of two compartments—exocrine and endocrine—each of which is believed to give rise to distinct types of tumors. The exocrine compartment, consisting of the acinar cells that produce digestive enzymes and the duct cells that carry pancreatic juice to the gut, comprises the bulk of the organ's mass, while endocrine cells, situated within the islets of Langerhans, produce hormones such as insulin. PDAC, the most common and most lethal pancreatic malignancy, is presumed to arise from the exocrine compartment, while pancreatic neuroendocrine (NE) tumors (PNETs), a less common and less lethal form of pancreas cancer, are presumed to arise from the endocrine component.

There is strong evidence from both mouse and human studies that PDACs evolve from dysplastic lesions known as pancreatic intraepithelial neoplasias (PanINs). PanINs, in turn, are thought to arise from acinar cells following a process of acinar-to-ductal metaplasia (ADM), wherein acinar cells transdifferentiate into a duct-like state. ADM is part of the normal pancreatic response to injury (pancreatitis), which is a risk factor for PDAC. Accordingly, malignant progression in PDAC GEMMs is markedly accelerated by experimental pancreatitis (Guerra et al. 2007; Carrière et al. 2009). Treatment with the cholecystokinin (CCK) analog cerulein, or pancreatic ductal ligation, can induce pancreatitis and ADM in the mouse, which can be further enhanced by a “desmoplastic” stroma (Liou et al. 2013; Liu et al. 2016; Pitarresi et al. 2016). But as ADM occurs at a lower frequency in human PDAC than it does in the mouse, there remains uncertainty regarding the generalizability of an ADM → PanIN → PDAC sequence in patients. Nevertheless, these observations, and the lineage-tracing studies described below, have led to a consensus that many if not most PDACs arise from acinar cells that have undergone ADM.

During ADM, acinar cells lose expression of acinar-specific transcription factors and digestive enzymes and begin to express a variety of ductal cell markers. Single-cell RNA-seq experiments suggest that acinar cells are likely a heterogenous population, which may influence the degree to which they are able to undergo ADM and form PDAC (Wollny et al. 2016; Schlesinger et al. 2020; Tosti et al. 2021; Ma et al. 2022). Work in the 2000s indicated that broad activation of mutant Kras in the pancreas could result in ADMs, PanINs, and, eventually, PDAC (Grippo et al. 2003; Hingorani et al. 2003). Subsequent experiments targeting mutant Kras and Tp53 to the entire pancreatic epithelium were shown to result in invasive PDAC with multiple features of the human disease (Hingorani et al. 2005). Variations in this GEMM have been used in the many PDAC cell-of-origin studies that followed, where a strategy of directing these two mutant alleles to discrete cellular compartments allowed an evaluation of the malignant potential of each cell population.

An important study by Kopp et al. (2012) directly compared the ability of ductal or acinar cells to give rise to PanINs by misexpressing the mutant Kras allele in either the acinar compartment (via Ptf1a-Cre) or the ductal compartment (via Sox9-Cre). Whereas ductal activation of mutant Kras resulted in relatively few PanINs, acinar activation resulted in substantial PanIN formation, suggesting that acinar cells are more competent than ductal cells to initiate pancreatic neoplasia. Consistent with this finding, activation of Kras and/or Notch signaling specifically in acinar cells also led to ADMs and PanINs (De La et al. 2008; Habbe et al. 2008). Beyond influencing PanIN formation, the cellular origins of PDAC also have implications for disease severity, as concomitant activation of Kras and loss of Tp53 in ductal cells resulted in a more rapid progression to PDAC compared to activation of these mutations in acinar cells (Lee et al. 2019). However, acinar cell–initiated tumors grew more rapidly, suggesting that growth and not time to initiation was the dominant mechanism at play in this context. Cell-of-origin may also influence PDAC subtype, as a ductal origin has been associated with the highly aggressive (squamous/basal-like) PDAC subtype, while an acinar origin has been associated with a less aggressive (classical) subtype (Flowers et al. 2021).

Other studies have examined the potential of additional cellular subpopulations to initiate PDAC. For example, pancreatic centroacinar cells (CACs), which reside at the intersection of acini and ducts, were deemed candidate cells-of-origin for PDAC based on their unique anatomic position and ability to respond to loss of the Pten TSG (Stanger et al. 2005). Similarly, activation of Kras in Nestin⁺ “pancreatic progenitor cells” resulted in PanIN formation (Carrière et al. 2007). In another study, β cells were found to be competent to give rise to PanIN lesions and PDAC in the setting of Kras activation and chronic injury, whereas Kras activation in pancreatic β cells had minimal effects on its own (Gidekel Friedlander et al. 2009). Interestingly, targeting of mutant Kras to Keratin19⁺ (Krt19⁺) ductal cells either failed to induce neoplasia (Brembeck et al. 2003) or resulted in rare mucinous ductal metaplasia (Ray et al. 2011). Collectively, these studies implicate acinar cells as the major, but not the sole, source of PDAC. Given that most PDAC tumors harbor ductal rather than acinar features, this conclusion serves as a reminder that histology is a poor indicator of a tumor's cellular origins.

PNETs—so named because they share features of hormone-producing endocrine cells and neurons—comprise the second-most-common form of pancreatic cancer. The first genetically engineered PNET model placed the SV40 large T-antigen (Tag) under the control of the Rat Insulin-2 promoter (RIP), which drives expression exclusively in the β-cell lineage (Hanahan 1985; Gahete et al. 2019). Similar approaches followed, relying on different cell-type-specific promoters and oncogenic drivers. For example, RIP-MyrAkt1 animals developed β-cell hyperplasias, dysplasias, and insulinomas (Tuttle et al. 2001; Alliouachene et al. 2008) while Glu-Tag mice (bearing Tag under control of the α-cell-specific Glucagon promoter) formed glucagonomas (Efrat et al. 1988; Lee et al. 1992). These findings suggest that PNETs possess a more straightforward lineage relationship to their hormone-producing counterparts in pancreatic islets. Nevertheless, the observation that NE transdifferentiation is a common mechanism of resistance to chemotherapy in multiple cancers (Boumahdi and de Sauvage 2020) raises the possibility that significant cellular plasticity, yet unrecognized, may also contribute to the initiation of this class of tumors.

Liver Cancer

Like the pancreas, the liver contains two main cellular compartments: (1) hepatocytes, metabolically active cells that produce various serum proteins and bile, and (2) cholangiocytes, or biliary epithelial cells (BECs), which carry bile from the liver to the intestine. Accordingly, the two major types of liver cancer—hepatocellular carcinoma (HCC) and cholangiocarcinoma (CCA)—have long been presumed to arise from hepatocytes and BECs, respectively.

HCC, the most common liver malignancy, is associated with chronic liver damage (hepatitis), leading to an array of oncogenic mutations. Several lines of evidence from mouse modeling indicate that most HCCs arise from hepatocytes, the presumptive cell-of-origin. For example, GEMMs in which mutations in Tsc1/Sqstm1 (Umemura et al. 2016), Stat3/IL-6 (Zhou et al. 2017), or Tp53 (Tschaharganeh et al. 2014) were exclusively directed to hepatocytes all developed HCC. Likewise, studies combining lineage tracing with carcinogen-induced HCC also indicated that mature hepatocytes could give rise to HCCs and hepatocellular adenomas (Shin et al. 2016). Thus, HCC may represent a case in which the most straightforward prediction of cellular origins is the correct one.

In addition to hepatocytes, so-called hepatic progenitor cells (HPCs) have been evoked as a potential source of HCC (Schneller and Angel 2019). Early work in carcinogen-treated rats focused on HPCs as the origin of HCC (Farber 1956; Fausto 1990; Brown et al. 2018), a view that was reinforced by several genetic models (Tang et al. 2008; Lu et al. 2010; Villanueva et al. 2012; Mokkapati et al. 2014; Nikolaou et al. 2015; Matsumoto et al. 2017). However, controversy regarding the definition and identity of HPCs, along with the challenge of targeting mutations specifically to this cell population, has made it difficult to reach strong conclusions regarding the contribution of HPCs to liver cancer. Indeed, there is evidence that the presence of a progenitor signature in a tumor does not reflect a progenitor cell of origin but rather a dedifferentiation of tumor cells derived from the hepatocyte lineage (see Boxes 1 and 2; Mu et al. 2015).

BOX 1. WHY SHOULD WE CARE ABOUT THE CELLULAR ORIGINS OF CANCER?

A well-meaning oncologist or cancer biologist might well ask why it is important to identify the particular cell or cells within a normal tissue that are competent to give rise to cancer. After all, patients present with lesions, either premalignant or frank cancer, and knowing the specifics of the lesion's origins would seem to have little to offer in terms of treatment. We would argue that just as a cancer's tissue of origin—whether carcinoma of the breast, pancreas, or colon—has significant implications for tumor biology and therapy that extend far beyond anatomic site, the same principle applies to a tumor's cell-of-origin. Specifically, pancreas cancers that arise from acinar cells versus duct cells are likely to retain epigenetic, metabolic, and other phenotypic characteristics of the initiating cell that will influence the spectrum of vulnerabilities present within each tumor. As tissue-of-origin is the dominant factor determining a solid tumor's molecular features (Hoadley et al. 2018), a similar logic is likely to apply at the cellular level across cancer, leading to more precise therapeutic opportunities taking cell-of-origin into account.

Knowledge regarding cell of origin is also relevant for cancer-prevention strategies. This concept is best exemplified in the context of OC, where the emerging concept is that cells within the fallopian tubes, rather than the ovary itself, carry significant malignant potential. The implication is that prevention strategies for women at high risk of developing OC (e.g., carriers of BRCA1 or BRCA2 mutations) should include prophylactic fallopian tube removal (salpingectomy) in addition to, or instead of, ovary removal (oophorectomy) (www.nature.com/articles/d41586-021-03717-7). Studies are now underway to test this approach in BRCA carriers (NCT02321228, NCT02760849, NCT04294927). A better understanding of cancer-susceptible cell populations could, in other settings, lead to similar preventive efforts.

BOX 2. DISTINGUISHING THE CANCER CELL-OF-ORIGIN FROM THE CANCER STEM CELL.

An important distinction, often overlooked in the field, is the difference between the cellular origins of cancer and cancer stem cells (CSCs). The former (the subject of this article) describes the normal tissue-resident cells that give rise to premalignant lesions and/or cancer, while the latter refers to a specialized subset of cells within a tumor. Specifically, the CSC hypothesis is based on the notion that only certain cancer cells are responsible for propagating the tumor, with most cancer cells having a more limited capacity to divide. While some cancers may arise from undifferentiated tissue-resident stem cells (see section on Colorectal Cancer), existing evidence suggests that most cancers arise from cells with at least some specialized features. Importantly, neither scenario—a stem cell origin nor a non–stem cell origin—resolves the question of whether CSC function drives a given tumor's growth (Rycaj and Tang 2015).

CCA is less common than HCC but shares the same risk factor of chronic liver injury. Like the early PDAC models, in which cancer-causing mutations were directed to all lineages of the pancreas, early CCA models employed Albumin-Cre mice to generate liver-specific mutations. However, this strain mediates recombination in hepatoblasts, the bipotential cell that gives rise to all lineages in the liver (i.e., hepatocytes and BECs) (Stanger 2015). Consequently, CCA models generated using this strain leave open the possibility that CCA might have arisen from hepatocytes, BECs, or both (Xu et al. 2006; O'Dell et al. 2012; El Khatib et al. 2013; Zender et al. 2013; Saha et al. 2014; Ikenoue et al. 2016).

Cre strains specific to the BEC or hepatocyte lineage have proven more useful. One model, using lineage tracing to mark BECs, indicated that cholangiocytes can serve as the cells-of-origin origin of CCA in the setting of Tp53 deletion and inflammation (Guest et al. 2014). Conversely, other lineage-tracing studies indicated that hepatocytes give rise to CCA in the setting of Notch and AKT activation (Fan et al. 2012; Sekiya and Suzuki 2012). Hydrodynamic tail vein injection has been another means of targeting hepatocytes, and this method too supports a hepatocyte origin for CCA (Wang et al. 2018). Thus, it is likely that CCA can arise from either hepatocytes or BECs. Given that hepatocytes may give rise to either CCA or HCC, it is possible that the spectrum of acquired oncogenic mutations determines how hepatocyte-derived tumors manifest histologically (Vicent et al. 2019).

Esophageal Cancer

Esophageal cancers fall into two major categories—esophageal adenocarcinoma (EAC) and esophageal squamous cell carcinoma (ESCC). EAC, the more common form in the United States and Western Europe, often arises in the setting of Barrett's esophagus (BE), a metaplasia in which the normal squamous epithelium of the esophagus is replaced by a columnar intestine-like epithelium. BE can serve as a precursor to EAC, as evidenced by the progression of BE from low-grade to high-grade dysplasia and from high-grade dysplasia to carcinoma. There is also evidence to suggest that EAC can arise independently of BE (Sawas et al. 2018). The source of the metaplastic cells is controversial. Specifically, BE (and EAC) has been postulated to originate from the gastric cardia (a transitional zone between the stomach and esophagus) (Paull et al. 1976; Hu et al. 2007; Quante et al. 2012; Lavery et al. 2014; Vega et al. 2014), native esophageal cells (Hu et al. 2007), submucosal gland cells (Coad et al. 2005; Leedham et al. 2008; Lörinc et al. 2015; von Furstenberg et al. 2017), residual embryonic stem cells (Wang et al. 2011), or transitional basal cells (Jiang et al. 2017).

Recently, the debate over the cellular origin(s) of BE has been informed by studies of human esophageal tissue. In one study, significant transcriptional overlap was observed between BE samples and submucosal gland cell profiles (Owen et al. 2018). As mice lack esophageal submucosal glands (Macke et al. 2011), these findings cannot be confirmed in the mouse and highlight an important limitation of murine lineage-tracing approaches. More recently, an integrative approach—incorporating single-cell transcriptomics, DNA methylation, and somatic mutation analysis—suggested that BE originates in the gastric cardia through aberrant activation of c-MYC and HNF4A transcriptional programs (Nowicki-Osuch et al. 2021), consistent with other studies (Quante et al. 2012; Singh et al. 2021).

Within the squamous epithelium of the esophagus, several studies have shown that a subpopulation of esophageal squamous basal cells serve as tissue-resident stem cells (DeWard et al. 2014; Giroux and Rustgi 2017; Hayakawa et al. 2021). Early genetic models indicated that basal cells are a source of squamous dysplasia (Nakagawa et al. 1997) and ESCC (Opitz et al. 2002; Stairs et al. 2011), the latter representing the first genetic mouse models of ESCC. This was followed by lineage tracing, demonstrating that a subset of basal cells gives rise to differentiated suprabasal cells that can transform into ESCC (Liu et al. 2013). These and other studies (Matrka et al. 2018; Tang et al. 2021) strongly implicate squamous basal cells as a major cell-of-origin for ESCC.

Lung Cancer

Lung cancer encompasses a diverse set of histological subtypes with discrete driver mutations that fall into two major subgroups: non-small-cell lung cancer (NSCLC) and small-cell lung cancer (SCLC). NSCLC is further divided into adenocarcinoma (LUAD), squamous-cell carcinoma (LSCC), and large-cell carcinoma (LCC) based on histopathological features. Cellular plasticity is a striking characteristic of lung cancer. Clinically, plasticity manifests itself as the conversion of NSCLC to SCLC as a mechanism of drug resistance; in addition, tumors bearing features of both subtypes can exist (Adelstein et al. 1986; Mangum et al. 1989; Oser et al. 2015). Mouse models have provided insight into the role of different oncogenes/TSG combinations in the lung's many epithelial cellular compartments, including basal cells, alveolar type 1 or 2 (AT1 and AT2) cells, and Club (formerly Clara) cells.

LUAD is the most common form of NSCLC and is frequently associated with activating mutations in KRAS (Eberhard et al. 2005; Pao et al. 2005; Riely et al. 2008). Consequently, many LUAD mouse models rely on conditional activation of mutant Kras, which can be directed to discrete lineages via Cre-lox technology (as discussed above for PDAC). Early mouse models targeting Kras to all lung epithelial cells (via Cre delivered by adenovirus) led to LUAD-like tumors in the lung periphery and the bronchiole/alveoli border in cells expressing both AT2 and Club cell markers (Jackson et al. 2001; Meuwissen et al. 2001; Kim et al. 2005). More refined models, in which mutant Kras was restricted to AT2 cells, Club cells, or bronchioalveolar stem cells (BASCs), revealed that only AT2 cell–targeted Kras led to LUAD, lending support to the notion that AT2 cells serve as a major cell-of-origin (Rawlins et al. 2009; Sutherland et al. 2011, 2014; Xu et al. 2012; Mainardi et al. 2014). Similar results were obtained in mouse models that used Egfr activation, or included loss of Tp53, loss of Ink4a/Arf, or loss of Lkb1 to initiate tumorigenesis (Fisher et al. 2001; Politi et al. 2006; Han et al. 2014).

Importantly, cellular plasticity can confound some of these inferences. For example, Kras activation with concomitant Lkb1 deletion in AT2 cells shifted tumors toward an adenosquamous phenotype (Han et al. 2014). Likewise, single-cell transcriptomic analysis of lineage-traced Club cells revealed that they can adopt an AT2-like phenotype after oncogenic transformation, a finding compatible with a Club cell origin for LUAD (Chen et al. 2022). Further support for this idea comes from the finding that tobacco-exposed mice bearing Kras mutations in the airway go on to develop tumors that originate (by lineage tracing) from Club cells (Spella et al. 2019). Concordantly, ablation of Club cells prevented such carcinogen-induced lung tumors (Spella et al. 2019). Taken together, these studies suggest that Club cells, in addition to AT2 cells, can serve as LUAD cells-of-origin, due in part to the lineage plasticity between these two cell types. In this respect, LUAD bears a resemblance to PDAC and CCA, where lineage plasticity between acinar cells and ducts or hepatocytes and BECs, respectively, allows multiple cell types to serve as cellular substrates for histologically similar tumors.

Lung squamous cell carcinoma (LUSC), the second-most-common subtype of lung cancer, is associated with a diverse set of driver mutations and is marked by the expression of basal cell lineage genes p63, Sox2, and Keratins Krt5 and Krt14 (Cancer Genome Atlas Research Network 2012; Campbell et al. 2016; Pan et al. 2021). Similar to LUAD, targeting various LUSC-associated oncogenic drivers broadly throughout the lung epithelium was sufficient to drive tumorigenesis but gave little insight into the cell of origin (Mukhopadhyay et al. 2014; Xu et al. 2014). More refined approaches targeting basal, AT2, or Club cells have not resolved the issue, with some studies reporting that all three lineages have the potential to give rise to LUSC (Ferone et al. 2016) and other studies reporting that only the Club cell lineage is permissive for LUSC formation (Ruiz et al. 2019). Mixed adenosquamous lesions have also been reported in mice using various genetic drivers, further complicating the cellular origins of LUSC and emphasizing the high degree of cellular plasticity in lung epithelial cells (Camolotto et al. 2018). Notably, these studies employed different genetic drivers to model LUSC, which may account for some of the discrepancies. Despite the caveats of these mouse models, there is clinical evidence that LUADs can undergo a histological transformation to LUSC as a result of therapeutic intervention (Schoenfeld et al. 2020).

Thus, despite many years of study, there remains limited consensus as to whether differences among the subtypes of NSCLC are related to differences in cellular origins, differences in oncogenic drivers, both, or neither.

SCLC, as noted above, can emerge de novo or following a treatment-induced conversion from NSCLC to SCLC, a recognized mechanism of resistance to EGFR-inhibition therapy (Zakowski et al. 2006; Sequist et al. 2011; Yu et al. 2013). Before the era of EGFR-targeted therapies, a small percentage of NSCLC patients treated with chemo- or radiotherapy also developed SCLC at relapse (Adelstein et al. 1986), indicating that NSCLC to SCLC conversion is a more generalizable mechanism of therapeutic resistance. Tumors can also present with combined SCLC and NSCLC histology, complicating the interpretation of NSCLC to SCLC conversions that may be misinformed from limited initial diagnostic biopsy material (Adelstein et al. 1986; Mangum et al. 1989; Oser et al. 2015). Because of this substantial plasticity between small- and large-cell tumors, extra care is needed for the interpretation of cell-of-origin studies in SCLC.

Because SCLCs arising de novo express a variety of NE markers (and, as their name suggests, exhibit histological features of small NE cells), SCLCs have long been presumed to arise from the pulmonary NE compartment. This relationship is supported by experiment, as the inactivation of disease-relevant TSGs in NE, AT2, and Club cell lineages revealed NE cells as the most permissive for transformation (Meuwissen et al. 2003; Park et al. 2011; Sutherland et al. 2011; Song et al. 2012; Semenova et al. 2015). However, the conclusion that SCLC arises from the lung's NE compartment remains tentative, as the ability of other epithelial cell compartments to initiate SCLC has not been fully investigated. Adding to this uncertainty is the emerging notion that SCLC is not a single disease but is instead comprised of molecular subtypes (George et al. 2015; Borromeo et al. 2016; Mollaoglu et al. 2017; Rudin et al. 2019).

Colorectal Cancer

Colorectal cancers (CRCs) arise from neoplastic precursor lesions (typically, adenomatous polyps), with the prevailing view that a stem cell or stem cell–like compartment constitutes the most prevalent cellular origin (Barker et al. 2009; Nassar and Blanpain 2016; Dekker et al. 2019). Intestinal stem cells reside in the colonic crypt and are competent to initiate neoplastic transformation that eventually proceeds to carcinoma (Barker et al. 2007, 2009). APC mutations are believed to be the first step of carcinogenesis in most CRCs; consequently, mouse models of intestinal cancer have relied on APC inactivation, which aberrantly activates WNT signaling. Apart from Apc, the perturbation of other genes has yielded only minor phenotypes with very long latencies suggesting that WNT disruption is a critical node in CRC tumorigenesis. Remarkably, reactivation of APC in APC/Tp53/Kras triple-mutant mice with established CRC causes a reversion back to histologically normal tissue (Dow et al. 2015).

Regarding the origin of CRCs, intestinal stem cells (marked by diverse markers: Lgr5⁺ (Barker et al. 2009), Lrig⁺ (Powell et al. 2012, 2014), CD133/Prom1⁺ (Zhu et al. 2009), Bmi1⁺ (Sangiorgi and Capecchi 2008), or Krt19⁺ (Asfaha et al. 2015) can give rise to intestinal cancers, with most mouse-modeling approaches resulting in adenomas in the small intestine (Huels and Sansom 2015). Colon tumorigenesis can be favored by using alternative Cre drivers, such as Cdx2-Cre, or by inducing panintestinal Cre drivers in the colon through a 4-OHT enema (Feng et al. 2013; Boutin et al. 2017).

In humans, the finding that in kindreds with familiar CRC, crypts composed entirely of dysplastic cells can be observed has been used in support of a stem cell–based “bottom-up” histogenesis model (Preston et al. 2003). By contrast, a “top-down” model of CRC has been proposed based on the observation that dysplastic cells with APC mutations reside mainly in surface epithelial cells, removed from the stem cells at the crypt base (Shih et al. 2001). More recent work suggests that cells outside of the stem cell niche can initiate colonic tumorigenesis (Davis et al. 2015). Thus, as with several other tumor types discussed already, it appears that CRCs may not arise from a single cellular source; rather, various cellular compartments may give rise to CRC under the right molecular (and perhaps microenvironmental) conditions.

Ovarian Cancer

Ovarian cancers (OCs) comprise a collection of related tumors classified based on pathological or genetic features, the most common being epithelial ovarian carcinomas (Kurman and Shih Ie 2010). Given this heterogeneity, and the absence of obvious precursor lesions like PanINs or adenomatous polyps, the cellular origin of OC has been the subject of much debate, with two principal theories emerging (Bell and Scully 1994; Vaughan et al. 2011; Bowtell et al. 2015). The first suggested that the ovarian surface epithelium (mesothelium) undergoes metaplasia, giving rise to epithelial cells with the ability to develop into diverse tumor types (with serous, endometrioid, clear-cell, mucinous, and transitional tumors comprising various subtypes of the disease). The second theory suggested that tumor cells with Müllerian phenotypes (serous, endometroid, and clear cell) are derived from Müllerian ducts, an embryonic lineage that gives rise to the fallopian tubes, uterus, and cervix, but not the ovaries (Fathalla 1971; Lauchlan 1972; Dubeau 1999; Vaughan et al. 2011).

The view that many OCs have their origins outside the ovary has gained credence over the past decade and is now the prevailing dogma (Piek et al. 2003; Kindelberger et al. 2007; Kurman and Shih Ie 2010; Kuhn et al. 2012; Erickson et al. 2013; Perets et al. 2013; Perets and Drapkin 2016). Early findings in support of this model targeted genetic drivers to the fallopian tube lineage, resulting in ovarian tumors resembling high-grade serous carcinoma (HGSC) (Kim et al. 2012; Perets et al. 2013; Sherman-Baust et al. 2014). Competence of the fallopian tube lineage to give rise to OC was confirmed when deletion of either Brca1 or Brca2 along with Tp53 and Pten in fallopian tube cells resulted in HGSC or its rare precursor, serous tubal intraepithelial carcinoma (STIC) (Finch et al. 2006; Powell et al. 2011; Perets et al. 2013). Consistent with these findings, expression of the SV40 Tag under the control of the Müllerian-specific Ovgp-1 promoter resulted in a full spectrum of OC morphological phenotypes, including STIC and metastatic HGSC (Sherman-Baust et al. 2014). Thus, OC represents another example of how long-held notions regarding a tumor's origin are being reconsidered in light of murine lineage-tracing data.

Lineage Plasticity in Other Cancers

As the above examples show (Fig. 3), cellular plasticity can confound a determination of the cancer “cell-of-origin.” Plasticity is commonplace in other tumor types as well. In the prostate, for example, deletion of the Pten TSG in either the basal or luminal population led to histologically similar neoplastic lesions, indicating that cellular plasticity led to a convergence of histopathological phenotypes (Wang et al. 2013). Plasticity reemerges as a mechanism of resistance, as histological transformation to an NE phenotype was found to confer resistance in both mouse models of prostate cancer (Zou et al. 2017) and in up to 20% of patients treated with androgen-blocking therapies (Aggarwal et al. 2018) like the SCLC/NE-like conversion accompanying therapy resistance in NSCLC. Similar changes in cell phenotype confer therapy resistance in bladder cancer, where partial squamous differentiation (“semi-squamatization”) is associated with loss of sensitivity to chemotherapy (Wang et al. 2022). In this case, treatment with a cathepsin inhibitor, which induces a full squamous differentiation, was associated with a return to sensitivity. Cellular plasticity can also drive heterogeneity in bladder cancer without the pressure of therapy, as described in bladder cancer (Sfakianos et al. 2020; Warrick et al. 2022). Thus, in a variety of settings, cellular plasticity is a major source of phenotypic variation in cancer, which in turn can allow tumor cells to circumvent the lethal effects of therapy (Box 3).

Figure 3.

Plasticity at the cellular level. Cellular plasticity is evident in tumorigenesis across multiple organs. In pancreatic ductal adenocarcinoma (PDAC), acinar cells transdifferentiate into ductal-like cells. In cholangiocarcinoma (CCA), hepatocytes form cholangiocyte-like cells. In high-grade serous carcinoma (HGSC) of the ovaries, Müllerian cells become serous tubal cells. (Figure prepared with BioRender.com.)

BOX 3. GLOSSARY OF TERMS.

Cancer stem cells (CSCs)

A putative population of cells within a tumor that possesses the capacity to propagate future tumor growth. In practice, CSC activity is generally measured by measuring the tumor-initiating potential of a cell after transplantation.

Cell-of-origin

The cell (or cells) within a normal tissue that gives rise to a tumor. Tumors may arise from tissue-resident stem cells or acquire a stem cell–like phenotype through plasticity.

Cholangiocarcinoma

Cancer of the biliary tract, thought to arise from either cholangiocytes/biliary epithelial cells (BECs) or hepatocytes.

Genetically engineered mouse models (GEMMs)

Cancers arising in mice with alterations in the genome (e.g., transgenes, gene “knockins,” or gene “knockouts”), in contrast to implantable mouse models or patient-derived xenografts in which already-established tumor cells are injected into a mouse.

Lineage tracing

The use of cell-marking methods to determine cellular ancestry in cell-of-origin studies. Common techniques in the mouse include Cre-lox, Dre-rox, and Flp-frt methods to induce irreversible recombination events in the DNA of a particular cell type.

Metaplasia

A form of “plasticity” whereby cells are interconverted from one histological type to another. Examples include acinar-to-ductal metaplasia (ADM) in the pancreas and Barrett's transformation (squamous-to-intestinal metaplasia) in the esophagus.

CONCLUDING REMARKS

Just as it is hard to tell from a full-grown plant what its seed must have looked like, it is difficult to establish the cellular origins of a tumor by examining it at end-stage. By marking various cell populations and assessing their malignant potential, lineage tracing in mice has revealed that cancers of a single histological type can arise from distinct cellular compartments. Accordingly, tumors arising in a given tissue need not derive from a unique cellular source; rather, multiple cell types likely share the competence to give rise to histologically and clinically similar tumors. The malignant competence of a particular cell (with its unique epigenetic history) may depend on the spectrum of oncogenic mutations it experiences along with environmental insults. Under such a scenario, cancer becomes a product of which mutations befall which cells, although our understanding of the interplay between cell and mutation that gives rise to cancer remains limited.

The picture is even hazier in human beings, where the lineage-tracing tools that have proven so useful in mice and other model organisms are unavailable. Hence, researchers studying the cellular origins of human cancer have had to rely on indirect approaches to infer lineage, such as cellular methylation patterns (Bormann et al. 2018). But as we saw earlier, cellular plasticity can give a false impression of a tumor's origins, and such plasticity may extend to epigenetic states, which may change in parallel with histological states. Consequently, it may be some time before scientists can confirm whether the cellular lineage relationships being mapped out in mice also apply to human cancer.

For nearly four decades, mouse models have been a critical tool for targeting specific mutations in defined cell populations, thus clarifying the role of lineage plasticity in cell-of-origin studies. More recent studies using refined mouse modeling approaches to model intraclonal plasticity (Rios et al. 2019) have shown promise for deconvoluting the often complex de novo tumor initiation across multiple cellular compartments. These approaches, combined with a deeper understanding of cellular plasticity in humans using single-cell RNA sequencing, will likely broaden our definitions of lineage plasticity in cancer as we continue to discover new cell types and transitional states.