Lineage factors and differentiation states in lung cancer progression

Abstract

Lung cancer encompasses a heterogeneous group of malignancies. Here, we discuss how the remarkable diversity of major lung cancer subtypes is manifested in their transforming cell of origin, oncogenic dependencies, phenotypic plasticity, metastatic competence, and response to therapy. More specifically, we review the increasing evidence that potentially links this biological heterogeneity to the de-regulation of cell lineage specific pathways and the transcription factors that ultimately control them. As determinants of pulmonary epithelial differentiation, these poorly characterized transcriptional networks may underlie the etiology and biological progression of distinct lung cancers, while providing insight into innovative therapeutic strategies.

Pathologists have long been familiar with the existence of various lung cancer histological subtypes and the challenges in accurately classifying them. Thoracic cancer can be divided into two major histotypes: small cell lung cancer (SCLC) and non-small cell lung cancer (NSCLC). Most lung cancer patients are diagnosed with NSCLC, which can be further subclassified into adenocarcinoma (LUAD), squamous cell carcinoma (LUSC) and large cell carcinoma (LCC). Updated histological sub-classes of LUAD have been proposed for diagnosis and treatment¹. As revealed by recent genome-wide studies, individual lung tumors also possess highly diverse genomes, which further underscore the biological complexity of these diseases. Untangling the relationship between lung tumor histotype, phenotype and molecular heterogeneity represents a critical barrier to improving the clinical outcome of thoracic malignancies, which collectively account for most cancer-related deaths to date².

Lung cancer: pleiotropic origins and drivers

Current views on the cellular origins of different lung cancers are shaped by our understanding of normal pulmonary development and homeostasis. The lung epithelium arises from the ventral side of the anterior foregut endoderm, where primary lung buds are formed³. Following extensive branching of the proximal conducting airways including the trachea, bronchi and bronchioles, cells at the distal branch tips differentiate into alveolar type I (AT1) and II (AT2) cells which constitute the gas-exchanging alveoli. In the developing and adult lungs, multiple regional epithelial cell types can serve as pools of progenitor cells. In the trachea and main bronchi, basal cells give rise to secretory and ciliated cells of the luminal layer, whereas in bronchiolar epithelium, club cells (formerly known as Clara cells) can self-renew and generate ciliated cells. In the distal airway, AT1 and AT2 cells arise directly from a bipotent progenitor during embryogenesis⁴. A recent single-cell transcriptome analysis identified several maturation intermediates during this process⁵. In the post-natal lungs, AT2 cells also acquire progenitor-like functions to generate AT1 cells^{4, 6}. Following severe injury and inflammation, distal epithelial regeneration can also occur from putative lineage negative stem cells^{7, 8}.

Having compiled a working catalogue of pulmonary epithelial stem/progenitor lineages, recent studies have attempted to determine which of these cells are targets for transformation in various lung cancers by using genetically engineered mouse models (GEMMs)⁹. Several GEMMs exploit cell type specific promoters, such as surfactant protein C (SPC) in AT2 cells, club cell 10-kDa protein (CC10) in club cells and calcitonin gene-related peptide (CGRP) in pulmonary neuroendocrine cells (PNECs), to express mutations commonly seen in lung cancer patients, in a lineage specific manner¹⁰. In sum, both the oncogenic mutation and epithelial cell type in which it is expressed can influence which lung cancer subtypes will form in mice. This topic has been extensively reviewed elsewhere^11–14, and recent updates are discussed here (Figure 1).

Figure 1.

Known and predicted influence of the cell of origin vs. oncogenic mutation in the formation of different lung cancer subtypes.

Small cell lung cancer (SCLC)

Because SCLC arises in the central airways and expresses neuroendocrine markers, it has long been postulated that this lung cancer type originates from PNECs. These progenitors are rare and usually cluster as neuroendocrine bodies in the bronchioles¹⁵. Most human SCLCs harbor inactivating mutations in the prototypical tumor suppressors TP53 and the retinoblastoma 1 (RB1) genes¹⁶. By using cell-type specific conditional knockouts of these two tumor suppressors¹⁷, independent groups have confirmed that PNECs are the predominant cells of origin of SCLC^18–20. Interestingly, inactivation of Tp53 and Rb1 in a subset of AT2 cells, but not in club cells, may also lead to SCLC at reduced efficiency¹⁸. The progression of Tp53/Rb1 null SCLC can be accelerated by the additional loss of Pten, which has recently been proposed as a frequent, but late genetic event during SCLC tumorigenesis^{20, 21}.

Lung squamous cell carcinoma (LUSC)

LUSC expresses basal cell markers (including KRT5, p63 and SOX2) and frequently occurs in the proximal airway. Therefore, it has been proposed that LUSC arises from basal progenitors. Until recently, the lack of LUSC GEMM has precluded the identification of its cellular origin, and several attempts have been made to model this malignancy. Although rarely mutated in human LUSC, inactivation of the tumor suppressor Stk11 (also known as Lkb1) in Kras mutant mice can generate LUSC, but also creates a wide spectrum of NSCLCs including mixed adenosquamous carcinoma and LCC²². In addition, IKKα kinase inactivation in mice generates spontaneous LUSC exclusively, albeit at a relatively low efficiency²³. The comprehensive catalog of genomic alterations in LUSC identified recurrent PTEN mutations²⁴. Concomitant inactivation of Lkb1 and Pten leads to LUSC in GEMM with 100% penetrance²⁵. Notably, the resulting tumors resemble many histological and transcriptional features of the human LUSC basal subtype. Given this result, it is expected that the cellular origin(s) of LUSC will soon be identified using conditional knockouts of Lkb1 and Pten in GEMM.

Lung adenocarcinoma (LUAD)

Among different lung cancer subtypes, the cellular origins of LUAD are relatively well studied owing to the availability of multiple GEMMs. Several of these rely on the spatial and temporal induction of an oncogenic Kras mutant allele. In humans, KRAS is most frequently mutated in tissues of endodermal origin, including the lung epithelium²⁶. Independent groups have shown that Kras-driven murine LUAD preferentially arises from AT2 cells, consistent with its predominant site of diagnosis at the distal airway and the immunohistochemical staining of alveolar markers in early stage patient tumors^27–30. In contrast, Kras activation driven by the CC10 promoter, presumably in club cells, bronchioalveolar stem cells (BASCs) and a small percentage of AT2 cells, leads to hyperplasia at the bronchoalveolar duct junction (BADJ), but not frank LUAD²⁸. In principle, tumor progression in the terminal bronchioles can also occur if the Tp53 tumor suppressor gene is concomitantly deleted in CC10⁺ cells³⁰. Inflammatory responses may also enhance the overall transforming effects of associated genotypes/cell phenotypes²⁷. It is also known that mutant Egfr expression in CC10⁺ cells can give rise to LUAD that rarely metastasizes³¹. Moreover, the cell surface antigenic profile of tumor-initiating cells varies when comparing LUADs driven by Egfr and Kras mutations³². Altogether, the experimental evidence to date implicates a combination of unique genetic and environmental contexts within the distal airway that is required for LUAD initiation. The function of other driver mutations in LUAD might be equally variable across regional epithelial cell lineages, contributing even further to the heterogeneous nature of this NSCLC subtype.

Switching paths to progress

Following tumor initiation from specific cell types, lung cancers can seemingly adopt various aberrant differentiation states. In solid epithelial tumors, one of the most extensively studied manifestations of this phenomenon is epithelial to mesenchymal transition (EMT). EMT is a developmental process that is not only linked to tumor cell invasion, but also to oncogenesis and tumor initiation³³. In human lung cancers, clinical evidence for EMT is seen in a fraction of LUADs that have acquired resistance to tyrosine kinase inhibitors (TKIs)^34–36. Independent of EMT, lung cancer histopathological variations may correlate with cell lineage states that are unique or selective for the airways. For instance, at the genomic level, resected human NSCLCs can be distinguished by gene expression profiles associated with different stages of pulmonary development³⁷. Despite arising from the alveolar epithelium, LUADs can convert to SCLCs, the latter being a neuroendocrine cancer more typically located in the central lungs³⁸. LUAD conversion to an “SCLC-like” phenotype is associated with drug resistance^{35, 39–41}. Furthermore, a significant proportion of high-grade LUADs histologically resembles LUSCs, which express markers of proximal airway basal cells. Mixed adenosquamous tumors retain epithelial markers yet are more invasive and have a worse prognosis^42–44. Within the LUAD subtype specifically, a gene module of normal alveolar differentiation stratifies tumors with distinct grades, biological properties and clinical outcome⁴⁵. The transcriptomic portrait of lung epithelial specification is ultimately reflected in the nomenclature adopted by The Cancer Genome Atlas (TCGA) to sub-classify human LUADs and LUSCs^{24, 46}. This remarkable molecular heterogeneity seen between and within human lung cancer subtypes may reflect the relative abundance of various progenitor cell lineages in a tumor mass. Instead of being fixed over time, tumor histopathology can also deviate due to the capacity of a tumor-initiating cell population to dedifferentiate or transdifferentiate relative to its original lineage. Significantly, molecular aberrations or extracellular signals can directly drive this phenotypic plasticity.

In GEMMs, lineage-tracing experiments have demonstrated that Kras transformed AT2 cells self-renew without additional changes in cell fate, at least up until the formation of adenomas⁴. By contrast, Kras-mutant CC10⁺ hyperplasias can switch to a CC10⁻/SPC⁺ like phenotype to develop into LUAD in the terminal bronchioles²⁷. This may be induced by local inflammation or oncogenic RAS-mediated AT2 differentiation, a lineage state that is more conducive to LUAD progression. Phenotypic conversion of AT2-derived LUADs is seen when mutating the tumor suppressor Stk11 in conjunction with Kras, which generates tumors that transdifferentiate into LUSCs^{22, 47}. LUSC metaplasia from LUADs is regulated by conserved developmental signals including the Hippo/Yap⁴⁸ and transforming growth factor/SMAD pathways⁴⁹. Another developmental program required for respiration is the canonical WNT/β-catenin pathway^50–52. Mutations in components of this pathway such as APC, GSK3B and CTNNB1 are detected in 15% of human LUADs⁵³. Multiple studies report even more frequent overexpression of various WNT pathway components and epigenetic silencing of WNT pathway antagonists in human LUADs⁵⁴. In GEMM, lung specific expression of a constitutively active mutant β-catenin is not sufficient to initiate LUAD, but may increase the pool of progenitor cells available for transformation^{55, 56}. Deletion of the WNT antagonist APC cooperates with Kras to induce dedifferentiated invasive LUADs⁵⁷. Moreover, transcriptional activation of the WNT/TCF effector LEF1 is a marker and mediator of LUAD metastasis to the brain^{58, 59}. The mechanism by which WNT/TCF promotes LUAD may involve both the expansion of the cell of origin and epigenetic reprograming of transformed cells into a primitive state that has increased metastatic potential.

Lineage transcription factors in context

Underlying the diverse pathways, molecular perturbations and differentiation states in lung cancers are transcription factors (TFs) and co-factors whose expression and activity can be restricted for certain cell lineages^{3, 60}. As will be discussed below, the genes encoding several such lineage TFs can be directly mutated, activated or silenced in particular lung cancer subtypes, suggesting that they are context dependent drivers of tumor progression and phenotypic heterogeneity.

Neuroendocrine factors in SCLC

Inactivating mutations in RB1 are ubiquitous to all SCLCs¹⁶. The extended pRb family of pocket proteins interacts with a myriad of DNA binding transcription factors and co-factors required for cell cycle progression, self-renewal, apoptosis, senescence and cellular differentiation⁶¹. Although components of the pRB pathway are frequently inactivated across many cancers, the strong selection for mutations in RB1 exclusively within SCLC suggests that pRb itself is a lineage-specific tumor suppressor. Remarkably, in the murine lungs, Rb is required for neuroendocrine cell fate, while the specification of other airway cell types can occur in the absence of Rb due to compensation from other pocket proteins⁶². This selective requirement for RB1 during mammalian lung development is consistent with the neuroendocrine origins of SCLC and the potent SCLC phenotype generated via Rb and p53 mutations in GEMMs.

Basal factors in LUSC

In LUSCs, Sry-related HMG box 2 (SOX2) is one of the most frequently amplified and overexpressed genes^{24, 63–65}. SOX2 encodes for a TF best known for its ability to reprogram embryonic stem cells. During normal development however, SOX2 is essential for basal cell commitment in the proximal airways⁶⁶. As an oncogene, SOX2 is required for the survival of LUSC cells, but its aberrant expression alone is not sufficient for transformation⁶⁵. A combination of genetic alterations, such as the loss of Lkb1⁶⁷, might be required for Sox2-driven LUSC in mice. SOX2 is usually co-amplified with the basal lineage TF p63 at chromosome 3q26⁶³. Their interaction and co-localization at target gene loci support oncogenesis instead of pluripotency⁶⁸. Moreover, within the same amplicon as SOX2 is PRKCI, which encodes for protein kinase Cl that phosphorylates SOX2, leading to the activation of Sonic hedgehog (Shh) signaling and LUSC self-renewal⁶⁹. In contrast to its role in promoting LUSCs, high levels of Sox2 in CC10⁺ cells restrict the formation of LUAD in the bronchioles by suppressing Notch activation⁷⁰.

Alveolar factors in LUAD

In distal epithelial cells, a different SOX family member, SOX9, marks the branch tips of early lungs and functions downstream of receptor tyrosine kinase signaling to suppress premature alveolar differentiation^{71, 72}. SOX9 is overexpressed in human LUAD, and its expression correlates with poor patient survival^{73, 74}. Induction of Sox9 expression has been observed in murine LUADs harboring activating mutations in Kras and β-catenin and this over-expression is associated with high grade tumors^{56, 57}. Moreover, SOX9 is preferentially upregulated in human KRAS mutant adenomas and is also induced by Notch activation⁷⁵.

NK2 homeobox 1 (NKX2–1), also known as thyroid transcription factor 1 (TTF-1), is expressed in AT2 cells and a subset of bronchiolar cells⁷⁶. NKX2–1 is essential for lung morphogenesis and alveolar cell differentiation⁷⁷. TTF-1 protein is a biomarker of thymic cancers and LUADs. About 15% of LUADs harbor NKX2–1 amplification^78–80, which correlates with poor outcome^{81, 82} and is required for tumor cell viability^{80, 83}. NKX2–1 can support pro-tumorigenic signaling downstream of mutant EGFR⁸⁴ and is required for EGFR mediated transformation in vivo⁸⁵. Paradoxically, NKX2–1 expression also correlates with better outcome since its expression is frequently suppressed in high-grade LUADs with wild-type/non-amplified NKX2–1^{81, 82, 86, 87}. Importantly, in murine LUAD models, wild-type Nkx2–1 inhibits LUAD progression driven by Kras^{85, 88}. This tumor-suppressive effect is closely linked to epithelial differentiation states, since Nkx2–1 haploinsufficiency promotes mucinous LUAD⁸⁵ and activation of enteric lineage markers that are normally repressed in alveolar cells⁸⁸. Repression of Nkx2–1 is also required for stochastic dissemination and metastasis by tumors with Kras and p53 mutations⁸⁶.

The contradictory role of certain lineage TFs likely depends on the epigenetic context in which they function. One determinant of such contexts is the relative abundance and accessibility of different co-factors in a given progenitor cell. The levels and dependencies of co-factors can be initially dictated by cell lineage states (e.g. tumor cell of origin or stage) and further dysregulated by somatic alterations (e.g. mutation, gene amplification, DNA methylation). NKX2–1 for instance can interact with multiple DNA binding transcriptional repressors or activators to expand or limit the range of target genes⁸⁹. Wild-type levels of NKX2–1 maintain alveolar differentiation and inhibit proliferation by restricting the activity and genomic target loci of FOXA1/2⁸⁸ and AP1 factors⁸⁵ respectively. In this regulatory network, decreases in NKX2–1 causes de-repression of genes that mediate cell proliferation⁹⁰, EMT⁹¹, motility⁹², and resistance to anoikis⁸⁶. Conversely, NKX2–1 amplification is often accompanied by overexpression of the lineage TF FOXA1⁹³. In this latter context which includes aberrantly high levels of NKX2–1 and FOXA1, both TFs can cooperatively access and transactivate de novo genes required for cancer cell survival⁹³. Thus, genes that would otherwise not be targeted by physiological levels of a given lineage TF can be co-opted for malignant progression under aberrant epigenetic settings (Figure 2).

Figure 2.

Epigenetic and cellular context dictates the effect of lineage TFs in lung cancer. Given a specific cellular origin, lung cancer progression can be driven by repression of a lineage TF and its tumor suppressive gene targets. Conversely, amplification of a lineage TF may activate oncogenic genes following de novo TF complex formation at novel target promoters. The tumor suppressive and oncogenic activities may be adopted by the same TF in different epigenetic contexts. These contexts are defined by the overall abundance and availability of TFs and their partnering co-factors. TF thresholds may be dependent on cooperating mutations and different cellular origins (not depicted).

An expanding network of alveologenic tumor suppressors

In addition to NKX2–1, several other alveolar specifying TFs can limit LUAD progression. Many of these factors genetically and physically cooperate with one another to regulate the expression of prototypical alveolar differentiation markers, including surfactant proteins. This suggests that the transcriptional network that controls alveologenesis is broadly linked to tumor suppression in the LUAD subtype.

The GATA family proteins are DNA binding TFs that can prime target promoters for repression or activation in specific tissues⁹⁴. GATA6 is abundantly expressed in the distal lungs where it is required for lung branching morphogenesis and alveolar specification and it restricts the expansion of adult progenitor cells at the BADJ^95–97. Gata6 directly binds to Nkx2–1 and induces the expression of surfactant proteins in AT2 cells^98–100. Similar to NKX2–1, GATA6 has both oncogenic^101–105 and tumor suppressive effects^{106, 107} in a variety of cancers. The oncogenic function of GATA6 may depend on its amplification or other cooperating gene amplifications¹⁰³ and availability of interacting TF partners, such as SP1¹⁰⁵, KLF5 and GATA4¹⁰⁴. Elevated levels of various GATAs correlate with EMT in a murine LUAD model¹⁰⁸. On the other hand, expression of GATAs is generally silenced in a significant proportion of human LUADs^109,46. In a subset of high grade LUADs, GATA6 is co-repressed with other alveolar TFs with whom it may cooperate to inhibit LUAD metastasis⁴⁵. Hypermethylation and mutation of GATA6 itself are uncommon in LUADs¹⁰⁹, but, as in other cancers, alternative mechanisms may contribute to its down-regulation¹¹⁰.

The homeodomain only protein X (HOPX), reported as a marker of AT1 cells^{5, 6}, is an atypical homedomain protein that is essential for lung maturation. Although Hopx lacks a DNA binding domain, it acts downstream of Nkx2–1 and Gata6 to modulate alveolar specific gene expression in a histone-deacetylase-dependent manner¹¹¹. HOPX exhibits tumor inhibitory effects in a wide range of cancers^112–116. In LUADs, HOPX expression correlates tightly with NKX2–1⁹³, and it can restrain tumor cell invasion and metastasis⁴⁵. Moreover, ectopic HOPX expression can induce senescence and is preferentially silenced via DNA methylation in the LUAD subtype, where decreased HOPX levels correlate with poor clinical outcome¹¹⁷. Reducing both HOPX and GATA6 in LUAD cells not only reverses alveolar identity but also activates the expression of squamous basal cell markers and pro-invasive genes⁴⁵. Thus, HOPX and its cooperating TFs may function similarly to NKX2–1, by controlling latent differentiation programs that limit overall metastatic competence¹¹⁸.

CCAAT enhancer binding protein α (C/EBPα) is a family member of basic leucine zipper TFs. C/EBPα is required for perinatal AT2 differentiation and surfactant gene expression^{119, 120}. In the adult lungs, loss of C/EBPα impairs epithelial regeneration¹²¹. Although rarely mutated in lung cancer¹²², C/EBPα is often suppressed by DNA hypermethylation¹²³, and its re-expression was shown to inhibit the proliferation and survival of human NSCLC cells¹²⁴. In mice, Cebpa is dispensable for spontaneous LUAD initiation, however its loss accelerates stress-induced LUAD formation, indicative of a tumor suppressive role in response to environmental stimuli¹²⁵. Further underlying the complexity of respiratory TF networks, C/EBPα binds to NKX2–1¹²⁶ and Cebpa expression is also regulated downstream of Nkx2–1 during development¹¹⁹.

FOXA1 and FOXA2, previously termed hepatocyte nuclear factor 3α/β (HNF3 α/β), belong to the subfamily of forkhead box TFs. FOXA1/2 have similar temporal and spatial expression patterns in developing lungs. With partially overlapping functions, they play critical roles in lung morphogenesis and alveolar differentiation^{127, 128}. Both FOXA1/2 are binding partners of NKX2–1 in LUADs^{88, 129}, and they can activate latent epithelial differentiation genes when NKX2–1 is suppressed⁸⁸. FOXA2 also cooperates with GATA6 to regulate the expression of cell adhesion genes in the developing lung epithelium¹³⁰. In both normal epithelial cells and lung cancer cells, FOXA2 induces the expression of C/EBPα¹¹⁹, and vice versa¹³¹. Similar to other alveolar TFs, FOXA2 silencing is observed in LUADs through promoter hypermethylation¹³². Finally, overexpression of FOXA2 in human NSCLC cells inhibits cell growth, invasion and EMT^{131, 133}.

The runt-related transcription factors (RUNX) are required for the development of several mammalian tissues and are dysregulated in epithelial cancers¹³⁴. In the murine lungs, Runx3 is expressed by distal epithelial cells including AT2 cells, and its targeted loss causes alveolar hyperplasia^{135, 136}. RUNX3 is yet another alveolar gene noted to be methylated at a significantly high frequency in lung malignancies¹³⁷. RUNX3 methylation status is an independent prognostic factor in NSCLC patients¹³⁸ and is associated with chemoresistance¹³⁹. Other modes of RUNX3 loss of function, such as EZH2-mediated H3K27 trimethylation¹⁴⁰ and MDM2-mediated ubiquitination¹⁴¹, have been reported in other cancers, but have yet to be explored in LUAD. In GEMMs, heterozygous loss of Runx3 causes pre-neoplastic lung adenoma^{135, 136}. Notably, Kras activation only results in non-mucinous LUADs, whereas Runx3 loss cooperates with mutant Kras to initiate both mucinous and non-mucinous LUADs¹⁴². This further suggests that deregulation of cell fate-determining TFs can drive transition to different lung tumor subtypes. The molecular connection between RUNX3 and other prototypical alveolar TFs is not yet known, but its ability to abrogate the ARF-p53 pathway by interacting with the bromodomain protein BRD2 implicates an epigenetic modulatory role¹⁴².

MYC revisited

The MYC family proto-oncogenes, which encode C-myc (Myc), N-myc and L-myc, regulate numerous biological processes. Although not restricted to pulmonary lineages, the MYC proteins are nonetheless potent transcriptional regulators of lung epithelial differentiation. N-myc is expressed at high levels in the distal airway epithelium of mice¹⁴³ and is controlled by Wnt/β-catenin signaling⁵². Loss of N-myc and C-myc can delay lung branching morphogenesis^{144, 145}. C-myc is amplified in human LUAD^{46, 146}. All three MYC members are frequently amplified in SCLC^{16, 147}, but not in LUSC²⁴. Moreover, recent genomic studies have identified loss of function mutations in MGA and MAX, two genes that encode for MYC inhibitory proteins^{46, 148}. In human lung cancers, alterations in MGA, MAX and MYC are mutually exclusive^{46, 148}, indicating a more pervasive activation of the MYC pathway via mutations than was initially appreciated.

In GEMMs, expression of Myc in SPC⁺ or CC10⁺ cells induces LUAD^{149, 150}. In addition, Myc-induced LUAD requires Kras activating mutation¹⁵⁰, which can be substituted by Notch1 overexpression¹⁵¹. Myc is also required for tumor maintenance in a variety of Ras pathway initiated NSCLC models^{152, 153}. MYC can activate or repress specific genes by recruiting different cofactors to target promoters¹⁵⁴. Myc has also been suggested to broadly amplify gene transcription in cancer cells¹⁵⁵. In resected human LUADs, MYC overexpression is linked to increased DNA methylation of known tumor suppressor genes⁴⁶. Activation of other specific C-myc targets in human LUADs correlates with perturbations in lineage TF networks and poor clinical outcome^{45, 156, 157}. Consistent with these in silico analysis of human tumors, Myc overexpression from SPC⁺ AT2 progenitors in mice is sufficient to induce macrometastases that have undergone a mixed lineage switch¹⁵⁸. Finally, human mutations in the MYC pathway are mutually exclusive to SMARCA4 (or BRG1)^{148, 159}, which encodes a major component of the SWI/SNF chromatin-remodeling complex¹⁶⁰ and is a mediator of SCLC neuroendocrine differentiation¹⁴⁸. Therefore, despite its multiple functions, the major consequence of MYC activation in lung cancers may be to epigenetically reprogram specific lineage states.

Lineage plasticity and metastasis: cause or consequence?

Epithelial lineage plasticity has been associated with cancer metastasis. As core mediators of cell fate, lineage TFs may therefore regulate genes required for lung cancer cell dissemination and distant organ colonization. However, considering the diverse functional targets of these TFs, is their capacity to regulate metastasis directly connected to the lineage state in a given cell? In lung cancer, this remains an open question, but some insight may be drawn from recent studies.

The over-expression of EMT inducing TFs in epithelial cancer cells is often interpreted as a marker of increased cell invasiveness and metastasis¹⁶¹. Overexpression of EMT inducing TFs has been observed in various models of metastatic LUAD¹⁶². However, this is confounded by the fact that EMT promoting TFs also target genes that are independently required for other cellular functions such as cell survival, proliferation, DNA damage responses and self-renewal^{163, 164}. In some carcinoma models, EMT is neither sufficient nor required for cell invasion, while disseminating cancer cells that undergo the reverse process, mesenchymal to epithelial transition (MET), have a greater advantage for tumor re-initiation and metastatic colonization. Intriguingly, some studies suggest that mesenchymal carcinoma cells may not invade beyond the local tumor, but rather enhance the dissemination of other epithelial cell populations with tumor re-initiating capacity^{165, 166}. An analogous phenomenon has been described in GEMMs of SCLCs, where a fraction of malignant cells can transition from their neuroendocrine (NE) origins into non-NE cells that express mesenchymal genes¹⁶⁷. Although these non-NE cells are clonally related to the rest of the SCLC, they do not metastasize. On the other hand, their presence in the tumor mass enhances the ability of the NE lineage positive cells to spread. Thus, modulation of lineage states by TFs may indirectly influence metastatic potential by controlling intra-tumoral clonal dynamics, presumably via paracrine signals (Figure 3).

Figure 3.

Hypothetical mechanisms by which cell lineage plasticity and heterogeneity promote lung cancer metastasis. (A) Epithelial lineage positive malignant cells, such as NE cells in SCLC, constitute the majority of distant metastasis, but their dissemination is enhanced via non-cell autonomous signals from mesenchymal transitioned tumor cells in the primary tumor. (B) LUAD cells may disseminate via a classical EMT/MET model. (C) Alternatively, LUAD cells may undergo multiple switches in the expression of airway epithelial lineage specific markers that promote metastasis via cell autonomous mechanisms.

An alternative mechanistic consequence of tumor metaplasia is that it may lead to the activation of lineage gene programs that directly control the metastatic capacity of a given cell. The activation of latent basal epithelial markers, such as cytokeratins (K) has been reported in more invasive tumor cells. For example, cells at the invasive border of luminal breast tumors express the basal epithelial marker K14 and basal TF p63. These invasive basal cells arise from the well-differentiated luminal cells and are responsible for collective tumor cell invasion in a manner that is independent of EMT¹⁶⁸. In liver cancer, the biliary/progenitor marker K19 is expressed in a subset of hepatocellular carcinomas with a high incidence of microvascular invasion and metastasis¹⁶⁹. Similarly, LUAD cells may acquire invasion potential by down-regulating alveolar differentiation genes and upregulating the squamous basal cell markers K6A/B⁴⁵. Importantly, knockdown of these cytokeratins in their respective disease models inhibits tumor cell invasiveness, suggesting an unexpected direct role for basal/biliary epithelial markers in malignant invasion and metastasis.

Harnessing lineage states for therapy

The TFs discussed here are but a few examples linking the molecular control of cell lineage fate to lung cancer progression. Despite the broad biological significance of these observations, their clinical and therapeutic impacts remains underexplored. Part of the reason for this is the intrinsic technical barriers in pharmacologically manipulating pathways or TFs that are not only tissue specific, but also likely to have dosage and stage specific effects during tumor progression. Nevertheless, exploiting aberrant tumor differentiation states or lineage TFs for therapy has been shown to be clinically beneficial in leukemia and neuroblastoma¹⁷⁰ and pre-clinically feasible in melanoma^{171, 172}, glioma^{173, 174}, nasopharyngeal carcinoma¹⁷⁵, and rhabdomyosarcoma¹⁷⁶. Driving the tumors toward more differentiated states may not only delay their growth but also potentially increase their sensitivity to commonly used cytotoxic agents.

Several novel compounds have recently been explored to target the expression or activity of TFs. For example, drugs that inhibit the chromatin-remodeling BET bromodomain proteins are being actively investigated as a strategy to target the lineage-dependent activity of TFs in multiple cancers, including LUAD^{177, 178}. Another potential approach is to use covalent inhibitors of cyclin-dependent kinase 7, which preferentially mediates the expression of super-enhancer associated genes^{179, 180}. Since such genes include oncogenic MYC and neuroendocrine TFs, this class of transcription targeting drugs may be particularly effective against SCLC¹⁸¹.

Assessing the lineage state of the bulk tumor may have more immediate clinical benefits for anticipating dynamic therapeutic responses. This is relevant in the treatment of KRAS mutant LUADs, which are highly heterogeneous. Notably, a subset of KRAS mutant LUADs that are well-differentiated and retain epithelial markers respond specifically to a combination of drugs that inhibit MAP kinase and BCL-Xl¹⁸². In contrast, LUADs that have undergone lineage transition become drug refractory, even when those treatments target their driver oncogenes. At the same time, such lineage transitions may have foreseeable shifts in signaling dependencies. For example, the survival of TKI resistant EGFR mutant tumor cells that are more mesenchymal-like requires activation of the receptor tyrosine kinase AXL^{34, 183}, which can be inhibited with small molecules.

Given the fact that lung cancers can adopt various differentiation states, new biological insight may be revealed by comprehensively defining their underlying transcriptional circuits. These circuits may in turn modify the oncogenic dependencies of a given tumor, providing novel therapeutic opportunities. Major challenges remain in directly inducing or reverting these lineage specific pathways at different stages of disease progression. Nevertheless, several lung cancer differentiation states, defined either transcriptionally or histologically, may already be used as both prognostic and predictive biomarkers. Moving forward, a systematic search for efficacious agents tailored not only to the oncogenic drivers of various lung cancer subtypes, but also to the cell lineage of their malignant compartment, holds the promise of abating metastatic progression.