Mutations, Bottlenecks, and Clonal Sweeps: How Environmental Carcinogens and Genomic Changes Shape Clonal Evolution during Tumor Progression

Abstract

The transition from a single, initiated cell to a full-blown malignant tumor involves significant genomic evolution. Exposure to carcinogens—whether directly mutagenic or not—can drive progression toward malignancy, as can stochastic acquisition of cancer-promoting genetic events. Mouse models using both carcinogens and germline genetic manipulations have enabled precise inquiry into the evolutionary dynamics that take place as a tumor progresses from benign to malignant to metastatic stages. Tumor progression is characterized by changes in somatic point mutations and copy-number alterations, even though any single tumor can itself have a high or low burden of genomic alterations. Further, lineage-tracing, single-cell analyses and CRISPR barcoding have revealed the distinct clonal dynamics within benign and malignant tumors. Application of these tools in a range of mouse models can shed unique light on the patterns of clonal evolution that take place in both mouse and human tumors.

Animal models have been used to explore mechanisms of carcinogenesis for more than 100 years, since the first demonstration by Yamagiwa that coal tar, when applied repeatedly to rabbit ear skin, could induce the development of tumors (Yamagiwa and Ichikawa 1915). Subsequent studies in the 1940s refined the concept of multistage carcinogenesis, showing that mouse skin tumors initially developed as benign lesions, some of which progress to malignant and metastatic carcinomas, thus giving rise to the now widely accepted notion of carcinogenesis as a multistep process that is relevant to human cancer development (Berenblum and Shubik 1947). These pioneering experiments demonstrated that there are different types of exogenous environmental factors that can cause cancer in mice, but they appeared to act through different mechanisms. Some were proposed to act as mutagens, although the evidence for this claim was not provided until decades later, whereas others were associated with extensive inflammation and proliferation, essentially mimicking a wound response. Berenblum and Shubik (1947) showed that the order of events was important: Initiation, presumed at that time to be caused by a mutational event, had to take place first, followed by a promotion stage induced by chronic exposure to an irritant substance, otherwise no tumors were induced. The same authors also demonstrated that the initiating event was essentially permanent, as promotion started more than one year later still induced papillomas (Berenblum and Shubik 1949).

The discussion of the relative importance of mutations and promoting factors in cancer was resurrected more recently by whole-genome sequencing of a series of tumors induced in mice by known or suspected environmental carcinogens (Riva et al. 2020). It was noted that the majority of carcinogenic chemicals did not appear to be mutagens, and therefore were more likely to act as promoters by stimulating the outgrowth of cells carrying specific, but spontaneously arising, driver gene mutations. A recent study of patients and murine lung cancer models demonstrated that air pollutants also promote lung cancer through nonmutagenic effects on tumor cells and their microenvironment (Hill et al. 2023). These different mouse model studies carried out over a period of almost 80 years (Berenblum and Shubik 1947, 1949; Riva et al. 2020) support the conclusion that tumor promotion is not only an essential rate-limiting step in mouse skin and lung carcinogenesis, but that environmental or lifestyle factors that do not act as mutagens may play an important causal role in increasing cancer incidence in human populations (Balmain 2022; Hill et al. 2023).

The models discussed above, reflecting the roles of exogenous DNA-damaging mutagens and environmental promoting factors in causing cancer, were largely superseded by the discovery of oncogenes and tumor suppressors in the 1970s and early 1980s, which laid the foundation for our understanding of the critical role played by genetic mutations in driving cancer. The development first of transgenic mice (Stewart et al. 1984; Hanahan 1985; reviewed in Hanahan et al. 2007), followed by knockout mice (Koller et al. 1990; Zijlstra et al. 1990; Jacks et al. 1994), conditional gene targeting using Cre-Lox technology, and most recently by CRISPR-mediated introduction of almost any desired germline alteration (Wang et al. 2013), have allowed the interrogation of tissue-specific gene function in unprecedented detail, thus contributing to a much deeper understanding of how mutations impact multiple stages of cancer development. Each of these approaches to modeling cancer in the mouse—using exogenous chemical carcinogens or germline manipulation—can provide invaluable but also complementary insights into the processes that drive clonal evolution of cancer. Tumors induced by environmental exposures generally have many point mutations and other genomic events that are caused directly or indirectly by the agents used. Genetic models, with few exceptions, have a very low mutational burden, but these can provide precise information on the specific targeted mutations, and their combinations, that drive tumor progression and metastasis. The present review will focus on clonal evolution of cancer using both of these approaches, to highlight the complementary insights that can arise from each.

INITIATING MUTATIONS DO NOT MAKE CANCER

Statistical analyses of the relationship between age and cancer incidence in the 1950s suggested that four to six rate-limiting events were needed for cancer to arise (Armitage and Doll 1954). The nature of these “events” was not defined, but a common assumption was that these represented genetic events required to drive progression between different biological stages. Although the first analysis of the epidemiological data emphasized the four- to six-stage model, other interpretations of the same data, even from the same group, suggested alternative two-stage models if clonal selection and stochastic effects were taken into account (Armitage and Doll 1957; Moolgavkar and Knudson 1981). Discussions of the need for “two hits” or “multiple hits” to explain cancer incidence thus date back several decades (reviewed in Ashley 1969; see also Balmain 2022). As our understanding of tumor genomics has advanced, and we have now extensively cataloged the genomic alterations that constitute these “hits,” this same tenet has held: One mutation typically is not sufficient, by itself, to induce tumors. However, if followed by chronic wounding or tissue damage through exposure to promoters, one-point mutation would appear to be enough for development of benign lesions, and these can then acquire additional genomic alterations leading to promoter independence and malignant growth. Alternatively, combinations of two or three cancer-promoting mutations can be extremely potent in driving malignancy, even if acquired over an extended period of time spanning multiple decades. An “initiating” cancer mutation may be acquired early in life, or even be present at birth, but this initiated cell has to undergo a significant evolutionary process to progress to something we would recognize and diagnose as cancer. This evolutionary process of clonal selection can be accelerated by exposure to promoting factors, such as chemical promoters (McCreery and Balmain 2017) or air particulates (Hill et al. 2023), or can be engineered to take place simultaneously or sequentially in the same mouse target cells by introduction of specific genetic changes into mouse embryonic stem cells or directly in mouse embryos.

It is quite fortunate that a single mutated oncogene or loss of a single tumor-suppressor gene is itself not sufficient to cause cancer. Just how fortunate is increasingly becoming apparent, as next-generation sequencing of normal tissues has revealed that pathologically normal skin (Martincorena et al. 2015), esophagus (Martincorena et al. 2018), and colon and liver (Blokzijl et al. 2016) are riddled with cancer-associated genomic alterations. Up to 20% of normal skin cells and 80% of esophageal epithelium cells contained cancer-associated mutations in NOTCH1. Further, mutations in TP53, NOTCH2, NOTCH3, and FAT1—all implicated in both cutaneous and esophageal squamous cell carcinomas—were also common (Martincorena et al. 2015, 2018). Approximately 75% of moles (melanocytic nevi) carry activating oncogenic mutations in BRAF or NRAS (Ascierto et al. 2012; Zeng et al. 2020)—and yet, it is estimated that less than 1 in 33,000 of these benign nevi will progress to malignant melanoma (Tsao et al. 2003). Additionally, chronic conditions such as inflammatory bowel disease and endometriosis have been linked to an increased accumulation of somatic mutations in the colon and endometrium, respectively (Suda et al. 2018; Olafsson et al. 2020).

Although a single cancer-driving mutation may not lead to full-blown malignant disease, cells that acquire such mutations can gain a growth advantage and expand within the tissue, even while it remains histologically normal. These “fields” of mutations are described by the concept of “field cancerization,” wherein a tissue accumulates a large number of cells that carry “initiating” mutations that are poised to progress to malignancy. However, the “field” remains overtly cancer-free until cells within it either acquire additional genetic hits or are exposed to endogenous or environmental factors that induce inflammation (Coussens and Werb 2002; Hill et al. 2023) or increased epigenetic plasticity (Feinberg and Levchenko 2023). Field cancerization is particularly relevant to some cancer types—including lung, colon, skin, prostate, and bladder—and can lead to patients presenting with multifocal disease, with multiple tumor nodules arising out of a “field” of cancer-predisposed cells (Curtius et al. 2017). Taking advantage of recent technological advances, spatial copy-number analysis of prostate tissues revealed that high-grade prostate cancer and adjacent histologically benign and normal regions shared the same copy-number alterations—clearly demonstrating that the tumor has arisen from a field of genomically altered cells, which themselves were not (yet) cancerous (Erickson et al. 2022).

What do tumors need, then, beyond acquisition of a cancer-driving mutation? As mentioned above, the acquisition of additional genetic hits can help drive an initiated cell down the path to malignancy. During this progression phase, from initiated cell to fully malignant cancer, tumors undergo substantial evolution, and the subsequent sections of this review will discuss what we know about the genomic changes—including acquisition of both point mutations and of larger-scale chromosomal alterations—and evolutionary pressures that take place during this progression.

MULTIPLE “HITS” IN MOUSE MODELS OF CANCER

The principle that “more than one hit” is needed to drive cancer has been very much borne out in the development of mouse models of cancer. Many elegant models of a wide range of cancer types have been developed by engineering activated oncogenes or tumor-suppressor deletions into a targeted cell type, collectively referred to as genetically engineered mouse models (GEMMs). For example, engineering a Cre recombinase-activated Kras^G12D mutation and p53 deletion (widely referred to as the “KP” model) into pancreatic cells using Pdx1-Cre leads mice to develop pancreatic ductal adenocarcinoma (PDAC) (Hingorani et al. 2005), whereas having these KP mice inhale viral Cre leads them to develop lung adenocarcinoma (LUAD) (Jackson et al. 2005).

As a general principle in GEMM development, combining multiple oncogenes leads to a more aggressive disease. The Kras^LA and Braf^CA lung cancer models—which contain a single Kras^G12D- or Braf^V600E-activating mutation, respectively, that is “turned on” when mice inhale viral Cre—develop mostly preneoplastic atypical adenomatous hyperplasia (AAH) and benign lung adenomas (Jackson et al. 2001; Dankort et al. 2007). However, when these mice were crossed with mice that also lost a tumor suppressor (p53 or Ink4a/Arf), mice developed tumors faster, and also developed tumors with more aggressive malignant features that look histologically more like lung adenocarcinoma (Jackson et al. 2005; Dankort et al. 2007). This paradigm, in which the addition of more “hits” leads to more aggressive cancer, has played out across many GEMM cancer models. Another example—to pick one of many—is seen in a series of models of colorectal cancer. An Lgr5-driven Cre was matched with eight combinations of four driver alleles: Apc loss (Apc^fl/fl), Kras^G12D, Tgfbr2^fl/fl, and p53^fl/fl (Tauriello et al. 2018), and it was found that aggressiveness and tumor stage generally increased in models that had higher numbers of driver alleles: The “4×” combination led to a majority of lesions being metastatic, whereas the three “3×” combinations led to the majority of lesions being classified as benign (Tauriello et al. 2018). Of course, not all “hits” are equal. Rational development of GEMMs typically involves combining cancer-promoting alleles that are observed together in patient cancers. Further, different alleles may drive different directions of progression: Adding a loss of either Apc (“A”) or Lkb1 (“L”) to the KP lung cancer model leads to accelerated tumor development, but “KPA” versus “KPL” tumors show distinct paths of transcriptional changes as they progress (Yang et al. 2022).

The concept of “multiple hits” can also be seen to play out in carcinogen models of cancer, in which mice are exposed to a chemical carcinogen to initiate tumorigenesis (McCreery and Balmain 2017). Some chemicals act as “complete carcinogens,” meaning that treatment with these chemicals as a single agent is sufficient to induce cancer when applied repeatedly (Peto et al. 1975), or as a single high dose. As discussed above, a single low dose of a carcinogen is sufficient for initiation, but tumor development requires further treatment with a promoting agent. In the dimethylbenzanthracene (DMBA) skin carcinogenesis model, DMBA is commonly used in conjunction with 12-O-tetradecanoylphorbol-13-acetate (TPA) as a promoting agent. DMBA induces A > T mutations, including activating Q61L mutations in Hras (Quintanilla et al. 1986). As shown by Berenblum and Shubik (1947, 1949) addition of a promoter such as TPA can be started as far out as a year later and will promptly lead to the outgrowth of benign tumors. The original studies carried out in the 1940s have been replicated in our laboratory, with very similar results (YR Li, E Kandyba, K Halliwill, et al., unpubl.). However, whole-genome sequencing showed that tumors that arose in young initiated mice or in those that had been initiated with DMBA one year earlier were indistinguishable. The number of DMBA-induced signature A > T mutations ranged from a few hundred to almost 50,000, but even cells carrying many thousands of mutations, including the potent Hras Q61L mutation, remained in a latent state and were not capable of giving rise to tumors without going through the promotion stage.

This model thus nicely recapitulates dynamics in human tissue, where multiple genetic “initiating” events can take place long before a tumor arises. The events that lead to conversion of otherwise latent mutated cells into actively growing early-stage tumors are unknown, but it would seem that the simple sequential acquisition of multiple point mutations in rare single cells is unlikely to be the entire answer (Fig. 1). One possibility is that promoters or tissue damage could induce larger-scale genomic changes, as suggested decades ago (Fürstenberger et al. 1989). Indeed, cytogenetic (Aldaz et al. 1989) and exome sequencing studies (McCreery et al. 2015) have shown evidence for early changes in chromosome numbers, particularly for chromosome 7, on which the activated Hras gene is located, and chromosome 6, which harbors the Kras, cRaf, and Met oncogenes, all of which are implicated in signaling pathways leading to tumorigenesis. Sequencing of DMBA/TPA-induced malignant carcinomas revealed additional candidate driver alterations—namely, loss of Cdk2na or amplification of chromosome 1, as well as a number of single-nucleotide mutations in genes such as Trp53 that clearly impact the progression stage of carcinogenesis (Kemp et al. 1993; McCreery et al. 2015; Nassar et al. 2015). Significant genomic changes occur between benign and malignant states, indicative that evolution occurs during progression, even in cases where it is not clear which—or if—a particular genetic event potentiated malignancy.

Figure 1.

Multiple paths to “multiple hits” in the evolution of tumors. The concept that cancer is more than a single genetic mutation has been around now for more than a century. Studies with mouse models have illustrated that many factors, ranging from genomic damage to chemical exposures to inflammation, can contribute to tumor progression, and moreover they do so through multiple mechanisms. Some of these influences lead to mutations and chromosomal changes that can be tracked through genomic evolutionary studies, but others, such as changes in the cytokine milieu of the microenvironment, are genetically invisible.

GENOMIC EVOLUTION DURING PROGRESSION

Genomically, an initiated cell, a benign tumor, and a full-blown malignant tumor do not look the same. Although they all may carry a common oncogenic mutation—such as a BRAF V600E mutation in initiated melanocytes, benign nevi, and melanomas in patients, or the Hras Q61L mutation in DMBA-induced murine skin and skin tumors—the rest of the genome evolves substantially as tumors progress.

Comparing benign and malignant tumors as general categories, malignant tumors have on average more genomic alterations than their benign counterparts, both in terms of single-nucleotide variants (SNVs) and copy-number alterations (CNAs). This is true across multiple patient tumor types and multiple mouse models. In the DMBA/TPA skin carcinogenesis model, malignant carcinomas were found to have more SNVs than benign papillomas (averaging a 65% increase) and contained significantly more whole-chromosome gains and losses (McCreery et al. 2015; Nassar et al. 2015). This remained true even when the carcinomas and papillomas of the same age within the same cohort of mice were compared (McCreery et al. 2015). In a model of urethane-induced lung chemical carcinogenesis, it was similarly found that fully malignant adenocarcinomas had a higher incidence of SNVs than age-matched benign adenomas (Westcott et al. 2015), and in a 4-nitroquinoline-1-oxide (4NQO)-induced model of oral squamous cell carcinoma, invasive oral carcinomas were found to have more than threefold higher mutations than lesions classified as hyperplasia or dysplasia (Sequeira et al. 2020). Of note, these trends of increasing mutation burden with tumor stage are seen when comparing average mutations across tumors of different stages. However, outlier tumors are observed—for example, fully malignant DMBA-induced carcinomas that have a strikingly low mutation burden (McCreery et al. 2015; YR Li, E Kandyba, K Halliwill, et al., unpubl.)—underscoring the point that, in spite of these trends, there is not a specific number of mutations or mutation threshold required for malignancy.

Genomic evolution during the benign growth phase has been less well-studied, but several studies suggest that benign tumors also accumulate SNVs and CNAs over time prior to—or even in the absence of—transition to malignancy. In the 4NQO model of oral carcinoma, regions of mild dysplasia were found to have significantly fewer SNVs than regions of moderate to severe dysplasia (Sequeira et al. 2020). A study of early-stage lung lesions from patients also reached the same conclusion: Sequencing of small lung lesions, divided into four progressive categories (premalignant atypical adenomatous hyperplasia [AAH], adenocarcinoma in situ, minimally invasive adenocarcinoma, and adenocarcinoma), found that premalignant AAH lesions exhibited the lowest mutation burden of all stages examined and had virtually no copy-number alterations (Hu et al. 2019). Both SNVs and CNAs increased with each progressive tumor stage, and this held true within lesions from nonsmokers as well as lesions from smokers (Hu et al. 2019), suggesting the increase was not merely a surrogate of more prolonged carcinogen exposure.

Significant changes in CNAs during the benign growth phase are also seen in the DMBA/TPA skin carcinogenesis model. In a study of “early” (∼3 mo) versus “late” (∼1 yr) papillomas, it was found that there was a dramatic increase in the number of CNAs in “late” papillomas, suggesting these had been continuously acquired (McCreery et al. 2015). This interpretation is supported by single-cell RNA-seq analysis of a series of papillomas and carcinomas, which suggested sequential accumulation of large-scale CNVs involving chromosomes 7, 6, 1, and 10 in papillomas (MA Taylor, E Kandyba, and A Balmain, in prep.). Although these changes were seen only in subpopulations of cells in the papillomas, they were clonal and present in all cells in the carcinomas analyzed. Interestingly, some aneuploidy events in other chromosomes were seen in multiple cells within papillomas, but were not detected in any carcinomas, suggesting that certain events lead to evolutionary dead ends and are not required for or compatible with clonal progression to the carcinoma stage.

CLONAL DYNAMICS OF BENIGN AND MALIGNANT LESIONS

A great deal of outstanding work has led to the generation of multiregion profiles of human tumors, in particular in renal carcinoma (Gerlinger et al. 2012; Turajlic et al. 2018; Zhao et al. 2021) and lung cancer (Jamal-Hanjani et al. 2017; Hu et al. 2019). Where mouse models have been able to lend particular insight to this field has been in the ability to introduce lineage tracing tools into tumors and follow the clonal dynamics of tumor growth over time. A key finding from such modeling is that the clonal dynamics and growth patterns of benign tumors significantly differ from those of malignant tumors. Using K14-driven yellow fluorescent protein (YFP) lineage tracing in DMBA/TPA-induced skin carcinomas and quantification of YFP⁺ clone sizes, Driessens et al. demonstrated that the majority of cells within a benign tumor have limited proliferative capacity, mirroring the architecture of normal skin in which growth is driven primarily by a small number of stem cells (Driessens et al. 2012). This benign tumor growth was later identified to be driven primarily by a small number of Lgr6⁺ cells with stem-like features in the benign tumor (Huang et al. 2017). Multicolor lineage tracing of papilloma growth supported these findings, revealing that although they are clonal in origin (Reeves et al. 2018), benign DMBA/TPA-induced tumors are comprised of a patchwork of regionalized clones, each presumptively driven by a local stem cell, which for the most part retain structured boundaries between neighboring clones (Reeves et al. 2018).

By contrast, growth within malignant tumors is driven by a large proportion of the cells in the tumor, rather than by rare stem cells (Driessens et al. 2012; Huang et al. 2017). Adjacent clones within malignant tumors in the DMBA/TPA model no longer form clean boundaries, but intermix with one another, perhaps having gained more migratory phenotypes (Reeves et al. 2018). This increased intermixing of clones in malignant tumors echoes findings from sequencing of individual glands within patient colorectal tumors, in which physical intermixing of genetically distinct clones was found in carcinomas, whereas adenomas by contrast harbored subclones with clear spatial delineations (Sottoriva et al. 2015).

The topic of stem cells within tumors has been the subject of some amount of controversy. Lineage-tracing studies have unequivocally demonstrated that tumor cells—at steady state, unperturbed by therapy—exhibit hierarchical relationships, with a disproportionate amount of tumor growth being driven by only a subset of cells within the tumor (Driessens et al. 2012; Huang et al. 2017; Reeves et al. 2018). However, whether these growth-driving “stem-cell-like” cells are special and occupy a privileged state or whether the majority of cells within a tumor have the capacity or plasticity to adopt the “stem cell” state is still unclear. A recent study of melanoma growth, using the Tyr::Nras^Q61K/°Ink4a^–/– GEMM model, proposes an intriguing model, in which stem cell state is determined by proximity to blood vessels (Karras et al. 2022). Stem cells maintaining their stem cell fate based on environmental and niche cues is a widely observed phenomenon in the biology of healthy tissues, but the pro-stem-cell niche in tumors is not as well-understood. A drawback to this study is that it was carried out in transplanted melanomas, which do not always fully recapitulate the architecture of spontaneous tumors, and future work will be required to clarify the location and architecture of such niches in spontaneously evolving tumors.

A BOTTLENECK AT THE BENIGN TO MALIGNANT TRANSITION

Although we have discussed the dynamics within benign and malignant tumors, the transition from benign to malignant is a crucial step in tumor progression and merits its own attention. Both multicolor lineage tracing and genomic analyses give complementary windows into clonal dynamics, and both approaches suggest that the benign-to-malignant transition is typically carried out by a “sweep” of a particularly fit clone that drives malignancy. In the DMBA/TPA-induced skin carcinogenesis model, multicolor labeling of clones within benign papillomas led to papillomas that are a patchwork of colors; however, after progression to malignancy, all carcinomas were comprised of only a single color—corresponding to the single clone that drove progression (Reeves et al. 2018). Sequencing of both 4NQO-induced murine oral carcinomas of progressive stages (hyperplasia, mild and moderate/severe dysplasia, and invasive carcinoma) found that the earliest stage lesions were comprised of more tumor subclones and more low-variant allele frequency mutations than invasive carcinomas, which by contrast had more clonal mutations (Sequeira et al. 2020). These findings mirror those of a study of early patient lung lesions, in which the frequency of subclonal mutations was highest in lesions at the premalignant AAH stage, whereas tumors of later stages (adenocarcinoma in situ, minimally invasive adenocarcinoma, and adenocarcinoma) had a higher frequency of clonal mutations, suggesting that progression to these more advanced stages was associated with a sweep of the fittest clone (Hu et al. 2019).

CRISPR-based continuous lineage tracing, in which cells accumulate ongoing mutations in a barcode region, has now enabled tracing of cell hierarchies over a dozen cell generations (Baron and van Oudenaarden 2019; Wagner and Klein 2020). One such CRISPR-based lineage tracing technology was combined with the Kras^LSL-G12D/–; p53^fl/fl (KP) GEMM model of lung adenocarcinoma to map clonal relationships at single-cell resolution (Yang et al. 2022). The authors found that tumors progress by moving through highly plastic states, in which a fit clone is able to rapidly expand, and then stabilizes. The majority of tumors analyzed had been through one or two “expansions,” in which a single subclone had come to dominate the tumor (Yang et al. 2022). After such an expansion, future progeny would evolve neutrally. Because this study only looked at tumors harvested at late time points (5–6 mo) and did not evaluate histological tumor stages, it was not possible to compare evolution patterns between tumors with more invasive versus more benign pathologies. Nonetheless, these findings are consistent with other data pointing to the occurrence of punctuated clonal sweeps or clonal expansions during tumor progression (Fig. 2).

Figure 2.

Clonal evolution involves continuous accumulation of genetic damage and punctuated clonal sweeps. Evolution of a tumor involves the accumulation of an increasing number of genomic alterations, along with selection of specific tumor subclones. In the diagram, the initiated cell (red) accumulates additional mutations (teal, purple, blue). The tumor cells containing the teal mutation gain a competitive advantage, and this clone eventually sweeps the tumor. The teal and purple subclones develop additional mutations (light cytoplasm, dark cytoplasm). The teal light cytoplasm subclone outcompetes other teal subclones, leading to a second sweep of the tumor. This subclone accumulates another genomic alteration (dark granules), and the evolutionary process continues. The purple, blue, purple light cytoplasm and teal dark cytoplasm mutations represent evolutionary “dead ends” of clones that are eventually eliminated as the tumor progresses. This model depicts the evolutionary dynamics that enable the emergence of dominant tumor subclones simultaneously with the maintenance of genetic heterogeneity within the tumor. It should be noted that the accumulation of genetic changes takes place in a complex microenvironment where promoting factors are simultaneously inducing inflammation, invasion, angiogenesis, epigenetic changes, and other hallmarks of cancer. Selection of particular clones is shaped by these environmental pressures and is not merely the product of accumulated mutations. In some cases, the mutations distinguishing a tumor subclone may not even be particularly advantageous, but nonetheless serve to genetically mark a population of cells that gained a nongenetic advantage that enabled their expansion.

There is genomic data supporting the idea that the transition to malignancy can be driven by a specific tumor subclone that acquires the “right” second (or third or fourth) genomic alterations. Some combinations of mutations are almost never found in benign tumors, suggesting that a benign tumor may have one (or more) of these components, but gaining the “final” hit is incompatible with the tumor staying benign. In the DMBA/TPA skin carcinogenesis model, malignant carcinomas are frequently found to harbor losses of the tumor-suppressor genes Cdkn2a/b or Trp53, but neither of these events is generally found in benign papillomas (McCreery et al. 2015; Nassar et al. 2015). Thus, it would appear that the Hras Q61L mutation can support benign tumor growth, but the combinations of Hras Q61L plus loss of Cdk2na or Trp53 strongly potentiate full-blown malignancy. Gain of chromosome 1 is also strongly implicated in progression, as this genomic alteration is seen only in few cells within papillomas but is a common event in malignant carcinomas (McCreery et al, 2015; MA Taylor, E Kandyba, A Balmain, et al., unpubl.). Interestingly, the Lgr6 gene, which is located on this chromosome, acts as a clonogenic stem cell marker in skin tumors (Huang et al. 2017) and is also increased in expression at the benign–malignant transition, suggesting that self-renewal of Lgr6-positive stem cells may be a driver of malignant progression. This echoes the observations in GEMM models, discussed above, in which specific combinations of genetic alterations lead to more aggressive cancers, with the key difference that in carcinogen models these hits are stochastically acquired rather than engineered. Supporting that this model is also true in patients, CDKN2A loss is one of the most common events in several human tumor types including squamous cell carcinomas (SCCs), and in melanomas is specifically associated with progression: CDKN2A is almost always found intact in benign melanocytic nevi, but monoallelic and biallelic loss are found with increasing incidence in tumors and tumor regions classified progressively as melanoma in situ, thin invasive, and thick invasive melanoma (Zeng et al. 2018). Although this does not mean that progression to malignancy is always potentiated by a genetic event, it would appear that in many cases a genetic event plays a substantial role in the sudden expansion of a malignancy-driving tumor subclone. In the case of the tumor suppressor Trp53, its loss or inactivation can provide many routes to tumor progression because of its multifaceted roles in control of DNA repair, homologous recombination, genomic stability, and stem cell fate decisions (Charruyer et al. 2021).

LESS OF A BOTTLENECK DURING METASTASIS

Progression from benign-to-malignant stages and progression from a primary malignant tumor to metastasis can be thought of as the two landmark transitions that a tumor undergoes. Of course, these actual processes can be complex, with benign tumors moving through a range of increasingly invasive stages, and metastatic dissemination occurring in an ongoing fashion rather than being confined to a single point in time. Those nuances acknowledged, now that we have looked at the clonal dynamics accompanying progression to malignancy, we will now examine the dynamics that accompany metastasis.

In comparison to the benign-to-malignant transition, metastasis appears to be a bottleneck, but a less restrictive one. Multicolor lineage tracing of metastasis in the DMBA/TPA skin carcinogenesis model (Reeves et al. 2018) and in the Pdx1-CreER;Kras^LSL-G12D;p53^fl/+model of PDAC (Maddipati and Stanger 2015) found that metastases are frequently polyclonal, with contributions by more than one cell from the primary tumor. Maddipati and Stanger were able to isolate polyclonal (multicolor) clusters of circulating tumor cells in the blood of the PDAC-bearing mice, suggesting that polyclonal metastases arose from a seeding event that included multiple tumor populations; however, they did not rule out the possibility that in some cases metastases arising from a single tumor clone were colonized by additional clones at a later time (Maddipati and Stanger 2015). However, in support of the polyclonal seeding hypothesis, circulating tumor cell clusters have been found to have higher metastatic potential (Aceto et al. 2014; Maddipati and Stanger 2015), and such clusters have also been found to have an advantage in evading immune-mediated killing during transit (Lo et al. 2020).

In our work in the DMBA/TPA skin carcinogenesis model, we noticed a trend toward polyclonal metastasis occurring with higher frequency in the lymph node than in distant sites (MQ Reeves and A Balmain, unpubl.). Notably, these same findings are seen in patients. Following the publication of these lineage-tracing-based findings of polyclonal metastases in mice, improvements in sequencing technology allowed the same questions to be asked in patients. Analyses of prostate, colorectal, lung, and breast tumors from patients have all found evidence of polyclonal metastases (Gundem et al. 2015; Naxerova et al. 2017; Hu et al. 2020). When lymph node metastases and distant metastases were compared, it was observed that lymph node metastases were twice as likely to be polyclonal—specifically, one pan-cancer study found that 60% of lymph node metastases versus 30% of distant metastases were polyclonal (Hu et al. 2020).

Exploiting the KP tracer mouse described above, which combines CRISPR continuous lineage tracing with the KP GEMM model of lung cancer, the phylogenies of metastasis were also interrogated. The authors found that metastases could be mapped back to specific spatial regions of the originating primary tumor. Intriguingly, the three liver metastases in the main case study example mapped to one end of the primary tumor, whereas the soft-tissue metastasis mapped to the other side of the same primary tumor (Yang et al. 2022). These results recall a nearly decade-old patient case report in which multiregion sequencing of a primary prostate tumor identified one specific, low-grade region (Gleason pattern 3; amid a tumor that was mostly Gleason pattern 4) of the primary tumor as the origin of metastasis (Haffner et al. 2013). Multiregion sequencing of renal carcinomas has also suggested that metastases arise from specific spatial locations, particularly from regions that are located in the hypoxic interior of the tumor (Zhao et al. 2021).

It is of note that the conclusions reached with the KP-tracer mouse, which align with observations in patients, are in some contrast to an earlier application of CRISPR-based lineage tracing to a xenograft model using human lung adenocarcinoma A549 cells (Quinn et al. 2021). This earlier study had focused on the relationship between a primary lung xenograft and its metastases and found that a very large proportion of primary tumor cells gave rise to progeny that seeded metastasis and that these metastatic cells were highly mobile, frequently recolonizing the primary tumor or populating another metastatic site (Quinn et al. 2021). This difference in conclusions highlights a major caveat in transplantation models: Transplanting a large pool of already-aggressively malignant cells can obscure the evolutionary processes and clonal dynamics that occur during spontaneous tumor development. Thus, although transplant models have a very wide range of powerful uses, evolutionary dynamics are best investigated in models in which tumors must undergo the full evolutionary process of progression.

TUMOR EVOLUTION TO AVOID IMMUNE RECOGNITION

A review of tumor progression would not be complete without discussion of the role of immune pressure on shaping tumor growth. In the early 2000s, Robert Schreiber proposed the “3E” model of immune selection, postulating that tumors go through progressive phases of “elimination,” in which the immune system keeps tumor growth in check; “equilibrium,” during which the immune pressure prunes immunogenic clones and the tumor evolves in response; and ultimately “escape,” in which a tumor successfully grows despite the presence of the immune response against it.

The first demonstration of immunoediting was in a carcinogen (methylcholanthrene)-induced sarcoma model in 2012. Sarcomas were induced in Rag2^−/− mice, which lack functional T and B cells, and then transplanted into mice with fully competent immune systems (Matsushita et al. 2012). The immunocompetent mice largely rejected the tumors that had been growing in immunodeficient mice. However, in rare cases, the sarcomas from immunodeficient mice did grow, and Matsushita et al. (2012) showed that it was consistently the genetic loss of a specific highly immunogenic mutation in spectrin-β2 that permitted this “escape” from immune control. In other words, some tumors were successful in evolving to “edit out” the immunogenic spectrin-β2 mutation, and these were the ones that could grow in the presence of an intact immune response.

This model of immunoediting has now been widely validated in patient studies, as immunotherapy has become a pillar of cancer treatment and brought new focus to the role of immune pressure on tumor development and progression. A recent study of early, untreated lung lesions in patients found abundant evidence of immunoediting, both in the form of immunogenic mutations that were epigenetically silenced, and in the form of loss of heterozygosity (LOH) of human leukocyte antigen (HLA) alleles (Rosenthal et al. 2019). A similar analysis of multiple samples from 394 tumors across 22 tumor types provided evidence of ongoing karyotype remodeling through common subclonal LOH at the HLA locus, indicating constant adaptation to the immune environment (Watkins et al. 2020). All antigens must be “presented” to T cells on HLA molecules, and patients carry several different alleles that enable them to present a specific set of antigens; thus, loss of an HLA allele represents a reduction in the diversity of antigens that can be presented to the immune system to mediate an antitumor response. In a similar vein, a multiregion sequencing study of lung cancer found that regions of a tumor that were “immune cold”—meaning they had a low abundance of T cells—were phylogenetically closely related, suggesting that a common “immune cold” tumor clone had undergone positive selection and come to dominate several regions of the growing tumor (Abduljabbar et al. 2020).

It is now being recognized that biological sex and age both play a role in the extent to which immune selection shapes tumor evolution. A pan-cancer analysis of patients stratified by age and sex found that younger patients and female patients had tumors with fewer immunogenic mutations than their older and male counterparts, and these effects were cumulative (Castro et al. 2020). These findings offer an explanation for reports that younger, female patients respond more poorly to immunotherapy: Their tumors may have evolved to be more “invisible” to the immune system even before the start of therapy.

Although immunoediting is now widely accepted, we are only beginning to understand how it fits in with other evolutionary dynamics of tumor progression and what biological factors influence the strength and extent of immune pressure. Mouse models with intact immune systems and physiologically relevant mutation burdens—such as those induced by carcinogens—will be important and powerful tools for future studies digging deeper into immune-mediated pressures on tumor evolution, for many questions remain as to how and when immunoediting occurs during tumor development.

ADVANTAGES OF MOUSE MODELS IN STUDYING TUMOR EVOLUTION

Mouse models have made significant contributions to our understanding of tumor evolution. Modeling cancer in mice offers the unparalleled ability to probe both the events contributing to tumor progression and the timing with which specific features of tumor evolution occur. Human patients, by the time their cancer is diagnosed, have typically experienced a wide range of carcinogenic and environmental exposures—ranging from sunburns to air pollution to dietary exposures—that are virtually impossible to recount. Each patient's tumor has its own unique genomic architecture that is the culmination of inherited genetics, the range of environments the patient has lived in, and the patient's own health history. Deconvoluting how each of these factors shapes tumor evolution using patient samples alone is thus challenging, if not intractable. Mouse models offer the unique opportunity to observe the evolutionary trajectories of tumors that have experienced precisely defined genetic manipulations, chemical exposures, or a combination thereof. Further, biopsies and specimens from human patients are typically collected at specific points that align with their care, and samples of very early tumors as well as late-stage metastatic tumors can be logistically challenging to collect. Throughout this review, we have made an effort to highlight the long history of mouse models providing insights into tumor evolution that were later shown, once technological advances permitted, to also drive human cancer, underscoring the value of these models.

CONCLUDING REMARKS

The recent demonstration that all human tissues are essentially complex mosaic structures harboring cells with many different types of spontaneous or carcinogen-induced mutations has forced us to reconsider how normal cells can maintain tissue function in the presence of multiple strong oncogenic mutations and the forces that stimulate these otherwise latent mutated cells to undergo neoplastic transformation. Normal tissues are themselves constantly evolving during the aging process, with many tissues showing high clonal diversity at a young age, with fewer but larger clones being selected with increasing age. Similar evolutionary processes are seen in tumors, with constant generation of new clones being balanced by clonal sweeps as “winner” clones are selected. Mouse models of cancer can make a unique contribution to analysis of the processes and exact mechanisms that underlie the cell–cell competition that results in winner and loser cells (Morata 2021). Combinations of mutagens, administered either constantly at low levels (Colom et al. 2020) or by a single treatment followed by promotion (see above), can mimic the clonal diversity and selection processes that operate during normal tissue growth and aging or that lead to outgrowth of malignant clones. Genetic models of cancer are less well-suited to these approaches as they generally have a very low point mutation burden and simultaneously induce identical mutations in many target cells in the same tissue, thus altering cell dynamics and competition. However genetic approaches are essential for tracing individual clones to follow cell fates and for identifying the precise mechanisms that control cell competition and emergence of malignant and metastatic clones. Combinations of these different approaches, particularly when complemented by new barcoding technologies for single-cell lineage tracing (Yang et al. 2022), offer great hope that in the not-too-distant future, mouse models will help us to reach a more complete understanding of the plasticity and cell state transitions that govern the evolution of both mouse and human cancers.