Review

Metastasis as an evolutionary process

See allHide authors and affiliations

Science  08 Apr 2016:
Vol. 352, Issue 6282, pp. 169-175
DOI: 10.1126/science.aaf2784

Abstract

Therapeutic advances in oncology have not fully translated to the treatment of metastatic disease, which remains largely incurable. Metastatic subclones can emerge both early and late in the life of the primary tumor. A better understanding of the genetic evolution of metastatic disease has the potential to reveal differences in the therapeutic vulnerabilities of primary and metastatic tumors, shed light on the temporal patterns of and routes to metastatic colonization, and provide insight into the biology of the metastatic process. Here we review recent comparative studies of primary and metastatic tumors, including data suggesting that macroevolutionary shifts (the onset of chromosomal instability) contribute to the evolution of metastatic disease. We also discuss the practical challenges associated with these studies and how they might be overcome.

Despite recent advances in the treatment of cancer, metastatic disease remains largely incurable and the main cause of cancer-related deaths. Metastases are the end result of a multistage process that includes local invasion by the primary tumor cells, intravasation into the blood or lymphatic system, survival in circulation (hematogenous and/or lymphatic), arrest at a distant organ, extravasation, survival in a new environment, and metastatic colonization. Each of these steps relies on specific phenotypic features of the tumor cell, as well as interactions with the host microenvironment and the immune system (1, 2).

Genetic and epigenetic alterations acquired by primary tumor cells and metastases probably contribute to these phenotypes and host interactions. Studying the genetic similarities and differences between primary tumors and metastases has the potential to provide new insights into the biology of metastasis. It may also reveal differences in the therapeutic vulnerabilities of local and systemic disease and shed light on the temporal patterns of and routes to metastatic colonization—information that could ultimately improve treatment and suggest strategies for preventing metastasis. Here we review recent studies examining the genetic evolution of metastases. We first describe some of the methodological challenges associated with determining the clonal relationship between the primary tumor and its metastases. We focus particularly on the challenge posed by intratumor heterogeneity (the occurrence of multiple genetically and geographically distinct tumor cell subpopulations within a single tumor), which has been documented in most solid primary tumors (3). We then synthesize and interpret the results acquired to date in this burgeoning research area.

Current models of the evolution of metastatic disease and challenges in distinguishing between them

There are two general models of metastatic dissemination: the linear progression model and the parallel progression model. Both models assume that the primary tumor and its metastases are clonally related, in that they derive from a common ancestral cell. The two models are distinguished principally on the basis of (i) the relative timing of the emergence of the metastatic precursor population in the primary tumor and (ii) the expected genetic divergence between the primary tumor and its metastasis. The latter (termed P-M genetic divergence) is the number of independent single nucleotide variants (SNVs) accumulated by the primary tumor and the metastasis after the appearance of the most recent common ancestor (Fig. 1).

Fig. 1 Phylogenetic relationships in paired primary tumors and metastases, based on sampling of one, two, or three regions of the primary tumor.

The phylogenetic trunk (blue) represents ubiquitous mutations, intermediate branches represent shared mutations (yellow), and terminal branches (red) represent private mutations. Metastases that disseminate from the primary tumor early show a substantial degree of genetic divergence (top left panel), whereas the late-arising metastatic subclone is very similar to the primary tumor (top right panel). Metastases can be identical to a subclone in the primary tumor (P3) or related to a subclone with evidence of additional alterations (P2), which is indicative of ongoing evolution. Limited sampling of the primary tumor can give the illusion of genetic divergence (top left versus bottom left panel) by failing to detect the metastasis-driving subclone in the primary tumor.

In the linear progression model (4), the metastasis-competent clone arises late in tumorigenesis and disseminates just before clinical detection of the primary lesion (5). The degree of P-M genetic divergence is expected to be small, because the metastasis is seeded by the most advanced primary clone or subclone (Fig. 1, top right panel). In the parallel progression model, the metastatic clone or subclone disseminates from the primary tumor early, and both the primary tumor and the metastasis continue to evolve in parallel, resulting in substantial genetic divergence between the primary and metastatic lesions (Fig. 1, top left panel). The parallel progression model assumes that multiple waves of dissemination of metastasis-establishing clones occur long before the primary tumor is clinically detectable, and that genetic alterations that are required for metastatic colonization evolve outside the primary tumor (5).

Interpreting genomic data within these frameworks is problematic, given that neither model accommodates the subclonal complexity (intratumor heterogeneity) of the primary tumor or the inevitable P-M divergence that arises through the inherent genomic instability of tumor cells (6). In comparative studies of P-M pairs, spatially limited sampling of the primary tumor (e.g., by a single biopsy) incompletely resolves its clonal structure and can lead to erroneous inferences of P-M divergence. This concept is illustrated in Fig. 1. The apparent genetic divergence between the primary tumor (P) and the metastasis (M) would decrease if two or more regions of the primary tumor (P1 and P2), rather than only one (P1), were sampled. Sampling a third region of the primary tumor (P3) would potentially decrease the apparent P-M divergence further if, for example, this region proved to be the origin of the metastasis. Inferences related to the timing of the emergence of the metastatic precursor population in the primary tumor, which are critical for distinguishing between the linear and parallel progression models, are likewise vulnerable to sampling bias. Thus, in the left column of Fig. 1, which depicts early divergence of the metastatic precursor population from the primary tumor, comparison of one primary tumor sample (P1) with the metastasis would support the parallel progression model, whereas data derived from comparison of three primary tumor samples (P1, P2, and P3) with the metastasis would support the linear progression model.

Several other caveats need to be considered when evaluating data from comparative genetic studies of primary tumors and metastases. Critically, the number and the breadth of clonal markers directly affect the inferred degree of genetic divergence in P-M pairs. Comparisons based on whole-exome sequencing (WES) or gene panels provide limited resolution relative to those based on whole-genome sequencing (WGS), and they may skew the data, because they assess only the coding regions of genes or known driver genes, respectively. Comparative studies in colorectal cancer (CRC) illustrate this point. In several studies that were based on the profiling of 45 to 1321 cancer-related genes (7, 8), P-M concordance rates were found to be 80 to 85%, rising to over 90% if only recurrently mutated genes were included (9). When the same P-M pairs were compared at the whole-genome level, the degree of P-M divergence increased (7). This observation is unsurprising, considering that gene panels are informed by our knowledge of driver genes. Current approaches for identifying driver genes are heavily skewed toward identifying early (clonal) drivers, which, unlike late (subclonal) drivers, are almost always shared between a primary tumor and its metastases (unless they are lost during tumor progression). Additionally, comparisons of P-M pairs that are limited to a single type of somatic alteration [usually SNVs and small insertions and deletions (indels)] also present an incomplete picture, especially given that somatic copy number alterations (SCNAs) affect a larger fraction of the cancer genome than any other type of alteration does (10).

Additional factors that can confound the inference of the mode of metastatic spread include exposure to systemic therapy or known carcinogens. For example, systemic drug treatment of patients before tumor sampling can reduce or enhance clonal diversity in both primary and metastatic sites (11), thereby affecting estimates of divergence between the primary tumor and metastases. Chemotherapy in particular has an impact on the mutational landscape, as has been shown for temozolomide-treated glioma (12) and platinum-treated esophageal cancer (13). In addition, the number of mutations accumulated before malignant transformation directly affects the apparent degree of divergence between the primary tumor and the metastasis (14). Carcinogen-induced tumors have a high mutational burden, but most of the mutations are probably acquired before the onset of tumorigenesis. For example, ultradeep sequencing of cancer genes in sun-exposed normal skin revealed an ultraviolet-induced mutational burden similar to that seen in many cancers (15). Thus, in cancers such as melanoma (16, 17) or smoking-related lung cancer (11), a high proportion of mutations are shared by the primary tumor and its metastasis, potentially masking their actual divergence.

Having outlined these caveats, we next evaluate the data generated by recent WES- or WGS-based comparative studies of primary tumors and metastases, with a focus on the analyses of multiple spatially and/or temporally separated biopsy samples. We restrict our discussion to clonally related P-M pairs or groups.

Comparative genetic studies of primary tumors and metastases

Large-scale comparisons have been limited to cohorts of unpaired primary and metastatic tumors and are frequently based on a small number of known genomic alterations. In the past 5 years, much smaller but paired cohorts have been profiled in a less biased fashion across a number of solid tumor types. Although descriptive in nature, these reports have provided unprecedented biological insights, and we therefore review them in detail.

Colorectal cancer

Genotype-based studies in CRC have consistently shown high levels of P-M concordance (18), leading some researchers to conclude that CRC conforms to the linear model of metastasis evolution. In the first prospective multiregional sampling study of CRC, Kim et al. (19) WES-profiled two to five primary regions and two to six liver metastases from five patients with microsatellite-stable metastatic CRC. Four of the five cases showed clear evidence of intratumor heterogeneity in the primary tumors, with up to 50% of coding mutations not shared by all the regions sampled. The primary tumor region that spawned the metastasis was identifiable in two cases, with enrichment for the metastatic clone evidenced by a higher cancer-cell fraction in the metastasis. “Private” genomic alterations (i.e., those that are found exclusively in one lesion) were observed in both primary and metastatic sites, although it is conceivable that these were in fact shared minor clones that evaded sampling or detection. Metastasis-private events included chromothripsis (a phenomenon in which one or a few chromosomes in a cancer cell bear dozens to hundreds of clustered rearrangements) and focal amplification of MYC. Chromothripsis was a shared event in another case that had evidence of additional SCNAs acquired in the rearranged chromosomal segments in the metastasis, indicating ongoing chromosomal instability (CIN). No alterations in known drivers of CIN or genome instability (GIN) were observed exclusively in the chromothripsis-containing metastasis; TP53 clonal mutations were observed in all the cases studied. These data suggest that not all cases of CRC conform to the linear model of progression and raise the possibility that CIN plays a role in metastasis.

Breast cancer

A WES comparison study of a basal-like breast cancer primary tumor and a metachronous brain metastasis (20) revealed 48 SNVs that were shared by the primary tumor and the metastasis and only two SNVs that were private to the metastasis. Greater than 90% of the SCNAs detected in the primary tumor were propagated in the metastasis, whereas ~80% of SCNAs in the metastasis were not shared by the primary tumor. Because these additional SCNAs were also observed in a xenograft model derived from the primary tumor, it is likely that they were present in a minor subclone in the primary tumor that seeded the metastasis and was also selectively engrafted. This observation further supports the role of CIN in metastatic progression. In a recent multiregional profiling study by Yates and colleagues (21), metastatic subclones were identifiable in a primary basal breast tumor that gave rise to a lung metastasis and an estrogen receptor–positive primary breast tumor that gave rise to a lymph node metastasis. In that study, SCNAs and TP53 mutations were detected both clonally and subclonally across all histological subtypes of primary breast cancer. Through profiling of 52 single cells from a primary basal breast tumor and 48 cells from its liver metastasis, Navin and colleagues (22) found a high level of concordance at the level of SCNAs, which is consistent with limited P-M divergence or reseeding of the tumor by the metastatic cells (23) (the direction of travel is discussed further below). Only one report provided evidence of true parallel progression, with considerable P-M divergence observed in a case of lobular breast cancer. However, this inference was limited by select SNV profiling of the primary tumor and exclusion of SCNAs from the analysis (24).

Pancreatic cancer

Two studies in pancreatic cancer based on SNV (25) and SCNA (26) data traced the metastatic subclones to specific regions of the primary tumor. Yachida and colleagues found that most deleterious mutations and CIN were accumulated in the primary tumor, with limited private alterations evident in the liver, peritoneal, and lung metastases, suggesting that the lethal subclone emerged late (25). Through modeling these data, the authors estimated an average of 7 years between the birth of the cell that gave rise to the parental clone and the seeding of the index metastasis, representing a substantial window of opportunity to prevent metastatic disease (25). In their analysis of primary pancreatic tumors and lung, liver, and peritoneal metastases, Campbell and colleagues (26) reported that structural rearrangements often occurred as shared events, indicating that these alterations occurred early in tumorigenesis. However, a varying degree of P-M divergence was noted, owing to ongoing evolution in both primary and metastatic lesions; the data were consistent with parallel progression involving known oncogenes (amplification of KRAS and MYC) in some cases.

Renal cell carcinoma

Spatial and temporal resolution of a primary renal cell carcinoma (RCC) (27) revealed the origin of its metastatic subclone, which continued to evolve after dissemination to the chest wall and perinephric fat. Parallel evolution of distinct mutations in SETD2 (coding for a histone-lysine N-methyltransferase) and convergent SCNAs were observed in two cases with chest wall and liver metastases, respectively (28). Parallel evolution of distinct mutations in PBRM1 (coding for a component of the SWI/SNF-B chromatin remodeling complex) was reported in a primary RCC and its brain metastases (29). Similar patterns were observed in four CRC cases with respect to distinct mutations in TP53 (7). These findings indicate a strong selective pressure for (in)activation of these genes, both locally and systemically.

Prostate cancer

Several recent studies in prostate cancer P-M pairs have been particularly illuminating in terms of the wide range of patterns of metastatic progression. Comparative analysis of 333 primary prostate cancers (represented by single biopsies) (30) and an unrelated cohort of 150 bone and soft tissue metastases from castration-resistant prostate cancer (31) found the mutational and SCNA burden to be significantly higher in the metastases than in the primary tumors. These findings are consistent with the observation that patients with a high burden of SCNAs have an increased risk of relapse (32). Overall, the metastatic samples showed more frequent alterations in the AR gene (encoding the androgen receptor), TP53, RB1, the lysine N-methyltransferase genes KMT2C and KMT2D, and genes implicated in the DNA repair and phosphatidylinositol 3-kinase (PI3K) pathways. These results suggest substantial P-M divergence in prostate cancer, but they could also reflect the failure to sample the minor subclone in the primary tumor that spawned the metastasis. Accordingly, WGS analysis of multiple metastases that arose 17 years after resection of a primary prostate cancer traced their origin not to the bulk of the tumor but to a 2-mm low-grade region of it (33).

Using a combination of WGS and deep targeted sequencing, Hong and colleagues (34) profiled multiple regions of four primary prostate cancers and their associated metastases. Two primary tumors showed clear evidence of branched evolution. In the first case, the metastatic subclone was detected in the primary tumor, with additional variants involving TP53, AR, and MSN (coding for moesin) detected exclusively in the metastasis; in the second case, the primary tumor was largely unrelated to its metastasis, indicating early divergence of the metastasis-seeding clone and parallel progression in the primary tumor and metastasis. Enrichment for mutations in TP53, AR, and MYCN was also observed in the metastases in these four cases, as well as in an additional 19 cases in the extension cohort (34). Where paired primary lesions were available, TP53 mutations were detected in minor subclones of the primary tumor. Taken together with the data from CRC studies, these findings offer further evidence that the acquisition of TP53 mutations is associated with the expansion of subclones with metastatic potential. Microsatellite instability (MSI) and a mutation in BRCA2 were also observed as metastasis-exclusive in two cases (34). Last, there was evidence of two separate waves of metastatic spread from the primary tumor to the metastasis, suggesting repeated or continuous metastatic seeding. This observation has important clinical implications. If the primary tumor has the capacity to repeatedly seed metastases, then its removal, even if metastatic disease is already established, could halt further metastatic progression. It is this removal of the reservoir of diverse metastatic clones that is postulated to contribute to the survival benefit associated with palliative resection of the primary tumor in some advanced cancer, as observed in RCC (35).

Gundem et al. (36) characterized the subclonal architecture of ten cases of prostate cancer in which primary tumors had multiple paired metastases. In cases where the origin of the metastases was identifiable in the primary tumor, it was always a minor subclone, and both the primary and metastatic tumors continued to evolve after dissemination. In some cases, multiple subclones with different degrees of divergence from the primary tumor gave rise to the metastases, indicating independent and temporally separate seeding, as was observed by Hong and colleagues (34). AR gene amplifications were subclonal in 16 of 17 primary tumors, which is consistent with the development of secondary resistance to androgen deprivation therapy (ADT) (37); the sequential gains of the AR gene that were observed in some cases imply continuous pressure exerted by therapy that selects for alterations in this gene and/or pathway. A similar phenomenon has been observed in breast cancer. Treatment of a patient exhibiting PIK3CA (phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit alpha)–mutant metastatic breast cancer with a drug that inhibits the PI3Kα signaling pathway resulted in parallel evolution of six distinct PTEN mutations in the metastatic sites, which had progressed through therapy and which were sampled at autopsy (38). Thus, selective pressure from therapy can also play a role in shaping metastatic progression.

In a recent report, Zhao et al. presented 40 P-M WES analyses across 13 different tumor types, including 13 cases of lung cancer (39). A varying degree of P-M divergence was observed. Early divergence of metastases, consistent with the parallel progression model, was detected in 11 cases. However, no clear correlation was observed between the degree of P-M divergence and tumor type.

Several conclusions can be drawn from these analyses. First, evidence from these studies indicates elements of both linear and parallel metastatic progression within specific tumor types and even within individual cancers (Fig. 1, middle column). Although it may not be important to assign the model of progression correctly, an estimate of the timing of metastatic spread from the primary tumor has important clinical implications. If metastatic dissemination occurs before the primary tumor is clinically detectable, then early surgical resection may fail to reduce the risk of metastatic disease in the future.

Second, notwithstanding the small number of cases included, the studies reviewed here offer support for the idea that a macroevolutionary shift (defined as large-scale genomic alterations, such as copy number changes and structural chromosomal rearrangements manifested as CIN) is often evident at metastatic sites. Increases in CIN have been observed in parallel to metastatic progression in cases of prostate cancer (31, 32), pancreatic cancer (26), breast cancer (20, 21), CRC, and RCC (27). CIN is linked to poor outcomes in cancer treatment (40) and to metastatic progression in mouse models of skin carcinogenesis (41) and pancreatic adenocarcinoma (42). Evidence of parallel evolution was observed for TP53, as well as for other genes that play a role in maintaining genome stability: SETD2 through nucleosome stabilization, suppression of replication stress, and the coordination of DNA repair (43, 44) and PBRM1 through promoting cohesion of chromatin at centromeres (45). In RCC, SETD2 and PBRM1 loss-of-function mutations were frequently subclonal, suggesting that they are preferentially inactivated later in the evolution of the primary tumor, triggering CIN. The observed relationship between CIN and/or GIN and metastatic progression is reminiscent of the progression from pre-invasive to invasive disease, which is linked to the onset of CIN; two examples are the progression of Barrett's esophagus to esophageal carcinoma (46) and the progression of melanoma in situ to invasive melanoma (47). Karyotypic abnormalities such as genome doubling and chromothripsis were observed as a late event in some lung adenocarcinomas (11) and a metastasis-specific event in CRC (19), respectively. MSI and a BRCA2 deficiency were detected exclusively in the metastasis of two prostate cancer cases (34), suggesting that distinct forms of GIN can be associated with metastatic progression.

Genetic divergence between metastases: The route and destination of metastatic cells

Genetic approaches have also been used to determine the relationship between two or more metastases arising from the same primary tumor. Such studies can ascertain whether primary tumors contain multiple metastatic lineages that disseminate independently of each other and whether metastases that spread via lymphatics or blood (the route of travel) or to particular organs (destination of travel) are underpinned by genetic differences.

The relationship between metastases can also be described using the terms linear and parallel. In one hypothetical scenario, the same subclone of the primary tumor can seed both a metastatic lesion within a local lymph node (LN) and a metastatic lesion at a distant site, such as the liver or brain (Fig. 2). It could do so in a linear (P→LN1→M1 in Fig. 2) or a parallel and independent fashion (P→LN1 and P→M2 in Fig. 2). In another hypothetical scenario, the subclone that seeds the distant metastasis (M3 in Fig. 2) may be distinct from the subclone that seeded the lymph node. Differentiating these scenarios in reality has profound implications for patient management, not least in terms of the morbidity of lymph node dissection. It is widely accepted that the clearance of involved lymph nodes improves local disease control, but whether it influences systemic dissemination is less clear. In cases where systemic metastases are established through hematogenous spread, lymph node clearance should have no overall survival benefit; conversely, if involved lymph nodes are a gateway to distant metastases, then their removal should reduce the risk of dissemination. A small number of comparisons between paired lymph node and distant metastases have been reported. In a WES analysis of a primary breast cancer, its axillary lymph node metastasis, and the synchronous systemic metastases, Murtaza et al. showed that a minor subclone in the lymph node gave rise to the distant metastases (48). Using WGS, Haffner and colleagues (49) tracked the clonal evolution of a prostate cancer that relapsed 17 years after radical surgery. Liver, bone, nodal, and soft tissue metastases were not seeded by the involved lymph node that was removed at surgery, but rather arose from a minor subclone in the primary tumor and contained additional alterations, including AR amplification, which were probably therapy-selected. A comparative study of metastatic melanoma demonstrated that one subclone of the primary tumor seeded loco-regional (i.e., localized to the area of the primary tumor) and lymph node metastases, whereas another subclone seeded a brain metastasis; in contrast, in another patient with melanoma, the lymph node and brain lesions grouped together and were distinct from the loco-regional metastasis (16). In combination, these studies show that systemic metastases can occur via direct hematogenous spread, bypassing the lymph node; however, some involved lymph nodes can seed systemic metastases, presumably through lymphovascular shunts.

Fig. 2 Metastatic spread can take place through multiple routes and in different directions.

The primary tumor (P) is shown in the center of the diagram, with different colored circles representing distinct subclones. Lymph nodes can be seeded by single clones (LN1) or multiple clones (LN2). Lymph nodes can seed systemic metastases directly (M1), or the same primary subclone can seed a systemic metastasis (M2) in parallel to seeding the lymph node. A subclone other than the one that seeded the lymph node can seed individual systemic metastases in parallel (M3, M4, and M5; monophyletic metastases). Alternatively, it can seed just one metastatic site (M3), which in turn can seed other metastases (metastatic cascades; M6 and M7). Metastases can also be polyphyletic, with distinct subclones seeding different metastases (for example, M1, M3, and M8). Metastases can undergo parallel evolution, indicated by the double circles (M8 and M9). Metastases can continue to evolve after they have disseminated from the primary tumor, represented by the dual colors in M10. These metastases can then potentially re-infiltrate the primary tumor or surgical bed, a process called self-seeding. Cross-metastatic seeding can also occur, resulting in complex subclonal mixtures in the metastases themselves (M10→M9).

The same subclone in the primary tumor can seed multiple metastases in different organs; such metastases are termed monophyletic. The seeding of these metastases can occur in parallel (P→M3, P→M4, and P→M5 in Fig. 2), in which case the metastases share the parental clone but can each harbor additional private subclonal mutations. Metastases can also appear monophyletic as a result of a metastatic cascade (M3→M6→M7 in Fig. 2), in which one metastatic lesion gives rise to another, and thus they are all closely related. Experimentally, it is challenging to distinguish these two models, unless the primary tumor is profiled exhaustively and all the metastases are sampled (49). Data on multiple metastases from the same organ, including from the lung (26), skin (50), and brain (51), are consistent with metastatic cascades taking place within organ boundaries. When distinct subclones in the primary tumor seed metastases in different target organs in parallel (M2, M3, M8, and M10 in Fig. 2) and at different times, the metastases are termed polyphyletic. In one illustration of this, distinct subclones of a primary pancreatic cancer that seeded metastases in the lung and abdominal wall were found to harbor distinct mutations in PARK2 (52), indicating parallel evolution between metastases. In a study of multiple metastases across a range of tumor types, monophyly and polyphyly were not correlated with any particular tumor type (39).

Some tumor types have the tendency to metastasize to certain organs and not to others, a phenomenon termed organ tropism (53). An extreme example is uveal melanoma, which metastasizes preferentially to the liver. Epidemiological evidence supports a relationship between organ tropism and certain tumor genotypes; for example, KRAS-mutant CRC is more likely to metastasize to the lung than to the liver (54), and half of HER2-amplified breast cancers metastasize to the brain (51). However, it is not known how these or any other genomic alterations influence organ tropism within or across tumors. Brain tropism is of particular interest, both biologically, because of the presence of the blood-brain barrier, and as a clinical problem, because of the poor prognosis associated with the diagnosis of brain metastases. Brastianos and colleagues (29) used WES to examine 86 cases spanning a variety of primary tumors and single or multiple paired brain metastases. In about half of the cases, clinically informative genetic alterations (those that present actual or potential therapeutic targets), including CDKN2A, PTEN, and PIK3CA, were exclusive to the brain lesions. These observations are in keeping with the previous reports of increased activation of the PI3K-AKT pathway in brain metastases in general (5557), suggesting that therapies targeting this pathway could play a role specifically in the treatment of brain metastases.

Genetic divergence between metastases: The direction of metastatic spread

Another phenomenon that can obscure phylogenetic relationships in genomic studies of metastases is self-seeding (M10 in Fig. 2). Thus far, we have discussed the various routes of metastatic spread but have assumed that the spread is unidirectional—that is, primary tumor cells seed the metastases, and metastatic cells do not seed the primary tumor. However, self-seeding (23), a process by which metastatic cells can re-infiltrate their tumor of origin, has also been proposed. Hypothetically, this would make the primary and metastatic tumors appear to be closely related. Experimentally, self-seeding is difficult to differentiate from the conventional linear progression from primary tumor to metastasis, because both processes would result in limited P-M divergence.

A proof of multidirectional seeding was provided by a study of human prostate cancer (34): The recurrence at the site of the resected primary tumor was found to be divergent from the original primary tumor and was seeded by a clone derived from a bone metastasis (an example of reseeding of the primary tumor by the metastasis). Repeat sampling also revealed cross-metastasis reseeding (M10→M9 in Fig. 2) in response to ADT—that is, a resistance-bearing subclone from one metastasis invading an originally therapy-sensitive metastasis (34). Self-seeding could account for the observations made by Gundem and colleagues (36) and Haffner and colleagues (49), in which the origins of prostate cancer metastases were traced to a very small subclone in the primary tumor. The alternative hypothesis is that the metastases described in these two studies disseminated from the primary tumors, evolved further at distant sites, reentered the circulation, and reseeded the primary tumors, resulting in the detection of a minor metastatic subclone in the primary tumor.

Clonal dynamics between metastases: Do metastatic cells compete or cooperate?

When metastases are seeded by single cells or clones from the primary tumor (LN1 in Fig. 2), a reduction in genetic diversity is observed; that is, there is a decrease in the number of observed mutations in the metastasis relative to the primary tumor (provided that all of the subpopulations in the primary tumor were characterized) (M2, M3, and M8 in Fig. 2). This is analogous to a bottlenecking event in evolutionary biology. Further evolution can occur in the metastasis, resulting in subclonal mutations, but these are not shared with the primary tumor (M10 in Fig. 2).

Comparative studies, as well as profiling of unpaired metastases, indicate monoclonal patterns of seeding across many human cancers (19, 25, 26, 29, 34), suggesting that clones compete to metastasize. However, polyclonal seeding, in which multiple clones from the primary tumor seed the same metastasis, is also observed (LN2 and M9 in Fig. 2); this indicates that subclones might cooperate as well as compete to metastasize. Comparative genomic studies have found that polyclonal seeding results in primary and metastatic tumors sharing multiple subclonal populations (17, 36). Polyclonal seeding can occur in parallel (direct clonal cooperation) or in temporally separate waves (indirect clonal cooperation); for example, the initial clone can remodel the metastatic niche, making it attractive for additional clones to colonize later (16, 36). In an example illustrating polyclonal seeding, two subclones in a melanoma primary tumor that were characterized by distinct mutations in the catenin beta 1 gene CTNNB1 were found to seed the same loco-regional metastasis (16), suggesting that parallel evolution may support polyclonal seeding. Similarly, two subclones that contributed to the polyclonal seeding of a lymph node in a castration-resistant prostate cancer harbored distinct alterations associated with ADT resistance: MYC amplification and a pathogenic AR substitution (M9 in Fig. 2) (36).

Evidence from mouse models supports the notion that metastasis can arise from the collective migration and colonization of cancer cells (58, 59) and that cell clusters can be more efficient than single cells at forming metastases (58). Murine models of small cell lung cancer have demonstrated polyclonal seeding of lymph node metastases (60) and how one subclone can endow another with metastatic capacity through paracrine signaling (61). Recently, using lineage labeling, Maddipati and Stanger (57) demonstrated the polyclonal origin of metastases in a mouse model of pancreatic cancer. Cells were shed from the primary tumor as multiclonal aggregates, rather than one cell at a time. Although all metastatic sites were seeded polyclonally, monoclonal or polyclonal patterns of metastatic outgrowth were site-specific. Whereas peritoneal lesions remained polyclonal, liver and lung lesions drifted toward monoclonality, presumably reflecting different selective pressures at these metastatic sites. These data suggest that both clonal competition and clonal cooperation play a role in the evolution of metastases. In practical terms, the human tumor data indicate that a single biopsy of a single metastatic site may not accurately portray the polyclonality of that particular metastasis or the overall diversity of metastases.

Conclusions

Evidence from most comparative genomic studies of primary tumors and metastases indicates elements of both linear and parallel metastatic progression (Fig. 1, middle column), even within the same patient. The timing of active sampling and analysis of the primary tumor relative to the disease course could affect which model of progression is observed. Thus, the two models are not mutually exclusive and are part of a biological continuum. Temporal waves of early and late, multidirectional, monoclonal and polyclonal, and monophyletic and polyphyletic metastatic spread are observed across cancers and within individual cases. Overall, no clear relationship has emerged between the mode of metastatic progression (linear versus parallel, monoclonal versus polyclonal, or monophyletic versus polyphyletic) and the characteristics of the primary tumor, including tumor type and other clinical features. To date, there is little compelling evidence for the existence of specific metastasis-driving genes, with the exception of TP53. Mutations in TP53 appear to be linked to expansion of the metastatic subclones in prostate cancer (34) and in a subset of CRCs (28), possibly because TP53 dysfunction is associated with tolerance of chromosomal instability and rearrangements (62) and chromothripsis (63). In this regard, CIN is a genetic hallmark that is frequently identified in metastatic subclones. Thus, there appears to be a link between the onset of CIN and metastases, but for this link to be corroborated, it is essential that future comparative studies include analysis of somatic copy number and structural alterations, in addition to SNVs and indels assessed longitudinally through the disease course.

The multitude of phenotypes required for metastases would be hard to achieve in a microevolutionary stepwise manner (i.e., through small-scale genomic alterations, such as single nucleotide substitutions and small indels affecting single genes). It is conceivable that macroevolutionary leaps (large-scale genomic alterations) (64) could catalyze all the steps to metastases, especially in narrow time frames, such as in synchronous presentation of primary and metastatic disease.

Future large-scale comparative studies with longitudinal and spatial sampling and robust clinical annotation will be required for reliable documentation of the patterns of metastatic dissemination within and across tumor types. Inclusion of postmortem sampling will be critical, because this allows exhaustive sampling of metastatic sites; evidence from autopsy studies shows that subclinical metastases are very common (65). Liquid biopsies have been shown to reflect the clonal dynamics of primary and metastatic tumors (48, 66, 67) and are an important adjunct to tissue analyses. The results of such studies could illuminate the relationship between the mode of metastatic spread and important clinical characteristics. These characteristics include primary tumor size [considering, for example, that the risk of late recurrence is the same for small and large breast primary tumors (68)]; the timing of metastases [considering the wide range observed across tumor types, from early (non–small cell lung cancer) to late (melanoma, RCC, and breast cancer)]; and patterns of metastatic spread [organ tropism and whether metastases are limited to a single organ (oligometastatic disease) or widespread]. The ability to anticipate the route, direction, and timing of metastatic spread could inform many aspects of surgical management, including the value of metastasectomy (the surgical removal of metastases), lymph node removal, and palliative resection of primary tumors for disease control; management of small tumor masses (renal or prostate); and individualized follow-up care for patients who have had surgery with curative intent. Lastly, such large-scale studies will have the statistical power to identify metastasis-permissive and metastasis-suppressive processes in the tumor microenvironment and the immune system, thereby creating therapeutic opportunities to limit this almost intractable facet of cancer medicine.

References and Notes

  1. Acknowledgments: S.T. is a Cancer Research UK clinician scientist and is funded by Cancer Research UK (grant reference number C50947/A18176) and the National Institute for Health Research (NIHR) Biomedical Research Centre at the Royal Marsden Hospital and Institute of Cancer Research (grant reference number A109). C.S. is a senior Cancer Research UK clinical research fellow and is funded by Cancer Research UK (TRACERx project), the Rosetrees Trust, the NovoNordisk Foundation (grant ID 16584), the European Union 7th Framework Programme for Research and Technological Development (projects PREDICT and RESPONSIFY; grant ID, 259303), the Prostate Cancer Foundation, the Breast Cancer Research Foundation, the European Research Council [THESEUS (Tumour Heterogeneity and Somatic Evolution of Unstable cancer genomes) project], and the NIHR University College London Hospitals Biomedical Research Centre. We thank N. McGranahan and M. Jamal-Hanjani for their helpful review of the manuscript. We apologize to those authors whose work we did not cite because of space constraints. C.S. is a paid advisor for Janssen, Boerhinger Ingelheim, Ventana, Novartis, Roche, Sequenom, Natera, GRAIL, Apogen Biotechnologies, Epic Biosciences, and the Sarah Cannon Research Institute, and he is a stockholder in Apogen Biotechnologies, Epic Sciences, and GRAIL.
View Abstract

Stay Connected to Science

Navigate This Article