Intratumor Heterogeneity: Evolution through Space and Time

Intratumor heterogeneity and evolutionary bottlenecking

Through the analysis of structural variations and allelic frequencies in a primary basal breast cancer, a xenograft, and a brain metastasis from the same patient, Ding and colleagues showed that the metastasis may derive from a low-frequency subclone within the primary (17). The team noted a wide range of allelic variant frequencies in the primary tumor, indicative of substantial ITH, with less divergent mutational frequencies at the metastatic site, suggesting a process analogous to evolutionary bottlenecking through subclonal selection during the metastatic process (18). Substantiating the capacity for such heterogeneity to foster metastatic growth, medulloblastoma metastases from the same patient are relatively homogeneous and derive from a low-frequency subclone of the primary tumor (19).

Through our multiregion sequencing analysis of clear-cell renal-cell carcinomas (ccRCC), metastases could be traced back to a distinct region of the primary tumor (14). Consistent with observations from Ding and colleagues, there appeared to be a relative restriction of diversity at metastatic sites (17). Sector ploidy and allelic imbalance analysis of the tumor and metastatic sites suggested that the primary region that spawned the metastasis had become tetraploid with the metastatic sites harboring a chromosomally unstable pattern by allelic imbalance analysis with structural chromosomal complexities that differed between regions of the same metastatic site (Fig. 2). Conceivably, following subclonal selection and the restriction of diversity and bottlenecking, generation of tumor chromosomal instability (CIN) provides a route to rapidly initiate a further expansion in tumor heterogeneity. It is tempting to speculate that this may explain observations of increased CIN at metastatic sites to compensate for the transient restriction of diversity during the selection of a minority tumor subclone (20). The parallels with the work of Harris and colleagues on dynamic heterogeneity are notable as genomic instability driven by heterogeneous structural and numerical chromosomal changes, promoting extensive alterations in gene dosage, may be one such way of creating nonmutational routes to tumor metastases (4, 5). Conceivably, such mechanisms may also be reversible, as postulated in the dynamic heterogeneity model, as distinct alterations in gene dosage permissive for metastatic outgrowth may be lost in subsequent cell divisions because of spontaneous chromosome reassortments.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Evolutionary bottlenecking and restriction of diversity between primary and metastases. A, transient restriction of diversity through an evolutionary bottlenecking process, for example, during clonal selection through therapy or during metastatic progression may lead to the requirement for alternative mechanisms to generate ITH. One such mechanism can be generated through structural and whole CIN. Tetraploidy is thought to be a precursor of CIN. B and C, multiregion analysis of a primary renal-cell carcinoma and its metastatic sites revealed that the metastatic lesions were most similar to region 4 of the primary. Ploidy analysis revealed that the primary regions were all diploid with the exception of region 4, which was tetraploid. Metastatic sites were subtetraploid. D, allelic imbalance analysis revealed substantial heterogeneity in chromosome structure between 2 biopsies of the same metastatic site (M2a and M2b). Taken together with ploidy analysis, these data indicate the possible emergence of CIN at metastatic sites. Longitudinal studies will reveal to what extent metastatic sites represent outgrowth of multiple heterogeneous subclones from the primary tumor and genetic events that may be permissive for metastatic outgrowth during the bottlenecking process. Adapted from Gerlinger et al. (14, 18).

Tumor sampling bias

Tumor sampling bias may arise because of differences in somatic events within the primary tumor, between the primary and metastatic sites, between metastatic sites, or even within single biopsies (Fig. 1). Heterogeneity is also dynamic and evolves over time, as has been observed in elegant FISH studies in cytogenetically high-risk multiple myeloma (24). Dynamic changes in the subclonal architecture of the tumor, where tumor subclones may change and compete with each other for dominance during the disease course and through treatment, create challenges for predictive or prognostic biomarker efforts in which the tumor subclone that defines clinical outcome (e.g., secondary plasma cell leukemia) may not be readily detectable at diagnosis (24).

Therapeutic decision making in oncologic practice is often made with reference to the primary tumor lesion, diagnosed months or years previously, or in cases in which patients present with advanced disease, from one metastatic site. Such approaches are likely to be therapeutically tractable if these tumor somatic events occur in the tumor trunk and are present ubiquitously throughout all tumor subclones and continue to maintain tumor growth and survival at all sites of disease. However, the changing dynamics of tumor subclonal architecture over space and time may result in previously subdominant clones, perhaps either not present or barely detectable in the primary, gaining preeminence. Conceivably, differences in tumor environmental selection pressures at metastatic sites may result in regional variation of tumor subclone evolution as distinct environments select for certain subclones over others, further contributing to ITH (11). Therefore, alterations in the subclonal architecture of the tumor may result in changes in tumor molecular profile that may differ between sites of metastatic disease, distinct from the profile of the primary tumor.

Qualification of clinical biomarkers has been a notoriously difficult process with only 100 of the estimated 150,000 biomarkers qualified and implemented into clinical practice (25). Biomarker discovery approaches combining laboratory analyses of gene function with genetic or transcriptomic analyses of tumor tissue often rely on single tumor biopsies of the primary or metastatic lesions to prioritize the identification of candidate biomarkers for validation (26). Conceivably, a highly variable genetic and transcriptomic landscape across a primary and metastatic tumor might lead to tumor sampling bias confounding the validation of biomarkers because of spurious associations of heterogeneous tumor genetic events with clinical outcome in the discovery phase of these studies.

Capacity for therapy to augment intratumor heterogeneity

In contrast to mutator phenotypes and microenvironmental factors such as tissue hypoxia and acidosis that may enhance ITH (27), the capacity for cytotoxic therapies to augment ITH remains relatively underexplored. In the study by Keats and colleagues, the enhanced complexity of the relapsed subclone suggested that treatment with DNA damaging agents through myeloma therapy may potentially exacerbate genomic complexity (24). The potential for therapy to augment genome instability and ITH has been observed in relapsed methyl-guanine-DNA methyltransferase-deficient glioblastoma multiforme following alkylator therapy that provides a selection pressure to lose DNA mismatch repair function (28). Similarly, cytotoxic therapy was thought to be responsible for the increase in transversions witnessed in relapsed acute myeloid leukemia following exposure to DNA damaging agents (29). Given the association of ITH with poorer clinical outcome (30), such observations suggest the controversial concept that therapeutics may in some cases contribute to enhanced tumor diversity and adaptation.

Therefore, cancer cytotoxics may contribute to an enhanced tumor evolutionary rate through fostering genetic diversity and, as proposed by Gillies and colleagues, by initiating an adverse tumor microenvironment that enables small phenotypic changes to result in large variations in fitness (27).

Defining actionable mutations within a model of clonal dominance

Tumor deep-sequencing analyses have led to attempts to stratify therapeutics based on the occurrence of “actionable mutations” where a clinician matches a tumor mutation to a cancer drug. Somatic mutational heterogeneity, where distinct mutations may be present in some biopsies but not detectable in others, suggests a more cautionary approach to the clinical implementation of deep sequencing in circumstances where previously undescribed “actionable mutations” might be considered as suitable for clinical trial stratification. Given observations of the spatial separation of somatic mutations, it is conceivable that some actionable mutations, particularly ones with limited prior evidence of relevance in a disease subtype, may not be present within all tumor cells (present at submodal frequencies) or may be regionally separated, and thus represent relatively poor therapeutic targets (Fig. 1). Similarly, synthetic lethal approaches to drug discovery are likely to have optimal efficacy if the genetic dependency of the drug discovery approach occurs early in tumor evolution, present in all tumor subclones, as witnessed by the potent efficacy of PARP inhibition in tumors of BRCA germline carriers (31).

Perhaps, within the context of early-phase clinical trial development, consideration could be given to the clonal dominance of such “actionable mutations,” as sampling multiple sites of disease, to establish the ubiquitous nature of such mutations, is not only traumatic for the patient but also impractical clinically. For example, recent evidence, in glioblastoma and triple-negative breast cancer, indicates that commonly accepted “actionable mutations” in signal transduction pathway regulators such as PTEN or PIK3CA may represent somatic events occurring in the branches of the tumor rather than clonally dominant ubiquitous, trunk mutations (22, 32).

Such considerations should not imply that genetic approaches to treatment stratification from single biopsies are futile. Many of the established biomarkers in oncology derive from, and have been clinically qualified in, single tumor biopsies. Examples include HER2 amplification or overexpression to guide trastuzumab therapy in breast cancer, BRAFV600E mutation to guide BRAF inhibitor treatment in melanoma, and epidermal growth factor receptor (EGFR) activating mutations to guide gefitinib or erlotinib treatment in adenocarcinoma of the lung.

It is likely that many established drivers of disease biology occur early in tumor development and represent ubiquitous events in the “trunk” of the tumor, present at all sites of disease, less subject to tumor sampling bias (21). In the case of renal cancer, such trunk events would be represented by VHL or PBRM1 somatic mutations (14). However, low-frequency somatic events present in some subclones but not others (analogous to the branches of the tumor tree), generating ITH, are likely to contribute to the acquisition of drug resistance driven by Darwinian selection through treatment. Low-frequency gatekeeper mutations in the EGFR tyrosine kinase are associated with inferior progression-free survival on EGFR tyrosine kinase inhibitor therapy (33). Similarly, non–small cell lung carcinoma (NSCLC) exposure to EGFR tyrosine kinase inhibitors can result in the selection of resistant subclones harboring low-frequency mesenchymal epithelial transition factor amplification, present in the tumor before treatment (34).

Such examples suggest that biomarker efforts may have to adapt to the challenge of detecting heterogeneous somatic events present in the tumor at low frequency to predict therapeutic outcome or define combination approaches to limit the acquisition of drug resistance as well as understand the complex phenotypic interplay between heterogeneous cancer subclones. ITH supports a more cautious approach when defining the presence of an actionable mutation for clinical trial stratification to encompass a definition that incorporates clonal or interregional tumor dominance.

Cancer as a complex ecologic system dependent on population level heterogeneity

The clonal dominance model to define an actionable event does not necessarily preclude the possibility that tumor growth can be limited through the targeting of heterogeneous, low-frequency subpopulations. Evidence suggests that minority cancer cell populations may maintain ITH and by inference, targeting such low-frequency cells might have an impact on the bulk tumor population; Low-frequency glioblastoma subclones harboring mutated EGFR maintain ITH through paracrine activation of proliferation of the EGFR–wild-type cancer cell population (35). Considering tumors as ecologic niches with complex functional interdependencies may be necessary to further refine definitions of actionable mutations to attenuate tumor growth. Taking advantage of ITH through enforced competition between tumor subclones forms the basis of elegant models of tumor adaptive therapy (7).

Cancer cell phenotypic heterogeneity and drug sensitivity

ITH is likely to have direct phenotypic consequences on tumor behavior and the acquisition of drug resistance. For example, low-frequency subclones, detectable in the tumor before treatment, harboring resistance mutations in EGFR in NSCLC (33) as well as multiple distinct secondary mutations in c-KIT in different metastases that occur in gastrointestinal stromal tumor following KIT/platelet-derived growth factor receptor (PDGFR) tyrosine kinase inhibitor therapy (36) confer resistance to targeted approaches.

Our group has shown that a heterogeneous mutation near the kinase domain of mTOR promotes resistance to serum deprivation and hyperactivation of the mTOR pathway following everolimus therapy in regions of the primary tumor that harbor the mutation, but not in primary tumor regions with wild-type mTOR. Similarly, extensive ITH in DNA copy number events of “drivers” such as MET, PDGFRA, and EGFR have been shown to occur in a mutually exclusive manner in glioblastoma (37). Adjacent glioblastoma tumor cells display distinct copy number abnormalities in these “targetable” or actionable receptor tyrosine kinase amplification events. Heterogeneous copy number events present in the tumor branches, rather than trunk, suggest that targeting individual branched genetic lesions may prove relatively futile if ITH results in phenotypic heterogeneity in drug sensitivity. Szerlip and colleagues have confirmed this by demonstrating that cell lines grown from the same glioblastoma with heterogeneous PDGFR or EGFR amplification states require both PDGFR and EGFR inhibition for maximal phosphoinositide 3-kinase pathway attenuation and growth inhibition (38). Intriguingly, such heterogeneity is likely to be maintained and selected for by the presence of double-minute chromosomes harboring RTK amplification. Such double-minute chromosomes lack centromeres and are, therefore, unequally segregated during mitosis, resulting in the propagation of ITH. Phenotypic heterogeneity in drug sensitivity profiles that may be spatially separated or present at low frequency in minor subclones of the tumor suggests that profiling cancer cell phenotypes from single biopsies to guide therapeutic decision making from heterogeneous tumors may prove challenging.

In summary, although a proportion of tumors of the same histopathologic subtype may share common drivers, branched events initiating heterogeneity in potential resistance pathways to targeted therapeutics will likely result in the need to consider individual tumor-specific strategies to extend progression-free intervals. A future in which personalized medicine moves from the current status of patient cohorts defined by single trunk tumor driver events, to the single patient, in which both trunk and branch tumor events are characterized in advance of treatment, and in which no 2 tumors share the same characteristics, may be envisaged. This may have important regulatory, ethical, and health economic implications and raises the need to incorporate an understanding of tumor heterogeneity into clinical trial design. For these reasons, some rightly argue that focusing on the consequences of genetic diversity in terms of common tumor phenotypes, rather than the specific genetic aberrations themselves, may prove more beneficial (27).

Relationships between intratumor heterogeneity and clinical and pathologic parameters

The presence of extensive ITH within primary renal cell carcinomas suggests a rational basis for the improvements in survival outcome associated with palliative surgery to the primary site in patients with oligometastatic disease, through the removal of an evolutionary sink of primary tumor diversity with the capacity to seed further metastases. Longitudinal studies are primed to reveal further insight about the extent to which metastatic sites represent outgrowth of multiple heterogeneous subclones from the primary tumor and genetic events that may be permissive for metastatic outgrowth during the bottlenecking process.

Despite the emerging consequences of ITH on tumor adaptation, little is known about the relationship of ITH with standard tumor histopathologic prognostic parameters, nor how mechanisms generating ITH vary between primary tumors and metastatic recurrences. Developing robust methods to define ITH will be a critical step in this process (Fig. 1). Extensive evidence supports the association of clonal diversity with progression from preinvasive to invasive adenocarcinoma and CIN with poorer disease outcome (30, 39, 40). Prospective analyses of the association of ITH in the primary tumor with risk of early metastatic relapse following adjuvant cytotoxic or radiotherapy seem justified. Such approaches will address whether the degree of ITH might shed light on our ability to cure some primary tumors but not others.