PerspectiveTUMOR HETEROGENEITY

Intratumor Heterogeneity: Seeing the Wood for the Trees

See allHide authors and affiliations

Science Translational Medicine  28 Mar 2012:
Vol. 4, Issue 127, pp. 127ps10
DOI: 10.1126/scitranslmed.3003854

Abstract

Most advanced solid tumors remain incurable, with resistance to chemotherapeutics and targeted therapies a common cause of poor clinical outcome. Intratumor heterogeneity may contribute to this failure by initiating phenotypic diversity enabling drug resistance to emerge and by introducing tumor sampling bias. Envisaging tumor growth as a Darwinian tree with the trunk representing ubiquitous mutations and the branches representing heterogeneous mutations may help in drug discovery and the development of predictive biomarkers of drug response.

There has been much excitement over the past decade about technologies that measure RNA expression and DNA copy number in tumors, yet these technologies have often failed to deliver on their initial promise of yielding biomarkers that can predict the emergence of resistance to traditional drugs or new targeted therapeutics (1, 2). Here, we consider why reliable biomarkers that predict tumor responses to chemotherapy have not been validated successfully. We propose that intratumor genetic heterogeneity combined with the polygenic nature of drug resistance may impede both the implementation of predictive biomarkers and the development of new antitumor molecular therapeutics.

Despite the recent huge effort to sequence the complete genomes of a variety of tumors, only a few single-gene, transcriptomic, epigenetic, or structural genomic alterations in tumors have been discovered that could serve as clinically implementable biomarkers. For example, in breast cancer only a few prognostic gene signatures and three traditional single-gene predictive markers—the estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2)—are in routine clinical use. There are numerous reasons for the failure of predictive biomarker validation across different cancers (3). Technical issues, such as small and often clinically heterogeneous patient cohorts, differences in disease stage in the discovery and validation cohorts, and variable methodologies associated with the validation of numerous proposed predictors (either single-gene or multigene), have resulted in universal observations of the limited predictive value of these markers. Even though statistically significant associations with treatment outcomes are observed repeatedly, the sensitivities and specificities for the accuracy of prediction remain below those required for clinical decision-making. This suggests that these biomarkers only capture partial information regarding the biological basis of drug sensitivity in tumor tissue. This may appear surprising considering that current genomic methods are able to encompass a near-complete inventory of all genes expressed in a tumor and can survey the whole genome for structural abnormalities.

Two important lessons have emerged from the comprehensive genomic analyses of cancers, which may provide an explanation for difficulties with predictive biomarker development. First, each tumor contains an individual assortment of multiple genomic aberrations, few of which are shared between patients with the same histopathological tumor subtype. Second, emerging evidence suggests that these anomalies appear to vary within the tumor, indicating substantial intratumor heterogeneity (4).

Tumors resistant to therapeutic agents usually bear a “convergent phenotype,” in which multiple different genetic and epigenetic aberrations may potentially cause resistance to a particular agent (5, 6). For example, resistance to anti-HER2 agents may be driven by multiple mechanisms, including p95 expression, phosphatase and tensin (PTEN) loss of function, PIK3CA mutations, and overexpression of mucin 1 (MUC1). None of these mechanisms alone are sufficiently strong predictors to warrant clinical use as biomarkers of drug resistance. Sensitivity to a given antitumor agent also has a convergent phenotype involving a range of mechanisms, including mutations in multiple genes leading to sensitivity to inhibitors of poly(ADP-ribose) polymerase (PARP) (7). Of course, more than one of these mechanisms of drug resistance or sensitivity may operate within the same tumor.

Increasingly, molecular evidence suggests that intratumor heterogeneity may contribute to tumor growth through a branched (polytypic) rather than a linear pattern of tumor evolution, as first demonstrated by Enver, Greaves, and colleagues (8). Such branched evolutionary growth was first demonstrated in hematological cancers (8) and more recently in renal cancer through multiregion exome sequencing (4) and in medulloblastoma (9). Branched evolutionary growth and intratumor heterogeneity results in coexisting cancer cell clones with variegated genotypes and associated phenotypes that may be regionally separated within the same tumor (4). Intratumor heterogeneity has been documented in other solid tumors, exemplified by regional separation of subclones harboring different patterns of HER2 amplification in breast cancer (10).

Emerging evidence suggests that regionally separated heterogeneous somatic mutational events and DNA copy number aberrations can lead to sampling bias, which impairs the interpretation of genomics data derived from single-tumor biopsies (4). Intratumor heterogeneity and sampling bias, resulting from single biopsy-driven biomarker discovery and validation approaches, may contribute to the recently reported failures in implementation of robust biomarkers in the clinical setting (1). Contributing to yet more complexity, intratumor heterogeneity is increasingly recognized within a single breast cancer biopsy, with multiple coexisting and intermixed breast cancer cell populations that can be distinguished by differences in genomic content and chromosome number (12, 13).

Variegated phenotypes, resulting from intratumoral genetic heterogeneity, are likely to have important implications for developing new targeted therapies and for preventing the emergence of drug resistance (6, 8, 14). For example, there is emerging evidence that resistance to some targeted cancer drugs may result from the outgrowth of preexisting low-frequency cancer cell populations harboring somatic mutations that confer resistance to the targeted agent. Such subclonal resistance is exemplified by patients with non–small cell lung cancers (NSCLCs) who are treated with inhibitors of the epidermal growth factor receptor (EGFR). EGFR inhibitor treatment results in meaningful clinical benefit for patients with lung adenocarcinomas harboring activating mutations in EGFR (15, 16). However, the emergence of drug resistance, and in some cases reduced progression-free survival, is mediated by the presence of T790M gatekeeper mutations in EGFR (17, 18) or the amplification of the MET gene (19, 20), both of which may be present in the tumor at low frequency before the initiation of therapy. Therefore, low-frequency somatic events present in the tumor before drug treatment may determine treatment outcome—a clear challenge for biomarker identification and the clinical implementation of that biomarker. Furthermore, because distinct driver events may be regionally separated within tumor subclones, the targeting of such genetic aberrations with specific inhibitors may not make an impact on the growth dynamics of the bulk of the tumor. This problem is compounded by the likelihood that in a heterogeneous tumor, what amounts to a driver event may also depend on the environmental context, as well as the variable constellation of a large number of passenger mutations that characterize subpopulations. Therefore, given the lack of molecular tools to distinguish mutations as common drivers (shared by the entire tumor population), branched drivers (drivers only for a small subpopulation), or passengers (not sufficient on their own to drive tumor growth or survival), the interpretation of tumor mutation data remains a challenge. This challenge is exacerbated by the likelihood that mutations exist on a spectrum between the two extremes of drivers and passengers. Conceivably, the situation may be further complicated by the capacity for passenger events to become driver events (and vice versa) as environmental selection pressures change during tumor growth.

Given the evidence for intratumor heterogeneity, more attention needs to be paid to the clonal evolution of tumors over the course of the disease and during and following drug therapy (6). Clearly, recent technical advances in single-cell analytical techniques herald a new era in understanding tumor heterogeneity and the dynamics of subclonal tumor architecture that may occur during therapy and their contribution to drug resistance and disease progression (13, 2123).

A useful analogy when considering intratumor heterogeneity and the challenges of biomarker analyses is to consider a tumor as a growing tree (Fig. 1A). The “trunk” of the tumor harbors the ubiquitous founding driver mutations (level 1 complexity). The sprouting “branches” (representing different regions of the tumor or cells within a single tumor biopsy) carry heterogeneous mutations (level 2 and level 3 complexity) that may distinguish the biological behavior of subclones under distinct selection pressures. Additional driver events in the tumor branches are exemplified by multiple distinct inactivating mutations in SETD2 or PTEN that occur in different regions of the same renal tumor and converge to inactivate the same pathway (level 2 complexity) (4). A complicating factor is the possibility that fitness effects of passenger mutations or driver mutations in both trunk and branches may vary under changing environmental conditions and selection pressures (level 3 complexity). For example, with therapy or during the metastatic process original passenger mutations in the trunk may become driver events. Alternatively, driver events in the trunk may lose their status under changing environmental conditions and selection pressures in favor of new driver events, which previously had a neutral fitness effect. For example, in EGFR-mutated and anaplastic lymphoma kinase (ALK)–rearranged NSCLCs, drug-resistant tumors may result from low-frequency heterogeneous secondary mutations, yet they maintain the original genetic aberration present in the trunk of the tumor (11). Of course, it may also be the case that the branches gain additional genetic aberrations that are the seeds of resistance, but to which the tumor is not solely addicted. Protean tumor tendencies, imposed by heterogeneous genetic, transcriptomic, or epigenetic aberrations and their associated multifaceted phenotypes, render biomarker discovery efforts increasingly difficult unless each underlying mechanism is deciphered systematically in advance of therapy.

Fig. 1. A trunk-branch model of intratumor heterogeneity.

(A) The development of intratumor heterogeneity is analogous to a growing tree. The trunk harbors the founding ubiquitous driver mutations of a cancer present in every tumor subclone and region. The sprouting branches represent different geographically separated regions of the tumor or subclones present within single biopsies that carry heterogeneous mutations that are not present in every tumor cell or tumor region. Such mutations may distinguish the biological behavior of subclones and harbor the potential to become driver mutations under distinct selection pressures. Ubiquitous genetic events present in the trunk may provide more tractable biomarkers and therapeutic targets than heterogeneous events in the branches. We describe three levels of complexity: level 1, the trunk carries driver events, whereas the branches carry neutral mutations; level 2, the trunk carries driver events, whereas the branches carry neutral or additional driver events that may harbor convergent phenotypes (for example, distinct mutations in SETD2 or PTEN occur in different regions of the same renal cancer and converge on the same pathway resulting in its inactivation) (4); level 3, level 1, and level 2 events plus neutral mutations in the branches (or trunk) that become driver events under selection pressures (11, 1720). With level 1 complexity, one biomarker can be developed against one target; with level 2 and 3 complexity, a single biomarker is unlikely to be sufficient. The risk of drug resistance may increase with each level of complexity. (B) Clonal architecture as a biomarker. The polygenic nature of drug resistance and intratumor heterogeneity may exacerbate difficulties in predicting therapeutic outcome. Consideration of tumor growth within a Darwinian evolutionary tree framework may support the identification of new predictive biomarkers. The length of the trunk and size of the branches are analogous to the ratio of the number of ubiquitous and heterogeneous genetic events in the tumor, respectively. “Palm tree–like” tumors, harboring many more ubiquitous genetic events than those of heterogeneous mutations, may result in improved clinical risk profiles. In contrast, “Baobab tree–like” tumors in which the heterogeneous genetic events outnumber the ubiquitous mutations may result in poorer clinical outcomes and increased propensity for drug resistance. Meanwhile, the risk of treatment failure with “chestnut tree–like” tumors may lie between these two extremes.

CREDIT: Y. HAMMOND/SCIENCE TRANSLATIONAL MEDICINE

The challenges of biomarker discovery and validation can be considered within the framework of the trunk and branch model of tumor growth. Whole-genome sequencing and profiling techniques that currently only detect major clonal populations may miss rare or spatially and temporally separated subclones. Such subclones can be defined by mutations in the tree branches— that is, mutations within distinct regions or subpopulations of the primary tumor or the metastases (Fig. 1A). This may ultimately affect treatment outcome and the efficient prediction of drug response. This model of tumor progression also implies that a single tumor biopsy is unlikely to fully represent the complete molecular landscape of the disease at primary and metastatic sites in all cases, as has recently been demonstrated in renal cancer (4). Massively parallel DNA sequencing techniques have also demonstrated the occurrence of “strain enrichment” and a phenomenon analogous to evolutionary bottlenecks in individual subclones at sites of cancer metastases compared with primary tumors (4, 9, 24, 25). For example, studies in renal cancer, breast cancer, pancreatic cancer, and medulloblastoma have demonstrated substantial genetic differences between primary and metastatic sites of disease, with low frequency subpopulations in the primary tumor being enriched at the metastatic site (26, 27) or new mutations evolving in the metastasis (4, 9, 2628). These results illustrate the difficulties of biomarker studies restricted to analysis of primary tumor tissue in which genomic aberrations may differ substantially in frequency compared with metastatic sites. In addition, these results suggest a cautious approach when considering the use of single tumor biopsies combined with deep sequencing technologies for early phase clinical trial stratification of cancer patients with metastatic disease, particularly if biopsies are temporally separated or derived from the primary tumor (6, 9).

Thus, there is an urgent need for systematic analyses of the evolution of cancer clonal architecture during therapy to identify ubiquitous driver events present in all regions of the tumor that may be more efficiently targeted. Notably, despite such efforts, there may be practical limits to drug efficacy and the predictive accuracy of any drug response marker because of the polygenic nature of drug resistance combined with extensive and dynamic intratumor genetic heterogeneity that may alter during the course of the disease. Furthermore, it is important to consider tumor compensatory feedback loops or adaptations that may develop in cancer cells after anticancer treatment that enable the circumvention of drug efficacy, thus hindering the effective prediction of drug response (29).

It may be helpful to consider predictive biomarkers within three distinct categories that may also represent different degrees of complexity in terms of the discovery and validation process. (i) There are broad response markers that represent a fundamental biological process critical for the growth and survival of a tumor, which may be programmed relatively early during cancer development (present in the trunk). An example of such a marker in breast cancer is high proliferation rate, which is determined by measures including elevated expression of Ki67, an increased Oncotype DX recurrence score, and high genomic grade (2). Cancers with a high rate of proliferation are sensitive to different classes of chemotherapeutic agents that interfere with key steps in cell division. In the future, similar broad markers may emerge for agents directed against metabolism or for immunotherapies. (ii) There are drug class–specific predictors such as ER, HER2, or activating mutations in EGFR that may represent singular, early driver events present in the trunk of an evolving tumor. Similar driver events for distinct small subsets of cancers are likely to be discovered in the next decade and will yield clinically useful predictors. (iii) Single drug-specific response markers may remain elusive because of heterogeneous molecular mechanisms that converge to cause sensitivity or resistance to a particular molecule. These markers, present in the branches of the tumor, may be regionally separated and not detectable in a single tumor biopsy, subject to tumor sampling bias. This translates into limited sensitivity and specificity for predictive assays that focus on a few selected mechanisms of tumorigenesis.

Therefore, when the multifaceted and polygenic nature of drug resistance is superimposed on intratumor heterogeneity and branched tumor growth, with the recently described capacity for selection of multiple drug resistance mechanisms in individual tumors (11), predicting therapeutic outcome in some patients may prove intractable. Such considerations suggest that describing the shape of the tumor evolutionary tree in terms of the length of the trunk relative to the size of the branches (analogous to the ratio of ubiquitous to heterogeneous tumor mutations) may allow new prognostic and predictive tools to be developed (Fig. 1B).

More than a decade of genomics research has revealed great complexity in the genomic abnormalities of individual cancer cells and also substantial intratumor molecular heterogeneity that may evolve over the course of disease and exposure to treatment. These findings may explain the failure of some approaches for predictive biomarker development that are exacerbated by tumor sampling bias driven by intratumor heterogeneity, while also indicating the way forward. Identification of ubiquitous driver events present in the trunk of the tumor remains a promising strategy because it would simultaneously define the patient selection marker and molecular target for treatment. Examining tumor clonal architecture and its evolution through treatment and disease progression may enable identification of these common driver events present in tumor trunks and the heterogeneous somatic events present in the tumor branches that contribute to drug resistance and treatment failure. Distinguishing “the wood from the trees” in such a way may yield more tractable therapeutic targets for drug discovery and more robust predictive biomarkers to improve treatment outcomes and limit the acquisition of drug resistance.

References

View Abstract

Navigate This Article