Heterogeneity and Tumor History

See allHide authors and affiliations

Science  20 Apr 2012:
Vol. 336, Issue 6079, pp. 304-305
DOI: 10.1126/science.1222361

All cells are almost perfect copies of prior cells. Imperfect DNA replication creates random variation, which is the substrate for evolution. Such differences may be small [normally about one new mutation per human cell division (1)], but the accumulation of mutations over time can eventually transform a single cell. Progeny of this first transformed cell expand by bifurcating branching cell division (see the figure) to form visible billion-cell clonal tumor populations. Each one of these cancer cells is an almost perfect copy of the first transformed cell. Given this scenario, different genetic alterations in different parts of the same cancer should be found in most tumors. Such intratumor heterogeneity has been found in many cancers, but its true extent is becoming much more evident with the unprecedented ability to sequence genome-wide many times (deep sequencing).

Intratumor heterogeneity has been found whenever deep sequencing has been appropriately applied to different parts of the same cancer for the colon, pancreas, breast, and blood (25). The recent study by Gerlinger et al. (6) highlights this phenomenon, as over half the mutations in multiple different parts of the same advanced renal cell carcinoma (primary tumor and its metastases) were different. Topographical differences in chromosome copy-number variations and gene expression signatures were also readily found, indicating that with high-resolution methods, intratumor heterogeneity is present wherever one looks.

One conclusion drawn from this baffling intratumor heterogeneity is that applying molecular signatures of prognosis or therapy to individual patients will be extremely difficult because the “answer” will vary depending on what part of the tumor is sampled (7). Analyzing more biopsies and sequencing even deeper are likely to reveal even more abnormalities and heterogeneity. Because all genomes are almost perfect copies of prior genomes, potentially the history of a tumor is encoded by its heterogeneity. With a simple molecular clock hypothesis (the number of differences between two genomes increases with the number of divisions since a common ancestor), the heterogeneity within and between different parts of the same tumor can indicate how long ago transformation occurred and how cancer cells spread and migrate during progression. Heterogeneity can be a list of mutations and a language of change.

Mutation history.

Tumorigenesis can be represented by an ancestral tree that starts from the zygote and ends with present-day cancer cells. Mutations (circles) accumulate with cell division and can be classified as public or private. Tumor heterogeneity is an inherent outcome of branching cell division, with the amount of genomic heterogeneity dependent on the time after transformation and how cancer cells divide and migrate. It should be possible to infer the past by characterizing mutations in different parts of the same tumor. The past, as recorded in the multiple heterogeneous genomes within a tumor, may help predict its future.


One way to illustrate tumorigenesis is with ancestral trees, whose shapes can be inferred by comparing mutations in different parts of the same cancer (see the figure). Driver mutations (those that are causally implicated in oncogenesis) are therapeutic targets, but most of the historical information is encoded by the much more numerous passenger mutations (those that do not alter the fitness of a cell). Mutations can also be classified as founders or “public” (present in all cancer cells), “semiprivate” (present in a detectable fraction of cancer cells), or “private” (present in a single or few cancer cells). Public mutations accumulate between the zygote and the first transformed cell, whereas private mutations arise after transformation. The spread of a private mutation depends both on its selective value and when it was acquired after transformation. Current tumor sampling cannot easily determine whether private mutations are present in every cell (a single subclone) or in different cells (multiple subclones) within a small biopsy. Interestingly, at even finer physical resolution, genomic heterogeneity is still present within single cancer glands or between adjacent cancer cells (8, 9), suggesting that many private mutations lack sufficient driver value to confer much local dominance. In the absence of selection, heterogeneity becomes a function of time, error rates, and random drift.

Driver mutations that confer selective advantages reduce heterogeneity through clonal evolution. A newly emerged tumor cell population has relatively few private mutations, but more will accumulate. Therefore, private mutation frequencies can encode the relative age of a human tumor—the greater the heterogeneity, the greater the time since transformation. The amount of tumor heterogeneity has clinical implications for chemotherapy. Targeted therapies should be directed at public driver mutations present in all cancer cells, but the probability of drug-resistant variants increases with cell division (10). Therefore, a more heterogeneous tumor with many private mutations is more likely to fail chemotherapy. The renal carcinoma data (6) illustrate this point. Tumor size and tumor heterogeneity were maintained despite therapy with everolimus, an mTOR inhibitor. More effective chemotherapy would be expected to cause more severe genetic bottlenecks by killing the majority of cells and their private mutations, thereby reducing the genetic diversity of the surviving cancer cells. By contrast, less initial heterogeneity and a greater bottleneck effect were apparent after leukemia chemotherapy (5), a treatment that often results in cures.

Heterogeneity cannot be fully sampled because of the logistical difficulties of obtaining a complete tumor “physical,” but translating heterogeneity into inferred tumor “histories” provides an alternative roadmap for personalized cancer diagnosis and treatment. Initial studies of tumor heterogeneity have begun to reconstruct exactly how individual human cancers evolve, branch, and spread (26). For example, the topographical heterogeneity in present-day pancreatic cancer cells is consistent with initiation on average about 20 years ago and metastasis about 3 years ago (3). Instead of generating increasingly long lists of mutations, genomic alterations could be distilled into better patient-specific tumor histories with relatively simple outputs—how long ago transformation occurred, and how and when their tumors spread. Reading this fundamental information may provide actionable clues for treatment and prevention because like many human diseases, tumors with the same histories may have relatively predictable futures. Moreover, serial biopsies can help monitor treatment by measuring reductions in heterogeneity expected with more effective therapies. How to best reconstruct the histories written in tumor genomes is uncertain, with many challenges in developing optimal sampling schemes and algorithms. However, taking the additional time and effort to obtain a good tumor history may help make more sense of the multiple physical alterations in a cancer, and bring more effective personalized medicine to its unfortunate host.


Navigate This Article