A population genetics perspective on the determinants of intra...

Biochimica et Biophysica Acta xxx (2017) xxx–xxx

BBACAN-88143; No. of pages: 18; 4C: 2, 6, 8, 9, 10

Contents lists available at ScienceDirect

Biochimica et Biophysica Acta

j ourna l homepage: www.e lsev ie r .com/ locate /bbacan

A population genetics perspective on the determinants ofintra-tumor heterogeneity☆

Zheng Hu, Ruping Sun, Christina Curtis ⁎Departments of Medicine and Genetics, Stanford University School of Medicine, Stanford, CA 94305, USAStanford Cancer Institute, Stanford University School of Medicine, Stanford, CA 94305, USA

☆ This article is part of a Special Issue entitled: Evolutioncancer?, edited by Dr. Robert A. Gatenby.⁎ Corresponding author at: Department of Medicine an

School of Medicine, Stanford, CA 94305, USA.E-mail address: [email protected] (C. Curtis).

http://dx.doi.org/10.1016/j.bbcan.2017.03.0010304-419X/© 2017 Elsevier B.V. All rights reserved.

Please cite this article as: Z. Hu, et al., A popul(2017), http://dx.doi.org/10.1016/j.bbcan.20

a b s t r a c t
a r t i c l e i n f o
Article history:Received 16 December 2016Received in revised form 1 March 2017Accepted 2 March 2017Available online xxxx

Cancer results from the acquisition of somatic alterations in a microevolutionary process that typically occursover many years, much of which is occult. Understanding the evolutionary dynamics that are operative at differ-ent stages of progression in individual tumorsmight inform the earlier detection, diagnosis, and treatment of can-cer. Although these processes cannot be directly observed, the resultant spatiotemporal patterns of geneticvariation amongst tumor cells encode their evolutionary histories. Such intra-tumor heterogeneity is pervasivenot only at the genomic level, but also at the transcriptomic, phenotypic, and cellular levels. Given the implica-tions for precision medicine, the accurate quantification of heterogeneity within and between tumors has be-come a major focus of current research. In this review, we provide a population genetics perspective on thedeterminants of intra-tumor heterogeneity and approaches to quantify genetic diversity. We summarize evi-dence for differentmodes of evolution based on recent cancer genome sequencing studies and discuss emergingevolutionary strategies to therapeutically exploit tumor heterogeneity. This article is part of a Special Issue enti-tled: Evolutionary principles - heterogeneity in cancer?, edited by Dr. Robert A. Gatenby.

© 2017 Elsevier B.V. All rights reserved.

1. Introduction

Cancer results from the acquisition of alterations during somatic celldivision in a microevolutionary process that typically occurs over de-cades, much of which is occult andwith extended clinically latent inter-vals. For example, more than half of all detectable somatic mutations incolorectal cancers occur prior to transformation [1,2]. A set of initiating(epi)genetic events (so called drivers) provide a selectivefitness advan-tage (resulting in higher proliferation or lower apoptosis for example)in the target cell relative to other pre-malignant cells, resulting in clonalexpansion. These initiating events in the founding tumor cell are oftenspecific to certain tissue types or cells of origin as is the case for APC incolon and other gastrointestinal tumors [3–5] or DNMT3 in leukemias[6]. Hence, the fitness benefit that the mutation confers may dependon themicroenvironment or cellular context inwhich it occurs. Howev-er, a singlemutation is seldom sufficient to evoke a fully malignant phe-notype and the pre-malignant clones likely maintain functionality until

ary principles - heterogeneity in

d Genetics, Stanford University

ation genetics perspective on17.03.001

additional ‘hits’ accrue, eventually leading to transformation and uncon-trolled cell growth (Fig. 1).

Given that tumor cells continuously accrue mutations and are sub-ject to ever changing microenvironments, genetic and phenotypic het-erogeneity is expected. Indeed, according to the clonal evolutiontheory, heterogeneity is attributed to continued genetic and heritableepigenetic alterations and selection in diverse microenvironmentswithin tumors [7–9]. Hence mutation rates, genetic drift, populationstructure, microenvironment and selection impact the extent of tumorheterogeneity. As tumors progress, expanding into an environmentthat is resource limited, cells that are better able to replicate, utilize en-ergy, and migrate will have a competitive advantage and these ‘hall-marks’ are noted in most advanced cancers [10]. At this stage, theinitiating genetic lesions may no longer ensure cell survival. As such, itis essential to develop an understanding of the phenotypic conse-quences of putative oncogenes and tumor suppressors in a context-de-pendent fashion.

Although intra-tumor heterogeneity (ITH) has been appreciated fordecades [11], the advent of high-throughput technologies has enabledthe characterization of (epi)genomic, transcriptomic, phenotypic, andcellular heterogeneity at enhanced resolution [12–25]. The presence ofpervasive ITH poses significant challenges for precision medicine. Forexample, sampling bias due to solid tumor spatial structure withinlesions can obscure the interpretation of genomic profiles for patient

the determinants of intra-tumor heterogeneity, Biochim. Biophys. Acta

http://dx.doi.org/10.1016/j.bbcan.2017.03.001

mailto:[email protected]

Journal logo


http://www.sciencedirect.com/science/journal/0304419X

www.elsevier.com/locate/bbacan


Fig. 1. Schematic illustration of different phases and modes of tumor evolution. Somatic alterations (including SSNVs and CNAs) arise during cell division in normal/pre-neoplastic cells,where mutations that confer a selective growth advantage relative to the rest of the population (drivers) result in clonal expansions. Multiple such events promote transformation in aprocess that can occur over long time periods (indicated by hatch marks). The mutations (both drivers and passengers) that accumulate in the founding tumor cell lineage duringinitiation will be present in all cells in the tumor (public/clonal). After transformation, the tumor may grow in the absence of stringent selection, compatible with a Big Bang model[Ref 21] and effectively neutral evolution (A) or be subject to continuous selection resulting in clonal expansions (B and C). In the Big Bang model, the tumor grows as a single‘terminal’ expansion populated by numerous heterogeneous subclones that are not subject to strong selection (A). As such, the timing of a mutation is the primary determinant of itsfrequency in the final tumor and most detectable subclonal alterations (resulting in ITH) arise early during growth, whereas late-arising mutations will be largely undetectable. Incontrast, under selection, the so called ‘sequential’ or ‘linear’ evolution model describes clonal successions or sweeps that occur sequentially (B) while ‘branched’ evolutioncorresponds to a scenario where multiple subclonal alterations co-occur and compete during tumor growth (C).

2 Z. Hu et al. / Biochimica et Biophysica Acta xxx (2017) xxx–xxx

stratification and therapeutic decision-making. Elevated ITH mayalso be associated with disease progression [26] and poor prognosis[27,28]. Given the numerous clinical implications, the accurate quantifi-cation of heterogeneity within and between tumors from the same pa-tient is a major focus of current research. However, to date, theseanalyses have largely been descriptive in nature.

While human tumor evolution cannot be directly observed, spatio-temporal patterns of genetic variation amongst tumor cells are stably

Please cite this article as: Z. Hu, et al., A population genetics perspective on(2017), http://dx.doi.org/10.1016/j.bbcan.2017.03.001

inherited during cell division, and hence surreptitiously encode theirancestries. As such, the resultant patterns of ITH can be ‘read’ viamulti-region sequencing (MRS) of spatially and/or temporally separat-ed tumor regions or single-cell sequencing (SCS) and used to recon-struct the evolutionary relationships amongst the distinct cellpopulations (clones) that comprise a tumor, as has nowbeen performedin a variety of tumor types [13,16–19,21,22,25,29–34]. Intriguingly,these and other cancer sequencing studies suggest that distinct modes



3Z. Hu et al. / Biochimica et Biophysica Acta xxx (2017) xxx–xxx

of evolution are operative during tumor progression and in differenttumor types. However, this has yet to be systematically evaluated.More generally, while the elements of somatic evolution have been de-fined [7,8,35] and include mutation, genetic drift, migration, populationstructure, and selection, the evolutionary dynamics that are operative atdifferent stages of progression in individual tumors are poorly charac-terized. A quantitative understanding of tumor dynamics has the poten-tial to define cancers' evolutionary playbook and might hold clues to asto how to better prevent, detect, and treat cancers. For example, the ear-lier the initial clonal expansion is detected, the less diverse and likelyless fit the cell population, and the better the prognosis. Hence, knowl-edge of the initiating events and their relative temporal ordering mayinform strategies for earlier intervention. Likewise, computational andmathematical modeling can provide insights into mechanisms of pro-gression and enable the inference of patient-specific tumor dynamics.Such information may ultimately inform the design of rational treat-ment strategies.

In this review, we provide a population genetic perspective on theorigins of tumor heterogeneity and summarize approaches to quantifygenetic diversity, drawing from established theory for studying geneticvariation within and among natural populations in light of the forces ofmutation, genetic drift, selection, and demography. We summarize evi-dence for differentmodes of clonal evolution based on recent cancer ge-nome sequencing studies and the potential clinical relevance of theresultant heterogeneity. Finally, we discuss emerging evolutionarystrategies to therapeutically exploit tumorheterogeneity and steer clon-al dynamics.

2. Genetic and non-genetic tumor heterogeneity

2.1. Genomic heterogeneity within and between tumors

Conventionally, the distinct stages of tumor initiation and subse-quent growth have been assumed to proceed in a ‘linear’ sequentialfashion in which successively more fit ‘driver’ events arise within aclone, resulting in the replacement or clonal succession of less fitclones through selective sweeps [8,36]. While the sequentialmodel may accurately describe tumor initiation, it is not knownwheth-er this applies to most solid tumors. Sequencing of established primarytumors has revealed extensive ITH in diverse tumors as demonstratedvia multi-region profiling of clear-cell renal cell carcinoma (ccRCC)[13,18], colorectal cancer (CRC) [21,37], glioblastoma (GBM) [16,22],non-small cell lung cancer (NSCLC) [30], lung adenocarcinoma [29],cervical [38] and ovarian cancer [17], where somatic alterationsare often restricted to distinct tumor regions. Several of these studieshave reported evolutionary trajectories in which distinct driver alter-ations are ubiquitous (public/clonal) and therefore located on thetrunk of the tree, whereas other drivers are heterogeneous (private/subclonal) and restricted to certain branches of a phylogenetic tree.

As a result of these and other studies, ITH is now accepted as an in-herent feature ofmalignancy. The observation thatmany tumors exhibitgenetic variegation has therapeutic implications since actionable ‘driv-er’ mutations may be suboptimal targets if they are restricted tosubclonal branches of the tumor's ancestral tree. Moreover, ITH itselfmay represent a potential prognostic biomarker [26,39]. There are nu-merous excellent reviews on ITH [15,40–42], hence we do not attemptto comprehensively cover this vast literature here, but rather to intro-duce relevant concepts for quantifying, interpreting and exploiting ITHwithin an evolutionary framework. A crucial concept for understandingcancer genomic data is that clonal (public) mutations correspond toevents that accumulated in the founding tumor cell lineage during initi-ation andprior to transformation such that they are present in all cells ofthe tumor (Fig. 1). These include both pathogenic alterations that con-tribute to disease progression and benign passengers. Subclonal alter-ations occurring after transformation are by definition present in asubset of the tumor cell population and hence considered private. They


may be restricted to certain tumor regions or more pervasive, but asthese events initially occur in a single cell, they only become detectableafter subsequent rounds of cell division or as a result of a clonal expansion.

As discussed in subsequent sections, multiple factors influence theability to reliably detect subclonal variants, including the sensitivityand specificity of an assay, tumor purity (or the proportion of tumorcells), sequencing coverage, as well as the nature of the sample andthe mode of tumor evolution. To date, most large-scale cancer genomesequencing efforts have profiled a single bulk sample. This requiresdeconvolution to delineate the fraction of tumor cells harboring agiven alteration, and poses limitations due to tissue architecturein solid tumors, which can result in the spatial segregation ofsubclones and therefore sampling bias. Sampling of multiple spatiallyseparated bulk tumor regions or microdissected subregions such as indi-vidual glands can mitigate sampling bias and provide varying degrees ofspatial resolution. However, these approaches still require deconvolution.In contrast, single cell sequencing enables direct measurement of clonalgenotypes, but is not yet scalable to large numbers of cells and suffersfrommeasurement errors. Each of these methods requires tissue dissoci-ation so spatial information is lost if not recorded during sampling.

In situ analyses that retain tissue architecture have typically beenlimited to relatively low resolution techniques such as fluorescent insitu hybridization (FISH). However, these can be complemented byhigh-resolution spatial tumor sampling studies, where both approacheshave revealed the intermingling of heterogeneous subclones. For exam-ple, analysis of EGFR copy number states in GBMvia FISH exposed strik-ing patterns of mosaicism [43]. Copy number heterogeneity wassimilarly reported for both EGFR and PDGFRA and found to be associat-ed with differential growth factor responsiveness in GBM [44]. Thesefindings were recapitulated by genome-wide copy number profiling ofmultiple tumor sectors from a cohort of 11 GBM patients [16]. More-over, parallel transcriptional profiling revealed the presence of multiplemolecular subgroupswithin a given patient, highlighting the challengesthat ITH poses for precisionmedicine. In CRC, single gland neutralmeth-ylation tag profiling was consistent with the presence of numerousintermixed subclones [37]. More recently, multi-region, single glandand single cell analyses revealed extensive genetic variegation andsubclonemixing at the copy number andmutational levels in both ade-nomas and carcinomas [21], as discussed in greater detail in Section 4.Computational modeling of spatial tumor growth and statistical infer-ence on the genomic data revealed that most detectable ITH occursearly during tumor growth, long before the lesion is clinically evident.These findings were subsequently corroborated by reports thatsubclonal alterations originate early in colorectal adenomas/carcinomas[45–48]. Hence, ITH is already present in early lesions and continues toaccrue during progression. Subclone mixing has been reported in othersolid tumors, when detailedmulti-region samplingwas performed [49],whereas such patterns would be obscured by bulk profiling. Single cellsequencing has similarly revealed extreme levels of genetic heterogene-ity [32,50–53]. Intriguingly, early studies in GBM reported that geneti-cally distinct cell subpopulations can cooperate to promote tumorgrowth, progression, and maintenance [54]. Several recent studieshave similarly reported clonal cooperation in a breast cancer mousemodel [55] and xenografts [56]. Collectively, these findings highlightthe potential for a minor tumor cell population to promote ITH. Theyalso suggest that interdependencies amongst subclones may representpotential therapeutic vulnerabilities.

Despite the extensive diversity, convergent evolution is apparent insome tumors, suggestive of constrained evolutionary trajectories. This ismost notable in ccRCC, where on a background of clonal (truncal)inactivating mutations in the von Hippel Lindau (VHL) gene and lossof heterozygosity (LOH) of chromosome 3p, harboring the secondcopy of the VHL gene, spatially separated subclones were found to har-bor distinct mutations in multiple members of the SWItch/Sucrose nonFermentable (SWI/SNF) chromatin remodeling complex, includingPBRM1, SETD2, ARID1A and SMARCA4 or even distinct mutations in the




same gene [13,18]. Convergent evolution of EGFR mutations/amplifica-tions was also observed in GBM via single nucleus sequencing [57],and in chronic lymphocytic leukemia (CLL) [58]. These findings implythat alterations in these pathways are selectively advantageous andsuggest genetic canalization towards a particular phenotype. Such pat-terns have only been noted in a minority of tumors examined to date,although the integration of multiple ‘omic’ features may unmask this.In the context of targeted therapies, which impose a highly specific se-lective pressure, convergent evolution of on-target point mutations inkinase domains and the activation of compensatory pathways havebeen observed [59,60].

Most genomic profiling studies to date have focused on primary tu-mors, whereas relatively fewer have examined genetic divergence be-tween paired primaries and metastases or recurrences arising at localor distant sites after initial therapy. This is due, in part, to the challengein obtaining biopsies fromdistantmetastatic sites,which are sometimesinaccessible and the time lag between initial diagnosis and disease re-currence, wherein procedures may be performed at different centers.Given that therapy can alter a tumor's evolutionary trajectory, it is im-portant that the genomic data are interpreted in this context. Severalstudies have comprehensively profiled paired primary and recurrentdisease in glioma [19], GBM [22,61], ovarian cancer [62], AML [63],ALL [64], neuroblastoma [65].

Genomic concordance between paired primaries andmetastases hasbeen investigated in diverse solid tumors at the level of somatic singlenucleotide variants (SSNVs) [13,66–73], structural variants (SVs) [74],copy number aberrations (CNAs) [66,67,75,76] [73] and loss of hetero-zygosity (LOH) [77,78] with varying degrees of genomic divergence re-ported. The inferences that can be drawn from these efforts depend onthe study design, platform(s) used, depth of sequencing (when rele-vant), and extent of clinical and treatment information. For example,studies that are targeted in nature preclude the unbiased discovery ofglobal genomic similarities and differences. Sampling bias and tumorspatial structure, as well as intervening therapy can confound the inter-pretation of patterns of heterogeneity between lesions. Hence, all ofthese factors should be reported, even if they cannot practically be con-trolled for.

The genomic characterization of multiple metastases from the samepatient has been limited to a handful of studies [18,34,79,80], whichtend to suggest greater similarities amongst metastases than betweenthe primary and metastasis. For example, in a large study of paired pri-maries and metastases (n = 86), Brastiantos et al. [34] reported thatspatially and temporally separated brain metastases were often geneti-cally homogenous, whereas 53% of cases harbored potentially clinicallyinformative alterations in brain metastases that were not detected inthe matched primary. The authors also reported that extra-cranial me-tastases were often distinct from the brain metastases.

Although multi-region sampling is essential for the accuratereconstruction of tumor evolutionary histories [21,81], few studieshave performed MRS of both the primary and metastasis and thishas generally been restricted to one or a few cases [47,73,82]. Hence,there remains a need to characterize larger patient cohorts at a ge-nome-wide level in order to delineate concordance between primariesand metastases, as well as the timing of metastasis and patterns ofseeding.

2.2. Sources of tumor heterogeneity

Cancer is a disease of the genome, resulting from both endogenousand environmental damage to DNA, which can be propagated duringnormal cell division. In this review, we focus on genetic heterogeneityin cancer and the evolutionary forces that shape these patterns. Howev-er, tumor heterogeneity manifests not only at the genomic level, but atthe cellular, phenotypic andmicroenvironmental levels and they jointlyinfluence one another.


2.2.1. Mutational heterogeneityAsmuch of tumor progression is occult, by the time the lesion is clin-

ically evident (and composed of tens of millions to billions of cells), tensto hundreds of thousands of mutations can have accrued. Comprehen-sive cancer genome sequencing studies have revealed that the aberra-tion landscape is vaster than previously thought with different genesmutated in individuals with the same histological tumor type [36,83,84]. As such while a handful of genes are recurrently mutated at highfrequency in the population, the vast majority of alterations occur infre-quently, resulting in the so-called ‘long-tail’ distribution of putativedriver genes. Considerable (N1000-fold) variation in the frequency ofnon-synonymous mutations was observed across tumor types with pe-diatric cancers exhibiting the lowest mutational burden, andmelanomaand lung cancer the highest. In some cases, the elevated mutation fre-quency could be explained by exposures to known carcinogens, namelyUV radiation in the case ofmelanoma and tobacco smoke in lung cancer.Even amongst tumors of the same type, mutation frequencies variedmultiple orders of magnitude (0.01–100/Mb), highlighting diversity insomatic mutation rates.

2.2.2. Genomic instabilityIt has been more than 40 years since Loeb et al. [85] proposed the

mutator phenotype hypothesis, which posits that errors during DNAreplication promote malignant progression [86]. Two years later,Nowell's perspective piece on the clonal evolution of cancer proposedthat heterogeneity in cancer results from increased genetic instabilityduring disease progression [7]. Cancer genome sequencing effortshave revealed genomic instability resulting from point mutations(SSNVs), CNAs, and SVs, each ofwhich can contribute to elevated genet-ic diversity. For example, chromosomal instability (CIN) leading to CNAsand SVs [87,88] results in cells inwhich themechanisms that ensure thefidelity of chromosome segregation during cell division are compro-mised. As such, chromosomal aberrations can dramatically impact cel-lular fitness [89,90], which further propagates aneuploidy andfacilitates rapid evolution [91].

Another form of genome instability, known as microsatellite insta-bility, results from DNA slippage events at simple tandem repeat ele-ments present throughout the genome, and is common in tumorswith defective DNA mismatch repair (MMR). Although microsatelliteinstability (MSI) has been appreciated for over two decades beginningwith its discovery in familial CRC patients [92], the advent of next-gen-eration sequencing (NGS) has revealed fresh insights into its functionalimplications [93]. As the mechanisms and consequences of genome in-stability in relation to tumor heterogeneity have recently been reviewed[40,94], herewe focus on core concepts and points relevant to an under-standing of the population dynamics amongst tumor cells.

A particularly intriguing aspect of genomic instability is that it ap-pears to be associated with cellular fitness tradeoffs [89,95–97]. For ex-ample, aneuploidy has been shown to both initiate and inhibit tumorprogression depending on the extent and context [98]. On the onehand, aneuploidy appears to be associated with tumor cell fitness ad-vantages. For example, in CRCswhole genomedoubling events are com-monly thought to precede the acquisition of a CIN phenotype and toendow the tumorwith tolerance to additional genome instability, there-by driving further evolution [97]. Indeed, genomedoubling is associatedwith worse relapse free survival in colorectal [97] and ovarian [99] can-cers. CIN has also been associated with poor prognosis [100]. On theother hand, overexpression of MAD2, a gene contributing to chromo-some instability (CIN), usually leads to cell growth arrest or death, sug-gesting that it is disadvantageous to cells [90]. These apparentcontradictions may be explained in part by the findings of Laughney etal. [89] who demonstrated that cancer cells have evolved to exist withina narrow range of chromosomemissegregation rates that optimize phe-notypic heterogeneity and clonal survival, suggesting that CIN is underselective constraints.




Mutational burden has also been hypothesized to achieve saturation,where additional mutations are lethal, resulting in “error catastrophe”,that render the tumor more sensitive to insult and therefore confer fa-vorable outcome [95]. Indeed, higher total mutation burden does notuniversally associate with worse prognosis. This is perhaps best exem-plified in CRC, where mismatch repair (MMR) deficiency results in ahypermutator phenotype and excess (100-fold) SSNVs, and yet MMR-deficient microsatellite unstable (MSI-high) tumors tend to have favor-able prognosis compared to microsatellite stable (MSS) tumors [101].

Given that genome instability has been established as an importantdeterminant of genetic diversity [40,102], it is critical to understandhow tumors respond to these insults, and how they propagate diversityas this may inform strategies to limit ITH. While a common feature ofmany cancers, variable levels of mutational burden and genomic insta-bility are observed within and between tumor types [83,103]. In adulttumors, relatively genomically (copy number) stable subgroups ofbreast [104] and gastric cancers [105] have been reported. Moreover,some pediatric cancers have remarkably ‘quiet’ genomes, as is the casefor synovial sarcoma, which is commonly driven by the SS18-SSX onco-genic fusion, resulting in disruption of the SWI/SNF (BAF) chromatinregulatory complex [106]. Thus, while genome instability is a criticalsource of tumor heterogeneity in many cancers, it does not appear tobe universally essential for disease progression.

2.2.3. Cellular, phenotypic and microenvironmental heterogeneityTumor cell heterogeneity can manifest at the level of cellular mor-

phology, proliferation kinetics, cell motility, and cell metabolismamongst others [107,108]. These differences can be attributed to bothgenetic and non-genetic variation, where the latter includes epigeneticand transcriptional changes, as well as microenvironmental factors.For example, epigenetic changes can shape cellular phenotypes andmodulate cellular plasticity [109,110]. Phenotypic heterogeneity pro-vides cancer cells with immense plasticity to cope with environmentalstress during tumor progression in the face of limited resources andhence may also be important for allowing tumor cells to escape thera-peutic insult. Indeed, phenotypic heterogeneity is now recognized as amajor source of drug resistance [111,112]. Early work demonstratedthe presence of reversible drug tolerance associated with specific chro-matin states and the re-establishment of heterogeneity of the putativecancer stem cell (CSC) surface marker CD133 [109]. Subsequently,stem-like characteristicswere reported to contribute to the stabilizationof drug-tolerant populations for extended time periods [111]. These andother studies point to the fact that high-dose cytotoxic therapies can in-duce pathogenic cell state transitions in the residual tumor cell popula-tion, wherein they revert to a primordial (stem-like) developmentalstate [111,112]. Intriguingly, other studies have demonstrated pheno-typic divergence between distant metastases relative to primary breasttumors and lymph node metastases [113], consistent with the notionthat elevated diversity in advanced stage disease contributes to treat-ment resistance.

It is also evident that stochasticfluctuations can be functional, partic-ularly in constrained conditions such as development [114–116], stressresponse [117], apoptosis [118] and cancer progression [119,120]. Fur-ther, non-cell-autonomous factors can contribute to tumor growthand promote clonal interactions that may, in turn, lead to new pheno-types [56]. These non-genetic components occur on the backdrop of ge-netic variation and jointly shape a tumor's evolutionary trajectory. Astumors progress, irrespective of their tissue of origin, similarmicroenvi-ronmental cues appear to arise as a result of spatial constraints, limitednutrients and vasculature, as well as activation of the immune system[121]. This is in accordance with the long-held view that advanced can-cers exhibit similar defining hallmarks [10].

Microenvironment factors, including immune and stromal cells,acidification, oxygen, and geographical constraints also contribute tophenotypic heterogeneity within tumors, influencing signaling path-ways and regulatory circuitry. For example, cellular adaptations to


hypoxia and acidosis have been reported during breast tumor evolution[122] and invasiveness has been attributed to adaptive changes in themicroenvironment [123,124]. Additionally, cancer cells at the tumor pe-riphery versus necrotic core can have different phenotypes resultingfrom regional variation in microenvironmental selective forces, wherelocal cancer cell populations rapidly converge to the fittest phenotypegiven a stable environment [124].

In summary, the genetic, non-genetic, and environmental forces atplay during a tumors long evolution shape the detectable patterns ofITH in a tumor sampled at the time of diagnosis. While studies to datehave largely focused on genomic heterogeneity, in part due to greaterease in measuring it, new techniques to delineate the genotype to phe-notype map in cancer and to probe tumor microenvironment in situmay offer opportunities to characterize their relationship.

3. Quantification of genetic diversity in cancer genomes

3.1. Practical challenges in quantifying ITH in tumor samples

The quantification of ITH from current NGS data is challenging formultiple reasons. From a practical perspective, the power to detectsubclonal somatic single nucleotide variants (SSNVs) is limited in cur-rent bulk tumorWES studies, which often aim to achieve 80–100× cov-erage (with WGS often lower) [84]. This is particularly problematicwhen coupled with the low purity of many tumor samples, which con-tain a heterogeneous mixture of non-cancerous and cancerous cells.Ideally purity is estimated from adjacent hemotoxylin and eosin(H&E) stained tissue sections prior to nucleic acid extraction to guidethe target depth of coverage. When needed, flow sorting [125], singlegland isolation [21,37] or microdissection [126,127] can be performedto enrich for tumor cells. As pathology based purity estimates may notagree with molecularly inferred estimates obtained using tools such asPurBayes [128], ABSOLUTE [99], or TITAN [129], it can also be beneficialto perform initial sequencing to determine howmuch additional cover-age is needed to achieve a given power for detection. However, eventhen, genome sequence context, and high duplication rates can hinderuniform and optimal target coverage, the latter of which can be partic-ularly problematic for formalin fixed paraffin embedded (FFPE) clinicalspecimens. Nonetheless, given the wealth of clinically annotated archi-val tissue samples, it is important to optimize both the study design andbioinformatics approaches for such specimens.

Perhaps even more problematic is the issue of sampling bias, whichis inevitable in solid tumors defined by their unique tissue architectureand spatial structure. Multi-region sequencing (MRS) of spatially sepa-rated tumor regions can help mitigate the issue of sampling bias, andhas become more common in recent years [13,16–19,21,22,25,29,30,34]. Indeed, several recent studies have highlighted the need for MRSto accurately reconstruct tumor subclonal architecture [81,130–133]and infer tumor evolutionary dynamics [21,37]. From a practical per-spective, it is not feasible to randomly sample the entire surgicallyresected lesion and in most cases, the location of the specimen withinthe tumor is not known. For biopsies, the specimen itself is limitedand hence only provides a local sampling. Moreover, multiple factorscan influence the extent of sampling bias, including spatial constraints,the mode of evolution, and the rates of migration, which are as of yetpoorly characterized. Despite these complicating factors, it is evidentthat MRS aids the detection and quantification of ITH [21,134].

The distribution of variant allele frequencies (VAF) critically de-pends on the underlying mode of tumor growth [21,135]. For example,in colorectal tumors characterized by Big Bang growth dynamics, aftertransformation, the tumor expands in the absence of stringent selection,compatible with effectively neutral evolution [21]. Hence, in this model,the timing of a mutation is the fundamental determinant of its frequen-cy and later arising subclonal mutations will be present in increasinglydiminutive fractions of the tumor cell population (Fig. 1A). As such, ge-netic diversity will be vastly underestimated [21], as has been predicted




theoretically [127]. In amodel of effectively neutral evolution, subclonalvariants are not functionally relevant for primary tumor growth, butthey nonetheless provide a potentially rich substrate for subsequent se-lection in the context of therapy. In the alternate scenarios of continuousstrong positive selection (Fig. 1B) or branched evolution (co-occurrenceof multiple selectively advantageous clones, Fig. 1C) after transforma-tion, subclonal variants that confer a selective growth advantage mayattain high frequency (depending on various population parameters).In this case, they may be more readily detected, but in the absence ofmultiple sampling data, it is not always feasible to determine whetherthey are high frequency subclonal (private) SSNVs or public alterationsdue to the inherent noise associated with NGS (Fig. 2).

Numerous methods have been developed to perform somatic vari-ant calling from tumor-normal pairs, including deepSNV [136] andMuTect [137], which aim to achieve greater sensitivity for subclonal var-iants. However in a typical NGS study, the limits of detection corre-sponds to a VAF of ~10% [137]. Critically, subclonal variants alone areinformative in chronicling the dynamics of growth after transformation,as variants that were accrued prior to transformation will have eitherbeen lost or else present in the founding tumor and hence clonal inthe final tumor. On the other hand, it is possible to over-estimate ITHwhen calling variants frommultiple tumor samples in the same patientthat have non-uniform coverage or purity. Purity differences can be es-pecially notable in comparisons of paired primaries andmetastases [34]or paired pre versus post-treatment samples [138,139] and hence areimportant to account for. The increasing availability of MRS also moti-vates methods such as MultiSNV [140] that aim to jointly call SSNVsfrom multiple samples. By exploiting pseudo-repeated measurements,such an approach can potentially aid the detection of false positivesand false negatives. Deep targeted sequencing or digital PCR are com-monly used for orthogonal validation of candidate SSNVs discoveredthrough unbiased surveys. However, PCR based approaches (includingNGS library preparation) can suffer from ‘jackpot effects’ due to

Fig. 2. Schematic illustration of differentmodes of tumor evolution and the impact of spatial samthe absence of stringent selection (for example if the cells are already sufficientlyfit), compatiblselection. To illustrate these scenarios, Muller plots (http://doi.org/10.5281/zenodo.240589) dpanel) or strong selection (bottom panel) are shown. Descendant genotypes emerge withinrelative abundance of the genotype in the tumor cell population or cancer cell fraction) andsingle sample or multiple samples (denoted A, B) are sampled and subject to NGS. (B) Tufrequency spectrum (SFS) with a public mutational cluster centered at 0.5 and a large clusfrequency mutations cannot be reliably detected. This can be contrasted with the expected SFSclonal expansion of a selectively advantageous clone (purple shading).


stochastic fluctuations and errors introduced during early PCR stepsthat are amplified exponentially, resulting in differences in biased allelefrequency estimates [141–143]. Such artifacts can bemitigated to vary-ing extents by modifying library preparation techniques to incorporateunique molecular identifiers and via circle [144] or duplex sequencing[145], although these are not currently scalable to whole exome/ge-nome analyses.

Once SSNVs have been called, it is common to adjust their VAFvalues for tumor purity and copy number to obtain an estimate oftheir cellular prevalence or cancer cell fraction (CCF). However, this isa non-trivial task since multiple parameters are unknown (purity, ploi-dy, number of subclones) and often need to be estimated simultaneous-ly. Moreover, these estimates depend on the relative ordering of CNAs,SSNVs, and genome doubling events during tumor evolution.

The accurate identification of subclonal CNAs fromNGS is a challeng-ing task given that the distribution of short reads along the genome isnon-uniform,with high variability amongst adjacent loci at standard se-quencing depths. As a result, only CNAs with relatively high frequencycan be reliably detected from current NGS data, and this depends onthe segment length and copy number states. Methods to predict thesubclonal CNAs from tumor sequencing data include TITAN [129],TheTA [146], CloneHD [147], and CloneCNA [148]. Tools designed toinfer CNAs frommulti-region tumor sequencing data are currently lack-ing, but have the potential to improve the detection of subclonal CNAs,as well as the inference of tumor phylogenies, as discussed below.

Various tools have been developed to estimate the number ofsubclonal clusters from tumor sequencing data by grouping mutationson the basis of their allele frequencies, [133,147,149,150]. The majorityof these approaches aim to define ‘clonal’ relationships based on the as-sumption that mutations with similar frequencies descended from thesame ancestral cell and that a limited number of dominant subcloneswith distinct mutational profiles have undergone clonal expansion.However, a single bulk sample may fail to delineate clonal alterations

pling on patterns of ITH. (A) After transformation, an established tumormay propagate inewith effectively neutral growth. Alternatively, the tumormay be subject to ongoing clonalepicting tumor subclone composition under a model of effectively neutral growth (upperthe parental clone, where height indicates genotype frequency (corresponding to thethe horizontal axis indicates time (generations). At the time of diagnosis or surgery, amors arising under effectively neutral growth are expected to exhibit a bimodal siteter of low VAF subclonal mutations, where θ indicates the threshold below which lowunder strong selection in which private SSNVs can attain high frequency due to regional


http://doi.org/10.5281/zenodo.240589



versus subclonal alterations that attained high frequency to selection ordrift (Fig. 2). More generally, the inference of the number of subclones,their proportions and genotypes from single sample bulk tumor se-quencing is challenging with the solution underdetermined undermost conditions [81,147,151,152]. MRS data should facilitate these in-ferences and have recently used to reconstruct tumor phylogenies, asdiscussed below. Single cell sequencing enables the directmeasurementof clonal genotypes, and hence avoids the need for deconvolution, mak-ing it a powerful approach [153,154]. However, this is complicated bynoisy and incomplete measurements and remains impractical on alarge scale. Ultimately, a combination of bulk and single cell sequencingmay help to further resolve tumor subclonal architecture.

3.2. Phylogenetic approaches to interpret ITH

Phylogenetic trees are the canonical structure for describing thegenealogical relationship among tumor cells. They offer a potential-ly powerful and theoretically grounded approach to study heteroge-neity within and between lesions [155,156], drawing from the richfield of molecular phylogenetics, the principles and methods ofwhich are reviewed elsewhere [157]. Here, we briefly outline con-siderations for the analysis of cancer genome sequencing data.When modeling NGS data, it is important to account for systematicerrors and to increase robustness to model violations. In the caseof cancer genomes, not only is this complicated by uncertainty inSSNV and CNA estimates (as discussed above), but by extensive ge-netic diversity due to mixtures of cells for which both the number ofclones and their relative proportions are unknown. The issue ofsampling bias in solid tumors further necessitates that multiple re-gions of temporally separated lesions be assayed to robustly recon-struct their ancestral relationship, although this has seldom beendone in practice.

To date, various approaches have been employed to reconstructsample or subclone phylogenies from tumor sequencing data. Sev-eral studies have employed standard phylogenetic approaches, in-cluding parsimony [158], maximum-likelihood [159] anddistance-based methods such as the neighbor-joining algorithm[160] to delineate the relationship between tumor samples basedon SSNVs (often from CNA-neutral regions) or CNA breakpoints.For example, distance methods were used to cluster SSNVs acrossmultiple bulk samples in renal cell carcinoma [13] and CNAs in sin-gle breast cancer cells [32]. As distance-basedmethods take as inputsummary statistics, they cannot account for complex character(SSNV, CNA) states. Likewise, the exclusion of SSNVs in regions ofCNA to avoid the complex issue of phasing ignores potentially valu-able information, although for tumors that are largely diploid orhave sufficient events in non-diploid regions, this simplifying ap-proach may be reasonable. CNAs also represent a powerful featurefor phylogenetic inference, but require custom methods due to thehorizontal dependencies between adjacent loci and overlapping al-terations. An early method, termed TuMult [161] sought to over-come the issue of overlapping aberrations by employing copynumber breakpoints (which should be persistent). More recently,MEDICC was developed to jointly phase parental alleles and per-form tree reconstruction [162]. This approach takes as input integercopy number profiles and B-allele frequencies and has been used toreconstruct tumor phylogenies from multi-region copy numberprofiles [21,163].

Other studies have sought to perform the challenging task ofsubclonal deconvolution so that subclones can be modeled as speciesor taxa. For example, by performing in silico phasing of subclonalSSNVs and exploiting the pigeon-hole principle, a probable phylogenywas inferred from high coverage bulk WGS data of a breast tumor[164]. However, phylogeny reconstruction from a single bulk tumorsample will generally lead to an incomplete and potentially erroneoustree [37,165]. Numerous methods have been described to automate


deconvolution, including those that can take as input multiple samples[132,166,167]. However, even with multiple samples, there is no guar-antee that low VAF SSNVs can be accurately phased.More recently, sev-eral computational methods have been developed to jointly modelSSNVs and CNAs [81,131,133] towards more accurate reconstructionof clonal trees. This represents an important methodological advanceand importantly suggest that many possible trees may be consistentwith the data [81]. However, these methods often couple the task ofmutation clustering and phylogeny reconstruction and require priorclustering of CCF values derived from WGS/WES to reduce the featurespace to a limited number of subclones due to the computationally in-tensive nature of enumerating a large possible tree-space. Hence, thistoo can result in a loss of information. However, this new class ofmethods is in its relative infancy and is likely to remain an area of activeresearch. Likewise, several recent methods have been described to re-construct phylogenies from single cell data [168,169], where phase isknown, but the data are generally sparse and noisy due to allelic drop-out. Single cell WGS has the potential to directly resolve clonal geno-types, however this will require improved genome coverage andmutation sensitivity since errors can distort lineage reconstruction[170].

Given the absence of ground truth phylogenies for actual tumors,simulation studies provide the primary means to evaluate amethods performance. Hence, it is important that simulations usedfor benchmarking capture relevant aspects of tumor growth,which can be influenced by numerous factors (as discussed inSection 4). It is also essential to consider the underlying model as-sumptions and robustness to violations. For example, an infinite-site model is commonly assumed such that each site can mutateonly once and persists. However, in practice, convergent evolutioncould result in the acquisition of the same mutation more thanonce and mutations could be lost due to loss of heterozygosity(LOH), thereby violating this assumption. Some methods also re-quire assumptions about the underlying models of molecular evolu-tion, which may require refinement in cancer.

Despite the power of phylogenetic approaches to delineate tumorarchitecture and interpret ITH, there are limitations. In particular, phy-logenies alone do not report on the underlying dynamics [130]. Howev-er, genealogy-based statistical inference methods, stemming from thedevelopment of coalescent theory [171] and the advent of routinelarge scale sequencing efforts have transformedmodern computationalpopulation genetics approaches [172]. Techniques such as ApproximateBayesian Computing (ABC) have been used extensively in populationgenetics to infer posterior parameter distributions from moleculardata when using stochastic models for which likelihoods cannot be cal-culated [172,173]. Related approaches based on coalescent modelingandBayesian inferencehave been used to characterize stemcell dynam-ics in human colon crypts [174]. More recently, ABC has been integratedwithin 3-dimensional agent-based computational models of tumorgrowth that represent cancer as an evolutionary process to enable theinference of patient-specific tumor dynamics such as the mutationrate and CSC fraction [37], as well as subclone fitness differences andthe mutational timeline [21].

3.3. Ecological and population genetic measures of ITH

The relationship between genetic diversity in cancer and populationevolvability may provide prognostic and therapeutic information [26].Importantly, it is the total diversity (includingboth functional and selec-tively neutral diversity) that likely determines the evolvability of cancercells since in the absence of diversity natural selection is not operative.This is in-line with the Dykhuizen-Hartl effect proposed by Kimura, inwhich neutral mutations may later become adaptive in an altered envi-ronment [175]. Hence, ITH may be a proxy for the underlying rate ofevolutionary processes operative in a tumor. While it is not clear howto best sample a tumor or to quantify ITH, several measures have been




borrowed from ecology and population genetics for this task and othershave been defined based on recent cancer genomic data.

One widely used measure of genetic diversity within a population isheterozygosity [176], or the probability that two sequences chosen atrandom from the population have different allelic types, which is equiv-alent to Simpson's Diversity Index in ecology [177]:

D ¼ 1−∑i f2i ð1Þ

where fi is the frequency of the allelic type or cell type, i. For asexuallyreproducing populations such as cancer, the heterozygosity is the prob-ability that two cells chosen at randomhave different sequences. For ex-ample, Ling et al. [127] derived that the expected heterozygosity at timet (denoted by Ht) during tumor growth can be expressed by:

Ht ¼ 1−e−2u

Nt−1−∑t

j¼2e−2uj

Ntþ1− j−1∏ j−2

i¼0 1−1

Nt−i−1

� �� ð2Þ

whereNt is tumor size at time t startingwith a single transformed cell attime 0 and assuming a well-mixed neutrally growing population ofsynchronous cell generations. Here, u is the mutation rate per celldivision for the 30 Mb exonic coding regions, namely u = μ × 3 × 107

(e.g. corresponding to u = 0.03 when the per-base pair mutation rateis μ = 10−9). A cancer genome can diversify rapidly during tumorgrowth as illustrated in Fig. 3 for a logistic growth model. In particular,H N 0.8 when the tumor is barely detectable at 1 million cells. Thus,tumor growth may accompany rapid genetic diversification and asmall tumor can already harbor extensive diversity. This prediction isconsistent with recent findings that early tumors display high-levelsof genomic diversity [33,46,178].

Genetic heterogeneity within and between regional samples hasalso been quantified using the Shannon index [179]:

H ¼ −∑i f i � ln f ið Þ ð3Þ

where fi is the frequency of the allelic type or cell type, i. For example,the Shannon Index was used to measure microsatellite and copy num-ber heterogeneity based on FISH in Barrett's esophagus where high di-versity was predictive of progression to esophageal cancer [26]. It hassimilarly been used to and to quantify neutral methylation tag andcopy number diversity in colorectal tumors [21,37], where high ITHwas observed at different genomic levels within and between singleglands. The Shannon index was also used to quantify genetic diversitybased on immunoFISH of paired primary, lymph node and distant met-astatic breast tissue [113], where diversity was found to be highest indistantmetastases. Comparisons of pre versus post neoadjuvant treatedbreast tumors suggests that lower pre-treatment diversity was

Fig. 3. Genetic diversification resulting from random neutral mutations. (A) Tumor growth vcorrespond to the carrying capacity, growth rate and initial tumor size, respectively. (B) The extumor growth (shown in A) where the neutral mutation rate is u = 0.03 per cell division in ex


associated with pathologic complete response, whereas levels did notchange in tumors that failed to respond to treatment [20]. Importantly,the extent of ITH was comparable irrespective of the chromosomal re-gion analyzed. A survey of various measures of clonal diversity was re-ported by Merlo et al. [180].

More recently, theMATH indexwhich corresponds to the ratio of themedian absolution deviation (MAD) and the median of the VAF valueswas described [181] and used to quantify ITH in head and neck squa-mous cell carcinoma (HNSCC), where it was shown to correlate withworse outcome [27].

MATH ¼ 100� MAD VAFð ÞMedian VAFð Þ ð4Þ

Other approaches to quantify ITH include determination of the num-ber of subclones and their frequency based on shared CCF values [149,150], although this requires additional estimates and assumptions, asoutlined in the preceding section. Zhang et al. found that lung adenocar-cinomas with higher subclonal mutational burden (n = 3/11 cases)were more likely to relapse [29]. However, this was again based on alimited number of cases. Additionally, in post-chemotherapy treatedpediatric kidney cancer, tumors with microdiversity exhibited worsedisease-specific survival [182]. Other studies have reported an associa-tion between survival and various measures of ITH independent ofother clinical and molecular variables in diverse tumor types [183,184]. However, these data suggest a non-linear relationship betweenthe number of subclones and outcome, thereby complicating theinterpretation.

In summary, several studies have suggested that ITH is associatedwith disease progression [26] and poor prognosis [27,28], raising thepossibility that ITH may represent a general biomarker. However, lackof association is seldom reported and further studies are needed. More-over, as ITH is already present in early lesions, it is of interest to deter-mine whether its predictive utility depends on disease stage. Finally, itwill be important to understand which genomic features and measuresof ITH are most informative, and this may depend on the underlyingmodes of tumor evolution, which are as of yet poorly understood.

4. Evolutionary forces determining the extent of geneticheterogeneity

Mutation, genetic drift, migration and selection are the fundamentalforces that collectively shape genetic diversity within a population[185]. Clonal evolution, the process ofmutation accumulation and adap-tation in somatic cells, likely involves each of these forces, althoughtheir collective impact on ITH is seldom considered. Here we discuss

ia a logistic model, Nt = KN0ert/(K + N0(ert−1)), where K = 109, r = 0.05 and N0 = 1pected probability of heterozygosity, H (equivalent to Simpson's Diversity Index), duringonic coding regions. (C) The relationship between heterozygosity and tumor size.



Fig. 4. Genetic drift decreases genetic diversity in the whole tumor, while increasingdiversity amongst detectable mutations in NGS data. (A) The process of tumordiversification in the absence of genetic drift (and selection), in which the frequency of amutation (or cancer cell fraction, CCF) remains constant over time, corresponding to itsfrequency at the time of occurrence (1/n(t), where n(t) is the tumor size at time t). (B)The process of tumor diversification in the presence of genetic drift in which thefrequency of a mutation fluctuates stochastically during tumor growth. Early mutationswhose initial frequencies are high can reach detectable frequencies, while latemutations remain non-detectable under a model of effectively neutral evolution.


these factors in the context of classical population genetic theory and re-cent cancer genomic data.

4.1. Mutation

Mutation is the primary source of genetic diversity, the rate ofwhichis a key parameter in population genetics studies [185,186]. Althoughcancer genome sequencing efforts have provided a catalogue of somaticalterations across different tumor types [83,84], the per-cell divisionmutation rate is not knownand likely varies during the course of diseaseprogression. Mutational burden can also vary by orders of magnitudewithin and between tumor types [83,84]. Although the somatic muta-tion rate is generally assumed to be elevated relative to the germline[186,187], even germline mutation rates can vary considerably withinfamilies [188]. Additionally, the number of germ-cell divisions differsinmales (~102 divisions) and females (20–30 divisions) since ovumun-dergo twomeiotic and 22mitotic cell divisions and reachmaturity priorto birth, whereas in males the number of germ cell divisions increaseswith age [189]. Nonetheless, themutation rate for single nucleotide var-iants (SNVs) in the human germline has been estimated to be ~10−8 perbase pair per generation [188] or ~10−10 per base pair per cell division,whereas human somatic cells have been reported to accumulate 4 to 25timesmoremutations than germline cells, corresponding to an estimat-ed mutation rate of ~10−9 per cell division per base pair [186,187].However, there is no consensus regarding the mutation rate in humantumors. The finding that some tumors exhibit high genomic instabilityled to the mutator hypothesis, which posits that an elevated mutationrate is a prerequisite of tumor development [96,190,191]. Whilemutator phenotypes likely accelerate tumor growth and aremost effec-tive if acquired early, they do not appear to be universally required fortumorigenesis [192,193].

More generally, the extreme variation in cancer risk across tissuesremains poorly understood. While it has been proposed that this isdue to differences in the lifetime number of adult stem cell divisions,during which DNA replication errors occur [194], this simple explana-tion has been debated [195–197]. Recent studies have sought to charac-terize mutation rates in diverse mouse tissues [170] and human adultstem cells [198] by sequencing of normal clonal organoid cultures de-rived from primary multipotent cells. Intriguingly, the initial mousestudy reported differences in the mutation rate across tissue, as wellas distinct mutational processes, whereas the subsequent study similar-ly suggest tissue-specific activity of mutational processes but a steadyand consistent accumulation of mutations (approximately 40 de novomutations per year) in the human small intestine, colon and liver.While further investigation is needed, this approach suggests newways to quantify human somatic mutation rates.

4.2. Genetic drift

Genetic drift describes randomfluctuations in the frequency of geneticvariants in a population and is a major driving force of genetic diversityespecially for small populations [176,185]. The neutral evolution theoryposits that most fixed nucleotide substitutions in a population are drivenby random genetic drift rather natural selection [199]. The neutrality hy-pothesis has been foundational for statistical tests of neutrality and com-putational analyses of natural selection from population sequencing data.However, the role of genetic drift in cancer is oftenoverlooked as selectionis generally evoked to explain most aspects of tumor evolution. For in-stance, it is usually assumed that detectable subclones are under strongselection [13,49,200]. However, theory predicts that early neutral muta-tions during tumor growth may also be present at high frequency at thetime of diagnosis since random drift predominates when tumor size issmall. Much of the theory concerning genetic drift derives from theWright-Fisher or related models, which assume constant populationsizes (N genome copies) [185]. A well-known prediction is that the prob-ability of fixation in the population for a newneutralmutation is ρ=1/N.


For a growing population, the probability of fixation of a new mutationdepends on both the time of occurrence and the birth/death rate. Assum-ing a branching process, Bozic et al. [201] derived that the probability offixation of k-th surviving lineage is given by:

ρk ¼ u=u− log δð Þð Þk ð5Þ

where u is the neutral exonic mutation rate per cell division and δ theratio of cell death and birth rates. Although, there is limited experimentaldata, the ratio of death and birth rates in cancer has been estimated torange from δ = 0.72 in aggressive CRC metastases [202] to δ = 0.99 inearly tumors [203]. It can be seen from this formula that earlymutations can drift to fixation (clonality) with high probability duringthe early stages of clonal expansion when the cell death rate is highsuch that: ρ1 = 0.22, ρ2 = 0.05 and ρ3 = 0.01 for u = 0.03 (μ = 10−9)and δ=0.9. Neutralmutationsmay therefore be present at high frequen-cy in a tumor anddetectable byNGS. Thus,while genetic drift is a negativeforce that decreases whole-population diversity, it simultaneously in-creases the diversity of high frequencymutations due to random fluctua-tions (Fig. 4) and hence is an important, but often overlooked, feature oftumor sequencing data.

4.3. Spatial structure and migration

Spatial structure and migration can dramatically alter the genetic di-versity of a population [204,205]. In a spatially structured population, ge-netic drift and selection are restricted in localized regions thus giving riseto decreased genetic diversity within a local subpopulation while in-creasing genetic divergence between spatially segregated subpopula-tions. As shown in Fig. 5A, spatial partitioning within a tumor naturallygives rise to genetic divergence between samples from distant regions.This genetic segregation can be enhanced by local stochastic extinctionsdue to large demographic fluctuations at the expansion front and, hence



Fig. 5. Spatial structure and migration contribute to intra-tumor heterogeneity. (A) Schematic illustration of tumor biopsy or resection specimens derived frommultiple local regions of atumor expanding under stringent spatial constraints. Due to the spatial constraints, clones are partitioned and the genetic diversity within regions (R1, R2 or R3) is smaller than thediversity between regions (e.g. R1 vs R3). (B) While the three regions exhibit equivalent clonal composition, one cannot distinguish whether subclone mixing occurred early duringtumor growth. (C) Clone map derived from a virtual tumor simulated within an agent-based tumor growth model with spatial constraints and subclone mixing. Each colored regionrepresents an individual clone composed of cells that share the same mutations. Most clones are locally restricted, whereas the green clone spans distant regions of the tumor. (D)Migration and metastasis can significantly increase the divergence between tumors due to augmented genetic drift and clonal selection.


explains spatial heterogeneity in tumors [206]. These patterns are espe-cially relevant for interpreting NGS data from solid tumors characterizedby significant spatial constraints since biopsies and even resections areoften obtained from a localized region of the tumor. Such local samplingis inherently limited in its ability to capturemuch of the (detectable) ge-netic diversity in a tumor (Fig. 5). Recentmulti-region sequencing (MRS)studies have revealed extensive genetic diversity betweendifferent sam-ples from the same tumor suggesting that spatial constraints are signifi-cant in many tumor types [13,16,18,21,23,29,30,127,207]. Hence, tumorevolution may be driven by adaptations to distinct microenvironmentalniches amongst spatially partitioned populations.

Spatial structure thwarts selection and promotes the co-existence ofmultiple selective clones when selection is pervasive. Martens et al.[208] modeled crypt-structured tissues to study the effect of spatialstructure on colorectal tumor initiation where selection is expected tobe stringent. They report that spatial structure significantly increasesthe waiting time to transformation due to elevated clonal interferencebetween selectively advantageous clones compared with that in well-mixed populations such as leukemia. More generally, they proposedthat in somatic cells, the damaging effect of deleterious mutations canbe mitigated by differentiation and apoptosis and that tissue structureminimizes the risk of cancer [208]. However, the contributions of spatialstructure to these dynamics are complex [209] and warrant further in-vestigation. For example, spatial structure does not always impede se-lection. In contrast, it can accelerate complex adaption, a scenariowhere adaptation requires the accumulation of multiple mutations(e.g. a double hit to inactivate a tumor suppressor), depending on theepistatic fitness landscape [210].

CSC organization within tumors promotes spatial structure and in-creases phenotypic heterogeneity [211,212]. Due to the limited prolifer-ative potential of non-stem cells, only mutations in stem cells can result


in clonal expansions within the niche, regionally in the tumor, unlessthe mutations promote dedifferentiation of non-stem cells to stemcells [213]. CSC-driven tumor growth likely impedesmutational adapta-tion due to the decreased effective population size.Microenvironmentalheterogeneity can also promote the emergence of spatial structure intumors [214] and the plasticity of cellular phenotypes that are often reg-ulated by epigenetic factors. Importantly, the efficiency of selectionresulting from geneticmutationswould be dampened under high levelsof phenotypic plasticity [215].

Given the spatial constraintswithin solid tumors, spatial informationis important for resolving the evolutionary history of a tumor and thetiming of mutation occurrence. For instance, without knowledge ofsampling locations, subclonal composition can be explained by differ-ences in tumor growth dynamics, stringent spatial constraint (Fig. 5A)or early subclonemixing (Fig. 5B). However, knowledge of spatial loca-tion can help to constrain these scenarios. For example, if samples R1and R2 are spatially separated, early clonal mixing is more likely thana stringent spatial model. Therefore, the integration of spatial samplinginformation within a computational modeling framework (Fig. 5C) canaid the inference of tumor evolutionary dynamics.

Cellmigration has a dual role in the generation of tumor genetic diver-sity. On the one hand, within-tumormigration relaxes spatial constraints,thereby augmenting the efficiency of selection, giving rise to lower genet-ic diversity within a tumor. On the other hand, out-of-tumor migrationand metastasis will increase the genetic divergence between tumors(Fig. 5D). By simulating spatial tumor growthwith cellmigration,Waclawet al. [216] showed that out-of-tumormigration can dramatically acceler-ate tumor growth and increase tumor diversity. Recently, comparative ge-nomic sequencing studies revealed genomic divergence between pairedprimaries and metastasis [34,80,217] and multi-focal tumors [23,218],suggesting that migration drives tumor diversification.




4.4. Selection

4.4.1. Positive selectionCancer is traditionally viewed to result from the stepwise accumula-

tion of mutations, where cells transition from a normal health state topre-malignant, malignant and migratory phenotypes. The process be-gins when a single somatic cell acquires a heritable mutation that con-fers a fitness advantage, resulting in an increased proliferation orsurvival phenotype that enables it to outcompete less fit cells withinthe population. Natural selection operates on this phenotypic diversity,leading to sequential waves of clonal expansions and heterogeneityamongst subclones [7]. Within the current view, ongoing strong selec-tion is assumed to be the primary force and to govern all stages oftumor progression from initiation to subsequent primary tumor growth,aswell asmetastasis where the acquisition of additional ‘drivers’ resultsin continual selective sweeps [8,36,84]. In contrast, mutation, geneticdrift, population structure, and migration are often overlooked, al-though they represent key evolutionary forces that influence the expan-sion, fixation and extinction of subclones, as outlined in the precedingsections. The population dynamics and distribution of passenger, ad-vantageous and deleterious mutations can be understood through pop-ulation genetics models, where the relative importance depends on keyparameters such as population size. For example, genetic drift within asmall population can render selection less efficient, by allowing for thefixation of deleterious mutations and loss of advantageous mutations.

Various computational methods have been developed to identify‘driver’ mutations that confer a selective growth advantage (reviewedin [219]). Themajority of approaches evaluatewhether a gene is alteredmore frequently than expected by chance based on patterns of recur-rence across tumors, where it is critical to account for replication timingand variable mutation rates throughout the genome (or inhypermutated tumors) that can otherwise bias these patterns [83,220]. In contrast to these cohort-based approaches to define putativedriver genes, formal methods to detect selection in an individualtumor have been described in only a few instances [21,221]. For exam-ple, Sottoriva et al. [21] employed a spatial computational model oftumor growth and statistical inference framework to infer subclone fit-ness differences. Martincorena et al. [221] evaluated average mutantclone sizes and the dN/dS ratio, a measure of the relative abundanceof non-synonymous to synonymous mutations, and report evidencefor strong positive selection for mutations in normal skin, althoughthe interpretation has been debated [222,223]. More generally, it hasbeen assumed that the presence of subclonal known or putative driversimplies selection [13,29,30,164]. For example, in a breast cancer subjectto high-depth WGS high frequency subclonal mutational clusters wereobserved [164], compatible with the clonal expansion of a driver muta-tion accompanied by numerous hitchhiking passengers. Although this isa reasonable assumption, the presence of subclonal ‘driver’ mutationsalone does not necessarily guarantee that the mutation conferred a fit-ness advantage. Fitness estimates for specific mutations are unknownin human tumors, butmutations can be context dependent and epistatic[224,225] with differential effects during tumor initiation versus subse-quent expansion, potentially confounding the interpretation of suchpatterns. Moreover, many putative ‘drivers’ have yet to be functionallyand phenotypically characterized. Another strategy is to evaluate evi-dence for convergent evolution, as has been done in ccRCC [13,18,226], GBM [57] and CLL [58], providing evidence for selection.

4.4.2. Negative selectionNegative selection by deleterious mutations is pervasive in the evo-

lution of natural species and can significantly decrease population di-versity through purifying selection [227,228]. Classic molecularevolution theory posits that most newly arising mutations are slightlydeleterious and purged from the population quickly [185,199]. Howev-er, the extent to which this applies to somatic cell evolution is unclearsince tumors can harbor high levels of genomic instability in which


alterations might be deleterious [229]. Indeed, as discussed in Section2.2, chromosomal aberrations can impact cellular fitness [89,90,97].

Given the complete genomic linkage of asexually reproducingpopulations (due to the lack of recombination) such as tumor cells, in-terference between positive and negative selection (known as theHill-Robertson effect) [230] may be particularly strong when accompa-nied by high mutation rates [231]. Intriguingly, McFarland et al. [232,233] found that deleteriousmutations significantly alter tumor progres-sion by impeding the efficiency of positive selection. They furthershowed via simulation studies that some pre-malignant lesions dieout due to negative selection, suggesting the possibility of novel thera-peutic strategies through dramatic elevation of the mutation rate. It istempting to speculate that the apparently paradoxical relationship be-tween variable levels of genomic instability and cellularfitness tradeoffsis compatible with mutational meltdown and Muller's Ratchet [234,235], wherein deleterious alterations drive population extinction.While this may be particularly acute for small asexually reproducingpopulations, it is not clearwhether this applies in highly adaptive cancercell populations or how it is influenced by the interplay between delete-rious and advantageous mutations [233].

4.5. Evidence for different modes of tumor evolution

4.5.1. Pre-tumor progression: multiple clonal expansions to transformationIt has been appreciated for nearly two decades thatmore thanhalf of

all somaticmutations in colorectal cancers occur prior to transformation[1,2]. Critically, the initiating (epi)genetic events provide a selective fit-ness advantage in the target cell relative to other pre-malignant cells,resulting in clonal expansion. Initiating events are often tissue type spe-cific as is the case for APC in gastrointestinal tumors [3–5] andDNMT3 inleukemias [6], suggesting that they are beneficial only in certain cellularcontexts. It has long been assumed that the cell of origin may be a self-renewing stem cell and that this is the “unit” on which selectionoperates in cancer [8,236], where cancer reflects a state in which thenormal balance of self-renewal and differentiation is lost. This is sup-ported by observations that higher stem cell activity is associated withpoor prognosis in multiple tumor types [9]. The subsequent expansionof a subclone depends on the rates of advantageous and deleteriousmu-tations, the fitness benefit they confer in a given background populationand the effective population size.

Several studies have characterized genetic diversity in precursor orpre-invasive lesions, including the seminal study by Maley et al. [26],demonstrating that clonal diversity predicts progression from Barrettsesophageal to esophageal adenocarcinoma (ESCA). More recently,WGS and targeted sequencing was conducted to delineate mutationalordering in pre-invasive esophageal lesions [237], and reported thatmutations in ESCA occur very early during disease development,where several additional drivers occur during later stages of progressionwith both diagnostic and therapeutic implications. Pre-tumor progres-sion in the colon has also been studied by examining clonal evolutionin stem cell populations [238,239]. In particular, Baker et al. [239] re-ported that the number of functional stem cells is 5–6 in both normalpatients and individuals with familial adenomatous polyposis (FAP) as-sociated with germline APC mutation. In contrast, in adenomatouscrypts (APC(−/−)), an increase in both functional stem cell numberand the loss/replacement rate was noted as was the rate of divisionfor colonic crypts. Other studies have sought to experimentally derivefitness measurements of common drivers (such as APC and KRAS) dur-ing tumor initiation in the mouse small intestine and report that whilethey each provide strong selective advantages, mutation fixation ishighly inefficient due to hierarchical tissue architecture, strong geneticdrift and stochastic stem cell replacement [240]. Multiple studies haverevealed the early acquisition of tumor heterogeneity in non-invasiveadenomas [21,45,47,241]. Breast tumors also have well defined precur-sor lesions, facilitating studies of tumor initiation and early growth,where genetic diversity was similarly reported to occur early [178]




and reconstruction of phylogenetic trees amongst pre-invasive lesionsrevealed lineage heterogeneity both within and between lesions [33].

Clonal labeling techniques have also been employed to delineatemechanisms of tumor initiation and growth by genetically trackingthe fate of clones [242–244]. In particular, Dreissens et al. [242] reportedthe identification of two populations in benign papilloma cells, corre-sponding to a small, but persistent stem-cell-like population that rapid-ly divides and a slower cycling transient population that gives rise toterminally differentiated tumor cells. In parallel, mathematical model-ing of tumor initiation has sought to determine the number of driversrequired to initiate a tumor [245], suggesting that only three eventsare required for lung and colorectal cancer. As the timing of genomicevents that occur prior to transformation is obscured when studyingestablished tumors, it will be of continued importance to characterizeprecursor lesions towards identification of the cell of origin, driversand temporal ordering of tumor initiating events as may inform treat-ment and prognostication.

4.5.2. Linear versus branched (selective) evolutionAs described above, cancer genome sequencing efforts have revealed

extensive ITH in diverse tumors. Although relatively few primary tu-mors of a given tissue type have been analyzed in detail viamulti-regionor single cell sequencing, there have beenmultiple reports of ‘branched’evolution [13,17,18,22,25,29,30,33] (Fig. 1C). Hence, these data contrastwith the early view that after transformation, subsequent tumor growthtypically follows a linear clonal successionmodel [9] (Fig. 1B),wherein aselectively advantageous mutation has time to sweep through the pop-ulation prior to the acquisition of addition of additional advantageousmutations. Rather, branched clonal evolution occurs when multiple se-lectively advantageous subclones are present and complete via clonalinterference before fixation (Fig. 1C). It has been postulated that thismay be indicative of adaptation to different microenvironments orthat there are several routes from a normal cell to an advanced cancer[246]. Of note, phylogenies are branching diagrams and so called‘branched’ phylogenies do not imply branched clonal evolution. The to-pography of phylogenetic trees may differ depending on whether bothneutral (passenger) mutations and putative selectively advantageousmutations are considered or just the latter. Moreover, in spatially struc-tured populations, such as solid tumors, there is a high chance of detect-ing genetic divergence between regionally segregated samples due toenhanced genetic drift in localized regions.

4.5.3. Evidence for effectively neutral tumor evolution after transformationSelection is necessary for tumor initiation as the stepwise process of

mutation, selection and clonal expansion is what gives rise to a cell thatharbors the constellation of events required for unchecked growth andwhich becomes the founding tumor cell (Fig. 1).While it is generally as-sumed that strong selection is operative and detectable throughout allstages of tumor progression, this has not been systematically examined.Rather, recent findings have challenged the notion that strong positiveselection is continuously operative during the growth of a primarytumor after transformation [21,127,135]. For example, Sottoriva et al.[21] described a Big Bang model of colorectal tumor growth, whereinafter transformation, the tumor grows as a single terminal expansionpopulated by a large number of heterogeneous subclones that are notsubject to stringent selection. As such, in the Big Bangmodel, the timingof a mutation is the primary determinant of its frequency in the finaltumor and most detectable subclonal alterations (resulting in ITH)arise early during growth,whereas late-arisingmutationswill be largelyundetectable. The predictions of the Big Bang model were testedthrough single gland and bulk tumor profiling as well as spatial compu-tational modeling and statistical inference on the genomic data. Intrigu-ingly, inference on the genomic data in a patient-specific fashionrevealed signals of selection (corresponding to subclone fitness differ-ences) in the majority of carcinomas examined. However, selectionwas generally not sufficiently strong to result in hard selective sweeps


thatwould have dramatically altered subclonal architecture (by alteringclone frequencies), compatible with effectively neutral evolution.

Subsequent studies have corroborated aspects of ‘Big Bang’ dynam-ics [46,47,247], suggesting that they may be relatively common in colo-rectal tumors. Others have considered more strict models of neutrality.For example, Ling et al. reported neutral evolution in hepatocellular car-cinoma via in depth multi-region profiling [127]. Borrowing establishedapproaches from population genetics, they analyzed the site frequencyspectrum (SFS) [248] in a single liver tumorwhere 24 sampleswere sub-ject to WES and 300 samples were deeply genotyped and found that theobserved data could not reject the expected distribution arising under aneutral exponential growth model, assuming a well-mixed population.They also reported veryhigh levels of genetic diversity,where the hetero-zygosity between sampleswas estimatedH=0.941.Williams et al. [135]similarly employed a neutral exponential growth model and attemptedto fit somatic mutation data from bulk tumor samples in TCGA, resultingin the claim that 1/3 of tumors are compatible with neutrality.

4.5.4. Punctuated evolutionSeveral recent studies have reported punctuated clonal evolution

[31,32,83], karyotypic chaos [249], and other catastrophic events in var-ious tumor types, providing further evidence that the linear, sequentialclonal evolution model does not accurately describe the patterns of ITHfound inmanyhuman cancers. For example, chromothrypsis, a cataclys-mic event involving surges of chromosomal rearrangements that occurin a single or few cell cycles, has been reported in medulloblastoma, os-teosarcoma, pancreatic cancers, and other tumors [125,250,251]. Bygenerating multiple driver alterations in a single cell division, a greaterfitness benefit may be achieved thanwould be feasible through sequen-tial alterations [252]. Chromoplexy is another type of catastrophicevent, wherein chains of translocations are formed across multiplechromosomes, potentially due to double-stranded DNA breaks [253].In contrast, Kataegis results in localized hypermutation, preferentiallyat cytosine in TpC dinucleotides [254,255]. Collectively, these studiessuggest that mutation rates are not constant during the course oftumor progression and highlight the importance of understanding atumor's evolutionary trajectory (including the distinct stages of initia-tion and subsequent growth) in light of multiple sources of genomic al-terations, including SSNVs, CNAs and SVs.

5. Evolutionary strategies to exploit intra-tumor heterogeneitytherapeutically

5.1. Understanding, predicting and preventing tumor progression

Although tumor growth cannot be directly observed, patterns of ge-netic diversity present in a lesion reflect the evolutionary forces thatgave rise to it. It is therefore not surprising that ITH since a tumor's fu-ture behavior depends on its evolutionary course. In theory, the earlierthe initial clonal expansion is detected, the less diverse and less fit thecell population (since natural selection operates on diversity), and thebetter the prognosis. Even relatively early tumors exhibit high genomicdiversity [33,46], which may reflect processes that were already opera-tive for some time and likely will continue in the absence of interven-tion. The term “early” is indeed relative as the first clonal expansionswill be invisible and the lesion must be large enough to be detectedand sampled. In line with this view, we have previously proposed thatsome tumors may be ‘born to be bad’ [21], wherein invasive potentialmay be specified early. We have also observed that many private muta-tions originate from the first few cell divisions in human colorectaladenomas [45]. These findings suggest that targeting early clonal alter-ations in the relevant target cells and/or perturbing the microenviron-ment may alter the evolutionary trajectory of some tumors. Asdiscussed above, numerous forces shape tumor diversity and it is impor-tant to appreciate these contributions even when they cannot bedisentangled. More generally, understanding the mode of tumor




evolution may help to pinpoint the drivers of this process. This will re-quire a combination of detailed ‘omic’ analyses of clinical samples andrelevantmodel systems, aswell as computationalmodeling to delineatethe dynamics of this process.

Recently, several studies have described the use of computationaland mathematical modeling to gain insight into complex phenomenaduring tumor evolution [21,256], as well as strategies to prevent un-wanted clonal evolutionary processes [257–260]. For example, patternsof ITH can be analyzed using genealogy-based statistical inferencemethods within spatial computational models of tumor growth toinfer patient-specific tumor parameters such as the mutation rate,subclone fitness differences, and mutational timeline [21]. This ap-proach was employed to test the predictions of a Big Bang model of co-lorectal tumor growth. As it takes as input genomic data and requiresminimal assumptions about the underlying process, it can also revealunexpected features in the data.

Other efforts have sought to outline principles for preventing cancerby exploitinghost defenses against clonal evolution such as the immuneresponse, inhibition of angiogenesis and replicative senescence [258].For example, analysis of telomerase length regulation [261] illustratesthe need formathematicalmodeling in order to understand the dynam-ics of clones that could be restricted by replicative limits, as well as theefficacy of anti-telomerase therapies. Here, the authors demonstratethat knowing the replicative capacity of the founding tumor cell aloneis insufficient as it is also essential to understand the distributions ofthe number of cells, the balance between cell division and death, aswell as the probability of escaping replication senescence through telo-merase activation. These phenomena are unlikely to be understoodfrom experimental data alone; hence this highlights the importance ofintegrated experimentation and quantitative modeling.

Once tumor cells escape from natural defense mechanisms andachieve uncontrolled growth, disease progression inevitably ensues al-though the latency can vary considerably. Tumor reduction is oftenattempted via surgery, cytotoxic and/or cytostatic chemotherapy, andin some cases, molecular targeted agents. However, therapeutic resis-tance is virtually universal in solid tumors. Therapy itself can imposenew selective pressures on the tumor cell population and it will be im-portant to optimize treatment strategies by exploiting evolutionaryprinciples and context-dependencies, as discussed below.

5.2. Evolutionary and ecological strategies to circumvent resistance

The fact that cancer is a clonal evolutionary process poses challengesfor its effective control. Resistance can arise via several mechanisms, in-cluding the selection of pre-existing resistant subclones, de novo or ac-quired resistance, and competitive release, wherein elimination oftreatment sensitive cells results in a substantial reduction of tumor bur-den and a small population where resistant clones can readily grow outwithout competition [262]. Hence, the interplay between cancer cellpopulation size and their ecological niche impacts treatment response.Competitive releasemay be a commonphenomenon given that conven-tional treatment strategies aim to achievemaximal tumor cell kill underthe assumption that this is optimal for patient outcomes, whereas itmay actually promote resistance [257]. To circumvent this, Basanta etal. [257] proposed a game theoretical framework based on an evolution-ary double bindwhere two therapies are used sequentially such that re-sistance to one drug comes at a fitness cost, leaving the cells moresensitive to the other drug. They then investigated the impact that thisstrategy has on cancer control. A related approach, termed adaptivetherapy, administers treatment in pulses wherein removal of the drugresults in a fitness penalty to the resistant cell population, thereby keep-ing the population size in check. Adaptive therapy was found to yieldextended tumor control in preclinical breast cancer models [260] withefforts underway to evaluate this in clinical trials.

Some of the early efforts to understand resistance employedmathe-matical modeling to optimize drug dosing, for example in the context of


EGFR-mutant NSCLC [263]. Other studies sought to explore the evolu-tionary dynamics of treatment response. For example, Bozic et al. dem-onstrated that combination targeted therapy is superior to sequentialtherapy and can lead to long-term disease control, but even one poten-tial cross-resistancemutationwill dampen its benefit [264]. Others havefocused on understanding the impact of ITH on combinatorial treatmentstrategies. For example, Zhao et al. demonstrated that drugs selected asbeing the most beneficial single agent on the basis of clonal dominanceare often suboptimal for treating heterogeneous tumors via combina-tion therapy [265]. Rather, ITH reduces the advantage that a targetedtherapy has on a specific cell population, thereby altering the efficacyof intuitive combination therapy. Zhao et al. subsequently sought to ex-ploit temporal collateral sensitivity to circumvent resistance [259]. Theyreport the existence of a predictable and opportunistic therapeutic win-dow during a tumors' evolutionary trajectory and determine mecha-nisms of drug sensitization through a combination of genomic,phenotypic, signaling, and binding assays in conjunctionwith computa-tional modeling to identify novel vulnerabilities. This integrated evolu-tionary and pharmacological approach highlights the potential todevelop evolutionary informed treatment strategies with clinicalbenefit.

Collectively, these studies highlight key considerations for therapeu-tic strategies in light of ITH and suggest rational approaches for the de-sign of drug combinations and dosing. Ultimately, strategies such asthese that harness both evolutionary and ecological principles to targettumors may have the greatest benefit [206]. While most aspects oftumor ecology are hard to probe in vivo and therefore poorly under-stood, new technologies promise to enable fresh insights into tumormi-croenvironment and spatial structure, which can in turn, inform thesemodels. It is also clear that the full armament of strategies to combattumor progression have yet to be explored. As these efforts mature, itwill be important to continue to examine model assumptionsconcerning tumor dynamics, as well as to generate testable predictionsand to experimentally evaluate them. A combination of iterative exper-imentation and computational modeling should help to solidify thisframework.

5.3. Implications of ITH for clinical trial design and personalized medicine

Molecularly targeted therapies have revolutionized clinical oncologyand the discovery of new targets and biomarkers for patient stratifica-tion remains an active area of research. In the era of targeted therapies,the extensive evidence for ITH in diverse tumors accompanied by evi-dence for pre-existing resistance conferringmutations raises significantconcerns about the implications for achieving a durable response [60].In themetastatic setting, the acquisition of drug resistance can be partic-ularly rapid. To date, multiple examples of such pre-existing subclonalresistant mutations have been described, including KRAS mutations inthe context of anti-EGFR targeted therapy in colorectal cancer [202,266,267], rare MEK1 mutations in BRAF mutant melanomas receivingBRAF targeted therapy [268], chemo-resistant TP53 mutations in acutemyeloid leukemia [269]. These clinical examples are corroborated byin vitro studies using multiplexed barcode libraries to enable the detec-tion of extremely low frequency variants that confer resistance [270], aswell as by mathematical modeling, which predicts the presence of pre-existing resistant subclones [202,264]. Complicatingmatters further, ac-quired resistance can occur in the context of treatment selective pres-sure, as exemplified by convergent loss of PTEN during PI(3)Kαinhibition in breast cancer [271] and secondary EGFR mutations inNSCLC during anti-EGFR targeted therapy [272]. Non-genetic mecha-nism can also drive resistance, and have recently begun to be explored[109,111].

Given the diversity of mechanisms a cell can employ to evade ther-apy, the emergence of resistance should be anticipated and activelymanaged. Rare subclonal events pose a particular challenge as they fallbelow the limits of most diagnostic tools, and hence are not generally




considered for treatment stratification. The earlier detection of thesevariants could inform patient-tailored adaptive treatment strategies,particularly if this can be achieved in aminimally invasivemanor via liq-uid biopsy. Increasing data suggest that circulating tumor DNA (ctDNA)can be used to detect the emergence of resistance [273,274], monitormetastatic progression [275], predict recurrence and track minimal re-sidual disease [276]. Such liquid biopsies may also help to overcomesome of the challenges associated with spatial tumor heterogeneity.However, this will require detailed comparison of detailed MRS andctDNA. In the metastatic setting it will also be important to understandwhether distant metastases shed ctDNA at the same rates. For example,metastases to the brain (and primary brain tumors) are unlikely to shedsignificant cell free DNA into the blood due to the blood-brain barrier.Here, cerebrospinal fluid may provide an alternate reservoir of ctDNA[277].

In practice, the clinical community is grappling with how to detectsubclonal somatic alterations, whether the presence of a subclonalresistance mutation detected at low frequency should necessitate theuse of an alternate therapy (should one exist) [267] and whetheractionable clonal mutations are preferred targets. As discussed above,quantitative modeling can make predictions concerning both ofthese scenarios [259,265] and it will be important to pursue these ap-proaches further. It is also of interest to evaluate clinical actionabilityand utility in light of VAF values and quantitative measures of ITH. TheTRACERx (TRAcking non-small cell lung Cancer Evolution through ther-apy [Rx]) study, which aims to prospectively study cancer evolution vialongitudinal and spatial sampling, coupled with analysis of ctDNA andcirculating tumor cells (CTCs) should help to address some of thesequestions [278]. A chief aim of this effort is to determine whether theclonal dominance of molecularly targetable alterations portends im-proved progression-free survival in advanced disease. In parallel, thisstudy seeks to identify high-risk subclonal ‘drivers’ implicated in meta-static progression and to track clonal dynamics in the context oftherapy.

A systematic understanding of ITHwill be particularly important forthe improved management of metastatic disease, which is the ultimatecause of death for most patients. To date, therapeutic advances have yetto be fully translated to the metastatic setting. Here, the analysis of ge-netic diversity is particularly complicated due to the challenge inobtaining clinically annotated paired primaries and metastases, thecommon administration of intervening therapy, and potential for differ-ences in purity.Moreover, the extent of genetic divergence between pri-maries and metastases may vary according to tissue of origin withimplications for the relevant targets of systemic therapy. For example,colorectal cancers routinely share canonical drivers in the primary andmetastasis [66,67,69–71,75–77,279], whereas greater divergence andde novo putative drivers have been identified in other solid tumors[34]. To date, few studies have performed comprehensive genomic pro-filing of paired primaries and metastases with detailed treatment infor-mation (as therapy can confound the interpretation of the resultantgenomic patterns) and the limited available data derives from smallnumbers of cases. Thus, additional efforts to profile metastatic diseaseare needed and should similarly account for the spatial sampling biasesreported in primary tumors. Only then can rigorous comparisons ofpaired primaries and metastases be made with the potential to revealnew therapeutic vulnerabilities and to inform the dynamics of metasta-tic colonization [280].

It is clear that advances in precision oncologywill require tight coor-dination between laboratory scientists, computational biologists, oncol-ogists and clinical trial communities, as well as patient engagement.Despite the challenges that pervasive ITH and clonal dynamics pose,they beckon for evolutionary and ecological therapeutic strategies,which have only begun to be explored. Hence, there is an armamentof tools that were developed to understand population dynamics forwhich further investigation in the context of cancer is warranted to-wards the goal of improving patient outcomes.


6. Conclusions

In this review, we have outlined some of the recent advances incharacterizing ITH, as well as technical and methodological challengesin quantifying and interpreting these patterns. Although muchfocus to date has been on genotypic heterogeneity, emergenttechniques may facilitate the characterization of heterogeneity in situ[281]. In parallel, functional heterogeneity amongst tumor cellscan be assessed inpatient-derived xenograft [282,283] or organoid [284,285]models.Moreover, clinical imaging techniquesmay enable non-in-vasive, longitudinal monitoring of ITH at a whole organ level [286].Collectively, these approaches will help to resolve the genotype tophenotype map in cancer. We have also highlighted the power ofinterpreting tumor cell diversity within a population geneticsframework and the need for mathematical and computational modelsto infer tumor dynamics. The holy grail of such an approach would bethe development of predictive models to forecast and steer tumorevolution in order to achieve durable patient responses, thereby render-ing cancer a chronic disease. Before this can be achieved a systematicunderstanding of the different modes of evolution and fitness land-scapes is needed. It is also evident that the failure to accurately detectsubclonal alterations will impede precision medicine. Hence, it remainsof central importance to quantify both spatial and temporal genetic het-erogeneity towards the goal of improved predictive and prognosticbiomarkers.

Transparency document

The Transparency document associated with this article can befound, in the online version.

Acknowledgements

This work was funded by awards from the NIH (R01CA182514),Susan G. Komen Foundation (IIR13260750), and the Breast CancerResearch Foundation (BCRF-16-032) to C.C. Z.H. is supported by anInnovative Genomics Initiative (IGI) Postdoctoral Fellowship.

References

[1] J.L. Tsao, et al., Genetic reconstruction of individual colorectal tumor histories, Proc.Natl. Acad. Sci. U. S. A. 97 (3) (2000) 1236–1241.

[2] C. Tomasetti, B. Vogelstein, G. Parmigiani, Half or more of the somatic mutations incancers of self-renewing tissues originate prior to tumor initiation, Proc. Natl. Acad.Sci. U. S. A. 110 (6) (2013) 1999–2004.

[3] E.R. Fearon, B. Vogelstein, A genetic model for colorectal tumorigenesis, Cell 61 (5)(1990) 759–767.

[4] S. Nakatsuru, et al., Somatic mutation of the APC gene in gastric cancer: frequentmutations in very well differentiated adenocarcinoma and signet-ring cell carcino-ma, Hum. Mol. Genet. 1 (8) (1992) 559–563.

[5] A. Horii, et al., Frequent somatic mutations of the APC gene in human pancreaticcancer, Cancer Res. 52 (23) (1992) 6696–6698.

[6] L.I. Shlush, et al., Identification of pre-leukaemic haematopoietic stem cells in acuteleukaemia, Nature 506 (7488) (2014) 328–333.

[7] P.C. Nowell, The clonal evolution of tumor cell populations, Science 194 (4260)(1976) 23–28.

[8] M. Greaves, C.C. Maley, Clonal evolution in cancer, Nature 481 (7381) (2012)306–313.

[9] M. Greaves, Evolutionary determinants of cancer, Cancer Discov. 5 (8) (2015)806–820.

[10] D. Hanahan, R.A. Weinberg, Hallmarks of cancer: the next generation, Cell 144 (5)(2011) 646–674.

[11] G.H. Heppner, Tumor heterogeneity, Cancer Res. 44 (6) (1984) 2259–2265.[12] K. Anderson, et al., Genetic variegation of clonal architecture and propagating cells

in leukaemia, Nature 469 (7330) (2011) 356–361.[13] M. Gerlinger, et al., Intratumor heterogeneity and branched evolution revealed by

multiregion sequencing, N. Engl. J. Med. 366 (10) (2012) 883–892.[14] S.P. Shah, et al., The clonal and mutational evolution spectrum of primary triple-

negative breast cancers, Nature 486 (7403) (2012) 395–399.[15] A.Marusyk, V. Almendro, K. Polyak, Intra-tumour heterogeneity: a looking glass for

cancer? Nat. Rev. Cancer 12 (5) (2012) 323–334.[16] A. Sottoriva, et al., Intratumor heterogeneity in human glioblastoma reflects cancer

evolutionary dynamics, Proc. Natl. Acad. Sci. U. S. A. 110 (10) (2013) 4009–4014.


doi:10.1016/j.bbcan.2017.03.001

http://refhub.elsevier.com/S0304-419X(17)30046-X/rf0005



































[17] A. Bashashati, et al., Distinct evolutionary trajectories of primary high-grade serousovarian cancers revealed through spatial mutational profiling, J. Pathol. 231 (1)(2013) 21–34.

[18] M. Gerlinger, et al., Genomic architecture and evolution of clear cell renal cell car-cinomas defined by multiregion sequencing, Nat. Genet. 46 (3) (2014) 225–233.

[19] B.E. Johnson, et al., Mutational analysis reveals the origin and therapy-driven evo-lution of recurrent glioma, Science 343 (6167) (2014) 189–193.

[20] V. Almendro, et al., Inference of tumor evolution during chemotherapy by compu-tational modeling and in situ analysis of genetic and phenotypic cellular diversity,Cell Rep. 6 (3) (2014) 514–527.

[21] A. Sottoriva, et al., A Big Bangmodel of human colorectal tumor growth, Nat. Genet.47 (3) (2015) 209–216.

[22] H. Kim, et al., Whole-genome and multisector exome sequencing of primary andpost-treatment glioblastoma reveals patterns of tumor evolution, Genome Res.25 (3) (2015) 316–327.

[23] P.C. Boutros, et al., Spatial genomic heterogeneity within localized, multifocal pros-tate cancer, Nat. Genet. 47 (7) (2015) 736–745.

[24] M.R. Corces, et al., Lineage-specific and single-cell chromatin accessibility chartshuman hematopoiesis and leukemia evolution, Nat. Genet. 48 (10) (2016)1193–1203.

[25] A. McPherson, et al., Divergent modes of clonal spread and intraperitoneal mixingin high-grade serous ovarian cancer, Nat. Genet. (2016).

[26] C.C. Maley, et al., Genetic clonal diversity predicts progression to esophageal ade-nocarcinoma, Nat. Genet. 38 (4) (2006) 468–473.

[27] E.A. Mroz, et al., Intra-tumor genetic heterogeneity and mortality in head and neckcancer: analysis of data from the Cancer Genome Atlas, PLoS Med. 12 (2) (2015),e1001786.

[28] D. Wangsa, et al., Phylogenetic analysis of multiple FISH markers in oraltongue squamous cell carcinoma suggests that a diverse distribution ofcopy number changes is associated with poor prognosis, Int. J. Cancer 138(1) (2016) 98–109.

[29] J. Zhang, et al., Intratumor heterogeneity in localized lung adenocarcinomas delin-eated by multiregion sequencing, Science 346 (6206) (2014) 256–259.

[30] E.C. de Bruin, et al., Spatial and temporal diversity in genomic instability processesdefines lung cancer evolution, Science 346 (6206) (2014) 251–256.

[31] R. Gao, et al., Punctuated copy number evolution and clonal stasis in triple-negativebreast cancer, Nat. Genet. 48 (10) (2016) 1119–1130.

[32] Y. Wang, et al., Clonal evolution in breast cancer revealed by single nucleus ge-nome sequencing, Nature 512 (7513) (2014) 155–160.

[33] Z. Weng, et al., Cell-lineage heterogeneity and driver mutation recurrence in pre-invasive breast neoplasia, Genome Med. 7 (1) (2015) 28.

[34] P.K. Brastianos, et al., Genomic characterization of brain metastases revealsbranched evolution and potential therapeutic targets, Cancer Discov. 5 (11)(2015) 1164–1177.

[35] J. Cairns, Mutation selection and the natural history of cancer, Nature 255 (5505)(1975) 197–200.

[36] L.R. Yates, P.J. Campbell, Evolution of the cancer genome, Nat. Rev. Genet. 13 (11)(2012) 795–806.

[37] A. Sottoriva, et al., Single-molecule genomic data delineate patient-specific tumorprofiles and cancer stem cell organization, Cancer Res. 73 (1) (2013) 41–49.

[38] S.L. Cooke, et al., Intra-tumour genetic heterogeneity and poor chemoradiotherapyresponse in cervical cancer, Br. J. Cancer 104 (2) (2011) 361–368.

[39] E.A. Mroz, et al., High intratumor genetic heterogeneity is related to worse out-come in patients with head and neck squamous cell carcinoma, Cancer 119 (16)(2013) 3034–3042.

[40] R.A. Burrell, et al., The causes and consequences of genetic heterogeneity in cancerevolution, Nature 501 (7467) (2013) 338–345.

[41] C. Swanton, Intratumor Heterogeneity: evolution through space and time, CancerRes. 72 (19) (2012) 4875–4882.

[42] L.J. Barber, M.N. Davies, M. Gerlinger, Dissecting cancer evolution at themacro-het-erogeneity and micro-heterogeneity scale, Curr. Opin. Genet. Dev. 30 (2015) 1–6.

[43] M. Snuderl, et al., Mosaic amplification of multiple receptor tyrosine kinase genesin glioblastoma, Cancer Cell 20 (6) (2011) 810–817.

[44] N.J. Szerlip, et al., Intratumoral heterogeneity of receptor tyrosine kinases EGFR andPDGFRA amplification in glioblastoma defines subpopulationswith distinct growthfactor response, Proc. Natl. Acad. Sci. U. S. A. 109 (8) (2012) 3041–3046.

[45] H. Kang, et al., Many private mutations originate from the first few divisions of ahuman colorectal adenoma, J. Pathol. 237 (3) (2015) 355–362.

[46] C.K. Sievers, et al., Subclonal diversity arises early even in small colorectal tumoursand contributes to differential growth fates, Gut (2016).

[47] R. Uchi, et al., Integrated multiregional analysis proposing a newmodel of colorec-tal cancer evolution, PLoS Genet. (2016) 12(2).

[48] Y. Suzuki, et al., Multiregion ultra-deep sequencing reveals early intermixing andvariable levels of intratumoral heterogeneity in colorectal cancer, Mol. Oncol. 11(2) (2017) 124–139.

[49] L.R. Yates, et al., Subclonal diversification of primary breast cancer revealed bymultiregion sequencing, Nat. Med. 21 (7) (2015) 751–759.

[50] N.E. Potter, et al., Single-cell mutational profiling and clonal phylogeny in cancer,Genome Res. 23 (12) (2013) 2115–2125.

[51] A.P. Patel, et al., Single-cell RNA-seq highlights intratumoral heterogeneity in pri-mary glioblastoma, Science 344 (6190) (2014) 1396–1401.

[52] M. Meyer, et al., Single cell-derived clonal analysis of human glioblastoma linksfunctional and genomic heterogeneity, Proc. Natl. Acad. Sci. U. S. A. 112 (3)(2015) 851–856.

[53] X. Zhang, et al., Single-cell sequencing for precise cancer research: progress andprospects, Cancer Res. 76 (6) (2016) 1305–1312.


[54] M.M. Inda, et al., Tumor heterogeneity is an active process maintained by a mutantEGFR-induced cytokine circuit in glioblastoma, Genes Dev. 24 (16) (2010)1731–1745.

[55] A.S. Cleary, et al., Tumour cell heterogeneity maintained by cooperating subclonesin Wnt-driven mammary cancers, Nature 508 (7494) (2014) 113–117.

[56] A. Marusyk, et al., Non-cell-autonomous driving of tumour growth supports sub-clonal heterogeneity, Nature 514 (7520) (2014) 54–58.

[57] J.M. Francis, et al., EGFR variant heterogeneity in glioblastoma resolved throughsingle-nucleus sequencing, Cancer Discov. 4 (8) (2014) 956–971.

[58] J. Ojha, et al., Deep sequencing identifies genetic heterogeneity and recurrent con-vergent evolution in chronic lymphocytic leukemia, Blood 125 (3) (2015)492–498.

[59] C. Holohan, et al., Cancer drug resistance: an evolving paradigm, Nat. Rev. Cancer13 (10) (2013) 714–726.

[60] M.W. Schmitt, L.A. Loeb, J.J. Salk, The influence of subclonal resistancemutations ontargeted cancer therapy, Nat. Rev. Clin. Oncol. 13 (6) (2016) 335–347.

[61] J. Wang, et al., Clonal evolution of glioblastoma under therapy, Nat. Genet. (2016).[62] M. Castellarin, et al., Clonal evolution of high-grade serous ovarian carcinoma from

primary to recurrent disease, J. Pathol. 229 (4) (2013) 515–524.[63] L. Ding, et al., Clonal evolution in relapsed acute myeloid leukaemia revealed by

whole-genome sequencing, Nature 481 (2012) 506–510.[64] D.A. Landau, et al., Mutations driving CLL and their evolution in progression and re-

lapse, Nature 526 (2015) 525–530.[65] A. Schramm, et al., Mutational dynamics between primary and relapse neuroblas-

tomas, Nat. Genet. 47 (2015) 872–877.[66] F. Al-Mulla, et al., Comparative genomic hybridization analysis of primary colorec-

tal carcinomas and their synchronous metastases, Genes Chromosom. Cancer 24(4) (1999) 306–314.

[67] H.E. Alcock, et al., Analysis of colorectal tumor progression by microdissection andcomparative genomic hybridization, Genes Chromosom. Cancer 37 (4) (2003)369–380.

[68] S.P. Shah, et al., Mutational evolution in a lobular breast tumour profiled at singlenucleotide resolution, Nature 461 (7265) (2009) 809–813.

[69] E. Miranda, et al., Genetic and epigenetic alterations in primary colorectal cancersand related lymph node and liver metastases, Cancer 119 (2) (2013) 266–276.

[70] J.S. Vermaat, et al., Primary colorectal cancers and their subsequent hepatic metas-tases are genetically different: implications for selection of patients for targetedtreatment, Clin. Cancer Res. 18 (3) (2012) 688–699.

[71] A. Brannon, et al., Comparative sequencing analysis reveals high genomic concor-dance betweenmatched primary andmetastatic colorectal cancer lesions, GenomeBiol. 15 (8) (2014) 454.

[72] L. Ding, et al., Genome remodelling in a basal-like breast cancer metastasis and xe-nograft, Nature 464 (7291) (2010) 999–1005.

[73] P. Savas, et al., The subclonal architecture of metastatic breast cancer: results froma prospective community-based rapid autopsy program “CASCADE”, PLoS Med. 13(12) (2016), e1002204.

[74] P.J. Campbell, et al., The patterns and dynamics of genomic instability in metastaticpancreatic cancer, Nature 467 (7319) (2010) 1109–1113.

[75] T. Knosel, et al., Chromosomal alterations during lymphatic and liver metastasisformation of colorectal cancer, Neoplasia 6 (1) (2004) 23–28.

[76] J.K. Jiang, et al., Genetic changes and clonality relationship between primary colo-rectal cancers and their pulmonary metastases—an analysis by comparative geno-mic hybridization, Genes Chromosom. Cancer 43 (1) (2005) 25–36.

[77] J.C. Weber, et al., Allelotyping analyses of synchronous primary and metastasis CINcolon cancers identified different subtypes, Int. J. Cancer 120 (3) (2007) 524–532.

[78] T.E. Becker, et al., The genomic heritage of lymph nodemetastases: implications forclinical management of patients with breast cancer, Ann. Surg. Oncol. 15 (4)(2008) 1056–1063.

[79] M.C. Haffner, et al., Tracking the clonal origin of lethal prostate cancer, J. Clin. In-vest. 123 (11) (2013) 4918–4922.

[80] G. Gundem, et al., The evolutionary history of lethal metastatic prostate cancer, Na-ture 520 (7547) (2015) 353–357.

[81] M. El-Kebir, et al., Inferring themutational history of a tumor usingmulti-state per-fect phylogeny mixtures, Cell Syst. 3 (1) (2016) 43–53.

[82] T.M. Kim, et al., Subclonal genomic architectures of primary and metastatic colo-rectal cancer based on intratumoral genetic heterogeneity, Clin. Cancer Res. 21(19) (2015) 4461–4472.

[83] M.S. Lawrence, et al., Mutational heterogeneity in cancer and the search for newcancer-associated genes, Nature 499 (7457) (2013) 214–218.

[84] B. Vogelstein, et al., Cancer genome landscapes, Science 339 (6127) (2013)1546–1558.

[85] L.A. Loeb, C.F. Springgate, N. Battula, Errors in DNA replication as a basis of malig-nant changes, Cancer Res. 34 (9) (1974) 2311–2321.

[86] L.A. Loeb, Human cancers express a mutator phenotype: hypothesis, origin, andconsequences, Cancer Res. 76 (8) (2016) 2057–2059.

[87] C. Lengauer, K.W. Kinzler, B. Vogelstein, Genetic instability in colorectal cancers,Nature 386 (6625) (1997) 623–627.

[88] C. Lengauer, K.W. Kinzler, B. Vogelstein, Genetic instabilities in human cancers, Na-ture 396 (6712) (1998) 643–649.

[89] A.M. Laughney, et al., Dynamics of tumor heterogeneity derived from clonal karyo-typic evolution, Cell Rep. 12 (5) (2015) 809–820.

[90] K. Rowald, et al., Negative selection and chromosome instability induced by Mad2overexpression delay breast cancer but facilitate oncogene-independent out-growth, Cell Rep. 15 (12) (2016) 2679–2691.

[91] R.A. Burrell, et al., Replication stress links structural and numerical cancer chromo-somal instability, Nature 494 (7438) (2013) 492–496.
















































































































































































[92] L.A. Aaltonen, et al., Clues to the pathogenesis of familial colorectal cancer, Science260 (5109) (1993) 812–816.

[93] T.M. Kim, P.J. Park, A genome-wide view of microsatellite instability: old stories ofcancer mutations revisited with new sequencing technologies, Cancer Res. 74 (22)(2014) 6377–6382.

[94] J.K. Lee, et al., Mechanisms and consequences of cancer genome instability:lessons from genome sequencing studies, Annu. Rev. Pathol. 11 (2016)283–312.

[95] E.J. Fox, L.A. Loeb, Lethal mutagenesis: targeting the mutator phenotype in cancer,Semin. Cancer Biol. 20 (5) (2010) 353–359.

[96] J.H. Bielas, et al., Human cancers express a mutator phenotype, Proc. Natl. Acad. Sci.U. S. A. 103 (48) (2006) 18238–18242.

[97] S.M. Dewhurst, et al., Tolerance of whole-genome doubling propagates chromo-somal instability and accelerates cancer genome evolution, Cancer Discov. 4 (2)(2014) 175–185.

[98] B.A. Weaver, et al., Aneuploidy acts both oncogenically and as a tumor suppressor,Cancer Cell 11 (1) (2007) 25–36.

[99] S.L. Carter, et al., Absolute quantification of somatic DNA alterations in human can-cer, Nat. Biotechnol. 30 (5) (2012) 413–421.

[100] N. McGranahan, et al., Cancer chromosomal instability: therapeutic and diagnosticchallenges, EMBO Rep. 13 (6) (2012) 528–538.

[101] F.A. Sinicrope, Z.J. Yang, Prognostic and predictive impact of DNA mismatch repairin the management of colorectal cancer, Future Oncol. 7 (3) (2011) 467–474.

[102] R.A. Burrell, C. Swanton, The evolution of the unstable cancer genome, Curr. Opin.Genet. Dev. 24 (2014) 61–67.

[103] T.I. Zack, et al., Pan-cancer patterns of somatic copy number alteration, Nat. Genet.45 (10) (2013) 1134–1140.

[104] C. Curtis, et al., The genomic and transcriptomic architecture of 2,000 breast tu-mours reveals novel subgroups, Nature 486 (7403) (2012) 346–352.

[105] Cancer Genome Atlas Research Network, Comprehensive molecular characteriza-tion of gastric adenocarcinoma, Nature 513 (7517) (2014) 202–209.

[106] C. Kadoch, G.R. Crabtree, Reversible disruption of mSWI/SNF (BAF) complexes bythe SS18-SSX oncogenic fusion in synovial sarcoma, Cell 153 (1) (2013) 71–85.

[107] V. Almendro, A. Marusyk, K. Polyak, Cellular heterogeneity and molecular evolu-tion in cancer, Annu. Rev. Pathol. 8 (2013) 277–302.

[108] C.E. Meacham, S.J. Morrison, Tumour heterogeneity and cancer cell plasticity, Na-ture 501 (7467) (2013) 328–337.

[109] S.V. Sharma, et al., A chromatin-mediated reversible drug-tolerant state in cancercell subpopulations, Cell 141 (1) (2010) 69–80.

[110] T. Mazor, et al., Intratumoral heterogeneity of the epigenome, Cancer Cell 29 (4)(2016) 440–451.

[111] R.H. Chisholm, et al., Emergence of drug tolerance in cancer cell populations: anevolutionary outcome of selection, nongenetic instability, and stress-induced ad-aptation, Cancer Res. 75 (6) (2015) 930–939.

[112] A.O. Pisco, S. Huang, Non-genetic cancer cell plasticity and therapy-inducedstemness in tumour relapse: ‘what does not kill me strengthens me’, Br. J. Cancer112 (11) (2015) 1725–1732.

[113] V. Almendro, et al., Genetic and phenotypic diversity in breast tumor metastases,Cancer Res. 74 (5) (2014) 1338–1348.

[114] J.E. Dietrich, T. Hiiragi, Stochastic patterning in the mouse pre-implantation em-bryo, Development 134 (23) (2007) 4219–4231.

[115] H.H. Chang, et al., Transcriptome-wide noise controls lineage choice inmammalianprogenitor cells, Nature 453 (7194) (2008) 544–547.

[116] R. Losick, C. Desplan, Stochasticity and cell fate, Science 320 (5872) (2008) 65–68.[117] M. Acar, J.T. Mettetal, A. van Oudenaarden, Stochastic switching as a survival strat-

egy in fluctuating environments, Nat. Genet. 40 (4) (2008) 471–475.[118] S.L. Spencer, et al., Non-genetic origins of cell-to-cell variability in TRAIL-induced

apoptosis, Nature 459 (7245) (2009) 428–432.[119] S.A. Frank, M.R. Rosner, Nonheritable cellular variability accelerates the evolution-

ary processes of cancer, PLoS Biol. 10 (4) (2012), e1001296.[120] P.B. Gupta, et al., Stochastic state transitions give rise to phenotypic equilibrium in

populations of cancer cells, Cell 146 (4) (2011) 633–644.[121] L.I. Shlush, D. Hershkovitz, Clonal evolution models of tumor heterogeneity, Am.

Soc. Clin. Oncol. Educ. Book (2015) e662–e665.[122] R.A. Gatenby, et al., Cellular adaptations to hypoxia and acidosis during somatic

evolution of breast cancer, Br. J. Cancer 97 (5) (2007) 646–653.[123] H.O. Lee, et al., Evolution of tumor invasiveness: the adaptive tumor microenviron-

ment landscape model, Cancer Res. 71 (20) (2011) 6327–6337.[124] M.C. Lloyd, et al., Darwinian dynamics of intratumoral heterogeneity: not solely

random mutations but also variable environmental selection forces, Cancer Res.76 (11) (2016) 3136–3144.

[125] F. Notta, et al., A renewed model of pancreatic cancer evolution based on genomicrearrangement patterns, Nature 538 (7625) (2016) 378–382.

[126] N.R. Bertos, M. Park, Laser capture microdissection as a tool to study tumor stroma,Methods Mol. Biol. 1458 (2016) 13–25.

[127] S. Ling, et al., Extremely high genetic diversity in a single tumor points to preva-lence of non-Darwinian cell evolution, Proc. Natl. Acad. Sci. U. S. A. 112 (47)(2015) E6496–E6505.

[128] N.B. Larson, B.L. Fridley, PurBayes: estimating tumor cellularity andsubclonality in next-generation sequencing data, Bioinformatics 29 (15)(2013) 1888–1889.

[129] G. Ha, et al., TITAN: inference of copy number architectures in clonal cell popula-tions from tumor whole-genome sequence data, Genome Res. 24 (11) (2014)1881–1893.

[130] Z. Hu, C. Curtis, Inferring tumor phylogenies from mulit-region sequencing, CellSyst. 3 (1) (2016) 12–14.


[131] Y. Jiang, et al., Assessing intratumor heterogeneity and tracking longitudinal andspatial clonal evolutionary history by next-generation sequencing, Proc. Natl.Acad. Sci. U. S. A. 113 (37) (2016) E5528–E5537.

[132] N. Niknafs, et al., Subclonal hierarchy inference from somatic mutations: automaticreconstruction of cancer evolutionary trees from multi-region next generation se-quencing, PLoS Comput. Biol. 11 (10) (2015), e1004416.

[133] A.G. Deshwar, et al., PhyloWGS: reconstructing subclonal composition and evolu-tion from whole-genome sequencing of tumors, Genome Biol. 16 (2015) 35.

[134] R. Kostadinov, C.C. Maley, M.K. Kuhner, Bulk genotyping of biopsies can create spu-rious evidence for hetereogeneity in mutation content, PLoS Comput. Biol. 12 (4)(2016), e1004413.

[135] M.J. Williams, et al., Identification of neutral tumor evolution across cancer types,Nat. Genet. (2016).

[136] M. Gerstung, et al., Reliable detection of subclonal single-nucleotide variants in tu-mour cell populations, Nat. Commun. 3 (2012) 811.

[137] K. Cibulskis, et al., Sensitive detection of somatic point mutations in impure andheterogeneous cancer samples, Nat. Biotechnol. 31 (3) (2013) 213–219.

[138] J.M. Findlay, et al., Differential clonal evolution in oesophageal cancers in responseto neo-adjuvant chemotherapy, Nat. Commun. 7 (2016) 11111.

[139] C. Swanton, et al., Consensus on precisionmedicine for metastatic cancers: a reportfrom the MAP conference, Ann. Oncol. (2016).

[140] M. Josephidou, A.G. Lynch, S. Tavaré, multiSNV: a probabilistic approach for im-proving detection of somatic point mutations from multiple related tumour sam-ples, Nucleic Acids Res. 43 (9) (2015) e61.

[141] E.N. Smith, et al., Biased estimates of clonal evolution and subclonal heterogeneitycan arise from PCR duplicates in deep sequencing experiments, Genome Biol. 15(8) (2014) 420.

[142] I. Kinde, et al., Detection and quantification of rare mutations with massively par-allel sequencing, Proc. Natl. Acad. Sci. U. S. A. 108 (23) (2011) 9530–9535.

[143] M. Costello, et al., Discovery and characterization of artifactual mutations in deepcoverage targeted capture sequencing data due to oxidative DNA damage duringsample preparation, Nucleic Acids Res. 41 (6) (2013), e67.

[144] D.I. Lou, et al., High-throughput DNA sequencing errors are reduced by orders ofmagnitude using circle sequencing, Proc. Natl. Acad. Sci. U. S. A. 110 (49) (2013)19872–19877.

[145] M.W. Schmitt, et al., Detection of ultra-raremutations by next-generation sequenc-ing, Proc. Natl. Acad. Sci. U. S. A. 109 (36) (2012) 14508–14513.

[146] L. Oesper, A. Mahmoody, B.J. Raphael, THetA: inferring intra-tumor heterogeneityfrom high-throughput DNA sequencing data, Genome Biol. 14 (7) (2013) R80.

[147] A. Fischer, et al., High-definition reconstruction of clonal composition in cancer,Cell Rep. 7 (5) (2014) 1740–1752.

[148] Z. Yu, A. Li, M. Wang, CloneCNA: detecting subclonal somatic copy number alter-ations in heterogeneous tumor samples from whole-exome sequencing data,BMC Bioinformatics 17 (2016) 310.

[149] A. Roth, et al., PyClone: statistical inference of clonal population structure in cancer,Nat. Methods 11 (4) (2014) 396–398.

[150] C.A.Miller, et al., SciClone: inferring clonal architecture and tracking the spatial andtemporal patterns of tumor evolution, PLoS Comput. Biol. 10 (8) (2014), e1003665.

[151] I.M. Lonnstedt, et al., Deciphering clonality in aneuploid breast tumors using SNParray and sequencing data, Genome Biol. 15 (9) (2014) 470.

[152] M. El-Kebir, et al., Multi-state perfect phylogeny mixtures for cancer sequencing,Cell Syst. (2016).

[153] Y. Wang, N.E. Navin, Advances and applications of single-cell sequencing technol-ogies, Mol. Cell 58 (4) (2015) 598–609.

[154] A. Roth, et al., Clonal genotype and population structure inference from single-celltumor sequencing, Nat. Methods (2016).

[155] K. Naxerova, R.K. Jain, Using tumour phylogenetics to identify the roots of metas-tasis in humans, Nat. Rev. Clin. Oncol. 12 (5) (2015) 258–272.

[156] Z.M. Zhao, et al., Early and multiple origins of metastatic lineages within primarytumors, Proc. Natl. Acad. Sci. U. S. A. 113 (8) (2016) 2140–2145.

[157] Z. Yang, B. Rannala, Molecular phylogenetics: principles and practice, Nat. Rev.Genet. 13 (5) (2012) 303–314.

[158] Z.H. Yang, Phylogenetic analysis using parsimony and likelihood methods, J. Mol.Evol. 42 (2) (1996) 294–307.

[159] J. Felsenstein, Evolutionary trees from DNA sequences: a maximum likelihood ap-proach, J. Mol. Evol. 17 (6) (1981) 368–376.

[160] N. Saitou, M. Nei, The neighbor-joining method: a new method for reconstructingphylogenetic trees, Mol. Biol. Evol. 4 (4) (1987) 406–425.

[161] E. Letouze, et al., Analysis of the copy number profiles of several tumor samplesfrom the same patient reveals the successive steps in tumorigenesis, GenomeBiol. 11 (7) (2010) R76.

[162] R.F. Schwarz, et al., Phylogenetic quantification of intra-tumour heterogeneity,PLoS Comput. Biol. 10 (4) (2014), e1003535.

[163] R.F. Schwarz, et al., Spatial and temporal heterogeneity in high-grade serous ovar-ian cancer: a phylogenetic analysis, PLoS Med. 12 (2) (2015), e1001789.

[164] S. Nik-Zainal, et al., The life history of 21 breast cancers, Cell 149 (5) (2012)994–1007.

[165] K. Sprouffske, J.W. Pepper, C.C. Maley, Accurate reconstruction of the temporalorder of mutations in neoplastic progression, Cancer Prev. Res. (Phila.) 4 (7)(2011) 1135–1144.

[166] S. Malikic, et al., Clonality inference in multiple tumor samples using phylogeny,Bioinformatics 31 (9) (2015) 1349–1356.

[167] V. Popic, et al., Fast and scalable inference of multi-sample cancer lineages, Ge-nome Biol. 16 (1) (2015) 91.

[168] E.M. Ross, F. Markowetz, OncoNEM: inferring tumor evolution from single-cell se-quencing data, Genome Biol. 17 (2016) 69.
















































































































































































[169] K. Jahn, J. Kuipers, N. Beerenwinkel, Tree inference for single-cell data, GenomeBiol. 17 (2016) 86.

[170] S. Behjati, et al., Genome sequencing of normal cells reveals developmental line-ages and mutational processes, Nature 513 (7518) (2014) 422–425.

[171] J.F. Kingman, The coalescent, Stoch. Process. Appl. 13 (1982) 235–248.[172] P. Marjoram, S. Tavaré, Modern computational approaches for analysing molecular

genetic variation data, Nat. Rev. Genet. 7 (10) (2006) 759–770.[173] M.A. Beaumont,W. Zhang, D.J. Balding, Approximate Bayesian computation in pop-

ulation genetics, Genetics 162 (4) (2002) 2025–2035.[174] P. Nicolas, et al., The stem cell population of the human colon crypt: analysis via

methylation patterns, PLoS Comput. Biol. 3 (3) (2007), e28.[175] D. Dykhuizen, D.L. Hartl, Selective neutrality of 6PGD allozymes in E. coli and the

effects of genetic background, Genetics 96 (4) (1980) 801–817.[176] J.H. Gillespie, Population genetics : a concise guide, second ed. Johns Hopkins Uni-

versity Press, Baltimore, Md., 2004 (xiv, 214 p).[177] E.H. Simpson, Measurement of diversity, Nature (1949).[178] D.C. Allred, et al., Ductal carcinoma in situ and the emergence of diversity during

breast cancer evolution, Clin. Cancer Res. 14 (2) (2008) 370–378.[179] C.E. Shannon, A mathematical theory of communication, Bell Syst. Tech. J. 27

(1948) 379–432.[180] L.M. Merlo, et al., A comprehensive survey of clonal diversity measures in Barrett's

esophagus as biomarkers of progression to esophageal adenocarcinoma, CancerPrev. Res. (Phila.) 3 (11) (2010) 1388–1397.

[181] E.A. Mroz, J.W. Rocco, MATH, a novel measure of intratumor genetic heterogeneity,is high in poor-outcome classes of head and neck squamous cell carcinoma, OralOncol. 49 (3) (2013) 211–215.

[182] L.H. Mengelbier, et al., Intratumoral genome diversity parallels progression andpredicts outcome in pediatric cancer, Nat. Commun. 6 (2015) 6125.

[183] L.G. Morris, et al., Pan-cancer analysis of intratumor heterogeneity as a prognosticdeterminant of survival, Oncotarget 7 (9) (2016) 10051–10063.

[184] N. Andor, et al., Pan-cancer analysis of the extent and consequences of intratumorheterogeneity, Nat. Med. 22 (1) (2016) 105–113.

[185] D.L. Hartl, A.G. Clark, Principles of Population Genetics, fourth ed. Sinauer Associ-ates, 2007.

[186] M. Lynch, Evolution of the mutation rate, Trends Genet. 26 (8) (2010) 345–352.[187] M. Lynch, Rate, molecular spectrum, and consequences of human mutation, Proc.

Natl. Acad. Sci. U. S. A. 107 (3) (2010) 961–968.[188] D.F. Conrad, et al., Variation in genome-wide mutation rates within and between

human families, Nat. Genet. 43 (7) (2011) 712–714.[189] J.F. Crow, The origins, patterns and implications of human spontaneous mutation,

Nat. Rev. Genet. 1 (1) (2000) 40–47.[190] L.A. Loeb, A mutator phenotype in cancer, Cancer Res. 61 (8) (2001) 3230–3239.[191] L.A. Loeb, Mutator phenotype in cancer: origin and consequences, Semin. Cancer

Biol. 20 (5) (2010) 279–280.[192] I.P. Tomlinson, M.R. Novelli, W.F. Bodmer, Themutation rate and cancer, Proc. Natl.

Acad. Sci. U. S. A. 93 (25) (1996) 14800–14803.[193] I. Tomlinson, W. Bodmer, Selection, the mutation rate and cancer: ensuring that

the tail does not wag the dog, Nat. Med. 5 (1) (1999) 11–12.[194] C. Tomasetti, B. Vogelstein, Cancer etiology. Variation in cancer risk among tissues can

be explained by the number of stem cell divisions, Science 347 (6217) (2015) 78–81.[195] A.I. Rozhok, G.M.Wahl, J. DeGregori, A critical examination of the “Bad Luck” expla-

nation of cancer risk, Cancer Prev. Res. (Phila.) 8 (9) (2015) 762–764.[196] R. Noble, O. Kaltz, M.E. Hochberg, Peto's paradox and human cancers, Philos. Trans.

R. Soc. Lond. Ser. B Biol. Sci. (2015) 370(1673).[197] S. Wu, et al., Substantial contribution of extrinsic risk factors to cancer develop-

ment, Nature 529 (7584) (2016) 43–47.[198] F. Blokzijl, et al., Tissue-specific mutation accumulation in human adult stem cells

during life, Nature 538 (7624) (2016) 260–264.[199] M. Kimura, The Neutral Theory of Molecular Evolution, Cambridge University

Press, Cambridge Cambridgeshire; New York, 1983 (xv, 367 p.).[200] D.A. Landau, et al., Evolution and impact of subclonalmutations in chronic lympho-

cytic leukemia, Cell 152 (4) (2013) 714–726.[201] I. Bozic, J.M. Gerold, M.A. Nowak, Quantifying clonal and subclonal passenger mu-

tations in cancer evolution, PLoS Comput. Biol. 12 (2) (2016), e1004731.[202] L.A. Diaz Jr., et al., The molecular evolution of acquired resistance to targeted EGFR

blockade in colorectal cancers, Nature 486 (7404) (2012) 537–540.[203] I. Bozic, et al., Accumulation of driver and passenger mutations during tumor pro-

gression, Proc. Natl. Acad. Sci. U. S. A. 107 (43) (2010) 18545–18550.[204] M. Slatkin, Gene flow and the geographic structure of natural populations, Science

236 (4803) (1987) 787–792.[205] L. Excoffier, Patterns of DNA sequence diversity and genetic structure after a range

expansion: lessons from the infinite-island model, Mol. Ecol. 13 (4) (2004)853–864.

[206] K.S. Korolev, J.B. Xavier, J. Gore, Turning ecology and evolution against cancer, Nat.Rev. Cancer 14 (5) (2014) 371–380.

[207] T.M. Kim, et al., Regional biases inmutation screening due to intratumoural hetero-geneity of prostate cancer, J. Pathol. 233 (4) (2014) 425–435.

[208] E.A. Martens, et al., Spatial structure increases the waiting time for cancer, New J.Phys. 13 (2011).

[209] L. Hindersin, et al., Should tissue structure suppress or amplify selection to mini-mize cancer risk? Biol. Direct 11 (2016).

[210] N.L. Komarova, Spatial interactions and cooperation can change the speed of evo-lution of complex phenotypes, Proc. Natl. Acad. Sci. U. S. A. 111 (Suppl. 3) (2014)10789–10795.

[211] A. Sottoriva, et al., Cancer stem cell tumor model reveals invasive morphology andincreased phenotypical heterogeneity, Cancer Res. 70 (1) (2010) 46–56.


[212] J. Poleszczuk, P. Hahnfeldt, H. Enderling, Evolution and phenotypic selection of can-cer stem cells, PLoS Comput. Biol. 11 (3) (2015), e1004025.

[213] K. Sprouffske, et al., An evolutionary explanation for the presence of cancernonstem cells in neoplasms, Evol. Appl. 6 (1) (2013) 92–101.

[214] C. Carmona-Fontaine, et al., Emergence of spatial structure in the tumor microen-vironment due to theWarburg effect, Proc. Natl. Acad. Sci. U. S. A. 110 (48) (2013)19402–19407.

[215] J. Gallaher, A.R. Anderson, Evolution of intratumoral phenotypic heterogeneity: therole of trait inheritance, Interface Focus 3 (4) (2013) 20130016.

[216] B. Waclaw, et al., A spatial model predicts that dispersal and cell turnover limitintratumour heterogeneity, Nature (2015) http://dx.doi.org/10.1038/nature14971.

[217] S. Yachida, et al., Distantmetastasis occurs late during the genetic evolution of pan-creatic cancer, Nature 467 (7319) (2010) 1114–1117.

[218] C.S. Cooper, et al., Analysis of the genetic phylogeny of multifocal prostate canceridentifies multiple independent clonal expansions in neoplastic and morphologi-cally normal prostate tissue, Nat. Genet. 47 (4) (2015) 367–372.

[219] B.J. Raphael, et al., Identifying driver mutations in sequenced cancer genomes:computational approaches to enable precision medicine, Genome Med. 6 (1)(2014) 5.

[220] E. Hodis, et al., A landscape of driver mutations in melanoma, Cell 150 (2) (2012)251–263.

[221] I. Martincorena, et al., Tumor evolution. High burden and pervasive positive selec-tion of somatic mutations in normal human skin, Science 348 (6237) (2015)880–886.

[222] B.D. Simons, Deep sequencing as a probe of normal stem cell fate and preneoplasiain human epidermis, Proc. Natl. Acad. Sci. U. S. A. 113 (1) (2016) 128–133.

[223] B.D. Simons, REPLY TO MARTINCORENA, et al., Evidence for constrained positiveselection of cancer mutations in normal skin is lacking, Proc. Natl. Acad. Sci. U. S.A. 113 (9) (2016) E1130–E1131.

[224] B. Bauer, R. Siebert, A. Traulsen, Cancer initiation with epistatic interactions be-tween driver and passenger mutations, J. Theor. Biol. 358 (2014) 52–60.

[225] C.A. Ortmann, et al., Effect of mutation order on myeloproliferative neoplasms, N.Engl. J. Med. 372 (7) (2015) 601–612.

[226] R. Fisher, et al., Development of synchronous VHL syndrome tumors reveals con-tingencies and constraints to tumor evolution, Genome Biol. 15 (8) (2014) 433.

[227] D. Charlesworth, B. Charlesworth, M.T. Morgan, The pattern of neutral molecularvariation under the background selection model, Genetics 141 (4) (1995)1619–1632.

[228] B. Charlesworth, M.T. Morgan, D. Charlesworth, The effect of deleteriousmutationson neutral molecular variation, Genetics 134 (4) (1993) 1289–1303.

[229] R.A. Beckman, L.A. Loeb, Negative clonal selection in tumor evolution, Genetics 171(4) (2005) 2123–2131.

[230] W.G. Hill, A. Robertson, The effect of linkage on limits to artificial selection, Genet.Res. 8 (3) (1966) 269–294.

[231] C.I. Wu, et al., The ecology and evolution of cancer: the ultra-microevolutionaryprocess, Annu. Rev. Genet. 50 (2016) 347–369.

[232] C.D. McFarland, et al., Impact of deleterious passenger mutations on cancer pro-gression, Proc. Natl. Acad. Sci. U. S. A. 110 (8) (2013) 2910–2915.

[233] C.D. McFarland, L.A. Mirny, K.S. Korolev, Tug-of-war between driver and passengermutations in cancer and other adaptive processes, Proc. Natl. Acad. Sci. U. S. A. 111(42) (2014) 15138–15143.

[234] M. Lynch, et al., The mutational meltdown in asexual populations, J. Hered. 84 (5)(1993) 339–344.

[235] W. Gabriel, M. Lynch, R. Burger, Muller's ratchet and mutational meltdowns, Evo-lution 47 (6) (1993) 1744–1757.

[236] A. Kreso, J.E. Dick, Evolution of the cancer stem cell model, Cell Stem Cell 14 (3)(2014) 275–291.

[237] J.M. Weaver, et al., Ordering of mutations in preinvasive disease stages of esopha-geal carcinogenesis, Nat. Genet. 46 (8) (2014) 837–843.

[238] P. Calabrese, S. Tavaré, D. Shibata, Pretumor progression: clonal evolution ofhuman stem cell populations, Am. J. Pathol. 164 (4) (2004) 1337–1346.

[239] A.M. Baker, et al., Quantification of crypt and stem cell evolution in the normal andneoplastic human colon, Cell Rep. 8 (4) (2014) 940–947.

[240] L. Vermeulen, et al., Defining stem cell dynamics in models of intestinal tumor ini-tiation, Science 342 (6161) (2013) 995–998.

[241] C.K. Sievers, et al., Understanding intratumoral heterogeneity: lessons from theanalysis of at-risk tissue and premalignant lesions in the colon, Cancer Prev. Res.9 (8) (2016) 638–641.

[242] G. Driessens, et al., Defining the mode of tumour growth by clonal analysis, Nature488 (7412) (2012) 527–530.

[243] J. Chen, et al., A restricted cell population propagates glioblastoma growth afterchemotherapy, Nature 488 (7412) (2012) 522–526.

[244] A.G. Schepers, et al., Lineage tracing reveals Lgr5+ stem cell activity in mouse in-testinal adenomas, Science 337 (6095) (2012) 730–735.

[245] C. Tomasetti, et al., Only three driver gene mutations are required for the develop-ment of lung and colorectal cancers, Proc. Natl. Acad. Sci. U. S. A. 112 (1) (2015)118–123.

[246] M. Gerlinger, et al., Cancer: evolution within a lifetime, Annu. Rev. Genet. 48(2014) 215–236.

[247] Y. Suzuki, et al., Multiregion ultra-deep sequencing reveals early intermixing andvariable levels of intratumoral heterogeneity in colorectal cancer, Mol. Oncol.(2016).

[248] R. Nielsen, Molecular signatures of natural selection, Annu. Rev. Genet. 39 (2005)197–218.

[249] H.H. Heng, et al., Stochastic cancer progression driven by non-clonal chromosomeaberrations, J. Cell. Physiol. 208 (2) (2006) 461–472.

































































































http://dx.doi.org/10.1038/nature14971















































































[250] P.J. Stephens, et al., Massive genomic rearrangement acquired in a single cata-strophic event during cancer development, Cell 144 (1) (2011) 27–40.

[251] T. Rausch, et al., Genome sequencing of pediatric medulloblastoma links cat-astrophic DNA rearrangements with TP53 mutations, Cell 148 (1–2) (2012)59–71.

[252] J.O. Korbel, P.J. Campbell, Criteria for inference of chromothripsis in cancer ge-nomes, Cell 152 (6) (2013) 1226–1236.

[253] S.C. Baca, et al., Punctuated evolution of prostate cancer genomes, Cell 153 (3)(2013) 666–677.

[254] C. Cheng, et al., Whole-genome sequencing reveals diverse models of structuralvariations in esophageal squamous cell carcinoma, Am. J. Hum. Genet. 98 (2)(2016) 256–274.

[255] L.B. Alexandrov, et al., Signatures of mutational processes in human cancer, Nature500 (7463) (2013) 415–421.

[256] B. Waclaw, et al., A spatial model predicts that dispersal and cell turnover limitintratumour heterogeneity, Nature 525 (7568) (2015) 261–264.

[257] D. Basanta, R.A. Gatenby, A.R. Anderson, Exploiting evolution to treat drug resis-tance: combination therapy and the double bind, Mol. Pharm. 9 (4) (2012)914–921.

[258] I.A. Rodriguez-Brenes, D. Wodarz, Preventing clonal evolutionary processes in can-cer: insights from mathematical models, Proc. Natl. Acad. Sci. U. S. A. 112 (29)(2015) 8843–8850.

[259] B. Zhao, et al., Exploiting temporal collateral sensitivity in tumor clonal evolution,Cell 165 (1) (2016) 234–246.

[260] P.M. Enriquez-Navas, et al., Exploiting evolutionary principles to prolong tumorcontrol in preclinical models of breast cancer, Sci. Transl. Med. 8 (327) (2016)327ra24.

[261] I.A. Rodriguez-Brenes, C.S. Peskin, Quantitative theory of telomere length regula-tion and cellular senescence, Proc. Natl. Acad. Sci. U. S. A. 107 (12) (2010)5387–5392.

[262] B. Zhao, M.T. Hemann, D.A. Lauffenburger, Modeling tumorclonal evolution fordrug combinations design, Trends Cancer 2 (3) (2016) 144–158.

[263] J. Chmielecki, et al., Optimization of dosing for EGFR-mutant non-small cell lungcancer with evolutionary cancer modeling, Sci. Transl. Med. 3 (90) (2011) 90ra59.

[264] I. Bozic, et al., Evolutionary dynamics of cancer in response to targeted combinationtherapy, elife 2 (2013), e00747.

[265] B. Zhao,M.T. Hemann, D.A. Lauffenburger, Intratumor heterogeneity alters most ef-fective drugs in designed combinations, Proc. Natl. Acad. Sci. U. S. A. 111 (29)(2014) 10773–10778.

[266] S. Misale, et al., Emergence of KRAS mutations and acquired resistance to anti-EGFR therapy in colorectal cancer, Nature 486 (7404) (2012) 532–536.

[267] P. Laurent-Puig, et al., Clinical relevance of KRAS-mutated subclones detected withpicodroplet digital PCR in advanced colorectal cancer treated with anti-EGFR ther-apy, Clin. Cancer Res. 21 (5) (2015) 1087–1097.


[268] K. Kemper, et al., Intra- and inter-tumor heterogeneity in a vemurafenib-resistantmelanoma patient and derived xenografts, EMBO Mol. Med. 7 (9) (2015)1104–1118.

[269] T.N. Wong, et al., Role of TP53 mutations in the origin and evolution of therapy-re-lated acute myeloid leukaemia, Nature 518 (7540) (2015) 552–555.

[270] H.E. Bhang, et al., Studying clonal dynamics in response to cancer therapy usinghigh-complexity barcoding, Nat. Med. 21 (5) (2015) 440–448.

[271] D. Juric, et al., Convergent loss of PTEN leads to clinical resistance to a PI(3)Kalphainhibitor, Nature 518 (7538) (2015) 240–244.

[272] M.N. Balak, et al., Novel D761Y and common secondary T790M mutations in epi-dermal growth factor receptor-mutant lung adenocarcinomas with acquired resis-tance to kinase inhibitors, Clin. Cancer Res. 12 (21) (2006) 6494–6501.

[273] M. Murtaza, et al., Non-invasive analysis of acquired resistance to cancer therapyby sequencing of plasma DNA, Nature 497 (7447) (2013) 108–112.

[274] A. Romanel, et al., Plasma AR and abiraterone-resistant prostate cancer, Sci. Transl.Med. 7 (312) (2015) 312re10.

[275] S.J. Dawson, et al., Analysis of circulating tumor DNA to monitor metastatic breastcancer, N. Engl. J. Med. 368 (13) (2013) 1199–1209.

[276] J. Tie, et al., Circulating tumor DNA analysis detects minimal residual disease andpredicts recurrence in patients with stage II colon cancer, Sci. Transl. Med. 8(346) (2016) 346ra92.

[277] Y. Wang, et al., Detection of tumor-derived DNA in cerebrospinal fluid of patientswith primary tumors of the brain and spinal cord, Proc. Natl. Acad. Sci. U. S. A.112 (31) (2015) 9704–9709.

[278] M. Jamal-Hanjani, et al., Tracking genomic cancer evolution for precisionmedicine:the lung TRACERx study, PLoS Biol. 12 (7) (2014), e1001906.

[279] N. Knijn, et al., KRAS mutation analysis: a comparison between primary tumoursand matched liver metastases in 305 colorectal cancer patients, Br. J. Cancer 104(6) (2011) 1020–1026.

[280] S. Turajlic, C. Swanton, Metastasis as an evolutionary process, Science 352 (6282)(2016) 169–175.

[281] M. Angelo, et al., Multiplexed ion beam imaging of human breast tumors, Nat. Med.20 (4) (2014) 436–442.

[282] A. Bruna, et al., A biobank of breast cancer explants with preserved intra-tumorheterogeneity to screen anticancer compounds, Cell 167 (1) (2016) 260–274(e22).

[283] P. Eirew, et al., Dynamics of genomic clones in breast cancer patient xenografts atsingle-cell resolution, Nature 518 (7539) (2015) 422–426.

[284] M. van de Wetering, et al., Prospective derivation of a living organoid biobank ofcolorectal cancer patients, Cell 161 (4) (2015) 933–945.

[285] M.A. Cantrell, C.J. Kuo, Organoid modeling for cancer precision medicine, GenomeMed. 7 (1) (2015) 32.

[286] J.Y. Kim, R.A. Gatenby, Quantitative clinical imaging methods for monitoringintratumoral evolution, Methods Mol. Biol. 1513 (2017) 61–81.



























































































A population genetics perspective on the determinants of intra...

Documents

Transcript of A population genetics perspective on the determinants of intra...