Keywords

Introduction

Cancer is a genetic disease, from the predisposing alleles carried in the constitutive genome to the random somatic events selected for during tumorigenesis . In the last 15 years, the analysis of cancer genomes has dramatically improved in scope and level of detail. Low resolution and low-throughput methods such as G-banded karyotyping and comparative genomic hybridization have been superseded, first by array-based and more recently by sequencing-based technologies that enable affordable genome-wide single nucleotide resolution analysis of hundreds and even thousands of tumors. In a research setting this has led to novel insights regarding the initiation and evolution of cancer, and the genetic events detected are increasingly having clinically relevant implications.

This chapter introduces the main classes of genetic events that are commonly seen in cancer genomes and discusses the contemporary methodologies with which they are detected. Applying these methods has led to a number of discoveries with implications for molecular pathology, including using genetic events to evaluate cancer risk, refine diagnoses, provide prognostic information, and most critically, determining genetic events against which molecular therapeutics can be targeted.

Classes of Genetic Events in Cancer

Cancer has long been recognized as a genetic disease , since the earliest observations of deranged chromosomes [1] and familial clustering of cases [2–5]. Predisposition to cancer, initiating events, and progression are all influenced by genetics whether they be constitutive or somatic aberrations. There are many different types of genetic alteration that can occur, each of which arise through different mechanisms and each having varying consequences. The vast majority of somatic changes that occur in a tumor are thought to have little functional effect and are consequently described as “passengers,” carried along by coincidence upon selection of a co-existing “driver” in the same cell [6, 7]. Discerning driver mutations from passenger mutations remains a major challenge in translating genomic data into the clinic.

Somatic Mutation

Acquired changes in the constitutional DNA sequence are common in most cancer types and include base-pair substitutions and small (<1 kb) insertion–deletions (indels). Mutations are caused by a failure of one or more of the DNA repair pathways to recognize or accurately repair DNA following a genetic insult, which can include inherent replication errors, deamination of methylated cytosine, and mutagenic exposures such as UV light. The rate of such mutations per cancer genome varies greatly depending on the cancer type and has been estimated at 0.57/Mb of coding sequence for acute lymphoblastic leukemia [7], 0.19/Mb for breast cancer [7], ~1.8/Mb for high grade ovarian cancer [8], ~18/Mb for mutagen-exposed cancers like melanoma and non-small cell lung cancer [7, 9], and as high as 400/Mb for cancers with a DNA repair defect such as loss of mismatch repair in colorectal cancer [10]. Different cancer types often have unique mutation signatures in terms of the type of mutation, e.g. UV-exposed cancers have high rates of C:G > T:A transitions resulting in an enrichment of dipyrimidines.

The impact of a somatic mutation will vary widely depending on its location (coding/non-coding/splice site/regulatory) and type of change following translation (missense, nonsense, frameshift, etc.) (Fig. 1). Some mutations will have an immediate impact and are considered dominant while others may require loss of the remaining normal allele.

Fig. 1
figure 1

Mutation types . (a) Wildtype transcript. Shaded boxes depict coding exons, white boxes depict the untranslated regions (UTRs) of the transcript, and intervening grey lines indicate intergenic and intronic regions (b) Coding variants: Frameshift (triangle) and nonsense (X) variants are often overtly deleterious due to protein truncation. Missense variants may be deleterious depending on the function of the specific amino acid changed and the effect on protein folding, or have no effect. Synonymous mutations do not change the amino acid identity, but may influence splice site function or binding of regulatory proteins, or have no effect. (c) Essential splice site variants (2 bp ± intron–exon boundary, arrow) can result in exon skipping or cryptic exons being transcribed. (d) Non-coding variants (intergenic, UTRs, intronic, arrows) may have an effect on transcription regulatory regions, transcript splicing, and mRNA stability, or often no effect on transcript function. (e) Translocations resulting in an in-frame fusion can produce functional protein

Full size image

Copy Number

An abnormal chromosome number (aneuploidy) is a common feature of many carcinomas. The subsequent imbalance in gene copies is thought to lead to global changes in gene expression with wide-ranging effects on cell phenotype. Aneuploidy is caused by errors in chromosome segregation during mitosis or cytokinesis, leading to gain or loss of whole chromosomes, and not uncommonly duplication of the entire chromosome complement leading to tetraploidy.

Copy number aberrations can also occur at a sub-chromosomal level through various mechanisms, often involving compromised repair of double strand (ds) DNA breaks [11, 12], a breakage-fusion-bridge cycle subsequent to dsDNA breaks or telomere dysfunction [13], and less commonly chromothripsis [14]. Copy number changes include losses of material, either hemizygous or homozygous deletions, and gains of material, which can be low level changes, such as a duplication, or high-level amplification (from five to possibly hundreds of copies) (Fig. 2), as is often observed with ERBB2 in breast cancer.

Fig. 2
figure 2

Copy number aberrations and mutations. The boxes indicate the types of aberrations each technology is capable of detecting. The resolution of the technology indicates the smallest size of aberrations that can be robustly detected. (Asterisk) Detected as a copy number aberration by array methods, however, the site of the translocation fusion cannot be determined

Full size image

Loss of Heterozygosity

Loss of heterozygosity (LOH ) refers to the change in genotype from heterozygous to homozygous of polymorphic alleles that arises through chromosome loss, sub-chromosomal deletion or gene conversion via homologous recombination and DNA repair. LOH is often associated with copy number loss; however, gene conversion or duplication of chromosomes can lead to copy number neutral LOH (Fig. 2). LOH is distinct from allelic imbalance (AI), in which both alleles are still present but in a ratio different from 1:1 following copy number gain. The effect of LOH is to unmask recessive alleles, either inherited (e.g., BRCA1) or somatic (e.g., TP53), leading to loss of tumor suppressor gene function.

Structural Chromosome Changes

Any event involving inappropriate repair of a dsDNA break can lead to structural changes in chromosomes, including not only the aforementioned sub-chromosomal copy number changes, but also inversions and translocations. These latter events lead to the novel juxtaposition of genetic material, which can cause inappropriate gene regulation or novel protein products, such as the BCR-ABL translocation in chronic myelogenous leukemia [15, 16].

Germline Variation

The constitutive genetic variation carried by individuals is extensive and encompasses many of the same forms observed as somatic events. The most common class of germline aberrations most relevant to cancer are considered to be single nucleotide polymorphisms (SNPs) and small indels. It is likely that larger copy number variations and structural changes such as inversions are also important, but there are few conclusive incidences reported to date. SNPs and indels can vary widely in population frequency, from common (>10 % minor allele frequency) to rare (1–10 % frequency) to extremely rare (<1 % frequency). Most are inherited from parents, but some occur de novo, at a frequency estimated at 13 × 10−3/Mb for SNPs and 0.78 × 10−3/Mb for indels per generation [17, 18].

Methods of Genomic Analysis

High-resolution screening for genetic lesions on a genome-wide scale has only recently become feasible, i.e., on a kilobase-down to base-pair level. Prior to the invention of array and massively parallel sequencing-based methods (MPS; also referred to as next-generation sequencing, Sanger sequencing is considered first generation sequencing), analysis resolution was limited to tens of megabase-pairs using comparative genomic hybridization (CGH) or G-banded karyotyping. In addition, these methods were time-consuming and required a high degree of individual skill to interpret the chromosome spreads, limiting the number of samples that could be studied. Other, more targeted techniques could be readily applied to multiple samples such as fluorescence in situ hybridization (FISH), Sanger sequencing, and microsatellite genotyping by PCR, but these were not easily scaled up to a genome-wide analysis. These methodologies have now been supplemented by several whole genome techniques that have delivered myriad research findings and are increasingly been applied in clinical settings (Table 1).

Table 1 Methods of genomic analysis
Full size table

Karyotyping

The advent of fluorescence-based chromosome painting in the 1990s enabled a more automated karyotyping procedure compared to traditional G-banding. Variously known as Spectral karyotyping (SKY), M-FISH, and 24-color FISH, this technique uses paints made from individual flow-sorted chromosomes each labeled with a different mix of fluorophores. This paint is applied to metaphase chromosome spreads usually generated from primary tumors after short term cell culture. SKY is able to resolve complex marker chromosomes and is the only method discussed here that can measure exact ploidy (Fig. 3a). It can also give some indication of tumor genetic heterogeneity, as each nucleus is individually analyzed. However, it is still a low-resolution method (~10 Mb) and relies heavily on good quality metaphase spreads. Thus its use is limited to fresh tumor material and to laboratories with a cell culture facility.

Fig. 3
figure 3

Genomic analysis data. (a) Spectral karyotyping of the breast cancer cell line VP229, demonstrating aneuploidy and extensive translocations and structural rearrangements. (b) Array CGH method overview, with data analysis of one representative chromosome (log ratio data plotted), indicating where regions of chromosome have been gained (red) or lost (green). (c) SNP CGH method overview, with Partek® plot of Affymetrix® SNP6™ data from an ovarian tumor, showing chromosomes linearly mapped from 1 through X showing total copy number (upper) and allele-specific copy number (lower). (d) Circos plot demonstrating typical genomic copy number aberrations and structural rearrangements in the 94778 cell line derived from a retroperitoneal relapse of a well-differentiated liposarcoma [19] 94778 cells were provided by Florence Pedeutour (Laboratory of Solid Tumors Genetics, Nice University Hospital, Nice, France). The data was analyzed and figure generated by Anthony Papenfuss (Bioinformatics Division, The Walter and Eliza Hall Institute of Medical Research, Parkville, VIC, Australia, courtesy of D. Thomas and D. Garsed. CGH comparative genomic hybridization, SNP single nucleotide polymorphism

Full size image

Array-Based

All array-based methods for genomic analysis operate on the same basic principle: hybridization of a labeled DNA sample to complementary probe sequences that are immobilized to a solid substrate in known locations. The strength of the signal is proportional to the amount of each target sequence in the sample, however, with most platforms, the dynamic range tends to become compressed at high copy number and generally cannot accurately distinguish, for example, 8 copies from 12 copies.

Array-based systems all require some kind of normalization strategy to calculate copy number, where the signal intensity of the tumor sample is converted to a ratio using normal diploid samples. Normalization can be performed against matched normal DNA, which is useful for discriminating constitutional copy number variants, or against the average of multiple normal DNA samples. All data for a sample are median- or mean-centered, which means that it is not possible to distinguish between perfectly tetraploid and diploid samples, and exact ploidy cannot be determined unless genotype information is also available (see below).

The level of genomic resolution of array-based platforms is inherently limited by the number and type of probe sequences selected. Initially, detection of copy number was done using cDNA arrays produced for expression analysis but these were quickly superseded by superior platforms including bacterial artificial chromosome (BAC) arrays , oligonucleotide arrays, and SNP arrays.

BACs are large plasmid vectors with inserts of tens to hundreds of kilobases but because of the way they are constructed, they tend to be irregularly spaced across the genome, although whole genome tiling arrays with better uniformity have been produced. The resolution of BAC arrays is also limited by the large insert size. However, the signal-to-noise ratio and dynamic range are generally good.

As a consequence of the completion of the sequence of the entire human genome in the early 2000s, synthetic oligonucleotide-based arrays have become possible. These arrays utilize relatively short (25–60 bp) synthesized oligonucleotides, overcoming issues with appropriately spacing probes throughout the genome, and increasing the resolution of breakpoint detection. Array comparative genomic hybridization (aCGH; Fig. 3b) typically refers to the use of non-polymorphic oligonucleotide arrays, which can detect changes in total copy number but not changes in allelic ratios and therefore cannot identify copy number neutral LOH (Fig. 2). Single nucleotide polymorphism (SNP) arrays , a type of oligonucleotide array, allow the detection of both total copy number and LOH by designing probes spanning known common polymorphisms in the human genome (Fig. 3c).

Oligonucleotide-based arrays perform well on good quality DNA from fresh or frozen tissues and lymphocyte DNA; however, they are less useful for degraded DNA such as that obtained from formalin-fixed, paraffin-embedded (FFPE) tissue.

DNA from FFPE sources is predominantly highly fragmented and alternative approaches have been successfully developed for its analysis (Fig. 4). Molecular inversion probe (MIP) technology has been incorporated into the Affymetrix® OncoScan™ assay. This approach involves circularizable “padlock” probes with two terminal sequences that bind to homologous sequences either side of an SNP, followed by highly specific closing of the padlock through incorporation of a nucleotide complementary to the SNP. This approach circumvents the issue of having to directly digest, ligate, and amplify the fragmented DNA. An alternative approach, taken by Illumina®, is to “restore” the FFPE DNA by ligating the fragmented DNA together to generate fragments large enough for whole genome amplification prior to labeling and hybridization. This DNA can then be hybridized to a standard bead array.

Fig. 4
figure 4

Copy number analysis of FFPE DNA. DNA extracted from ductal carcinoma in situ assayed using OncoScan™ (Affymetrix®). The upper panel illustrates total copy number, while the bottom panel illustrates allele ratios. Detectable copy number aberrations include whole chromosome and chromosome arm gains and losses, focal deletions, and high-level amplifications

Full size image

Sequencing-Based

Sanger Sequencing

In the mid-1970s, Fredrick Sanger first described “Sanger Sequencing,” and since this time huge advances in the technology and its applications have been made, with the technology notably underpinning the Human Genome Project. The current iteration of the technology is based on sequencing by synthesis, with the products resolved using capillary electrophoresis and laser optics. Selective incorporation of fluorescently labeled, chain terminating dideoxynucleotides occurs during modified PCR amplification, and following purification of the products, they are resolved based on size using a capillary sequencer. Lasers excite the fluorescent dyes, and sequences of up to 700–900 bases can be determined, which are exported as a chromatogram (Fig. 5a).

Fig. 5
figure 5

Sequencing data. (a) Geneious (Biomatters) browser view of Sanger sequencing traces of wildtype KRAS (lower) and KRAS c.34G > A, p.G12D mutant (upper). (b) IGV browser view of MPS reads mapped to KRAS demonstrating wildtype and mutant (T) reads (KRAS c.34G > A, p.G12D mutant)

Full size image

Massively Parallel Sequencing

The advent of massively parallel sequencing (MPS) , or the so-called next-generation sequencing , represents an enormous step forward in the genomics field. While continuing to “sequence by synthesis,” these technologies vastly increase throughput by simultaneously sequencing multiple DNA strands (in “parallel”). Unlike Sanger sequencing, MPS generates millions of random short reads (35–700 bp) that must then be mapped to a reference genome (Fig. 5b). The most commonly employed technologies in cancer research are ion semi-conductor sequencing and optics-based dye sequencing. Semi-conductor sequencing (Ion Torrent Systems Inc.) measures the hydrogen release following incorporation of a nucleotide as determined by the template strand of DNA. Conversely, dye sequencing (Illumina®) relies on the immobilization of template DNA clusters onto a solid surface, upon which fluorescently labeled nucleotides competitively bind; a laser excites the label, and images of incorporated bases are recorded. Prior to sequencing, DNA samples are prepared into libraries, representing all of the desired DNA target sequences, be they the entire genome (whole genome sequencing, WGS), only the exons (whole exome sequencing, WES), or a targeted panel (Fig. 6). A targeted panel allows for specific enrichment of certain sequences; the enrichment can be performed using either a nucleic acid bait (“capture”) or through PCR-based amplification. It is at the library preparation stage that samples can be barcoded and pooled to increase throughput.

Fig. 6
figure 6

MPS sequencing. Schematic of the input genome (intronic DNA in red, exons in shades of blue, green and orange) and the sequences represented in libraries generated for the three major types of sequencing; whole genome, whole exome, and targeted

Full size image

Using MPS, it is possible to simultaneously detect single nucleotide mutations and SNPs, as well as small indels, large copy number aberrations, LOH, and structural rearrangements, depending on the type of sequencing performed (Fig. 2). Variants are identified (“called”) by programs that assess the sequence evidence for the particular variant (read depth, base quality, etc.) and are exported into a mutation annotation format (MAF) file. This calling procedure is typically performed against a matched normal for the detection of somatic variants as the number of non-reference germline variants in any given individual is substantial. Copy number and LOH events are assessed by comparing read depths and allele frequencies from the tumor to those from the matched normal sample. Structural rearrangements (translocations) and large indels are identified by assessing paired reads that did not map within the expected distance from each other (determined by the average fragment length of the DNA library prepared for sequencing) or that mapped to different chromosomes (Fig. 3c).

Limitations of Genomic Analyses

All genomic analyses of tumors using dissociated and homogenized tissues suffer potential dilution of tumor-derived genomic events by the genomes of surrounding non-neoplastic cells. Estimating the percentage of tumor cells in a sample and enriching for tumor cells using laser or needle microdissection or selective enrichment using a tumor-specific cell surface marker (e.g., EpCam) is an important process upstream of genomic analysis. Heterogeneity of genomic events within the tumor cell population can also contribute to dilution of signals, resulting in subpopulations not being discernible at any great resolution.

Sanger sequencing and array-based copy number outputs are an average of the genomes of all of the cells from which DNA has been isolated (Figs. 3b and 5a) and therefore have limited sensitivity to detect events occurring only in a subpopulation of tumor cells. MPS, particularly at very high read depth, offers greater scope to resolve events occurring in a small subpopulation of cells and gives a digital count of variant reads (Fig. 5b). Paired-end sequencing also allows the mapping of translocation events, which cannot be resolved using array technology where chromosomes are linearly mapped (Fig. 3b, c). MPS is not without its own issues; a problematic area of MPS is the accurate mapping of reads to a reference genome and the calling of variants. Reads become difficult to correctly map to areas of the genome that are highly homologous to other regions, have repetitive sequence, or when there is an indel in the read or region relative to the reference.

Whole genome and exome sequencing provide the possibility of simultaneously detecting all variants in the genome or all coding variants. Along with real variants, PCR and sequencing artifacts are also detected, resulting in a huge number of potential variants to analyze for validity, recurrence, and functional impact. Differentiating between mutations that are driving tumorigenesis (“drivers”) from those that are not anticipated to have any involvement in the development of a tumor (“passengers”) is difficult and is made more onerous because passenger mutations are predicted to far outnumber driver mutations. The bioinformatic analysis burden of MPS should not be underestimated, although efforts to improve software design and usability are ongoing (see Chapter “Bioinformatics Analysis of Sequence Data”).

Applications of Genomic Analysis in Cancer

Genome-Wide Association Studies

One of the most common uses of SNP arrays has been to the application of genome-wide association studies (GWAS), where linkage of SNPs to an increased (or decreased) risk of disease is assessed across the genome. Due to cost constraints, these studies are most often performed in a staged process, where a few hundred or thousand individuals with features suggestive of a genetic predisposition to cancer such as family history or early age of onset are first compared to age- and ethnicity-matched controls. Significant hits from this analysis are then validated in tens of thousands of cases and controls. This strategy has been applied to all common cancer types, with multiple predisposing SNPs identified in breast, colorectal, lung, and ovarian cancers. The risks associated with individual SNPs are usually low (1.2–1.5-fold above the general population), however, the polygenic risk when multiple low-risk SNPs are inherited together can reach much greater significance. In addition, because the resolution of the studies is limited by the array density, the SNPs with the highest risk association may not be the causative variant, but only closely linked. Thus, fine mapping is required for more precise information on the gene affected and the possible mechanism of the increase in risk. Nonetheless, some risk alleles, which have been validated in multiple independent cohorts, are now being utilized for testing in familial cancer clinics [20].

More recently, as the cost of MPS continues to fall, exome and genome sequencing of large cohorts are being undertaken. For example, large databases have been established with a focus on thoroughly characterizing common cancers [International Cancer Genome Consortium (ICGC) , The Cancer Genome Atlas (TCGA) ] and providing a population baseline for common and rare variants [Exome Variant Server (EVS), 1000 Genomes, dbSNP, Exome Aggregation Consortium (ExAC)]. These studies will have the advantage of being able to detect rare variants and causative alleles; however, it may be some years before they are powerful enough to identify rare, low to moderate risk alleles.

Mapping of Oncogenes and Tumor Suppressor Genes

A major goal of genomic analyses in the research setting is the discovery of the full repertoire of genes with a role in tumorigenesis. These genes can be tumor promoting when their activity is deregulated (oncogenes) or when they are inactivated (tumor suppressors [TSG]). Some genes can act as either oncogene or TSG depending on the cellular context and the pathways driving tumorigenesis; for example, NOTCH1 is targeted by activating mutations in hematopoietic malignancies [21] and inactivating mutations in solid tumors such as head and neck squamous cell carcinoma [22].

Oncogenes commonly act in a dominant fashion, with the genetic aberration ranging from copy number increase (e.g. ERBB2), recurrent activating point mutation (e.g. KRAS, BRAF) (Fig. 7a, b), translocation (e.g. BCR-ABL), or other structural chromosomal changes leading to loss of transcriptional (e.g. MYC) or post-translational control (e.g. EGFR). These types of recurrent activating events typically make oncogenes easier to design clinical tests for (compared to TSGs) because there is a limited number of functionally relevant mutational events.

Fig. 7
figure 7

Oncogene and tumor suppressor gene mutation patterns . Distribution of mutations in the oncogenes (a) KRAS and (b) PIK3CA and TSGs (c) ARID1A and (d) RB1. Green arrowheads indicate missense mutations, black arrowheads indicate nonsense mutations, grey arrowheads synonymous mutations, while blue arrowheads indicate truncating indels. Mutation patterns are based on 50 randomly selected mutations from the COSMIC database (COSMIC http://cancer.sanger.ac.uk/cosmic)

Full size image

Methods for discovering new oncogenes include mapping regions of copy number gain, exome sequencing for somatic mutations, and karyotyping or genome sequencing for structural chromosome changes. Regardless of methodology, a common challenge is distinguishing the driving genetic events from benign passenger events.

Identifying genes affected by copy number alterations has been most effectively achieved using array technologies. The increases in copy number are mapped in multiple samples, and those regions of the genome that most often display increases in copy number are short-listed as potential sites of oncogenes. However, this is complicated by the degree of copy number change—should any increase be investigated even if only a single copy, or should only high-level amplifications be considered? Both methods have been applied, and bioinformatic techniques that balance both possibilities have been developed (e.g. GISTIC [23, 24]). The list of genes in minimal regions of copy number change can still be long, and expression and functional analyses are then required to identify putative drivers. For example, integrated copy number and expression analysis identified novel growth promoting genes in ovarian carcinomas [25] and candidate oncogenes driving ovarian cancer were functionally investigated using RNA interference [26].

Full genome sequencing is the most comprehensive and sensitive method to identify structural chromosome changes, although to date fusion genes have also been detected using the much cheaper approach of RNAseq to short-cut to those translocations with an expressed gene moiety. For example, RNAseq analysis identified the MHC II transactivator CIITA as a recurrent fusion partner in lymphoid cancers [27].

Tumor suppressor genes are characterized by loss-of-function genetic events (Fig. 7c, d). Apart from a few examples where dominant negative mutations can be selected for (e.g. TP53), it is usually expected that both copies of a tumor suppressor gene must be inactivated, either through bi-allelic point mutation, homozygous deletion, methylation, or a combination of mutation and LOH. Thus, mapping of copy number loss and LOH has been applied to try and identify new TSGs. While early successes included genes where the initial mutation event was inherited (e.g. RB1 [28]), in the genomic age there have been very few genes identified through this method [29].

Exome sequencing studies have been applied to multiple cancer types to identify both oncogenes and tumor suppressor genes. Initially, only a small number of samples were investigated, with candidates followed up in larger cohorts (e.g., CANgenes [30]). More recently, as sequencing has become relatively cheap, cohorts of hundreds of samples have been analyzed. Interestingly, apart from a few histologically defined tumor types (e.g., granulosa cell tumors) there have been very few genes identified that are mutated at high frequency. It seems that for solid tumors each tissue type has 1–5 commonly mutated genes (>10 % frequency), and a long tail of genes each with a mutation frequency of just a few percent. Thus, the issue of identifying drivers versus passengers is again a problem. One strategy used to enrich for potential driver mutations is the employment of algorithms to predict the deleteriousness of an SNV given the nature of the amino acid change, the position within the protein sequence and the level of conservation of the protein sequence compared to other species [31–33]. Another strategy is the use of statistical methods to assess the mutation rate for a given gene relative to the background mutation rate and gene size [34, 35]. Increasingly, gene discovery studies are applying algorithms to identify common pathways that are affected which can assist in identifying the likely driver genes. For example, pathway analysis identified axonal guidance pathway aberrations in pancreatic cancer, revealing novel tumorigenic roles for these proteins [36].

Association of Genetic Events with Clinical Features

Genetic events are intrinsic to the development of malignant characteristics, thus, it is logical to assume that differences in clinical behavior may be attributed to specific genetic aberrations. Many studies have investigated the association of clinical with genetic features on a genome-wide scale. Associated features may then assist in prediction and risk management, diagnosis, prognosis, or treatment.

Germline Predisposition to Cancer

Many of the well-known cancer predisposition genes , such as APC (familial adenomatous polyposis), BRCA1 and BRCA2 (hereditary breast and ovarian cancer), and MLH1 (hereditary non-polyposis colon cancer), were identified through linkage analysis and candidate gene approaches [37–40]. This was possible because of their relative commonness and high penetrance in these hereditary conditions. Identifying additional candidates is now primarily undertaken through large-scale exome and genome sequencing of multiple members of high-risk families. However, the task of identifying these genes remains difficult since pathogenic mutations are often vanishingly rare, as encountered with RAD51C mutations in BRCA1/2 mutation-negative breast/ovarian cancer families [41–43], and definitive classification as a cancer predisposition requires very large case and control cohorts to achieve sufficient power.

Genome-wide association studies (GWAS ) have identified many more common genomic variants with much lower individual effect on cancer risk. Although the functionally relevant genes may not be identified, the SNPs from these studies can prove useful for risk prediction. Common risk alleles may act in concert to produce a multiplicative polygenic risk or act as risk modifiers [20].

In order to effectively incorporate new risk alleles into the clinic, current practices for genetic testing are undergoing a shift towards gene panels, where all known cancer susceptibility genes and SNPs can be sequenced simultaneously using MPS. This approach substantially decreases the time and cost per gene tested.

Molecular Subtyping and Diagnostics

Most subtyping studies have used expression microarrays to determine classes of tumors with distinct characteristics; however, it is becoming clear that these expression subtypes are often correlated with specific underlying genetic profiles. For example, a number of prognostic tests have been developed for breast cancer subtyping (e.g., OncotypeDX, MammaPrint, and PAM50) and these reflect both histological and genetic differences [44–46]. Targeted sequencing panels are increasingly being used to inform clinical decision-making by matching patients with appropriate conventional therapies or to direct patients to relevant clinical trials. Many biotechnology companies offer companion diagnostic cancer gene panels, enriched for the so-called druggable mutations and those associated with prognostication, including, for example, Illumina® (TruSight® Cancer/Tumor; TruSeq® Cancer Amplicon [47]); Foundation One™ [48]; and Ion Torrent (IonAmpliSeq™ Comprehensive Cancer/Cancer HotSpot [49]).

Prognostic Markers

Treatment of cancer tends to be aggressive, with side effects that can have a severe impact on the quality of life both in the short and long term, including radical surgery leading to scarring and loss/reduction of organ function, radiotherapy-induced burns and increased risk of subsequent malignancy, systemic cytotoxics leading to hair loss, nausea, etc. The consequences of disease progression or recurrence are sufficiently severe that these outcomes are accepted as a necessary evil. However, not all patients are at the same risk of progression, even after controlling for known prognostic factors such as stage, grade, and histological subtype. Genomic analysis has been applied to attempt to identify robust prognostic markers that may indicate that an aggressive treatment regime may not be necessary. For example, the presence of microsatellite instability in colorectal cancer has been shown to be a good prognostic indicator, identifying a proportion of colorectal tumors that do not respond to 5-fluorouracil (5-FU) systemic treatment, the mainstay of colorectal cancer systemic therapy [50, 51]. At the clinical level these data mean that individuals with stage II mismatch-deficient colorectal cancer are unlikely to be treated with systemic 5-FU treatment compared with mismatch repair proficient tumors as the clinical benefits do not outweigh the complications associated with this treatment.

Pharmacogenomics

Response to Conventional Therapies

In a similar manner to identifying prognostic markers of general tumor aggressiveness, studies have also tried to find markers that indicate a likely response to chemotherapies. Such a marker could be constitutional, for example, polymorphisms in cell transporter channels that affect the rate of drug efflux are strong determinants of chemotherapy toxicity and tolerable dosage [52–54]. Alternatively, deleterious germline mutations in BRCA1 and BRCA2 that are cancer predisposing paradoxically tend to improve the patient’s response to treatment due to a heightened susceptibility to the DNA damage caused by chemotherapy. Alternatively, response to therapies could be tumor-intrinsic, for example, CCNE1 gene amplification was determined to be an intrinsic resistance mechanism to platinum-taxol-based chemotherapy in high grade serous ovarian cancer [55].

Targeted Molecular Therapeutics

Recently, targeted molecular therapies have emerged with the potential to transform cancer treatment by personalizing drug regimens to the genetic “Achilles heels” of each tumor. Genome-wide analyses are key to identifying such targets in a research setting, and could be used clinically in the future, especially to identify the cause of therapy resistance. Obvious candidates for targeted therapies are over-active oncogenes as reducing activity is theoretically straightforward. Some prime examples of successful targeted treatments are imatinib (Glivec), which acts as an inhibitor of several tyrosine kinases including the BCR-ABL fusion, trastuzumab (Herceptin), targeted against overexpressed HER2 (first used in HER2+ breast cancers), and PLX 4032 (Vemurafenib), which targets the constitutively active form of BRAF (BRAF V600E) frequently mutated in melanoma. Targeted therapies for gene products where function has been lost tend to rely on unique weaknesses arising as a side-effect of the loss of gene function. For example, deleterious mutations in homologous recombination repair genes BRCA1/BRCA2/PALB2 that impede the efficient repair of DNA double-stranded breaks leave the cells susceptible to both conventional DNA damaging chemotherapies (e.g. cisplatin) and more molecularly targeted poly ADP ribose polymerase (PARP) inhibitors, which affect alternative DNA repair pathways and impede subsequent DNA replications [56].

Despite the breakthroughs in targeted molecular therapies, these almost always induce drug resistance and are often not directly transferable to other tumor types characterized by the same mutation or pathway alteration. For example, attempts to treat BRAF V600E positive colon cancers with the same BRAF inhibitors that had been successful in melanoma resulted in poor clinical response rates due to feedback activation of EGFR in response to BRAF inhibition [57]. However, in this case combination therapy with BRAF inhibitors and EGFR or PI3K inhibitors looks more promising [58].

Cancer cells can become resistant through a range of mechanisms, finding alternative ways to activate pathways or undergoing secondary mutations that reverse susceptibility. For example, initially successful treatment of colon cancer patients with EGFR inhibitors has been found to select for cancer cells with activating KRAS mutations, leading to bypassing of the receptor tyrosine kinase signal and resistance to EGFR inhibition [59]. Secondary mutations in BRCA1 and BRCA2 have been detected in chemotherapy resistant ovarian cancers, that result in restoration of protein function through reestablishment of the reading frame, mutation of deleterious nonsense codons to missense codons, and gene conversion where the mutant allele is lost [60, 61]. Detection of these resistance mechanisms is crucial for patient prognosis and identifying effective treatments for patients to progress.

Future and Near-Term Clinical Applications

New technologies are expediting the identification of cancer driver genes and potential new therapeutic targets, leading genomics to take center stage in diagnosis, prognosis, treatment planning, and the search for new treatment options. The possibility of affordable whole genome sequencing is likely to result in many current clinical tests becoming ancillary and potentially redundant.

With the advent of accessible genome-wide molecular analysis, the molecular subtyping of all cancer types using next-generation DNA and RNA sequencing, and copy number and expression arrays is currently being realized. This offers the possibility of mutation, copy number, and expression profiles superseding histological classification, particularly concerning selection of the most effective treatment options and prediction of recurrence risk.

Whole genome and whole exome sequencing of germline and tumor DNA are likely to become standard practice, both for the identification of predisposing genetic variation and to identify molecular targets for treatment and potential resistance mechanisms. Before these technologies become standard clinical techniques, however, there are the ethical and legal hurdles of incidental findings and patents concerning certain cancer predisposition genes. In the meantime, clinical tests are being converted to these modern technologies with the creation of high-throughput panels of cancer genes.

Advancements in genomics technologies that allow very limiting amounts of DNA to be sequenced are providing future potential for real-time monitoring of treatment response and development of resistance. Isolation of circulating tumor DNA from plasma offers a non-invasive “liquid biopsy” that gives an indication of tumor burden and provides a more representative sampling of the tumor cell population than traditional core biopsies [62]. Highly sensitive monitoring of patients at the molecular level as they progress through treatment and altering treatment based on resistance mutations as they arise could drastically alter the outcome for many cancer patients. These applications are currently under investigation and offer a paradigm shift in cancer screening and treatment in the near future [62].