The early history of cancer genetics

The elucidation of cancer’s genetic etiology and breadth of its diversity has paralleled seminal advances in genetics and genetic technology (Figure 3a). Many landmark studies in the late nineteenth and early twentieth century laid the groundwork for cancer genomics—development of the theory of evolution by natural selection,¹⁰ discovery of heritable biological units,¹¹ identification of chromosomes,¹² chromosomal heredity^{13, 14} and chromosomal genes,¹⁵ and modern evolutionary synthesis linking biology, genetics, and evolutionary theory.¹⁶ The subsequent discovery of DNA’s structure then provided a framework for understanding the genetic code and the mechanism for molecular transmission of genetic information.¹⁷ Meanwhile, reports of heritable breast and colorectal cancers had arisen¹⁸ and cancer was increasingly understood to be able to be caused by chemical agents such as coal tar¹⁹ and cigarette smoke,²⁰ physical agents such as radiation,²¹ and biological agents such as viruses.²² Discovery that these carcinogenic agents were also mutagenic drove speculation that cancer was a disease of mutation. Ultimately, although cancer had been suggested by some to be a chromosomal disease or a disease of mutation, the concept of cancer as a genetic disease did not gain traction until later in the twentieth century when cytogenetic and sequencing technology alongside experimental transformation of normal cell lines with oncogenes enabled assessment of cancer gene mutations at the molecular level and establishment of their causal role in tumorigenesis. Some of these landmark studies are shown in Figure 3a.

Molecular cytogenetics: Identification of recurrent chromosomal alterations in cancer

Mid-twentieth century developments in the field of cytogenetics such as chromosome staining in fixed leukocytes from peripheral blood finally enabled empirical observation of the chromosomal complement of a cell.^{23, 24} Using this approach, Nowell was the first to discover a specific recurrent genetic change associated with a cancer type—the “Philadelphia chromosome” fragment occurring in nearly all evaluated cases of chronic myelogenous leukemia (CML).²⁵ Over a decade later, Rowley then used chromosome banding to map the components of this fragment, determining the Philadelphia chromosome to be a balanced reciprocal translocation between chromosomes 9 and 22.^{26, 27} These discoveries drove a search for recurrent structural aberrations both within and across cancer types. They led to identification of characteristic translocations in sarcomas, leukemias, and lymphomas. However, most carcinomas were found to possess few recurrent alterations, instead displaying great variety in the number and type of chromosome aberrations.²⁸

After Nowell and Rowley’s discoveries, rapid innovation in chromosomal banding and spectral karyotyping (SKY) allowed progressively detailed views of chromosome aberrations in cancer cells using techniques based on fluorescence in situ hybridization (FISH), which relies on fluorescent DNA probes that hybridize to specific chromosomal regions.^29–38 Approaches for genome-wide cytogenetic analysis have also been developed including SKY,²⁹ multiplex-fluorescence in situ hybridization (M-FISH),³⁹ chromosome microdissection,³⁸ and comparative genomic hybridization (CGH).⁴⁰ CGH is a particularly powerful fluorescent molecular cytogenetic technique for screening genome-wide chromosomal copy number changes in tumor genomes.

Novel approaches to printing nucleic acids on slides (microarrays) soon revolutionized molecular cytogenetics by reducing cost and increasing throughput. Array comparative genomic hybridization (aCGH) originally involved spotting large probes (large-insert clones or artificial chromosomes)^{41, 42} onto slides in intervals (1–3 Mb resolution) across chromosomal regions. Now, high-resolution genome-wide platforms consisting of small, customizable oligonucleotide probes^{43, 44} have demonstrated increased throughput and sensitivity in characterizing cancers such as breast,^45–47 melanoma,^48–50 B-cell lymphoma,^{51, 52} and many others. Similarly, oligonucleotide arrays designed to genotype thousands of single nucleotide polymorphisms (SNPs) are useful for characterizing tumor genome complexity with rapid genotyping of over millions of SNPs now possible (Affymetrix, Illumina). Similar to CGH arrays, SNP arrays quantitate locus-specific hybridization signal and can be used to estimate copy number or loss of heterozygosity (LOH).^53–55 Notable cancer studies incorporating SNP arrays include genotyping of the NCI-60 cell line panel⁵⁶ as well as lung cancer,⁵⁷ AML,^{55, 58} neuroblastoma,⁵⁹ melanoma,^{56, 60, 61} basal cell carcinoma,⁵⁴ breast cancer,⁶² colorectal cancer,^62–64 glioblastoma,^{65, 66} and pancreatic cancer.⁶⁷

Both SVs and CNVs can be characterized using the tools of molecular cytogenetics and cytogenomics. Translocations are some of the most common SVs observed in cancer and, as in the case of the Philadelphia Chromosome in CML, are signature aberrations for many liquid cancers and sarcomas. They involve the movement of a chromosomal segment from one position to another, either within the same chromosome (intrachromosomal) or a different chromosome (interchromosomal). Other types of SV include chromosomal deletions, duplications, inversions, insertions, rings, and isochromosomes. These SVs may result from errors in double-stranded break (DSB) repair or other means of intra- or interchromosomal recombination.⁶⁸ Translocations and other SVs can either activate oncogenes (such as when the ABL tyrosine kinase is constitutively activated by fusion with the BCR gene) or inactivate TSGs (such as when translocations involving the TP53 tumor suppressor interrupt the gene and lead to loss of function). CNVs involving a gain or loss of an entire chromosome are some of the simplest and most common chromosomal aberrations. They result from defective chromosomal segregation. The functional consequence of CNVs can be difficult to establish because the aberrations may extend over tens of thousands of megabases and may affect hundreds to thousands of genes. It has been easier to establish the cancer relevance of more limited regions of chromosomal gain and loss, created by amplification or deletion, as these smaller aberrations have been shown to alter the dosage of known oncogenes or TSGs. Classic examples of oncogene amplification in solid tumors discovered through molecular cytogenetics include ERBB2 in breast cancers and MYC in a variety of tumor types, while loss of specific regions of the genome is often associated with loss of TSGs such as TP53, RB1, PTEN, and CDKN2A. Such deletions of TSGs are critical for promotion of tumorigenesis both in cases of heritable cancers in which germline mutation of a single TSG allele must be coupled with loss of the wild-type allele as well as in sporadic cancers in which both alleles must be inactivated.

Molecular genetics: Identification of recurrent sequence alterations in cancer

While some molecular cytogenetic techniques are capable of detecting chromosomal breakpoints at base pair resolution, advances in molecular genetics were required to allow detection of subtle sequence changes in DNA itself. These advances were enabled by developments in molecular cloning and invention of the polymerase chain reaction (PCR) capable of generating the millions of DNA copies necessary for detection of these mutations.⁶⁹ Coupled with the Sanger sequencing method based on dideoxynucleotide chain termination and agarose gel electrophoresis,⁷⁰ progressive advances in this methodology, including fluorescently labeled nucleotides, capillary electrophoresis, paired-end sequencing, shotgun sequencing, and improved laboratory automation have allowed scaling to sequence entire genomes. This technology enabled the Human Genome Project (HGP), launched in 1990 as a $3 billion dollar publicly funded international collaboration. The rough draft of the human genome, covering 94% of the genome, was published in 2001^{71, 72} with the complete draft finished ahead of schedule in 2003.⁷³ In addition to establishing genome-wide ground truth for the locations, organization, and sequences of all human genes, the technical and informatic lessons learned in the course of this project made possible the first ambitious cancer mutation screens on the level of entire gene families^74–77 and the majority of the protein-coding genome.^{62, 65–67} Nonetheless, genome sequencing projects using Sanger-based sequencing continued to be expensive and logistically challenging and thereby limited the application of this approach toward large-scale characterization of cancer genomes until the advent of massively parallel sequencing.

Gene expression microarrays: Cancer signatures and pathways

Microarrays such as those employed in aCGH studies were originally developed for high-throughput gene expression studies. Complementary DNA (cDNA formed from reverse transcription of RNA) and later oligonucleotide microarrays were implemented in high-throughput quantitation of mRNA transcripts. Advances in this technology paralleled improvements in automated Sanger sequencing in the 1990s and the turn of the millennium. They provided some of the first insights into gene signatures and pathways correlating to specific cancer subtypes. A parallel method, serial analysis of gene expression (SAGE) was developed to enable high-throughput digital quantitation of transcripts.⁷⁸ While originally designed to assay the expression of individual genes, gene expression profiling methods were later adapted to profile and map global transcription on the exon and genome levels,⁷⁹ as well as to profile the expression of mature human micro-RNAs (miRNAs).⁸⁰ Gene expression patterns have been widely used to subclassify cancers into homogeneous entities not easily discernible using traditional histopathologic or cytogenetic techniques, for instance, in diffuse large B-cell lymphoma (DLBCL),^{81, 82} lung cancer,⁸³ melanoma,⁸⁴ and breast cancer.^{85, 86} In some instances, subgroups have been shown to represent distinct disease states responding differently to standard therapies. Gene expression profiles have also been widely mined to specifically identify sets of genes predictive of disease progression, response to therapy, or metastasis,^87–89 as well as the presence of specific recurrent cytogenetic abnormalities,^90–92 or gene mutations.^93–96 Although advances in massively parallel sequencing have largely supplanted microarrays in sensitivity and robustness, if not cost and clinical readiness, gene expression profiling continues to play a significant role in cancer biology and medicine. In fact, commercially available clinical tests based on microarray signatures (OncoType DX and MammaPrint) are available for breast cancer profiling.

Massively parallel sequencing: Charting cancer genome landscapes

Explosive growth in our understanding of cancer genomics has been built on the backbone of the advances that emerged from the completion of the human genome at the turn of the twenty-first century. However, although these advances had driven the cost of one genome down from an estimated $1 billion to the $1 million mark at the time of the HGP’s completion, this extraordinary cost nonetheless held back large-scale genomic studies.⁹⁷ Thus, the National Human Genome Research Institute initiated a $70 million sequencing technology program intended to drive the price down to the $1000 mark. The cost of a human genome sequence now lies slightly below $2000 and is steadily approaching $1000 (Figure 3b).

The quest for the $1000 genome required fundamental rethinking of the approach to DNA sequencing. Sanger sequencing, even in its most expert execution, faced insurmountable bottlenecks including (1) the need for molecular cloning steps and/or individual PCR reactions to generate sequencing templates and (2) every sequencing reaction can only accommodate a relatively small genomic region. Therefore, cloning/amplifying and sequencing many regions, even when run in parallel in multiwell plates, requires large amounts of materials, equipment, and analysis time. The paradigm shift in sequencing cost and throughput was brought about by the invention of approaches that allowed generation of sequencing templates without cloning or individual PCR reactions, immobilization of entire sequencing libraries in a single reaction chamber, and massively parallel sequencing of millions of templates in that chamber. The ability of this approach to collect data on clonal populations derived from single DNA molecules additionally provides great power for detection of rare variants from heterogeneous tumor samples. These rapid, accurate, and inexpensive approaches are also referred to as next-generation sequencing (NGS).

Key inventions in the 1990s and early 2000s such as sequencing-by-synthesis, pyrosequencing, colony sequencing, and emulsion PCR enabled development of the first NGS platforms.^98–101 The first such platform was developed by Lynx Therapeutics (later acquired by Solexa) in 2000, although it did not achieve widespread adoption.¹⁰² 454 Life Sciences offered the first commercially available system—the 454 FLX Pyrosequencer—in 2004,¹⁰³ followed by the Solexa (now Illumina) Genome Analyzer in 2006, and the Supported Oligonucleotide Ligation and Detection (SOLiD) System from Applied Biosystems in 2007 that also incorporated polony sequencing.^{104, 105} The Solexa system was used in 2008 to sequence James Watson’s genome within a single laboratory in only 2 months.¹⁰⁶ These platforms at last made the possibility of sequencing and assembling multiple cancer genomes a reality as evidenced by the characterization of genome-wide SVs in two lung cancer genomes.¹⁰⁷ This study utilized a paired-end approach, demonstrating strikingly complex genomic rearrangements including those resulting in previously unreported fusion transcripts. Importantly, this study also demonstrated that DNA copy number can be estimated by the relative local abundance of genomic fragments sequenced, providing sensitivity comparable to aCGH platforms while at the same time providing DNA sequence information. Massively parallel sequencing methods have now dramatically decreased the cost of sequencing while greatly increasing throughput. Presently, sequencing of 500 billion bases (e.g., sequencing the equivalent of 156 human genomes or 156X average genome coverage) within a single 6-day commercial instrument run is standard, with throughput still steadily increasing.

Many massively parallel sequencing platforms exist today and are reviewed in more detail elsewhere.¹⁰⁸ Table 1 provides an overview of some of the most common platforms in use today. They have been adapted to a variety of purposes ranging from diverse DNA sequence, copy number, and structural analyses to transcriptomics and epigenomics. DNA sequencing applications include de novo sequencing, such as the first such application in determining the previously unknown genomic sequence of Acinetobacter baumannii.¹⁰⁹ In addition, resequencing of human genomes is performed in order to characterize normal variation or variation in specific disease states such as cancer. Resequencing is undertaken by mapping NGS sequencing reads to the reference genome and identifying differences between the sample and reference.⁶⁵ Of course, it is not always necessary or informative to assess whole genomes, and thus, a number of target-enrichment strategies have been conceived that enable more cost-effective and focused analyses. The first of these—direct genomic selection—was based on hybridizing defined genomic regions to biotinylated BACs and capturing these sequences on streptavidin beads before sequencing.¹¹⁰ New enrichment methods such as hybridization-based capture (using oligonucleotides either in solution or on microarrays), highly multiplexed PCR, and microdroplet PCR are common.^111–114 A relatively small fraction of the human genome has been functionally characterized, and therefore, it is often most expedient to assess the 1% of the genome containing coding regions (all exons—the exome). Thus, a number of commercially available human exome and other targeted capture kits (such as panels of common cancer genes) are available. This ability to target specific genes to enable rapid, accurate, affordable sequencing is aiding not only in biological discovery but also in clinical laboratory testing by enabling practical complex testing.

WGS, whole genome sequencing; E-S, exome sequencing; RNA-S, RNA sequencing; T-S, targeted sequencing; C-S, ChIP-sequencing; MG, MetaGenomics; and DN-S, de novo sequencing.

^a Additional versions are available for some models.

Landmark cancer genomic studies

Despite commonalities observed within and between individual types of cancer in historic genetic studies, the high-throughput genomic studies of the past decade have provided a much more complex view of cancer. The first large-scale cancer mutation screens implemented high-throughput Sanger sequencing of individual genes such as BRAF in large cohorts^{115, 116} or focused on gene families such as protein kinases in diverse cancer types.^{74–77, 117, 118} Notably, such screens have led to the identification of activating BRAF mutations in a large percentage of human melanomas¹¹⁵ and nevi,¹¹⁹ as well as ERBB2 mutations in human breast cancer.¹¹⁸ The first cancer gene panels (110 exons from 12 target genes) were completed in 140 acute myeloid leukemia (AML) samples in 2003.¹²⁰ This study identified six previously described and, strikingly, seven previously unknown coding mutations in AML genomes. Soon after, results of Sanger-based exome sequencing of 13,023 genes in 11 breast and 11 colorectal cancers were published in 2006.¹²¹ This study revealed a surprising total of 189 frequently mutated genes with an average of 90 mutant genes per tumor, although only a subset was thought to be causally related to tumorigenesis or progression (an average of 11 per tumor). In a follow-up analysis of an additional 7000 genes in the same tumors, it was confirmed that a handful of commonly mutated genes (the “mountains” in the genomic landscape) existed among a large number of less frequently mutated “hills”. In 2008, four Sanger-based exome projects additionally characterized over 20,000 genes in glioblastoma multiforme^{62, 65, 66} and pancreatic cancer.⁶⁷ Amidst an average of 63 and 47 mutations per pancreatic tumor or glioblastoma, these studies identified novel genes bearing low-frequency mutations such as IDH1 and PIK3R1 mutations in glioblastoma. Strikingly, they uncovered few unknown genes recurrently mutated at high rates. In these and future studies, it was found that the high-frequency mountains had already largely been identified using the low-throughput genetic technologies of the past decades. The growing challenge has since been interpretation and characterization of the many hills. However, these studies all show that multiple genes within core signaling pathways, rather than single genes, tend to be mutated in these cancers, suggesting an avenue for treating tumors based on aberrant pathways instead of individual aberrant genes.

Following these landmark studies, the number, breadth, and size of cancer genome projects rapidly accelerated in keeping with the advent of massively parallel sequencing. The first cancer whole genome sequences generated by massively parallel sequencing were those of two lung cancers published in 2008¹⁰⁷ followed shortly by those of individual AML cases published in 2008 and 2009.^{122, 123} These studies focused on assessment of genome-wide SVs and missense mutations, respectively. The first analysis of paired primary and metastatic tumors enabling genome-wide assessment of evolution was conducted in breast cancer in 2009, identifying both shared and exclusive mutations in each sample.¹²⁴ The first studies to systematically assess genome-wide mutations of all classes, published in 2010, were conducted in the melanoma cell line COLO-829¹²⁵ and a small cell lung cancer cell line.¹²⁶ These studies uncovered an extraordinary mutation burden (over 33,000 mutations in melanoma and 23,000 in lung) and mutational signatures supporting ultraviolet (UV) radiation etiology and tobacco smoke etiology in melanoma and lung, respectively. Discovery has since continued at a rapid pace and, as discussed later, large genomic analyses having been completed in all common cancers, many rare cancers, primary/recurrent/metastatic matched cohorts, and even cancers undergoing treatment.

Large-scale cancer genomic studies have also been powered by national and international consortia. The US National Cancer Institute (NCI) and National Human Genome Research Institute (NHGRI) launched a large collaborative project in 2005, the Cancer Genome Atlas (TCGA; see http://cancergenome.nih.gov/). Using multiple state-of-the-art genomic technologies to comprehensively identify all genomic alterations associated with multiple cancer types, the TCGA effort began with large pilot studies of glioblastoma multiforme⁶⁵ ovarian cancer,¹²⁷ and squamous cell lung cancer.¹²⁸ TCGA studies have since included colorectal,¹²⁹ breast,¹³⁰ endometrial,¹³¹ AML,¹³² clear cell renal cell,¹³³ expanded glioblastoma,¹³⁴ urothelial bladder,¹³⁵ lung adenocarcinoma,¹³⁶ gastric,¹³⁷ chromophobe renal cell,¹³⁸ papillary thyroid,¹³⁹ head and neck squamous cell,¹⁴⁰ low-grade glioma,¹⁴¹ cutaneous melanoma,¹⁴² and a suite of 18 pan-cancer studies.¹⁴³ Across 34 cancer types, data have been collected from more than 11,000 cases. Similarly ambitious studies have been undertaken by members of the International Cancer Genome Consortium (ICGC; see http://icgc.org)¹⁴⁴ launched in 2010 to comprehensively catalog all genomic aberrations associated with at least 50 different cancers. This consortium currently encompasses 55 projects covering 33 cancer types with over 13,000 donors and has produced publications on bisulfite sequencing techniques,^{145, 146} DLBCL,¹⁴⁷ Burkitt lymphoma,¹⁴⁸ X chromosome hypermutation,¹⁴⁹ signatures of mutational processes,¹⁵⁰ and primary CNS lymphoma.¹⁵¹ The ultimate success of these comprehensive, large-scale projects will continue to rapidly advance our understanding of cancer genetics and genomics and will potentially revolutionize our approach to the diagnosis and treatment of cancer.

Cancer genomic data repositories and analysis tools

Data generated by the TCGA, ICGC, and other curated genomic repositories are publicly available. Systematic analysis of these data remains challenging, but improved web-based tools for both novices and experienced bioinformaticians are greatly democratizing cancer genomics research (Table 2). One of the oldest and most comprehensive resources for mining such data is the Catalogue of Somatic Mutations in Cancer (COSMIC; see http://www.sanger.ac.uk/genetics/CGP/cosmic) maintained by the Wellcome Trust Sanger Institute Cancer Genome Project.¹⁵² COSMIC is manually curated from publications and therefore entails a broad reach across targeted studies and genome projects (currently more than 20,000). The current build (August 2014) describes over 2 million mutations in more than 1 million samples encompassing 12,000 cancer genomes. In addition to SNVs, it details over 6 million noncoding mutations, 10,000 fusions, 61,000 SVs, 700,000 CNVs, and 60 million expression variants. These data are easily queried using key words or by gene or cancer type. COSMIC also includes more advanced tools that are in some cases linked to other databases. These include a detailed census of all human genes that have been causally linked to tumorigenesis (Cancer Gene Census; see http://www.sanger.ac.uk/genetics/CGP/Census)^{5, 153} as well as tools to assess mutation signatures and drug sensitivity. Another highly versatile data portal is the cBioPortal for Cancer Genomics maintained by Memorial Sloan-Kettering Cancer Institute.¹⁵⁴ This portal primarily contains highly processed Cancer Cell Line Encyclopedia (CCLE) and TCGA data sets but also provides a powerful, intuitive web interface enabling queries by gene or cancer type as well as more advanced analysis through an application programming interface and also integration with R and MATLAB. A similar portal is available for ICGC data (see http://dcc.icgc.org).

While the above-mentioned portals enable advanced analysis with only modest bioinformatics training, additional repositories exist for advanced users to access various levels of raw data. The Cancer Genomics Hub (CGHub), hosted by the University of California Santa Cruz (UCSC), is a secure central repository for data generated through the NCI including TCGA, CCLE, and Therapeutically Applicable Research to Generate Effective Treatments (TARGET).¹⁵⁵ The European Genome-phenome Archive (see http://ega.crg.eu) similarly collects and distributes sequencing and genotyping data, primarily cancer data emerging from the ICGC. Data repositories and analysis tools are also available for other data types. Cytogenetic aberrations and fusions observed in over 65,000 human tumors have been compiled and are maintained online (The Mitelman Database of Chromosome Aberrations in Cancer at the US National Cancer Institute [NCI] Cancer Genome Anatomy Project [CGAP] Web site: http://cgap.nci.nih.gov).²⁸ Data from numerous gene expression microarray studies are warehoused by multiple entities, including the US National Center for Biotechnology Information (NCBI Gene Expression Omnibus; see http://www.ncbi.nlm.nih.gov/geo/) and the European Bioinformatics Institute (EBI Array Express; see http://www.ebi.ac.uk/microarray-as/ae/). Additional data repositories and web resources are outlined in Table 2.

The vast and varied landscapes of human cancers

Although few new mutational “mountains” have been discovered in the past 10 years, systematic study of cancer genomes has revealed many new “hills” as well as mutational processes and dysregulated pathways. The total mutation burden itself can reflect cancer etiology and have bearing on clinical course. This burden is much more variable than originally anticipated. In some pediatric cancers such as rhabdoid tumors¹⁵⁶ and small cell carcinoma of the ovary,¹⁵⁷ cases with only a single coding SNV have been discovered, while leukemias such as AML bear a median of nine coding SNVs (Figure 4). Conversely, cancers with mutagenic etiology such as bladder, lung, and melanoma bear high mutation rates with medians of 148, 217, and 254 coding SNVS, respectively, while tumors with mismatch repair defects can contain thousands of coding SNVs. Distinct mutational signatures can also reflect these external and internal mutagenic factors such as enrichment for C > T transition in dipyrimidines as seen in cancers from UV-damaged sites.¹⁵⁰ Overall, coding SNVs are generally more common in cancers than coding CNVs and SVs, although broad variability in CNV and SV profiles also exists across cancer types. Thirty-seven percent of cancers experience whole-genome doublings with a quarter of most solid tumor genomes containing large-scale chromosomal variations and 10% containing focal CNVs. Over 140 genomic regions have been found to contain recurrent CNVs, only 38 of which contain known cancer genes.^158–160 Most solid tumors also contain dozens of SVs although many of these occur in regions that do not contain genes and are likely passengers. Whole genome sequencing has also recently uncovered “chromothripsis” events, occurring in 2–3% of cancers when errors in chromosome segregation during mitosis lead to shattering of a single or a few chromosomes and massive rearrangement in a single cell generation.^161–163

In addition to mutagenic etiology, variability in mutation burden is the likely result of mutation timing, patient age, and the number of divisions that have occurred in precursor cells with mutations accumulating in dividing cells over time, although a “punctuated evolution” model has also been proposed.^164–166 Ultimately, only a few such mutations can “drive” cancer (driver mutations) insofar as they confer a selective growth advantage on the cells that carry them. Other mutations are “passengers” with no phenotypic effect on cell growth.¹⁶⁷ Several methods have been developed to classify mutations as drivers or passengers based on the pattern and frequency with which they arise across cancers,^{4, 168, 169} but experimental validation is required to confirm the role of these mutations in cancer.

It has been estimated that more than 138 cancer driver genes exist⁴ with 571 genes causally implicated in cancer catalogued in the Cancer Gene Census (see http://cancer.sanger.ac.uk/census).⁵ The top 10 most commonly mutated cancer genes are given in Table 3, while cancer genes associated with the most lethal cancers in the United States are given in Table 4. While this vast array of mutations may seem overwhelmingly complex, particularly from the therapeutic perspective, there is nonetheless some hope. For example, dramatic clinical responses are seen with agents targeting mutant proteins in highly complex cancers such as mutant BRAF and ALK in melanoma and lung cancer.^{194, 195} Further, multiple cancer genes often converge on an individual pathway and may be functionally equivalent. Thus, specific pathways may collectively be subject to genomic aberrations at a frequency far higher than any of their individual gene components. Indeed, most driver genes converge on the hallmark pathways and can ultimately be reduced to those impacting cell differentiation, cell proliferation, cell death, and genomic maintenance.² Thus, the integration of multiple types of genomic data will clearly be necessary to unravel the complexity of somatic cancer genetics.

^a Oncogene and TSG status were determined by the 20/20 rule⁴ using sample data in the COSMIC database (see http://www.sanger.ac.uk/genetics/CGP/cosmic/).

^b Amplifications shown for samples in which average genome ploidy ≤ 2.7 and total gene copy number ≥ 5 or average genome ploidy > 2.7 and total gene copy number ≥ 9. Homozygous deletions shown for samples in which the average genome ploidy ≤ 2.7 and total gene copy number = 0 or average genome ploidy > 2.7 and total copy number < (average genome ploidy − 2.7).

^a The seven cancers estimated to cause the greatest number of deaths in the United States in 2015.¹⁹³

While it seems possible that the majority of cancer genes have now been discovered, new genes are occasionally implicated in cancer, although at low frequency across many cancer types or at high frequency in a rare and previously uncharacterized cancer. However, most cancers require 5–8 driver mutations based on epidemiological studies¹⁹⁶ and only 3–6 drivers have been found in most cancers with notable exceptions (such as pediatric cases) in which three or fewer drivers have been discovered.⁴ Remaining knowledge gaps are likely due to both technical limitations of sequencing technology and study design as well as limitations in our growing understanding of mutations occurring in noncoding regions of DNA or through epigenomic mechanisms. Clearly, the catalogue of cancer-driving mutations is still incomplete and although the maps of cancer genome landscapes have been broadly charted, much work remains to characterize these vast landscapes in detail in order to guide clinical cancer management.

Clinical implications of cancer genome landscapes

Exhaustive information now available from cancer genomics projects has greatly improved our comprehension of the development, progression, and clinical behavior of human neoplasms. These data and the technology used to generate them are also impacting cancer screening, diagnostics, and treatment. As sequencing costs continue to decline, exome- and genome-wide population screening becomes increasingly viable and it is likely that each patient’s personal genome sequence will comprise a key component of their medical record. Such screening may reveal de novo mutations that would otherwise go undiagnosed until disease presentation and the power of this approach will grow exponentially as we continue to connect genetic variants to disease phenotypes. The NIH has acknowledged the importance of this new healthcare approach through the establishment of the Precision Medicine Initiative.¹⁹⁷ Targeted Sanger sequencing of small gene panels is already commonly used in familial cancer screening. In cases such as germline BRCA1/2 or MLH1/MSH2 mutations, which dramatically increase lifetime risk for breast/ovarian or colorectal cancer, genomic testing can guide clinical decisions about surveillance and prophylactic surgery. Now, NGS-based sequencing of hundreds of genes can be completed at a similar cost and thereby enable detection of lower penetrance rare variants not detectable with smaller panels. For example, in a study of 141 BRCA1/2-negative patients with a family or personal history of breast cancer, a panel of 40 additional genes identified 16 patients with pathogenic variants in nine genes other than BRCA1/2 such as TP53 and PTEN.¹⁹⁸ These findings have shaped new guidelines from the National Comprehensive Cancer Network recommending targeted panels for patients’ breast or ovarian cancer family history who are negative for common hereditary mutations.¹⁹⁹ Targeted NGS panels have also shown greater sensitivity and specificity than traditional diagnostic tools in genetic testing of familial cancer genes such as BRCA1/2, TP53, and APC in additional studies.^{170, 200–202} NGS panels may also facilitate differential diagnosis and patient stratification given that the diagnostic yield of NGS panels, exomes, or even whole genomes has been shown in multiple studies to outperform small Sanger-based panels.^{203, 204}

Cancer genetics discoveries have driven targeted drug development since the first report of a recurrent chromosomal abnormality—the BCR-ABL translocation in CML. The selective tyrosine kinase inhibitor imatinib, designed to target the BCR-ABL chimeric fusion gene and its constitutively active tyrosine kinase protein product in CML, was the first successful targeted therapy.^205–207 The paradigm of targeting cancer-specific mutations to vastly improve the therapeutic index of chemotherapy has now been repeated with dramatic success. Examples include expansion of imatinib to PDGFR– and KIT-mutated gastrointestinal stromal tumors (GIST) and hypereosinophilic syndromes^208–210; dasatinib and nilotinib in primary and imatinib-resistant CML^211–213; trastuzumab, a neutralizing antibody targeted to the Her2/ErbB2 tyrosine kinase receptor whose encoding gene ERBB2 is amplified and overexpressed in 25–30% of breast carcinomas^{214, 215}; sunitinib in renal cell carcinoma, GIST, and pancreatic neuroendocrine tumors^216–218; gefitinib and erlotinib with striking efficacy in the 5–10% of lung adenocarcinoma patients with European ancestry and 25–30% of Japanese patients who harbor activating mutations in EGFR^{219, 220}; crizotinib in ALK-rearranged lung cancer^{221, 222}; vismodegib in basal-cell carcinoma bearing hedgehog pathway mutations²²³; and vemurafenib in BRAF-mutant melanoma.²²⁴ Cancer gene mutations have also been associated with innate or acquired resistance to these targeted therapies. For example, amplification of the BCR-ABL gene is common in CML patients resistant to imatinib.^{205, 225} Preclinical studies and clinical trials are ongoing for numerous novel therapeutic agents directed against genomic targets in cancer in addition to agents capable of circumventing or treating drug-resistant disease.

The fundamental goal of oncology practice is recommendation of the most effective treatment supported by scientific evidence for an individual patient. Genomic data would be expected to provide just such data. However, although the HGP was completed in 2003, genomic medicine’s entrance into the care stream has been painstakingly slow. In 2010, results from one of the first pilot studies using personalized molecular profiling to guide treatment selection in refractory metastatic cancer were published. Despite considerable challenges including the absence of a precedent for the novel trial design, overall patient attrition, and a diverse pharmacopeia, it was found that in 27% of 68 patients, treatment selection guided by molecular profiling (gene expression microarray analysis) resulted in a longer progression-free survival than that during the most recent regimen on which the patient had progressed.²²⁶ The next major step in bringing genomic tools to the patient was a novel study combining whole genome sequencing and comprehensive RNA sequencing to individualize treatment of metastatic triple-negative breast cancer (TNBC) patients.²²⁷ TNBC is characterized by the absence of expression of estrogen receptor (ER), progesterone receptor (PR), and HER-2. This study identified somatic mutations in the RAS/RAF/MEK and PI3K/AKT/mTOR signaling pathways that have led to clinical trials combining agents targeting MEK and mTOR genes with encouraging results.

New trial designs, such as those above and including adaptive, basket, and umbrella trials, are aimed at extending individualized therapy to all cancer patients. Multiple large academic observational and interventional trials have now appeared in publication^228–233 or are ongoing (see http://clinicaltrials.gov). These include adaptive trials such as I-SPY 2 (Investigation of Serial Studies to Predict your therapeutic response with imaging and molecular analysis) in which treatment arms are modified in the course of the study based on patient response.²³⁴ Basket trials such as the NCI MATCH trial assign patients with various advanced cancers to treatment baskets based on mutational profiles, then treat with agents matched to their tumor’s mutations.²³¹ Umbrella trials are also ongoing in which multiple drugs are studied in a single disease. For example, the ongoing Stand Up To Cancer and the Melanoma Research Alliance Dream Team Clinical trial is assessing molecularly guided therapy in a single-histology patient population that lacks a universal therapeutic target—non-V600 BRAF metastatic melanoma. In a nontreatment pilot study (now a randomized treatment study), a combination of whole genome and whole transcriptome sequencing was used to identify molecular aberrations matched to an appropriate clinical treatment from a pharmacopeia comprised standard of care, FDA-approved and investigational agents.²³⁰ The challenge of returning actionable information during a clinically relevant time period is considerable. This includes patient consent, tumor biopsy, quality DNA/RNA extraction, DNA- and RNA sequencing, data integration, report generation, and tumor board review to formulate a treatment plan (Figure 5). Streamlining this process will be critical to expansion of these tools beyond major academic research centers.

Large-scale genomic characterization of tumors in the context of clinical trials has allowed identification of predictive biomarkers—recurrent genomic alterations that identify patients most likely to benefit from a particular drug. These provide tremendous clinical value by allowing patient stratification to the most relevant treatment option and have been rapidly commercialized. A broadening array of sequencing tests that may aid in diagnosis and treatment decisions is available to oncologists and their patients. Over 100 academic centers and 50 commercial laboratories have made such tests available. Examples of such commercial tests are provided in Table 5. Although cost is still prohibitive for widespread use, expanded panel-, exome-, or whole genome-based sequencing will continue to transform screening and diagnostics in coming years. However, routine clinical use will not only require further cost reduction but also more comprehensive data supporting genomics-correlated clinical outcomes to illustrate the benefit of such testing and off-label drug use. Accordingly, commercial sequencing laboratories will need to acquire FDA and Medicare approval. These data will ensure that insurance companies will cover these tests. Promising signs already include the addition of procedural terminology codes and a preliminary list of 21 sequencing tests included in the fee schedule of the Centers for Medicare and Medicaid Services. Finally, germline genomic analysis (whether part of hereditary cancer testing or conducted in tumor-matched normal tissue) entails the high likelihood of incidental secondary findings of putatively pathogenic genomic variants with unclear disease association and therefore presents a challenge. Widespread controversy exists over the balance between patient autonomy and the perception that patients knowledge of this information would result in physical or mental harm.²³⁵

^a These descriptions are derived from company web sites and may not be complete. They also do not constitute endorsement of any particular product.

Unlike relative differences in mRNA or protein expression between tumor and normal tissue, somatic mutations in cancer genes are exclusive and specific markers of cancer cells. This fact has been recently leveraged to use cell-free tumor-specific DNA in plasma (ctDNA) as an accurate circulating biomarker of tumor burden. The value of ctDNA for monitoring tumor burden, treatment response, and recurrence was first recognized in colorectal cancer.²³⁶ This study identified somatic mutations in patient tumors in recurrently mutated colorectal cancer genes (TP53, PIK3CA, APC, and KRAS) and retrospectively analyzed plasma samples using highly sensitive assays specifically design for each patient. The results showed ctDNA levels reflect changes in tumor burden during treatment when compared with imaging and carcinoembryonic antigen. The challenges in implementation of ctDNA as a routine biomarker for tumor monitoring include the need to identify somatic alterations in each patient, design of patient-specific molecular assays, and low abundance of mutations in circulation, particularly in early stages of disease. Another recent study used NGS of tumors and digital PCR and found ctDNA levels during treatment reflected disease progression in metastatic breast cancer.²³⁷ In these results, ctDNA was found to be more responsive and applicable to the largest fraction of patients when compared with enumeration of circulating tumor cells or CA-125, a glycoprotein biomarker of breast cancer. Similar results supporting the role of serial analysis of ctDNA for monitoring tumor burden have been published for lung cancer,^{238, 239} melanoma,²⁴⁰ and osteosarcoma.²⁴¹

Advances in molecular methods including the use of high-depth noise-corrected targeted NGS assays are enabling the study of patients with localized, potentially resectable cancers.^{242, 243} Recent work in localized breast cancer shows ctDNA can predict postoperative recurrence of cancer a median of 8 months before tumors become detectable on imaging.²⁴⁴ This study and additional papers describing ctDNA in localized cancers demonstrate novel opportunities for optimizing treatment of cancer with curative intent by individualizing therapeutic strategies.²⁴⁵

Prospective trials are needed to establish benefit of guiding clinical cancer management using ctDNA as a biomarker. Nevertheless, observational studies reported so far show superior performance of ctDNA to monitor tumor burden compared to a number of circulating tumor cell or glycoprotein biomarkers.²⁴⁶ With some variations between cancer types, ctDNA has wide applicability across patients as shown in a recent comprehensive survey of 640 patients.²⁴⁶ The same study found ctDNA was detectable in 55% of patients with potentially curable disease. These results suggest that improved molecular assays may enable detection of ctDNA in presymptomatic individuals with cancer, potentially leading to a screening test with higher specificity than traditional methods.^247–250

Background

Malignant melanoma is the sixth most common cancer in the United States and is one of few that is increasing in incidence.²⁶⁶ Although detection and surgical resection of early-stage disease can be curative, a tumor thickness of merely 4 mm portends metastatic disease and an abysmal prognosis. Dramatic recent progress in melanoma precision medicine powerfully illustrates the potential for genomics to impact clinical cancer management. In this section, we will focus on this progress and its impact on melanoma treatment while clinical features will be discussed in detail in Chapter 111. Melanoma is a disease of transformed melanocytes and is clinically subdivided by anatomic location, histopathology, sun exposure, and progression (using tumor-node-metastasis staging).²⁶⁷ By site, cutaneous melanoma most commonly originates on sun-exposed skin and is further categorized based on type of sun exposure of the primary site including areas with chronic sun exposure (e.g., face and forearms) and those with intermittent sun exposure (e.g., back). Less common subtypes include acral melanoma, occurring on non-hair-bearing areas of skin on hands and feet; ocular melanoma, which is predominantly uveal with a smaller proportion of conjunctival melanomas; mucosal (nasopharyngeal, bowel, anorectal, or vulvovaginal) and primary CNS melanoma. This categorization has implications for prognosis and treatment, although it is sometimes unable to predict outcome or treatment response. Further, until the discoveries of the past decade, metastatic melanoma was uniformly treated with cytotoxic chemotherapy to limited effect.²⁶⁸

The genetic basis of melanoma

Where variation in clinical course by melanoma histopathologic subtype is confounding, genomic characterization is now helping to clarify etiology, biology, and optimal treatment. Just as breast cancer can best be characterized according to ER, PR, and Her2 status, genotype should now take a prominent role in the clinical approach to melanoma owing to its implications for treatment. Although melanomagenesis is still poorly understood, melanoma is known to occur through the stepwise accumulation in melanocytes of mutations in cancer genes (Figure 7).²⁶⁹ Most melanomas are sporadic with 5–10% of cases due to familial predisposition predominantly driven by germline variants in the first familial melanoma gene identified—the tumor suppressor CDKN2A (40% of familial cases).^270–272 Additional rare variants in other cancer genes have since been found in familial cases (CDK4, BRCA2, BAP1, TERT promoter, MITF, and POT1).^273–278 Beyond heredity, UV radiation is recognized as the greatest risk factor for cutaneous melanomas, although not for sun-shielded mucosal, acral, or ocular disease.²⁷⁹ In keeping with UV etiology, additional melanoma risk factors include fair skin, increased freckling and benign nevi, MC1R germline variants, and tanning.^280–283

Early genetic and functional characterization of melanoma determined that it is driven both by mutations that inactivate cell cycle gatekeepers and by those that activate cell proliferation pathways. Building on discovery of familial mutations in CDKN2A, a locus that regulates the p53 and RB proteins, these tumor suppressors were found to be frequently disrupted by disparate mutations that enable unconstrained cell growth such as CDKN2A, CDK4, RB1, TP53, and MDM2 mutations.^284–288 Meanwhile frequent mutations in the proliferation pathway genes BRAF, KIT, NRAS, and PTEN were identified and found to cooperate with inactivation of cell cycle tumor suppressors to promote malignancy.^{115, 119, 289–302} Cytogenetic and molecular studies further pointed to as-yet unidentified genes in recurrently altered regions of chromosomes 1p, 6q, 7p, and 11q.^{288, 303–311} As with many early genetic analyses of human cancers, these studies were typically performed in disparate small cohorts or model systems, focused on only one or several genes in a single clinical subtype, and often produced inconclusive results.

Melanoma genomic landscapes

The striking genetic complexity between subtypes and within individual tumor genomes revealed by the above analyses was soon confirmed at high resolution through a succession of genomic studies. Genome-wide aCGH profiling and targeted sequencing of BRAF and NRAS in 126 tumors first revealed distinct patterns of somatic mutation based on subtype. BRAF and NRAS mutations were found to be enriched in intermittently sun-exposed and cutaneous melanoma but rare in chronically sun-exposed, mucosal, and acral melanoma. Patterns of mutation also varied by subtype with CNVs enriched in mucosal melanoma. Subsequent analyses identified patterns of CNV correlating with poor prognosis^{49, 289} as well as subtype-specific KIT mutations occurring in 10–20% of mucosal and acral as well as at low frequency in cutaneous melanoma in sun-damaged sites.^{48, 312} The existence of molecular subsets of disease was also borne out by gene expression microarray studies^{84, 313, 314} and such molecular subsetting of clinical melanoma categories seeded hope that these types could be tied to clinical outcome or exploited for therapeutic targeting.³¹⁵

A spate of massively parallel sequencing studies beginning in 2011 spurred explosive growth in the genomic understanding of cutaneous melanoma, recently culminating in two large multiplatform studies that jointly span over 500 cases.^{142, 275, 315–325} These studies not only corroborate prior identification of important melanoma drivers (BRAF, NRAS, NF1, TP53, CDKN2A, and RB1) but also point to new genes not previously linked to the disease including RAC1, PREX2, PPP6C, ARID2, TACC1, GRM3, MAP3K4, MAP3K9, IDH1, MRPS31, RPS27, and the TERT promoter. Melanoma is now predominantly categorized according to BRAF, RAS, and NF1 status, all of which, when mutated, constitutively activate the mitogen-activated protein kinase (MAPK) pathway, a central regulator of cell proliferation (Figure 8).¹⁴² These mutations tend to be mutually exclusive and account for all but 10% of cases, suggesting four genomic subtypes: BRAF, NRAS, NF1, and triple wild type (TWT). Genome-wide sequencing has also enabled characterization of mutation signatures in melanoma, shedding further light on the role of UV radiation in various subtypes. Melanomas on sun-exposed skin are characterized by UV signature mutations—cytosine to thymine (C > T) substitutions occurring in dipyrimidines as more than 60% of the total mutation burden.^{142, 318, 326} Curiously, most common mutations in BRAF and NRAS are not C > T substitutions despite their driver roles in the majority of sun-exposed melanomas.³¹⁷ While it is often noted that melanoma has the highest average mutational burden of any cancer type, the mutation frequency can range from 0.1–100 mutations per megabase.¹⁶⁹ This marked heterogeneity tracks with the clinical heterogeneity of the disease, with the highest mutational rate in melanoma on sun-exposed skin at >100/Mb, intermediate rates in mucosal melanoma at 2–5/Mb, and the lowest rates in uveal and CNS melanoma at <0.1/Mb.^{316, 327, 328}

Cutaneous BRAF melanomas account for over 50% of tumors and predominantly bear the hotspot missense mutation V600E, although other activating BRAF mutations, amplifications, and even fusions do occur. They are characterized by incidence in younger patients, UV signature in 91%, relatively fewer CNVs compared to TWT, amplifications of BRAF, MITF, PD-1, and PD-L1, enrichment for TP53 mutations, and more frequent PTEN deletions. RAS melanomas are the second most common (30%), typically NRAS Q61R/K/L and with occasional mutations in K– or H-RAS. These mutations are nearly always mutually exclusive with BRAF mutations although molecular features of this category otherwise resemble the BRAF subtype. NF1 melanomas comprise 15% of cases, the majority of which contain inactivating NF1 mutation. This subtype bears the highest SNV burden, occurs in older patients, and contains a UV signature. It can co-occur with BRAF and NRAS mutations and tends to show a pattern of co-occurrence with other mutated “RASopathy” genes such as RASA2.³¹⁹ Similar to BRAF and RAS mutations, most NF1 mutations drive RAS and MAPK pathway activation. Finally, TWT melanoma is a heterogeneous subtype that lacks recurrent driver mutations although KIT, CTNNB1, GNAQ, GNA11, and EZH2 mutations have been found at low frequency. Notable characteristics of this subtype include higher numbers of SVs, fusions and CNVs, only 30% containing UV signatures, amplifications containing KIT, PDGFRA, KDR, CDK4, CCND1, MDM2, and TERT, and only few (7%) with TERT promoter mutations.¹⁴² It is important to highlight that despite diversity across genomic subtypes, the vast majority of melanoma driver mutations converge on the MAPK and PI3K pathways, such that approximately 91% of cutaneous melanomas depend on aberrant activation of this single pathway.³²⁹

Despite exhaustive characterization of cutaneous melanoma, genomic landscapes of less-common clinical subtypes remain poorly characterized. Several small studies of uveal, acral, desmoplastic, and mucosal melanoma have identified recurrent mutations, but much still remains unknown. Uveal melanoma, the most common ocular tumor, has been shown in studies encompassing nearly 80 cases to have one of the lowest melanoma mutation burdens and to be characterized by mutations in GNAQ (50%) as well as frequent BAP1, EIF1AX, and SF3B1 mutations in tumors with high metastatic propensity.^{324, 325, 330–332} Acral melanoma occurs on non-hair-bearing skin of the palms, soles, and nail beds, is the most common melanoma subtype in non-Caucasians, and has a poorer prognosis than cutaneous melanoma on hair-bearing skin owing to later diagnosis.³³³ Although known for nearly a decade to be largely BRAF wild type, it is rarely included in larger cutaneous genomics studies and limited focused data sets are available. In total, 38 acral melanomas have been characterized across six studies that have identified the following characteristics: about 30% KIT-mutant, 20% BRAF-mutant, 10% NRAS-mutant, low SNV burden, high CNV and SV burden, and absence of UV signature.^{49, 316–318, 332, 334, 335} Data on mucosal melanoma, another rare and aggressive subtype that is predominantly BRAF wild type, is more sparse with only 10 cases comprehensively assessed in a single study.³²⁷ This study largely confirmed prior targeted and cytogenomic analyses, showing a low SNV frequency, high CNV and SV burden, absence of UV signature, and recurrent mutations in KIT, PTEN, and other putative cancer genes in two or fewer samples. Desmoplastic melanoma, a rare and invasive fibrous form of dermal melanoma has also recently been characterized (62 cases) and preliminary results reveal a notably high SNV burden (one of the highest to date at 62 mutations/Mb), UV signatures, absence of BRAF or NRAS mutations, and a diverse array of mutations that activate the MAPK and PI3K pathways.³³⁶ Notable similarities and differences between these rare tumor types support their potentially distinct etiology and may help guide therapeutic management.

Clinical implications of melanoma genomics

Although no definitive correlation with outcome has been identified based on the above-mentioned genomic subtyping or other molecular classifications in melanoma, these genomic subtypes have direct bearing on treatment of advanced metastatic melanoma (Table 6). The BRAF subtype carries the greatest array of clinical options. Mutant BRAF has now proven to be an effective therapeutic target, with BRAFV600E/K and K601 targetable by inhibitors of BRAF or MEK. As of 2015, three such small molecule inhibitors have demonstrated an overall survival (OS) benefit in randomized clinical trials. Vemurafenib and dabrafenib are BRAF inhibitors (BRAFi) with demonstrated response rates greater than 90% in BRAF melanoma.^{224, 337} Trametinib is a MEK inhibitor (MEKi) which, while less dramatically effective than BRAFi, nevertheless triples progression-free survival (PFS) versus chemotherapy.³³⁸ The best results to date from trials of targeted agents in BRAF melanoma have come from combined BRAFi and MEKi with dabrafenib and trametinib, a combination with marked improvement in PFS and OS with an overall reduction in toxicity.^{339, 340}

^a Phase III studies for immunotherapy have not included genotyping beyond BRAF. FDA approval for these does not specify genotype and thus is inclusive of all genotypes.

^b No trials to date have specifically identified TWT patients, although KIT and GNAQ/GNA11 cases do occur in this category.

Development of targeted agents for RAS, NF1, TWT, and rare histological melanoma subtypes has lagged behind that for BRAF melanoma, with no completed phase III studies. Nevertheless, early phase studies have shown sufficient efficacy to encourage ongoing efforts at developing agents for these patients. RAS melanoma has been treated with a variety of MEK inhibitors in clinical trials including trametinib,³⁴¹ MEK162,³⁴² and selumetinib.³⁴³ Of these, only MEK162 has shown a significant response rate (around 20%), with the others showing a best response of stable disease. No trials have been conducted specifically in NF1 melanoma, in part due to the emerging understanding of this subtype and overlap with RAF and RAS subtypes. Conflicting preclinical reports have suggested NF1 mutation and loss confers sensitivity or resistance to MEKi and that this relationship is also dependent on RAF/RAS status.^{319, 344–346} KIT-mutant melanoma (which is common in TWT cutaneous as well as acral and mucosal subtypes) has been shown in two randomized phase II studies to have an overall response rate to imatinib of 19–29%, with the best responses in cases with canonical KIT SNVs rather than amplifications.^{286, 347} The most robust data in GNAQ/GNA11-mutant disease comes from a randomized phase II study with the MEKi selumetinib.³⁴⁸ Selumetinib induced tumor regression in 49% of patients and led to a nonstatistically significant improvement in OS to 11.8 versus 9.1 months. Additional agents in development that are relevant to many subtypes and/or may motivate additional subtyping include CDK, MDM2, and PI3K/Akt/mTOR inhibitors in clinical trials in addition to ERK, IDH1, EZH2, and aurora kinase inhibitors in preclinical development.¹⁴²

Little is known about the impact of genomics on immunotherapy. Immunotherapy with the CTLA-4 inhibitor ipilimumab and the PD1 inhibitors pembrolizumab and nivolumab have revolutionized melanoma care with long-term survival even in advanced disease, but the role of genomics as predictive or prognostic for immunotherapy remains to be fully elucidated.^349–351 While there is interest in mutational burden as a predictor of response to immunotherapy based on the hypothesis that immunogenicity is a probabilistic aggregate that increases in proportion to total mutation burden, to date no prospective data bears this out. While the hypothesis is supported by the fact that PD1 inhibitors were first approved in melanoma and then lung cancer (the two cancers with the highest mutational burdens), renal cell carcinoma has a very low mutational burden but has demonstrated responsiveness to PD1 inhibitors in phase II studies. Only a minority of patients on the registration studies for pembrolizumab and nivolumab had noncutaneous melanoma and ocular melanomas were excluded. While 20–35% of patients had BRAF V600 mutations, genotyping was not routinely done so that the rate of NRAS mutant and TWT patients is unknown.

Drug resistance and evolution in melanoma

Despite the activity of targeted agents in BRAF melanoma, drug resistance (both innate and acquired) remains a key challenge. Only about 50% of patients respond to BRAFi and in those who do respond, acquired resistance to BRAFi invariably develops.²²⁴ BRAFi resistance is driven by a diverse range of mechanisms, most of which reactivate MAPK activity or redirect proliferative signaling through the PI3K pathway. Resistance can occur through further mutation in NRAS,³⁵² MEK1,^{353, 354} MEK2,³⁵⁴ NF1,³⁴⁴ or BRAF (via amplification).³⁵⁵ BRAF splice variants,³⁵⁶ EGFR activation,^{357, 358} and COT activation³⁵⁹ have also been implicated. Activation of other receptor tyrosine kinases such as PDGFR-β and IGF-1R or loss of PTEN has also been demonstrated in laboratory models.^{354, 360–364} Clinical trials are ongoing to test whether identification of these resistance mechanisms in the clinical setting will allow for targeted therapeutic approaches that may circumvent resistance to BRAFi. Hypothetical approaches to circumvent specific resistance mechanism after progression on dual inhibition with BRAFi and MEKi are given in Table 7.

With the recent approval of new targeted and immunologic therapies for metastatic disease, it is now critically important that the clinician clearly identify the genetic signature of the patient’s disease in order to optimize outcomes. Similarly, genomic testing can accelerate new drug development and target identification for the less common subtypes of melanoma and potentially lead to identification of resistance pathways to guide drug selection in individual patients. Thus, the genomic landscape of melanoma conforms to the clinical heterogeneity of the disease, with melanomas of different locations having markedly different genotypic signatures that speak to distinct etiologies. The challenge, then, is to use available genomic technologies in a real time clinical setting. In theory, one could identify the evolutionary mechanisms active in the individual patient to guide treatment selection.

In addition to molecular diversity across melanoma subtypes and even between individual patients with the same disease subtype, a higher order of complexity is becoming apparent. As outlined in the previous section on cancer evolution, recent data indicates that development of melanoma cannot be fully explained by a linear model of accumulating mutations leading to an increasingly aggressive phenotype. Instead, an emerging picture points to cancer as a complex ecosystem driven by the process of branched clonal evolution. As a result, multiple related, but distinct clonal lineages may exist within a single patient,²⁵¹ with some lineages being more likely to progress and metastasize.²⁵² Furthermore, this diversity may provide an escape mechanism leading to therapeutic resistance. This is well illustrated by almost universal development of resistance to BRAF inhibition in melanoma, where rapidly relapsing disease can harbor multiple resistance mechanisms differing between individual subclones even within a single patient.³⁶⁵ This challenge is potentiated by the fact that, while some mechanisms of resistance are driven by mutations, others appear to be mediated through epigenetic or posttranslational processes.³⁶⁶ The obvious and critical question raised by these observations is whether a single tumor biopsy in a metastatic setting can capture the full spectrum of disease heterogeneity, its drivers, and potential therapeutic targets. The solution to this conundrum may lie in approaches with the potential to monitor the clonal complexity on a larger scale, such as “liquid biopsies” that attempt to molecularly characterize circulating tumor DNA in the patient’s bloodstream. Emerging data describing detection of ctDNA and concordance of BRAF mutation status between tumor and plasma samples is encouraging.^{246, 367, 368} In addition, anecdotal evidence suggests that serial ctDNA analysis may be a useful marker of treatment response to immunotherapy.²⁴⁰

Summary

Detailed characterization of the genomic lesions underlying cancer has led to identification of biological pathways driving tumorigenesis, improved cancer diagnosis through molecular classification, enhanced selection of therapeutic targets for drug development, promoted the development of faster and more efficient clinical trials using targeted agents and sophisticated biomarkers, and created markers for early detection and recurrence surveillance. Yet, gaps nonetheless remain in our knowledge of the causative mutations underlying some cancers, in our understanding of the biology of the mutations we have identified, in developing drugs capable of targeting many of these diverse mutations, and in fending off the inevitable emergence of drug resistance. As summarized in this chapter, genomics has already made extraordinary contributions to our understanding of cancer biology and cancer medicine. Nevertheless, the enormous potential of cancer genomics has only just begun to be realized.

---

Only gold members can continue reading. Log In or Register to continue

Share this:

• Click to share on Twitter (Opens in new window)
• Click to share on Facebook (Opens in new window)

Related

Related posts:

Aberrant signaling pathways in cancer Disparities in cancer care Monoclonal serotherapy Neoplasms in acquired immunodeficiency syndrome Chemical carcinogenesis Infections in patients with cancer

Stay updated, free articles. Join our Telegram channel

Join