Integration of cancer genomics with treatment selection: from the genome to predictive biomarkers

Abstract

The field of cancer genomics is rapidly advancing as new technology provides detailed genetic and epigenetic profiling of human cancers. The amount of new data available describing the genetic make-up of tumors is paralleled by rapid advances in drug discovery and molecular therapy currently under investigation to treat these diseases. This review summarizes the challenges and approaches associated with the integration of genomic data into the development of new biomarkers in the management of cancer.

Introduction

We are currently in a new era of both cancer research and cancer therapy. It is truly remarkable to compare cancer treatment a half-century ago with the scientific and therapeutic innovations under consideration today. In 1953, the crystal structure of DNA was first described and postulated to be the carrier of heritable information¹. Only fifty years later in 2003, the Human Genome Project, a 13-year research endeavor, reported the base-pair sequence encoding an entire human genome. Today, technology is available that can interrogate the entire human genome in a matter of weeks, for orders of magnitude less cost. High throughput techniques can also detail complimentary epigenetic events such as gene expression, micro-RNA expression, and DNA methylation, providing hundreds-of-thousands of data points that can be used to evaluate the inner-workings of human cells and tissues.

Advances in the science of genomics have been paralleled by an increasing understanding of human cancer. In the last century, it has become clear that cancer is a disease of the genome. A growing understanding of the fundamental molecular biology of cancer has led to our current model in which genetic mutation leads to hallmarks of cellular dysregulation that allow cells to become cancerous². Improved insight into the role of genomic alterations in cellular transformation has led to the development of several new classes of drugs which target the molecular drivers that have been discovered in human cancers. The progress made in this new frontier of cancer research rests heavily on the ability of translational scientists to integrate new technology and increasing data with the vast array of new therapeutics available. The following article reviews these challenges and presents several strategies for improving the development of prognostic and predictive biomarkers.

Deciphering the cancer genome with modern technology

Over the past decade, several high-throughput techniques have been developed that allow rapid and comprehensive assessment of the cancer genome and epigenetic events. The following section summarizes several of these techniques and provides contemporary examples of how these technologies have been recently applied to cancer research.

Array-CGH

Array-based comparative genomic hybridization (array-CGH) allows evaluation of copy number variations (CNVs), such as microdeletions, unbalanced translocations, and amplifications across the whole genome in a high-throughput manner³. With this technique, a test DNA sample (usually from a tumor) and a reference sample (often from adjacent normal tissue or from blood cells) are digested and differentially labeled with fluorophores (usually red versus green), which are hybridized to an array containing thousands of probes corresponding to regions of the human genome⁴. The intrinsic resolution of the assay is based on the number and size of DNA probes on the arrays. Advances in array-CGH technology, such as representational oligonucleotide microarray (ROMA) analysis⁵, have sought to increase the resolution of the genomic regions assessed. Newer systems employing very small and overlapping probes allow CGH-microarray resolution down to a range of 10 kilobases to 200 base-pairs⁶. Analysis of global CNV’s can be used to classify tumors based on spectrum or number of CNV’s observed, or individual CNV’s can be explored. Identified unbalanced translocations or amplifications may yield targetable oncogenic events⁷. For example, a recent report by Morris and colleagues evaluated copy number variations using array-CGH in head and neck squamous cell carcinoma, which lead to their identification of common amplifications in PI3KCA, as well as novel deletion events in the PTPRS gene⁸. Several agents targeting the PI3K pathway exist⁹, and the authors demonstrate that PTPRS loss can diminish the efficacy of EGFR inhibition. Copy number variations appear to be a driving force in the promotion of carcinogenesis, and identification of these changes with Array-CGH can inform drug selection for targeted molecular therapy.

SNP Arrays

The human genome carries approximately 20 million conserved single nucleotide variations that occur with a defined regularity in the population, termed single-nucleotide polymorphisms (SNPs)^{10, 11}. These variations can be used to genotype individuals, do linkage analysis, carry out genome-wide association studies (GWAS), as well as detect CNVs and sites of loss of heterozygosity (LOH)^{6, 12}. Using current DNA microarray platforms¹³, up to approximately 1×10⁶ SNPs can be analyzed at one time. This technology allows evaluation of the genome at high resolution, and unlike array-CGH, does not require a reference sample. Copy number evaluation in head and neck squamous cell cancers by a multitude of techniques has been reviewed in great detail by Chen and Chen¹⁴, and recent studies have integrated copy number analysis using SNP arrays with gene expression data to develop prognostic signatures in oral cavity squamous cell carcinoma^{15, 16}. SNP genotyping has also been used to identify polymorphisms in germline DNA associated with risk of second primary tumors and recurrence among patients treated for early-stage head and neck squamous cell cancer¹⁷, demonstrating another application of SNP evaluation in patients- germline risk-profiling.

Next generation sequencing

Until very recently, DNA sequencing was dependent upon variations on methods originally developed by Frederick Sanger and colleagues in the late 1970’s^{18, 19}. With these techniques, fragments of DNA from the sample of interest are amplified with polymerase chain reaction (PCR) techniques, and each PCR reaction is randomly terminated with a chemically altered base-pair. The size of each pool of fragments generated is then measured (eg. with gel electrophoresis), and the last base added can be determined depending on which base terminated the reaction (eg. each nucleotide can be radio-labeled, or color-coded with a fluorescent marker). When each fragment is evaluated in aggregate, the entire sequence of the sample can be constructed. Sanger “base-by-base” sequencing techniques remained the state-of-the-art for genetic sequencing for three decades, and these methods were largely used to complete the Human Genome Project.

The term “next-generation” sequencing has been ascribed to a variety of techniques that parallelize sequencing reactions, which means that a large number of sequences (thousands-to-millions) from the DNA of interest are generated simultaneously and then aligned to compose the final sequencing result. Rapid advances in technology and computing power have produced a multitude of next-generation sequencing techniques (eg. 454, Ion semiconductor, sequencing by synthesis, sequencing by ligation)^{20, 21}, and a detailed review of these technologies are beyond the scope of this article. These sequencing techniques allow for gene or multi-gene sequencing in a very high-throughput and rapid manner. Next-generation sequencing has also led to the ability to sequence the entire coding region or entire genome of an individual or tumor in a matter of weeks.

Whole-exome sequencing and whole-genome sequencing

Exomes are regions of the genome that are transcribed into protein-coding RNA’s. There are approximately 180,000 exons in the human genome, composed of 30 megabases which yield roughly 20,000 protein-coding genes. Remarkably, this represents only ∼ 1% of the entire human genome. The premise of whole-exome sequencing relies on a method to enrich exomic DNA followed by next generation sequencing of these enriched targets. There are several enrichment methods, including PCR-based targeted amplification, the use of molecular inversion probes, hybrid capture, and in-solution capture²². Whole-exome sequencing can be used to identify mutations in coding genes, perform SNP genotyping from known SNPs in the exome, as well as identify translocations and determine copy number variations that involve exomic DNA ^{23, 24}. Whole-exome sequencing of head and neck squamous cell carcinoma (HNSCC) was recently reported in two large studies^{25, 26}. Findings confirmed mutations in genes known to be common players in this disease (eg. TP53, HRas, CDKN2A), and novel mutations were identified in genes not previously implicated in HNSCC, as well (eg. Notch 1, FBXW7, FAT1)^{25, 26}.

Whole exome sequencing is proving a valuable tool in cancer research and discovery, however non-protein-coding DNA, previously even referred to as “junk” DNA, is proving to be more important than once surmised. The ENCODE project has demonstrated that much of the genome is involved in the regulation of gene expression through three-dimensional conformation effects, interaction with coding elements, and through the production of non-coding RNAs, such as microRNA and long non-coding RNA (lncRNA)²⁷. Therefore, whole genome sequencing approaches are also proving important in cancer research. Whole genome analysis provides sequence for all (or most) DNA in an organism (eg. chromosomal and mitochondrial or chloroplast DNA). Older sequencing techniques based on Sanger methods were used to complete the Human Genome Project, but newer techniques, such as nanopore, nanoball, fluorophore, and pyrosequencing technologies combined with parallelization, as described above, have significantly reduced time and cost of whole genome sequencing.

Recent studies have just begun to apply whole-genome sequencing to analyze human cancers, such as a study of 39 pediatric low-grade gliomas²⁸, a study that included a subset of 15 esophageal adenocarcinomas²⁹, and 2 patients included in one of the HNSCC next-generation sequencing reports²⁶. These studies, however, all focused largely on events identified within protein-coding DNA. Our understanding of non-coding DNA is increasing, and the importance of these elements in human cancer is just being realized, as evidenced by the important lnc-RNA, HOTAIR, shown to play a role in cancer cell behavior³⁰. In another study, mutations in the TERT promoter region, which have been shown to potentially increase telomerase expression, were commonly identified in a subset of cancers, including gliomas, melanoma, and oral squamous cell cancer³¹. As more of the human genome is understood, more layers of complexity in gene sequence and regulation is uncovered, all of which have implications in human cancer.

High-throughput epigenetics and transcriptional analysis

This paper focuses on recent advances and applications of genomics in cancer research and biomarker development, however new technologies and applications in epigenetics, gene expression, and transcriptome analysis are inherently related, and deserve mention here. The newest technologies provide very comprehensive evaluation of gene expression, DNA methylation, and micro-RNA expression. High through-put gene expression analysis has now been available for approximately 2 decades, and several groups have identified gene expression signatures as potential biomarkers in cancer. Advances in breast cancer signatures have been notable³², including the development of a 21-gene signature³³ that has led to the clinical diagnostic, Oncotype dx®, used to prognosticate patients with estrogen receptor (ER)-positive, early-stage breast cancer. Gene expression evaluation has also been used to profile HNSCC, for example Chung, and colleagues used gene expression signatures to define 4 distinct groups of patients with head and neck squamous cell carcinoma and showed that these signatures could be used to predict outcome³⁴.

Until recently, cDNA microarray technology was the standard in gene expression analysis- the most common techniques involve purification of messenger RNA, reverse transcription to cDNA, and hybridization to an array carrying tens-of-thousands of probes that were used to determine the relative quantity of each target. In the last few years, next generation sequencing techniques have rivaled microarray technology for gene expression analysis. RNA-seq is a technique that uses next generation sequencing to evaluate the entire transcriptome. In this technology, RNA is purified according to the level of analysis that is desired (eg. messenger RNA can be isolated, or only ribosomal RNA can be eliminated, allowing evaluation of non-gene transcripts, such as micro-RNA and lnc-RNA), reverse-transcription is carried out, and sequencing commences³⁵. The entire transcriptome is reassembled from sequence reads, and coverage can be used to estimate expression level³⁵.

Microarray technology has also been developed to evaluate genome-wide methylation events. In these techniques, one DNA sample is treated with bisulfite conversion which only changes unmethylated DNA, and then bisulfite-converted sample is compared to untreated sample on the array in order to determine the degree of methylation for tens-of-thousands of probes across the genome.. Alterations of DNA methylation are common in human cancers, and individual methylation events or methylation signatures can be employed as potential prognostic or predictive biomarkers³⁶. Microarrays have also been used to profile micro-RNAs in cancer, with several biomarker approaches under development³⁷. Technology that allows high-throughput evaluation of gene targets of specific transcription factors have also been developed: Chromatin immunoprecipitation (ChIP) allows profiling of the potential gene targets of specific transcription factors. With this technique, a transcription factor of interest bound to genomic DNA is isolated, which is then fragmented and immunoprecipitated (chromatin-immunoprecipitation) to “pull-down” the DNA associated with the transcription factor. This DNA is then either hybridized to a microarray to assess and quanitify transcription targets, or fragments are sequenced (ChIP-seq)³⁸. Experiments examining protein-DNA interactions have been an important component of the ENCODE project³⁹ , and have led to a greater understanding of gene regulation and transcription factor activity.

The Cancer Genome Atlas

All of the technologies detailed above have been developed and improved over the last two decades. Table-1 summarizes these research techniques. The application of these methods has yielded a tremendous amount of data now available describing several human cancers. Perhaps the most comprehensive program applying modern high throughput technology and integrating the data generated is The Cancer Genome Atlas (TCGA) project. The TCGA was initiated in 2006 and is supported jointly by the National Cancer Institute (NCI) and the National Human Genome Research Institute (NHGRI). The goal of the program is to organize and support several expert centers with the common aim to characterize the genomes of more than 20 types of human cancers using the most up-to-date, high throughput technology available. To date, reports on glioblastoma, ovarian cancer, colorectal cancer, squamous cell lung cancer, breast cancer, and endometrial cancer have been published^40–44, and analysis of several other tumor types, including HNSCC, is underway. The data from hundreds of tumor samples submitted to the TCGA is available to the public, and includes information on CNV’s, somatic mutations, SNP’s, as well as epigenetic and trascriptome data, including methylation, gene expression, and microRNA expression. Modern technology and computing power have now allowed comprehensive genomic indexing of human cancers, and have made individualized tumor characterization a reality. The challenge at this point is to determine the optimal way to analyze these results and utilize these data clinically.

Table 1.

High throughput molecular biology techniques currently available for cancer research

Technique	Description
DNA evaluation
CGH arrays	DNA from a test sample and a reference sample are differentially labeled using different fluorophores. These are then hybridized to an array containing several thous and probes. Used to detect copy number changes at a resolution of 5 – 10 kb to 200 bp.
SNP arrays	DNA microarrays containing probes representing ~ 1×10^6 SNPs are used to explore SNP’s present in a test sample. Applications include genetic linkage analysis, examination of CNV’s and evaluation of LOH
Next- generation sequencing	Sequencing techniques that parallelize the process of DNA sequencing, producing massive numbers of sequences at once. Several techniques exist, including pyrosequencing (454), sequencing by synthesis, sequencing by ligation, ion-torrent, and single- molecule real- time sequencing. These sequencing techniques have shortened read time and lowered cost substantially.
Whole exome sequencing	Technique used to explore the entire coding region of human DNA for genetic variations. Typically, an enrichment strategy is employed to extract DNA coding for gene exons. These targets are then sequenced using next-generation techniques.
Whole genome sequencing	Technique used to explore the nucleotide base-pair sequence of an entire genome. Next generation sequencing techniques have allowed whole genome analysis to occure over a period of days- to -weeks and at a cost of approximately ~$1000 – $5000.
RNAevaluation
cDNA microarray	Messenger-RNA from a test sample is reverse-transcribed to cDNA. Labelled (eg. with fluorophore) cDNA’s are hybridized to DNA probes (typically ~ 28,000 - 44,000 probes) on the microarray which correspond to thousands of known genes. Intensity of the signal from hybridized cDNA’s are used to estimate gene expression level.
RNA-seq	RNA from a test sample is reverse-transcibed to cDNA. cDNA is then evaluated with next generation sequencing. Studies examining cDNA derived from non-coding RNA in addition to mRNA have yielded important regulatory RNA’s, such as lnc- RNA’s. RNA- seq analysis allows identification of genetic variations, such as mutations or fushion genes, if these are transcribed to RNA . Gene expression can be estimated from sequence coverage analysis.
Epigenetic evaluation
ChIP- on- chip	ADNA-binding protein (eg. a transciption factor) is cross- linked to DNA, and the DNA is fragmented. The protein and linked DNA are immunoprecipitated, and the DNA fragments are hybridized to a microarray for identification and quantification. Allows high throughput analysis to identify potential genetic targets of a protein of interest.
DNA methylation microarrays	Unmethylated DNA is segregated from methylated DNA with bisulfite-conversion. A bisulfite-converted and control DNA sample are differentially labelled, and relative abundance of methylated and unmethylated DNA for specific genes are evaluated using DNA microarrays.
Methylated DNA Immunoprecipitation.	Methylated regions of the genome are immunoprecipitated using an antibody directed toward 5- methylcytosine. Methylated genes are then evaluated using either microarray-based methods or with next-generation sequencing.

abbreviations: CGH (comparative genomic hybridization), SNP (single nucleotide polymorphism), DNA (deoxyribonucleic acid), RNA (ribonucleic acid), ChIP (chromatin immunoprecipitation), CNV (copy number variation), LOH (loss of heterozygozity), cDNA (complementary DNA), lnc-RNA (long non-coding RNA)

From Molecular Biology to Therapeutic Targets

Advances in molecular pharmacology, in parallel with an increased understanding of cancer genomics, have led to rapid expansion in new therapeutics available to treat cancer. During the mid-twentieth century, the first effective cancer therapeutics were realized with the use of aminopterin to treat childhood leukemia pioneered by by Sydney Farber⁴⁵. Subsequently, several cytotoxic therapies were developed and applied over the ensuing decades, many of which remain the cornerstone of most chemotherapeutic regimens today. These drugs, such as nucleotide analogues, DNA-damaging agents, and anti-folates, largely act upon rapidly dividing cells by disrupting the cellular machinery or molecular building blocks that drive cell division. A greater understanding of the molecular biology of cancer over the last three decades has led to the development of “targeted therapeutics” that act upon specific cancer cell proteins, such as surface receptors or intracellular signaling molecules. This vast array of new therapeutics that is currently available can be categorized based on their structure and mechanism of action.

Some of the earliest examples of “targeted” therapies were aimed at the inhibition of hormones. Examples include the utility of Tamoxifen to selectively block the estrogen receptor in breast cancer⁴⁶, and androgen deprivation via leuprolide or flutamide in prostate cancer⁴⁷. These approaches have proven to be remarkably effective for these diseases. As most cancers are not dependent on a hormonal driver, the concept of “targeted therapy” is used more commonly to apply to targeting of cancer-specific cell receptors or intracellular signaling molecules with either monoclonal antibodies or synthetic small molecules. A multitude of strategies have been developed, and several of these are highlighted in the following sections.

Monoclonal Antibodies that Target Cell Surface Receptors

The concept of targeting cancer-specific proteins with antibodies was established during the mid-to-late 20^th century, and the first therapeutics became a reality after monoclonal antibody (MCA) production became possible^48–50. Delivery of MCA’s to humans without an immune reaction became feasible as recombinant techniques led to chimerization, and then humanization of antibody products^{50, 51}. Several cell surface receptors have been found to be overexpressed in certain human cancers that can be targeted with MCA therapy. The epidermal growth factor receptor (EGFR/Her1) has been targeted with the MCA cetuximab, which is approved to treat colorectal cancer⁵² and head and neck squamous cell carcinoma⁵³. In breast cancer, overexpression of the human epidermal growth factor receptor 2 (Her2/Neu) in approximately 25–30% led to the development and application of trastuzumab in these tumors, which has proven to be remarkably effective in Her-2-expressing breast cancers⁵⁴. Angiogenesis, which is crucial to tumor growth and metastasis, has been targeted with bevacizumab, a monoclonal antibody against the vascular endothelial growth factor A (VEGF-A)⁵⁵ which was initially approved for use in the treatment of patients with metastatic colorectal cancer⁵⁶.

Kinase inhibition with Small Molecules

Self-sufficiency in growth signaling and insensitivity to anti-growth signaling have been described as hallmarks of cancer cells². It has been established that many cell-signaling proteins are activated or de-activated via cellular kinase activity or phosphatase activity, respectively, and these proteins can form cascades mediating signals from the extracellular space to the nucleus leading to transcription factor activation. Many of the kinases involved phosphorylate tyrosine molecules (tyrosine kinases) on downstream target molecules, but protein kinases that target other amino acids are not uncommon (eg. serine-threonine kinases). Tyrosine kinases (TKs) can be divided into receptor tyrosine kinases (RTKs), which are membrane-bound and carry an extracellular ligand-binding domain, and an intracellular kinase domain that is normally activated when the receptor domain is stimulated with a ligand. Non-receptor tyrosine kinases (nRTKs) are intracellular kinases located in the cytoplasm. They are often activated via phosphorylation by RTKs or other nRTKs, and propagate signal by phosphorylating downstream nRTKs or transcription factors.

Abnormal activation of kinase signaling pathways is ubiquitous in cancer, and this has been exploited with modern therapeutics through the development of small molecule inhibitors⁵⁷. These drugs usually act in the intracellular space, and can often be taken orally. One of the first tyrosine kinases targeted was the bcr-abl fusion protein, a constitutively active kinase created by the 9:22 translocation common in chronic myelogenous leukemia (CML)^58–60. Imatinib, a small molecule tyrosine kinase inhibitor (SMTKI) designed to target bcr-abl, has been shown to be effective in treating this disease⁶¹. As is common to many SMTKIs, imatinib inhibits other TKs, including c-Kit. C-Kit was found to be a major driver in gastrointestinal stromal tumors (GIST), and imatinib has been shown to have substantial efficacy in this disease, leading long-term durable responses in a majority of patients. More recently, vemurafenib, a SMTKI that targets BRAF, has been approved to treat advanced cutaneous melanoma containing the common BRAF V600E activating mutation^{62, 63}. Small molecule inhibtors that target key cancer pathways, such as the RAF-Mek-Erk cascade, EGFR signaling, and the PI3K/mTor pathway, are arguably the fastest growing class of cancer therapeutics, and a multitude of new drugs have been approved or are currently under development.

Targeted agents that facilitate immune mediated cell death

The immune system is constantly warding off cancers by attacking preneoplastic and neoplastic cells, as evidenced by an increase in cancer rates among immune-compromised individuals. At the same time, tumors commonly arise in the setting of chronic inflammation, and often promote an inflammatory response⁶⁴. Several newer therapeutics utilize the immune system to destroy cancer cells. There are many approaches to accomplish this⁶⁴. One approach is to use monoclonal antibodies to directly target and bind to tumor cell proteins, thus triggering a cytotoxic immune response. An example is Retuximab, aantibody MCA which binds CD20, a surface protein commonly found on B-Cells. Retuximab has been shown to be effective in treating CLL, and certain types of non-Hodgkin’s lymphoma ⁶⁵. Another approach to enhance immune-mediated tumor killing is to potentiate established responses that become suppressed by normal immune regulation or tumor modulation. This is the premise behind the utility of CTLA-4 inhibitors, such as ipilimumab. T-cells require co-stimulation of several receptors to potentiate activation by antigen presenting cells (APCs). CD28 presented by APCs stimulates CD80 and CD86 to co-stimulate T-cells. CTLA4, expressed on T-regulatory cells, can block co-stimulation with CD28, thus suppressing the immune response. Ipilimumab inhibits CTLA-4, thereby allowing co-stimulation and continued activation of the T-cell immune response⁶⁶. This approach has proven effective in the treatment of cutaneous melanoma⁶⁷, and its therapeutic efficacy is now being explored in several other cancers. Several agents and approaches that utilize or amplify the immune response to tumor antigens are either on the market already or are currently being studied, and in the near future genomic and epigenetic profiling of tumors and the tumor microenvironment may inform which cancers will be best treated with these approaches.

Other New Molecular Approaches

There are several additional therapeutic approaches based on molecular tumor biology that are either approved or under investigation. These include agents that target the cell proteasome⁶⁸, inhibit histone deacetylaces⁶⁹, and antibodies that recognize tumor antigens and deliver toxins⁷⁰ or radioactive particles⁷¹ to cancer cells. Full description of contemporary innovations in experimental therapeutics are beyond the scope of this article, but the plethora of current options both known to be effective in cancer and under active investigation can be appreciated by the list of approved targeted agents posted on the information web page by the National Cancer Institute (NCI) (http://www.cancer.gov/cancertopics/factsheet/Therapy/targeted), and are also summarized here in table-2. Several agents under active investigation can be found at the Cancer Therapy Evaluation Program (CTEP) website (ctep.cancer.gov/protocolDevelopment/docs/ctep_active_agreements.xls).

Table 2.

Molecularly targeted agents that are FDA approved to treat human cancer

Agent	Class	Primary Molecular Targets	Utility
Signal Transduction Inhibitors
Imatinib mesylate (Gleevec®)	SMTKI	c-Kit,bcr-abl, PDGFR, others	GIST, leukemia, dermatofibrosarcoma protuberans, myelodysplastic/ myeloproliferative disorders, systemic
Dasatinib (Sprycel®)	SMTKI	bcr-abl, Src	CML, ALL
Nilotinib (Tasigna®)	SMTKI	bcr - abl , Kit , LCK, EPHA3, EPHA8, DDR1, DDR2 , PDGFRB, MAPK11, ZAK	CM L
Bosutinib (Bosulif®)	SMTKI	bcr-abl, Src, HDACinhibitor, also	CML
Trastuzumab (Herceptin®)	MCA	HER- 2	Breast cancer, gastric/ GE junction adenocarcinoma
Pertuzumab (PerjetaTM)	MCA	HER- 2	met astatic breast cancer with trast uzumab and docetaxel
Lapatinib (Tykerb®)	SMTKI		breast cancer
Gefitinib (Iressa®)	SMTKI	EGFR	non-small cell lung cancer
Erlotinib (Tarceva®)	SMTKI	EGFR	non-small cell lung cancer, pancreatic cancer (unresectable/ metastatic)
Cetuximab (Erbitux®)	MCA	EGFR	head and neck squamous cell carcinoma, colorectal cancer
Panitumumab (Vectibix®)	MCA	EGFR	metastatic colon cancer
Temsirolimus (Torisel®)	SMSTKI	mTOR	advanced renal cell carcinoma
Everolimus (Afinitor®)	SMSTKI	immunophilin FK binding protein-12: binds and inhibits mTOR	advanced, progressive kidney cancer, subependymal giant cell astrocytoma in patients with tuberous sclerosis, advanced breast cancer, pancreatic neuroendocrine tumors
Vandetinib (Caprelsa®)	SMI	EGFR, VEGF, RET	metastatic medullary thyroid cancer
Vemurafenib (Zelboraf®)	SMSTKI	BRAF V600E	inoperable/ metastatic melanoma
Crizotinib (Xolkori®) protein	SMTKI	EML4-ALK fusion	locally advanced/ metastatic non-small cell lung cancer
Target proteinsthat regul ate key cell functions or gene expression
Vorinostat (Zolinza®)	SMI	HDACinhibitor	CTCL
Romidepsin (Istodax®)	SMI	HDACinhibitor	CTCL
Bexarotene (Targretin®)	retinoid	retinoid X recept or agonist	CTCL
Alitretinoin (Panretin®)	retinoid	reinoicacid and retinoid X receptor agonist	AIDS-related Kaposi’s Sarcoma
Tretinoin (Vesanoid®)	retinoid	retinoicacid receptor	Acute promyelocytic leukemia
Agents that induce apoptosis
Bortezomib (Velcade®)	PI	proteosome	Multiple myeloma, mantle cell lymphoma
Carfilzomib (Kyprolis®)	PI	proteosome	multiple myeloma
Pralatrexate (Folotyn®)	antifolate	selectively accumulates in RFC-1 expressing cells	peripheral T-cell lymphoma
Target Angiogenesis
Bevacizumab (Avastin®)	MCA	binds VEGF	glioblastoma, non-small cell lung cancer, metastatic colon cancer and kidney cancer
Ziv-aflibercept (Zaltrap®)	VEGFR-mimic / immune protein	binds VEGF	metastatic colon cancer
Sorafenib (Nexavar®)	SMTKI	VEGFR, PDGFR, C-Raf, B- Raf	advanced renal cell carcinoma, hepatocellular carcinoma
Sunitinib (Sutent®)	SMTKI	PGDF-R’s, VEGFR’ s, KIT, RET, CSF-1R, flt3	metastaticrenal cell carcinoma, Imatinib- resistant GIST, pancreatic neuroendocrine tumors
Pazopanib (Votrient®)	SMTKI	VEGR’ s, c-KIT, PDGFR	renal cell carcinoma, advanced soft tissue sarcoma
Regorafenib (Stivarga®)	SMTKI	VEGFR, angiopoietin-1 receptor (TIEF2), PDGFR, RET, c-KIT, RAF	metastatic colorectal cancer
Cabozantinib (CometriqTM) SMTKI		VEGF, RET, MET, TRKB, TIE2	metastatic medullary thyroid cancer
Agents that facilitate immune-mediated cell death
Rituximab (Rituxan®)	MCA	CD20	CLL, B-Cell lymphomas
Alemtuzumab (Campath®)	MCA	CD52	B-cell CLL
Ofatumumab (Arzerra®)	MCA	CD20	fludarabine and alemtuzumab resist ant CLL
Ipilimumab (YervoyTM)	MCA	CTLA-4 inhibitor	unresectable or metastatic melanoma
Monoclonal Antibodies that deliver toxic molecules to cancer cells
Tositumomab and 131I–tositumomab (Bexxar®)	MAB linked to I131	CD20-expressing B-Cells	non-Hodgkin’s B-Cell Lymphoma
Ibritumomab tiuxetan (Zevalin®)	MAB linked to radioisotopes	CD20-expressing cells	non-Hodgkin’s B-Cell Lymphoma
Denileukin diftitox (Ontak®)	IL-2 + diphteriatoxin proteins	IL-2 receptors	cutaneous T-cell lymphoma
Brentuximab vedotin (Adcetris®)	MAB + monomethyl auristatin E	CD3 0	anaplastic lymphoma, hodgkin lymphoma after chemotherapy and stem cell transplant

abbreviations: SMTKI (small molecule tyrosine kinase inhibitor), MCA (monoclonal antibody), SMSTKI (small molecule serine-threonine kinase inhibitor) SMI (small moleculeinhibitor), GIST (gastrointestinal stromal tumor), CML (chronic myelogenous leukemia), ALL (acute lymphoblastic leukemia) GE (gastro-esophageal), CTCL (cutaneous T-cell lymphoma), HDAC (histone deacetylase), AIDS (acquired immune deficiency syndrome) PI (proteosome inhibitor), CLL (chroniclymphocytic leukemia)

Integration of Genomic Data and Cancer Treatment to Develop Effective Biomarkers

The challenge we face is to incorporate an enormous amount of genomic information from human cancers into treatment strategies using the large number of molecular targeted therapeutic agents that are either currently available or under development. Cancer genomics must be translated into clinical biomarkers that can be used to prognosticate (ie. prognostic biomarkers) or predict response to therapy (ie. predictive biomarkers). Genomic biomarkers have several applications for patients with cancer or at risk for the development of cancer. Several of these applications are presented, with examples, in Table 3. There are many obstacles that must be surmounted if improvements in cancer treatment will continue efficiently and effectively, and the development of new genomic biomarkers requires several important steps.

Table 3.

Applications of genomic biomarkers in cancer.

Biomarker Utility	Example
Predict risk of cancer	BRCA mutation as a risk factor for development of breast cancer
Provide prognostic information	HPV and p16 positivity in oropharyngeal squamous cell carcinoma
Determine PK/PD of specific chemotherapeutics (pharmacogenomics)	CYP2D6 enzyme polymporphisms and metabolism of Tamoxifen
Predict response to specific therapy	BRAF V600E mutation in cutaneous melanoma
Tumor surveillance	Circul ating tumor DNA in breast cancer

abbreviations: PK (pharmacoki netics), PD (pharmacodynamics), HPV (human papilloma virus), DNA (deoxyribonucleicacid)

Development and application of bioinformatic approaches to identify candidate biomarkers

The first step in incorporating genomic data into clinical practice is to identify the genomic events most relevant to a given cancer type or subset of patients. The first-pass analysis of whole exome or whole genome data presents several challenges from the outset. Quality control and evaluation with appropriate normalization are essential in this process. Variant calls are highly dependent on sequence coverage and wild-type tissue contamination⁷². Several alignment tools are available to build a bioinformatic pipeline, and analysis requires alignment to a specific reference genome. Variant characterization then depends upon comparing identified variations to known polymorphisms in order to make bonafide mutation calls⁷³. Data can also be used to identify copy number changes and perform SNP genotyping⁷⁴.

The next step in translating genomic information to clinical biomarkers is to identify which genomic events are relevant: which are important to tumor biology, and which are relevant clinically. Several approaches exist, and are dependent on the scientific and translational questions being asked. Now that whole exome sequencing has been applied to a large number of tumors, it appears that the majority of mutations are “passengers”- seemingly irrelevant to a tumor’s development and behavior - while a select few appear to “drive” tumor biology. Several statistical methods and functional genomic approaches are being put forth to identify which genomic alterations in tumors are key oncogenic or tumor suppressive events crucial to cancer development, progression, and behavior⁷².

Another approach to identify the most relevant genomic events is to integrate genomic data with other platforms, such as epigenetic, proteomic, and metabolomic datasets. This systems biology approach can evaluate cellular signaling pathways and processes in order to identify common abberations that are caused by multiple insults along a specific pathway or network⁷⁵. Integrated analysis can clarify the picture of the molecular biology driving human cancer cells. For example, the pilot TCGA project integrated nucleotide sequencing, copy number variation, gene expression, and methylation data from glioblastoma, and uncovered a mutator phenotype linked to MGMT promoter methylation, and recurrent abberations among key pathways involving TP53, Rb, and a network of RTKs were described⁴⁰. Since this initial TCGA report, integrated analyses from other cancers have continued to shed light on several tumor types. In a recent project studying oral squamous cell cancer published by the senior author of the this review (JNM), an integrated analysis of mutation, copy number analysis, DNA methylation, gene expression, and microRNA expression⁷⁶ yielded insight not previously appreciated with whole exome evaluation alone^{25, 26}. These findings included four distinct pathways that were commonly altered by several events at multiple levels. Additionally, despite the fact that loss of tumor suppressor activity dominated the genomic landscape of these tumors, targetable oncogenic events were identified in most samples after integrated analysis was performed. Figure-1 provides examples of these findings after integrated analysis.

Figure 1.

An integrated analysis of mutation, copy number, methylation, and gene expression identified recurrent abberations in 4 dominant pathways: (A.) the Notch pathway (B.) cell cycle pathways (C.) pathways in mitogenic signaling, and (D.) the TP53 pathway. Several potentially actionable oncogenes in oral cavity squamous cell carcinoma were also identified after an integrated analysis (E.) . (Adapted by permission from the American Association for Cancer Research: Pickering CR, et al. Integrative genomic characterization of oral squamous cell carcinoma identifies frequent somatic drivers. Cancer Discovery. 2013 Apr 25. [Epub ahead of print] PMID: 23619168)

Bioinformatics and integration through systems biology approaches are helping us discover relevant genomic events to better understand human cancer. The next step in biomarker development is to determine which of these events are clinically relevant and have prognostic or predictive applications.

Preclinical approaches to select clinically relevant targets

With the bewildering amount of data now available and expected from future reports profiling cancers, it becomes difficult to focus on clinically actionable events. Although there are already several examples of genomic events that have been successfully developed as predictive biomarkers for treatment selection, the majority of cancers still lack clinically significant biomarkers. The history of the drug imatinib is a perfect example of how success depends on elements of directed and persistent development (ie. Imatinib’s proven efficacy against bcr-abl activity in CML⁶¹), and an element of intelligence-driven serendipity, as in the case of Dr. Jonesuu’s discovery of imatinib’s profound activity in GIST⁷⁷. Realistically, both will drive the identification and implementation of new and effective cancer treatments, but we must focus on improving the former rather than rely on the latter, as sound study design and investigation is in our control.

After stringent selection from bioinformatic and systems biology analyses, the functional effects of manipulation of specific genomic targets can be evaluated in several ways. In the current era, data utilizing tumor-derived cell lines in in vitro studies and tumor xenograft animal models remains important. Preclinical studies to date have largely focused on single genes or pathways, and analyzed small numbers of cell lines. In order to be translatable to the clinical setting, it is becoming increasingly important to perform global molecular assessments and/or utilize large numbers of cell lines in an attempt to recapitulate the heterogeneity observed in the reality of human cancer. For example, large-scale si-RNA screens applied in the context of a specific mutation or in conjunction with a targeted agent can identify aberrant pathways that might be synthetically lethal with each condition. Turner and colleagues used this method to identify factors related to PARP inhibitor-sensitivity in BRCA-mutant breast cancer cell lines⁷⁸, and Berns, et al. used similar methods to show that alterations in the PI3K pathway mediated resistance to herceptin in breast cancers⁷⁹. The authors of this review (TJO, VKS, HDS, JNM) performed genomic and phenotypic analysis on a large panel of immortalized head and neck cancer cell lines to show that TP53 disruptive mutations were associated with aggressive tumor growth and metastasis in an orthotopic xenograft model⁸⁰ as well as with in vitro radiation resistance⁸¹, providing preclinical corroboration with results observed clinically in this disease in two large patient studies^{82, 83}. In another very notable example, preclinical analysis of 602 cell lines showed that amplification or translocation of the ALK gene leading to activation was strongly associated with sensitivity to ALK inhibitors⁸⁴. This pre-clinical work led to a successful clinical trial of crizotinib, an ALK inhibitor, in non-small cell lung cancers with ALK rearrangements⁸⁵.

Large-scale genome projects, such as the TCGA, will likely continue to produce lists of novel genomic events that have not yet been functionally characterized. These events will need to be rigorously studied using preclinical functional evaluation in order to determine their potential as predictive biomarkers. Several platforms exist for this evaluation, including systematic mutagenesis using transposons or retroviruses, RNA-interference library screens (as described above), and high-throughput overexpression systems (eg. cDNA or open-reading frame libraries). These techniques can be used for comprehensive evaluation of the function and consequences of manipulation of target genes⁸⁶. Rigorous preclinical evaluation can optimize the selection of candidate biomarkers that will be most successful in the clinical arena.

Clinical Trial Design in the Current Era of Genomics

The typical phase III clinical trial compares a standard therapy against a new regimen (often standard of care + new drug), and typically hundreds of patients are evaluated in order to demonstrate a small, but statistically significant, improvement in outcome. The new era of cancer genomics and targeted therapy is changing approaches to the design of effective trials. Several targeted molecular therapies depend on the presence of a specific predictive biomarker. Examples include wildtype KRAS as a marker for cetuximab sensitivity in colon cancer⁸⁷ and BRAF V600E mutation as a marker for vemurafenib activity in cutaneous melanoma⁶². As more and more effective therapeutics are developed, clinical trials must be designed to deliver these drugs to their intended patient population, and they should be developed in a manner that demonstrates efficacy in the most expeditious manner. Several recommendations have been put forth to accomplish this goal.

First, molecular pre-screening should be applied wherever possible to select patients for clinical trials evaluating targeted agents⁸⁸. There are several barriers that exist hindering the translation of scientific findings into CLIA-certified diagnostics, including prioritizing the appropriate markers and drugs from preclinical data, logistical considerations such as the availability and accessibility of tissue for biomarker evaluation, and regulatory barriers to the establishment of certified diagnostics⁸⁹. These barriers must not prohibit evaluation of these markers in the clinical trial setting. Molecular analysis should also be incorporated into the design of the clinical trial to evaluate if molecular therapy is acting upon the proposed target, to determine if action on a target indeed correlates with clinical efficacy, and to identify secondary molecular events that correlate with sensitivity or resistance.

Second, the general consensus is that clinical trials assessing targeted therapy should become smaller and shorter^89–91. Genomic and molecular evaluation of cancer is identifying small cohorts of patients that have tumors with molecular characteristics that can potentially be affected by a specific drug, in some ways segregating cancer into a heterogeneous array of “orphan” diseases⁹⁰. Our approaches to trials should therefore concentrate effort into properly matching these small cohorts with appropriate drugs. This philosophy therefore lends itself best to phase I or phase II clinical trial designs targeting selected patients, where large numbers of patients are screened to identify a small cohort of patients who will likely benefit from receiving drug in the study. The endpoints should also focus on dramatic tumor responses in order to find drugs that have the best efficacy. Studies in the neoadjuvant or unresectable/metastatic settings represent the best opportunities to test markers/drugs under this paradigm. A notable example of this approach is the previously referenced trial examining crizotinib in non-small cell lung cancer with ALK rearrangements⁸⁵. In this study, approximately 1500 patients with advanced non-small cell lung cancer were screened for this genomic alteration. 82 patients were enrolled in the study, and the majority showed a response to crizotinib monotherapy with a very favorable side-effect profile⁸⁵. Another innovative approach was used in the The Biomarker-integrated Approaches of Targeted Therapy for Lung Cancer Elimination (BATTLE) trial studying patients with chemo-refractory non-small cell lung cancer published in 2011⁹². In this trial, mandated biopies were obtained to evaluate four biomarker profiles: EGFR mutation/copy number, KRAS/BRAF mutation, VEGF/VEGFR2 expression, RXRs/Cyclin D1 expression/CCND1 copy number. The initial cohort of patients was randomized to four treatments: erlotinib, vandetinib, erlotinib/bexarotene, and sorafenib. Disease control rates were determined for each biomarker group for each drug in order to create pre-test probabilities of response for each group. In the second phase of the study, a Bayesian adaptive randomization method was used to randomize patients to the four regimens. This study design allowed testing of pre-specified hypotheses regarding selected biomarkers and treatment efficacy, and showed for example, that the disease control rate in the KRAS/BRAF marker group treated with sorafenib was 79% versus only 14% with erlotinib. This study was a landmark trial in its strategy for studying targeted therapies in the context of relevant biomarkers.

With comprehensive genomic analysis of human cancers well underway, and with hundreds of new drugs to test either alone or in a near-infinite number of potential combinations, clinical trial strategies should focus on matching biomarkers and appropriate drugs, with the ultimate goal of dramatic efficacy. In order to accomplish these goals, the prevailing clinical trial paradigm will likely need to change. One might envision a comprehensive network of smaller-scale trials offering several drugs or combinations of drugs to patients selected by genomic or molecular criteria rather than by disease site and histology. Efficacy would then be confirmed in smaller Phase II trials, and ultimately in Phase III studies if deemed necessary. If we continue to accept marginal improvements after accruing hundreds of patients over several years of evaluation in large-scale phase III trials of unselected patients, we will never keep pace with genomic and therapeutic advances that are now available, and that offer potential cancer cures.

Several recent endeavors have been initiated and funded by the NCI with aims that parallel those highlighted in this review. The Cancer Target Discovery and Development (CTDD) program is made up of a network of 11 centers across the United States that are currently working both individually and collaboratively to drive 13 specific projects. The collective goal of the program is to translate the enormous wealth of data in cancer genomics and biology into improved clinical outcomes through biomarker development and improved treatments. The program can be reviewed at the following website (http://ocg.cancer.gov/programs/ctd2). A similar program, Therapeutically Applicable Research to Generate Effective Treatments (TARGET), is aimed at childhood cancers, and can be reviewed at the following NIH website (http://ocg.cancer.gov/programs/target). Planned for the near future is the NCI-MATCH (Molecular Analysis for Therapy Choice) program, which aims to enroll patients with solid tumors or lymphoma who have progressed after at least 1 standard therapy. This program proposes to require a biopsy and sequencing of tumor samples from enrolled patients, who will then be referred to a clinical center among the NCI-MATCH network to participate in one of several single-arm, phase II studies. The appropriate trial will be selected based on an agent targeting a genomic event profiled in the participant’s biopsy screen. This program is currently in the design phase, and will hopefully lead to significant advances in the endeavor to translate cancer genomics into predictive biomarkers for several cancers.

Other applications of modern genomics in cancer treatment and biomarkers

Finally, the application of high throughput genomic and epigenetic analysis is not limited to matching genomic biomarkers with the most efficacious drugs. Pharmacogenomics evaluates the different pharmacokinetics and pharmacodynamics of drugs due to genomic variation, which can help optimize drug and dose selection for specific patients. An example is identification of CYP2D6 polymorphisms that alter the metabolism of Tamoxifen, which have been shown to have a significant association with outcome among patients with ER-positive breast cancer who were treated with this drug⁹³. Another application of modern genomics in cancer biomarker development is the utility of circulating tumor DNA as a method of assessing tumor burden, response, and surveillance for recurrence. One recent study identified specific genomic alterations among biopsy samples taken from patients with metastatic breast cancer, and used next generation deep sequencing to identify and quantify circulating tumor DNA during treatment. Levels of circulating tumor DNA correlated with tumor burden and response to treatment⁹⁴. As new genomic information and techniques are developed, novel applications of this data will also continue to evolve.

One final consideration is how recent studies have further characterized the level of intratumoral genomic diversity. A recent review by Murugaesu, Chew, and Swanton⁹⁵ focuses on the current understanding of genomic diversity in individual tumors, clonal evolution, and the implications of these characteristics for cancer treatment. Pleomorphism among tumor cells is a hallmark of cancer, and tumor cell phenotypic heterogeneity and the degree of differentiation of cancer cells are often subjectively graded to aid in prognosis for many cancer types. Recent studies have characterized tumor cell heterogeneity at the genomic level. A study by Yachida and colleagues utilized whole-exome sequencing and copy number analysis to examine different regions within primary pancreatic cancers as well as among associated metastatic lesions. They demonstrated that the metastases evolved from sub-clones from the primary tumor, however the metastases were also, themselves, genetically evolved⁹⁶. Martinez, et al., have recently reported on copy number variations from 48 biopsies among 8 advanced renal cell carcinomas⁹⁷. Unsupervised clustering of these biopsies along with comparative evaluation of 440 tumors in the TCGA showed that there was significant heterogeneity within individual tumors, with clonal populations within each tumor that recapitulated clusters segregated among tumors in the TCGA. Studies such as these have significant implications for future approaches to targeted treatment.

Intratumoral heterogeneity and genomic heterogeneity between metastatic foci and primary disease pose obvious challenges to individualized therapy based on genomic evaluation. Future strategies will have to account for these hurdles, and several possible approaches exist. Intratumoral heterogeneity suggests multiple biopsies separated spatially and temporally may be necessary to optimize treatment, however there are obvious issues with practicality and morbidity. Treatment options will likely require tailored combinations of drugs to target dominant clonal populations of tumor cells. Strategies may attempt to selectively target clonal populations to “prune” tumors into a state that is genomically “manageable”. Perhaps new strategies will selectively allow a clonal unit with more benign features to dominate while first attacking smaller, yet more aggressive tumor populations prior to definitive intervention. The current era of genomics have certainly improved our understanding of the cancer genome, but with this knowledge comes new challenges facing cancer treatment.

The Future of Genomic Biomarkers

This is a very exciting time to be studying cancer genomics and to carry out translational cancer research. Modern genomics and contemporary therapeutics have already led to treatment strategies that were in the realm of science fiction three decades ago. However, there are still many challenges ahead. In several tumor sites, such as head and neck squamous cell cancer, no predictive biomarkers exist. Therefore, the scientific and medical community need to work collectively to employ the strategies proposed in this review (summarized in Figure 4) to develop more clinical biomarkers. With these efforts, new biomarkers and better treatment strategies will lead to improved outcomes for patients afflicted with cancer.

Figure 2.

Summary of steps necessary to translate cancer genomic information into successful biomarkers.