Strategies for Discovering Novel Cancer Biomarkers Through Utilization of Emerging Technologies
Vathany Kulasingam, BScH; Eleftherios P. Diamandis, MD, PhD
Summary and Introduction
Summary
The introduction of technologies such as mass spectrometry and protein and DNA arrays, combined with our understanding of the human genome, has enabled simultaneous examination of thousands of proteins and genes in single experiments, which has led to renewed interest in discovering novel biomarkers for cancer. The modern technologies are capable of performing parallel analyses as opposed to the serial analyses conducted with older methods, and they therefore provide opportunities to identify distinguishing patterns (signatures or portraits) for cancer diagnosis and classification as well as to predict response to therapies. Furthermore, these technologies provide the means by which new, single tumor markers could be discovered through use of reasonable hypotheses and novel analytical strategies. Despite the current optimism, a number of important limitations to the discovery of novel single tumor markers have been identified, including study design bias, and artefacts related to the collection and storage of samples. Despite the fact that new technologies and strategies often fail to identify wellestablished cancer biomarkers and show a bias toward the identification of high-abundance molecules, these technological advances have the capacity to revolutionize biomarker discovery. It is now necessary to focus on careful validation studies in order to identify the strategies and biomarkers that work and bring them to the clinic as early as possible.
Introduction
Cancer continues to be a major cause of morbidity and mortality among men and women. In the US in 2006, over 1.4 million new cases of cancer were diagnosed and over half a million people died from this disease; the disease accounts for approximately 25% of all deaths in the US each year.[1] With increasing life expectancy, the prevalence of many cancers will probably increase. Early detection of various forms of cancer before they spread and become incurable is an important incentive for physicians and research scientists.[2] One of the best ways to diagnose cancer early, aid in its prognosis, or predict therapeutic response, is to use serum or tissue biomarkers.
Cancer biomarkers can be DNA, mRNA, proteins, metabolites, or processes such as apoptosis, angiogenesis or proliferation.[3] The markers are produced either by the tumor itself or by other tissues, in response to the presence of cancer or other associated conditions, such as inflammation. Such biomarkers can be found in a variety of fluids, tissues and cell lines. Tumor markers can be used for screening the general population, for differential diagnosis in symptomatic patients, and for clinical staging of cancer. Additionally, these markers can be used to estimate tumor volume, to evaluate response to treatment, to assess disease recurrence through monitoring, or as prognostic indicators of disease progression (Box 1). Given the low prevalence of cancer in any given population, no marker has yet been discovered that meets all of these criteria.
A number of different types and forms of tumor markers exist. These markers include hormones, as well as different functional subgroups of proteins such as enzymes, glycoproteins, oncofetal antigens and receptors. Furthermore, other changes in tumors, such as genetic mutations, amplifications or translocations, and changes in microarraygenerated profiles (genetic signatures), are also forms of tumor markers. Regardless of the type of tumor marker or profile, the use of a tumor marker must be associated with proven improvements in patient outcomes, such as increased survival or enhanced quality of life, in order to be substantiated.[3] An ideal tumor marker should be able to be measured easily, reliably and cost-effectively by use of an assay with high analytical sensitivity and specificity (Box 2). A caveat concerning currently used tumor markers is that, generally, they suffer from low diagnostic specificity and sensitivity (Table 1). Only a few markers have entered routine use, and then only for a limited number of cancer types and clinical settings. In the majority of cases, the current markers are used in conjunction with imaging, biopsy and associated clinicopathological information before a clinical decision is made.
The first cancer marker ever reported was the light chain of immunoglobulin in the urine, as identified in 75% of patients with myeloma in an 1848 study.[4] The test for this marker is still employed by clinicians today, but with use of modern quantification techniques. From 1930 to 1960, scientists identified numerous hormones, enzymes and other proteins, the concentration of which was altered in biological fluids from patients with cancer. The modern era of monitoring malignant disease, however, began in the 1960s with the discovery of alfa-fetoprotein[5] and carcinoembryonic antigen (CEA),[6] which was facilitated by the introduction of immunological techniques such as the radioimmunoassay. In the 1980s, the era of hybridoma technology enabled development of the ovarian epithelial cancer marker carbohydrate antigen (CA) 125.[7] In 1980, prostate-specific antigen (PSA [KLK3]), considered one of the best cancer markers, was discovered.[8] Table 2 summarizes some currently used markers and their clinical utility.
Every era of biomarker discovery seems to be associated closely with the emergence of a new and powerful analytical technology. The past decade has witnessed an impressive growth in the field of large-scale and high-throughput biology, which has contributed to an era of new technology development. The completion of a number of genome-sequencing projects, the discovery of oncogenes and tumor-suppressor genes, and recent advances in genomic and proteomic technologies, together with powerful bioinformatics tools, will have a direct and major impact on the way the search for cancer biomarkers is conducted. Early discoveries of cancer biomarkers were based mainly on empirical observations, such as the overexpression of CEA. The modern technologies are capable of performing parallel rather than serial analyses, and they can help to identify distinguishing patterns and multiple markers rather than just a single marker; such strategies represent a central component and a paradigm shift in the search for novel biomarkers (Box 3).
These breakthroughs have paved the way for countless new avenues for biomarker identification. Very few serum tumor markers, however, have been introduced to the clinic over the past 15 years.[9] In this Review, we highlight some of the mechanisms behind biomarker elevation in biological fluids, and outline strategies for novel marker identification. These strategies should facilitate the delivery of potential candidate molecules for cancer diagnosis and prognosis and for prediction of therapy. These projected discoveries may be instrumental in substantially reducing the burden of cancer by providing prevention, individualized therapies, and improved monitoring following treatment.
Five major mechanisms exist by which molecules can be elevated in biological fluids during cancer initiation and progression. Such molecules could serve as effective cancer biomarkers. The mechanisms involved are outlined below; some of the different human body fluids that could be used as a source of biomarkers for specific types of cancers are shown in Table 3.
The protein encoded by a gene can be expressed in increased quantities as a result of increases in gene or chromosome copy number (i.e. gene amplification) or through increased transcriptional activity. The latter process could be the result of imbalances between gene repressors and gene activators. Epigenetic changes, such as DNA methylation, are also known to affect gene expression. On a larger scale, chromosomal translocations can result in gene regulation by promoters that are sometimes enhanced by steroid hormones;[10] transposons can serve a similar role.
An example of a putative biomarker is the protein human epididymal secretory protein 4 (HE4, also known as WFDC2), which is overexpressed in ovarian carcinoma. When complementary DNA microarrays were used to identify overexpressed genes in ovarian carcinoma, 101 transcripts were shown to be overexpressed in ovarian cancers compared with normal tissues.[11,12] Real-time polymerase-chain reaction (PCR) assay of an independent set of benign and malignant tissues confirmed that 12 of these transcripts were overexpressed in ovarian cancers. The transcripts HE4 and MSLN seemed to be the most differentially expressed between the tumor and normal tissues. Quantification of HE4 levels in serum revealed that this protein could be a potential biomarker for ovarian cancer,[13] although clinical evaluation is pending. Gene and protein expression of HE4 in a large series of normal and malignant adult tissues, however, showed that HE4 is present in pulmonary, endometrial and breast adenocarcinomas, in addition to staining positively in ovarian carcinoma.[14]
Given that 20-25% of all proteins are secreted, aberrant secretion or shedding of membranebound proteins with an extracellular domain (ECD) is another means by which molecules can be elevated in biological fluids. Alterations in the signal peptide of proteins as a result of single nucleotide polymorphisms can result in atypical secretion patterns.[15] Moreover, elevation of molecules in biological fluids can result from a change in the polarity of cancer cells, which can lead to the release of cancer-associated glycoproteins into the circulation. Increased expression of proteases that cleave the ECD portion of membrane proteins could also cause increased circulating levels.
Many proteins are secreted into the circulation; one example is alfa-fetoprotein, which is rapidly released from both normal and cancer cells.[16] A classic example of shedding of membrane proteins into fluids (and thus serving as a cancer biomarker) is HER2 (also known as ERBB2). HER2 is a cell membrane surface-bound tyrosine kinase that is involved in cell growth and differentiation.[17] Overexpression of this protein is associated with high risks of relapse and death from breast and ovarian cancers, and HER2 is the target of the therapeutic monoclonal antibody trastuzumab (Herceptin®; Genentech, San Francisco, CA).[18] The HER2 protein consists of a cysteine-rich extracellular ligand-binding domain, a short transmembrane domain, and a cytoplasmic protein tyrosine kinase domain. The ECD of HER2 can be released by proteolytic cleavage from the full-length receptor protein and can be detected in serum. High levels of HER2 in serum correlate with poor prognosis in patients with breast cancer.[19] In 2000, the FDA approved the serum HER2 test, which is the first blood test for measuring circulating levels of HER2 to have been approved for the follow-up and monitoring of patients with metastatic breast cancer.
Tissue invasion by the tumor might permit direct release of molecules into the interstitial fluid and subsequent delivery by the lymphatics into the blood. For epithelial cancer types, the proteins must break through the basement membrane of the invading tumor before they appear in the blood. For example, PSA is abundantly expressed by prostatic columnar epithelial cells and secreted into the glandular lumen, comprising a major component of seminal plasma (0.5-3.0 g/l) upon ejaculation. In healthy men, low levels of PSA enter the circulation by diffusing through a number of anatomic barriers, including the basement membrane, the stromal layer, and the walls of blood and lymphatic capillaries. This process gives rise to a normal serum PSA level in the range 0.5-2.0 ng/ml.
Prostatic carcinomas most often arise in the glandular epithelium of the prostate periphery. Although PSA (KLK3) gene transcription is downregulated in prostate cancer, PSA protein levels in the circulation of patients with prostate cancer increase through disruption of the anatomic barriers between the glandular lumen and capillaries. Concomitant to early-stage prostate cancer is the loss of basal cells, disruption of cell attachment, degradation of the basement membrane, initiation of lymphangiogenesis[20] and loss of the polarized structure and luminal secretion by tumor cells. Consequently, PSA levels in the serum can rise to 4-10 ng/ml. Late-stage prostate cancer is characterized by invasion of tumor cells into the stromal layers and the circulation, and by total loss of glandular organization. This situation enables considerable amounts of PSA to leak into the bloodstream, resulting in typical levels ranging from 10 ng/ml to 1,000 ng/ml (Figure 1).
Figure 1. (click image to zoom) Invasion and destruction of tissue architecture as a measure of biomarker elevation. (1) Healthy men display serum PSA in the range 0.5-2 ng/ml; low levels of PSA enter the circulation via diffusion through a number of anatomic barriers. (2) In early-stage prostate cancer, PSA levels in the serum rise to 4-10 ng/ml as a result of destruction of tissue architecture. (3) In late-stage prostate cancer, because of invasion of tumor cells, considerable amounts of PSA leak into the bloodstream; PSA levels typically range from 10 ng/ml to 1,000 ng/ml. Abbreviation: PSA = prostate-specific antigen.
Genomic and proteomic technologies have greatly increased the number of potential DNA, RNA and protein biomarkers under investigation. A paradigm shift has recently been realized, whereby single-biomarker analysis is being replaced by multiparametric analysis of genes or proteins. This development has triggered the question of whether cancer has a unique fingerprint (i.e. genomic, proteomic, or metabolomic). We outline a number of strategies for cancer biomarker discovery that utilize emerging technologies, and we discuss their merits and limitations (Figure 2).
Figure 2. (click image to zoom) Outline of strategies for biomarker discovery through utilization of emerging technologies. Abbreviation: MS = mass spectrometry.
Genomic microarrays represent a highly powerful technology for gene-expression studies. Microarray experiments are usually performed with DNA or RNA isolated from tissues, which are labeled with a detectable marker and allowed to hybridize to arrays comprised of gene-specific probes that represent thousands of individual genes.[21] The greater the degree of hybridization, the more intense the signal, thus implying a higher relative level of expression. The massive amount of data per experiment means that the molecular markers and their expression patterns need to be analyzed by elaborate computational tools, which add an additional layer of statistical complexity. Two basic forms of analysis are unsupervised and supervised hierarchical clustering algorithms;[22] the latter tools identify gene-expression patterns that discriminate tumors on the basis of predefined clinical information.[23] A third method, quantitative real-time PCR, is generally considered the 'gold standard' against which other methods are validated.
The cancer subclassification hypothesis states that gene-expression patterns identified with DNA microarrays can predict the clinical behavior of tumors.[24] The proof-of-principle for the cancer subclassification hypothesis has been provided for various malignancies, such as leukemias, breast cancers and many other tumor types.[25-31] For example, results from gene-array technologies have enabled breast cancers to be classified into prognostic categories dependent on the expression of certain genes. The 70-gene-panel microarray study of survival prediction led to the development of MammaPrint® (Agendia,
In 2005, Michiels et al. performed a metaanalysis of seven of the most prominent studies on cancer prognosis that used microarraybased expression profiling.[35] Surprisingly, in five of these studies the original data could not be reproduced.[36] The analysis of the other two studies provided much weaker prognostic information than given by the original data. The meta-analysis also indicated that the list of genes identified as predictors of prognosis was highly unreliable and that the molecular signatures were strongly dependent on the selection of patients in the training sets. This meta-analysis suggests that the results of the aforementioned studies are overoptimistic and that careful validation and larger sample sizes are needed before conclusions regarding the clinical utility of the data can be drawn. Despite promising proof-of-principle data, the successful use of gene arrays to discover novel subtypes of various carcinomas, and the utilization of these technologies for discovery of diagnostic markers, these new tools are not yet recommended for widespread clinical use by either organizations issuing clinical guidelines or expert panels.[37]
Proteomic-pattern profiling is a recent approach to biomarker discovery. Given that mRNA information does not best reflect the function of proteins, which are the functional components within organisms, the use of proteomic patterns to enable tumor diagnosis or subclassification seems more promising. The rationale is that proteins produced by cancer cells or their microenvironment may eventually enter the circulation and that the patterns of expression of these proteins could be assessed by mass spectrometry and used in combination with a mathematical algorithm for diagnostic purposes. Mass-spectrometry- based methods of proteomic analysis have improved and include more-advanced technology that brings higher mass accuracy, higher detection capability, and shorter cycling times, thereby enabling increased throughput and more-reliable data.[38] Technologies such as differential in-gel electrophoresis, two-dimensional polyacrylamide gel electrophoresis and multidimensional proteinidentification technology can be used for highthroughput protein profiling. The technology that has received considerable attention in the past involves the use of a minute amount of unfractionated serum sample added to a 'protein-chip', which is subsequently analyzed by surface-enhanced laser-desorption-ionization time-of-flight mass spectrometry (SELDI-TOF-MS) to generate a proteomic signature of serum.[39] These patterns reflect part of the blood proteome, but without knowledge of the actual identity of the proteins. The potential of proteomic pattern analysis was first demonstrated in the diagnosis of ovarian cancer.[40] In this study, exceptional results were seen, with a sensitivity of 100% (even for earlystage disease) and 95% specificity. These numbers are far superior to the sensitivities and specificities obtained with current serological cancer biomarkers. Subsequently, proteomic pattern analysis has been extended to a number of other cancer types, including breast, prostate, colon, liver, renal, pancreatic, and head and neck cancers.[41-47]
In spite of the optimism regarding this approach, a number of important limitations have been identified.[48] These shortcomings include bias from artefacts related to the clinical sample collection and storage, the inherent qualitative nature of mass spectrometers, failure to identify well-established cancer biomarkers, bias when identifying high-abundance molecules within the serum, and disagreement between peaks generated by different research laboratories.[49-51] Another limitation concerns possible bioinformatic artefacts. Baggerly et al. showed that signals that are detected that are a result of background noise can achieve a high level of discrimination between patients with cancer and those without.[52] Despite a substantial time lapse since the first report of this technology, no product has yet reached the clinic and no independent validation studies have been published. Guideline-developing organizations and expert panels do not currently recommend serum proteomic profiling for clinical use.[53]
The low-molecular-weight plasma or serum proteome has been the focus of recent attempts to find novel biomarkers.[54] Peptides are essential for many physiological processes, such as blood pressure (angiotensin II) and blood glucose (insulin) regulation. It has been suggested that "the low molecular-weight region of the blood proteome is a treasure trove of diagnostic information ready to be harvested by nanotechnology".[55] The low-molecular-weight serum proteome has been characterized by ultrafiltration, enzymatic digestion, and liquid chromatography coupled to tandem mass spectrometry[56,57] or via a top-down proteomics approach (whereby the intact peptide is distinguished directly by its fragment ions)[58] or by means of pattern profiling.[59] Informative diagnostic peptides that are generated after proteolysis of high-abundance proteins by the coagulation and complement enzymatic cascades can be identified by mass spectrometry. These proteomic patterns were claimed to distinguish not only controls from patients with cancer[60] but also between various types of cancer.[59]
One major consideration is that these peptides that are present in the serum are derived from a low number of high-abundance proteins. Koomen et al. studied peptides in serum and concluded that sample collection is of immense importance and could give rise to artefacts, and that serum is not ideal for proteomic experiments as it contains substantial endoproteolytic and exoproteolytic enzymatic activity.[61] These findings raise concerns regarding peptidomics data generated by profiling technologies. Peptidomic profiling might represent nothing more than peptides cleaved during coagulation or functions inherent to plasma or serum, including immune modulation, inflammatory response and protease inhibition.[62] Many of the aforementioned caveats associated with mass-spectrometry-based protein profiling technologies also apply to peptidomics.
The premise for the 'cancer biomarker family' approach is that if a member of a protein family is already an established biomarker, then other members of that family might also be good cancer biomarkers. For example, PSA is a member of the human tissue kallikrein family. Kallikreins are secreted enzymes with trypsin-like or chymotrypsin-like serine protease activity. This enzyme family consists of 15 genes clustered in tandem on chromosome 19q13.4.[63] PSA (KLK3) and KLK2 currently have important clinical applications as prostate cancer biomarkers.[64] Other members of the human kallikrein family have been implicated in the process of carcinogenesis and are being investigated as biomarkers for diagnosis and prognosis. For example, KLK6 has been studied as a novel biomarker for ovarian cancer.[65] It was found that elevated serum levels of this protein were associated with late-stage tumor, high grade and serous histotype, and with resistance to chemotherapy.[65] In general, increased levels of KLK6 were linked to decreased disease-free and overall survival, thus serving as an independent and unfavorable prognostic indicator. Similarly, KLK3, KLK5 and KLK14 have been shown to be increased in the serum of patients with breast cancer, thereby potentially serving as diagnostic markers. Being serine proteases, these proteins could be implicated in tumor progression through extracellular matrix degradation.
In theory, a candidate serological tumor marker should be a secreted protein, because secreted proteins have the highest likelihood of entering the circulation. Examination of tissues or biological fluids near to the tumor site of origin could facilitate identification of candidate molecules for further investigation. The increasing evidence that tumor growth and progression is dependent on the malignant potential of the tumor cells as well as on the microenvironment surrounding the tumor (e.g. stroma, endothelial cells and immune and inflammatory cells) further supports this approach.[66,67] A number of technologies can be utilized, but for systematic characterization of proteins in complex mixtures, mass spectrometry is the preferred technology. In the case of breast cancer, breast tissue, nipple aspirate fluid, breast cyst fluid, tumor interstitial fluid and breast cancer cell lines can all be explored. The tumor interstitial fluid that perfuses the tumor microenvironment in invasive ductal carcinomas of the breast was examined by proteomic approaches.[68] Over 250 proteins were identified, many of which were relevant to processes such as cell proliferation and invasion.
It should be noted that some of the widely used cancer biomarkers such as CEA, CA125 and HER2 are actually membrane-bound proteins, which are shed into the circulation. The identification of secreted proteins in tissues or other biological fluids does not necessarily imply that the proteins will be detectable in the sera of cancer patients. Serum-based diagnostic tests depend on the stability of the protein, its clearance, its association with other serum proteins and the extent of post-translational modifications.
A number of other strategies for detecting cancer biomarkers exist. One approach that is gaining popularity is based on protein arrays. Wang and colleagues have published data suggesting that autoantibody signatures might improve the early detection of prostate cancer.[69] Through use of a combination of phage-display technology and protein microarrays, this group identified new autoantibody-binding peptides derived from prostate cancer tissue. Another prevailing view is that tumor-associated antigens could serve as biosensors for cancer because tumors naturally elicit an immune response in the host. Moreover, breaking the cancer genetics dogma that hematologic malignancies result from chromosomal translocations[70] and that mutations underlie epithelial solid tumors, gene fusions as a result of translocations in prostate cancer have been identified through use of gene-expression data sets.[10] This translocation seems to be frequent (occurring in 40-50% of cases), may have prognostic value, and may be an early event in carcinogenesis. In addition, mass-spectrometrybased imaging of fresh-frozen tissue sections has yielded a number of potential candidate molecules.[71,72] Besides proteomic profiling of serum, attempts have been made to decipher the serum proteome via numerous fractionation schemes to simplify and reduce the dynamic range of molecules present in serum.[73] Finally, the use of animal models involving human tumor xenograft experiments has also shown promise for biomarker discovery.[74,75]
Biomarker development falls into five conceptual phases: preclinical exploratory studies; clinical assay and validation; retrospective longitudinal studies; prospective screening; and randomized control trials (Box 4).[76] Unfortunately, current studies of tumor markers are highly variable, not only in their methods of marker detection, but also in design and patient selection. Interpatient heterogeneity and intratumor heterogeneity are important confounding factors. In addition, the danger of bias and the problems of overfitting the data, as well as issues relating to the handling and storing of clinical specimens, are vital factors that need consideration before a study is conducted.[51] New tumor marker tests—single or multiparametric—must, therefore, undergo rigorous validation in order that their clinical value can be assessed.
New resources will most likely identify novel protein, genetic and low-molecular-weight cancer markers, which may impact on cancer care. Furthermore, with advances in genomic and proteomic technologies, human diseases may be classified on the basis of molecular rather than morphological analysis. Moreover, bioinformatics will serve to link scientific data with clinical information. Despite the optimism, ASCO and the National Academy of Clinical Biochemistry do not encourage the widespread use of tumor markers unless they affect patient outcome measures.[77] There is, however, a general agreement that a combination of multiple biomarkers may increase diagnostic sensitivity and specificity over use of individual markers. This is particularly important in relation to the recent development of powerful bioinformatic algorithms, which can interpret multiple parameters much more efficiently than can more-traditional approaches.[78,79] The most accurate, individualized, predictive assessment for patients might be attained through the use of artificial neural networks. There is no doubt that if these new technological advances prove to be successful in identifying cancer biomarkers for early cancer detection, the clinical benefits are likely to be enormous.
· Current cancer biomarkers suffer from low diagnostic sensitivity and specificity and have not yet made a major impact in reducing cancer burden
· The impressive growth of large-scale and highthroughput biology has resulted in increased popularity for the concept that novel biomarkers can be discovered through various emerging technologies
· A better understanding of the mechanisms behind biomarker elevation in biological fluids may facilitate the discovery of new tumor markers
· Some of the new promising strategies for biomarker discovery include microarray-based profiling at the DNA and mRNA level, and massspectrometry-based profiling at the protein or peptide level
· Study of tumor markers that include current biomarkers or examination of fluids and tissues that are in close proximity to the tumor might also assist in identification of novel tumor markers
· New tumor markers must undergo rigorous validation before they are introduced into routine clinical care