-Omics and Cancer Biomarkers: Link to the Biological Truth or Bear the Consequences

Before cancer tissue and body fluid samples are collected, it is important to recognize sources of bias in the size and stage of the cancer that is sampled for analysis, and the selection of the trial cohorts and controls (Table 1). Many investigators fail to plan for the intended use of the biomarker, and thus omit the appropriate control cohorts. Tumor tissue samples often do not control for the grade, stage, and size of the tumor. Too often the pathologic stage, grade, or tumor size is not recorded in a biomarker study. Many published biomarker candidates are never validated in well-controlled independent human clinical study sets, particularly those in which serum/plasma was collected in an asymptomatic group of subjects that later developed cancer. The inadequacies of past body fluid sample handling methodologies and procedures have created anxiety in the community as realization has set in that many of our retrospectively collected study sets with long-term follow up are likely fraught with hard-wired biases due to inconsistencies in how samples were collected and stored from site-to-site and study-to-study (1, 2). Many investigators fail to plan for the intended use of the biomarker, and thus do not collect specimens from the appropriate control cohorts. If the intended use of a biomarker is to uncover early recurrence of a cancer, then using a biobank of serum collected from patients with early-stage disease is not appropriate. Healthy patient plasma or sera is not a sufficient control for modern cancer biomarker research because it does not account for the overarching inflammatory and benign cellular processes that are coincident within the tumor microenvironment. Consequently, the proper controls must include patients who are sick with noncancer illnesses or harbor benign tumors and must include patients with inflammatory illnesses. The aggregate of known inflammatory disorders are much more highly prevalent in most population compared to any single cancer type, thus discovery and verification efforts must be designed to maximize the chances that a given candidate biomarker has the necessary specificity in the intended population for a given intended use. Failure to do so often has mortal consequences for the candidate marker as because it would have to be revalidated in population cohorts that take into account the intended use of the biomarker, adding tremendous cost and time.

Biobanking efforts are only as good as the design of the tissue banking itself and the intended use of the candidate biomarkers that are generated from the samples. Biomarker discovery efforts focused on uncovering mechanisms of aggressive tumor behavior or premalignant progression require tissue that contain/reflect that phenotype with adequate follow-up. Indeed, hard-wired bias can creep into tumor study sets simply by the analytic/technical requirements of the assays that will be used to study the tissue. Efforts such as The Cancer Genome Atlas (TCGA; ref. 3) are seeking to generate genomic, transcriptomic, and functional proteomic analysis on hundreds of tumor specimens with the hope that we can link together mutational data to functional molecular determinants and elucidate the “tumor circuitry.” In this time of tremendously constrained budgets and tight funding lines, these large-scale efforts require a proportionately huge amount of taxpayer money. Consequently, although these efforts are extremely exciting, we should be cautious and circumspect about maximizing the quality of science supported by every dollar. The TCGA effort uses broad-scale multiplatform molecular profiling efforts that require substantial tissue, and thus TCGA tumors are nearly all large tumors. This predetermination then hard-wires a tumor microecology bias into a high-profile molecular profiling effort. Of course, one of the best predictors of outcome is tumor size, regardless of treatment, and a biobank comprised solely of large tumors and biomarker candidates identified from such a set are then constrained in their potential impact. Although large tumor samples provide a high number of cells for analysis, this may bias the study because large tumors may have a different prognosis associated with their size or growth rate and may not reflect the most common clinical presentation for the cancer being studied. We now know that the molecular architecture of the primary tumor is much different than metastasis (4, 5), and because metastasis is the lethal aspect of the disease and cancer evolves from microscopic premalignant lesions, study large primary tumors will likely tell us little about either end of the tumorigenic spectrum. It is imperative that effective biobanking initiatives take into consideration and are matched with the intended biomarker discovery/validation effort.

Sample collection, handling, storage, and processing can significantly introduce bias into biomarker discovery and validation. With regard to tissue, the effects of preanalytic variables on the tissue microenvironment are often ignored. On the basis of recent publications (6–9), we now know that gene transcription, protein, and phosphoprotein levels change during the tissue collection and processing workflow before fixation and cell death. During the time period, the tissue is alive and many of the very biomarkers that we may use for early detection, prognosis, and prediction can change as a consequence of cellular hypoxia, acidosis, and so on, which occurs as the tissue is sitting on the pathologist's cutting table or as the surgical procedure is taking place. We need to now incorporate new workflows for rapid tissue collection so that prospective biobanking can take into consideration what the recent data is revealing while at the same time identify surrogate markers of tissue fidelity so that we can adequately assess the retrospective banks we currently have. A key component of these workflows will depend on rapid fixation and preservation of the molecular architecture of the tumor microenvironment. Formalin fixation of tissue is a century-old solution, which is incompatible with the current molecular profiling opportunities. Many retrospectively collected materials with attractive long-term clinical follow up should not be used for cancer biomarker work because of the methods used to collect the material. The slow (mm/hr) penetration time of formalin, well-known cross linking, and associated antigen retrieval issues greatly limit the molecular analysis and produce inaccurate data and are simply not acceptable. Just because we can generate data from these sets, does not mean we should and it can take years to realize that the data is not accurate after extensive validation. We urge caution in this regard.

Fortunately, there are new molecular fixatives (10, 11) and tissue-processing methods (12) that are being developed to eliminate or greatly minimize these sources of preanalytic variables. These new tissue fixatives are being incorporated and evaluated in large national biobanking efforts and specifically provide the molecular preservation equivalence of snap-frozen tissue so that labile biomarkers can be accurately measured concomitant with formalin-like histomorphology and detail (10). Incorporation of these fixatives into prospective rapid tissue procurement-based biobanking could greatly minimize the effects of hard-wired bias for tissue biomarker research.

Even if the tissue is collected and preserved properly, and the biobank designed to take into consideration all of the aforementioned issues, tissue and cellular heterogeneity is a major source of false positives and false negatives in the field of biomarker discovery. Recent publications (13) have revealed that a different mutational landscape is found depending on what part of the tumor piece you sample and points to significant inter- and intratumoral heterogeneity. At this time, from a biomarker standpoint it is not clear what the clinical impact of tumor heterogeneity is, or even if the DNA mutation differences seen because of sampling are simply due to the lack of knowledge of which mutations are the causal driving mutations and which are passenger mutations. The need to eliminate bias caused by tissue cellular heterogeneity has certainly been recognized for over 15 years (14), and this problem has become more acute recently because of the surge of interest in the role of the tumor microenvironment (15–17) and potential for stromal therapy (18, 19). Any given piece of tumor often contains a hodgepodge of tumor epithelium, normal appearing epithelium, nerve cells, immune cells, extracellular matrix, stromal cells, and vascular cells. We now know that all of these cell types contribute to the growth, invasion, and metastasis of the tumor mass. Importantly, all of these nontumor cells can produce biomarkers at a level similar to, or higher than, the tumor cells. When bulk tumor tissue is extracted, the cellular source of the biomarker, and the proportion of cellular subpopulations are unknown. Consequently, a marker elevated in the tissue extract may reflect the contribution of non-tumor cells. Conversely, biomarkers derived from the unknown proportion of tumor cells can be obscured by dilution. Silvestri and colleagues (20) and Wulfuhle and colleagues (21) have clearly shown that tumors with an equivalent high proportion of carcinoma cells can have a marked difference in total biomarker levels because of the contribution of the cellular stroma. “Grind-and-bind” type of tissue analysis may reveal some tumor architecture but its likely more akin to understanding the topography of a landmass by studying the shape and detail of the highest mountaintops. Given the recent findings that simply having tumors with differing stromal content have different outcomes, this issue is of great clinical importance. To truly understand the manifestation of genomic mutations in tumorigenic progression and metastasis, and map the aberrant and druggable protein signaling networks and then ultimately personalize therapy for a given patient, we will need to reorient our approaches to embrace the tissue microecology not ignore it, or pretend it is not that important. Many of the new molecular targeted therapies are directed against hyperactivated protein signaling networks, and the many of the drug targets within the implicated networks such as PI3K-AKT, RAS-ERK, JAK-STAT, and so on are ubiquitously expressed both in tumor epithelium and stromal compartments (endothelial cells, immune cells, fibroblasts, etc.). As we identify means of stratifying patients based on the underpinning molecular signature of their tumor and measure these drug target activities—the absolute onus is on us to get the measurement right. Fortunately bias because of tissue and cellular heterogeneity can be minimized by upfront tissue sample preparation, such as the routine use of Laser Capture Microdissection (LCM; refs. 14, 22, and 23). LCM, already well-established as a reliable and commercially available tool, is now becoming even more important as a means to study the biomarker content of individual subpopulations within the tumor microecology. Armed with LCM, we can now meet the challenge of the tumor ecology head-on. It is now routine to microdissect individual subpopulations of tumor epithelium, premalignant lesions, normal appearing epithelium, different stromal compartments, immune cells, and so on—all at once, from the same specimen, and study the biomarker content of each. Functional marker analysis of tumor microenvironment subpopulations by microdissection will be crucial for developing the next generations of immunotherapy and stromal therapy. This information is lost when the tissue is ground up and analyzed as a bolus sample.

Another key issue that is impeding progress in cancer biomarker translational research is the likely extremely low abundance of cancer biomarkers in the blood and body fluids and the minute amount of tumor cell material that is often obtained in clinical biopsy material. The issue of biomarker concentration is especially problematic in the very area that is one the most critical for cancer biomarker research: the study of premalignant progression and early-stage cancer detection. Protein biomarker discovery and quantitation by MS and multiple reaction monitoring (MRM) are powerful approaches to protein and metabolite biomarker research (24, 25) but are severely limited in their practical application for complex clinical samples because of their low effective sensitivity (26). The analyte detection sensitivity for MS or MRM applied directly to a complex body fluid is typically greater than 50 ng/mL (27). In contrast, the vast majority of diagnostic analytes measured in the clinical laboratory by immunoassay platforms fall in the range between 50 pg/mL and 10 ng/mL (2). Many cytokines and chemokines have even lower concentrations in the pg/ml range. Thus, the most important protein biomarkers, particularly those derived from early-stage disease are mostly invisible to conventional MS or MRM. MS and MRM lack practical sensitivity because of technical and physiologic constraints. The volume of sample introduced into the MS is very small (often <10 μL, depending on concentration of input sample) containing only a small number of target analyte molecules, thereby limiting the detection sensitivity. In addition, low abundance, low molecular weight proteins and peptides are masked by billion fold excess quantities of resident proteins such as immunoglobulin and albumin. Consequently, increasing the sensitivity is not simply a matter of concentrating the sample because this will overwhelm the total protein capacity introduced into the MS.

Tissue biomarker discovery and measurement is also hampered by low biomarker concentrations, and these issues are underpinned by the very low number of premalignant and tumor epithelial (carcinoma) cells in a given tissue specimen. Premalignant lesion and prevention-oriented biobanking efforts are increasing and certainly personalized therapy-based clinical trial sample collection efforts are expanding rapidly. Because of these opportunities, the banked pieces of tissue obtained from these efforts are mainly small-bore core biopsy and fine needle aspirate-sized samples—not large primary tumor pieces. Moreover, even within these tissue samples, the tumor epithelium content is often very low with much of the specimen being stroma, fat, or noncellular material. Thus, for translational and clinical specimen efforts, the total cellular input is often in the order of thousands to tens of thousands of cells, and not the millions to billions of cells that one routinely obtains in cell culture. Consequently, many of the MS analyte discovery and measurement methods that show such promise in cell culture biomarker work (28), cannot adequately measure low abundance tissue markers in these small-sized clinical specimens. Technologies such as the Reverse Phase Protein Microarray (RPPM) because of its tremendous analytic sensitivity (hundredths of cell equivalents per spot (29, 30) provide the opportunity to exploit these distinguishing characteristics and generate the necessary multiplexed pathway biomarker data even from LCM procured cells from core needle biopsy samples (21) and fine needle aspirates (29, 31).

Beyond low abundance, a still further barrier to biomarker discovery is the lability and perishability of candidate biomarkers ex vivo after clinical sample collection. Diagnostic proteins and peptides in body fluids are subject to rapid enzymatic degradation after collection (32). Moreover, when the level of biomarker is close to the level of detection of the analytic device, then by definition, CVs increase and error rates go up. What is needed for more effective biomarker discovery and measurement is to increase the level of the biomarker so it is squarely within the dynamic range of the assay method and thus, not only can it be detected in the first place but also is also more accurately and precisely measured. These efforts are distinct from past depletion and fractionation workflows that attempt to parse high abundance blood and body fluid proteins such as albumin and immunoglobulins away from the rest of the biomarker constituents so that MS can measure the lower abundance proteins that are masked by high-abundance analytes (33). With these methods, low-abundance analytes have not been concentrated and remain at levels still below the level of detection: fractionation and depletion does not equal concentration.

To overcome these concentration barriers, investigators have turned their attention toward upfront sample preparation techniques that concentrate the biomarker content from the outset. Immuno-MS methods (34, 35) use specific or class-specific antibodies (or even chemistries that preferentially bind specific classes of molecules such as phosphoprotein as discussed previously) to concentrate markers from larger volumes of body fluid samples before MS. Immuno-MS is a powerful enrichment process, but that technique has its best applications after discovery efforts have ended because you have to have identified the biomarker candidate ahead of time to generate the analyte capture molecule. New types of biomarker harvesting nanoparticles have been invented, which provide a rapid and facile means of binding and concentrating broad classes of analytes for broad-scale biomarker discovery efforts (36, 37). These porous, buoyant, core-shell hydrogel nanoparticles contain novel reactive chemical baits that can bind proteins, peptides, nucleic acids, hormones, and so on for concomitant harvesting, concentration, and preservation of cancer biomarkers in body fluids. The porous nature of the hydrogel particle shell has an effective pore size that acts as a molecular sieve with a cut-off determined by the cross-link density. Biomarkers small enough to enter the particle are captured by the high-affinity bait. The high affinity permits very rapid sequestration of target analytes and dissociation from carrier proteins. These nanoparticles effectively protected highly labile proteins such as interleukins and growth factors from enzymatic degradation in blood, sweat, and urine, and massively increased the effective detection sensitivity, while improving the precision, of multiple reaction monitoring analysis (37). Used in whole blood as a one step, in-solution preprocessing step, the nanoparticles greatly enriched the concentration of low molecular weight proteins and peptides while excluding unwanted albumin and immunoglobulins; this achieved a 10,000-fold effective amplification of the analyte concentration, enabling MS discovery of hundreds candidate biomarkers that were previously undetectable in the ng/mL and pg/mL range (37).

Assuming that analytic variables are minimized by the method of sample collection and the correct study design, if the biology is ignored, then the conclusions of the biomarker analysis may be difficult to validate, or may not have any clinical impact. Investigators should strive to mechanistically tie the presence of a candidate marker to the biology of the tumor itself. Reliance on correlation without causality diminishes confidence in the candidate marker(s) and makes it extremely difficult to understand its true clinical use. Linking the biomarker to a functional role in cancer premalignant progression, growth, invasion, and so on, and have the capability to study the biologic/biochemical effect of overexpressing or underexpressing the analyte in experimental models can provide greatly increased confidence in the use of the marker for early detection, high-risk screening, recurrence monitoring, or individualized therapy. The reliance on correlative linkage alone impacts on downstream aspects of biomarker development and implementation. For example, most -omic-based discovery efforts lead to the identification of dozens to hundreds of candidate genes, proteins, metabolites, and so on and without a firm biologic basis for ranking and prioritization, a rational approach to verification and validation is stymied and often the effort burns out at that step leaving the field with yet another paper on a promising signature/multi-omic fingerprint, without proper follow-up validation and the finding never reaches the bedside. Recent publications (38) have revealed the utility of linking the biology to biomarker discovery, which provide a good roadmap for these types of endeavors.

Tying the biology to tissue marker analysis can be enabled by focused approaches that are based on known biology and biochemistry. For example, an attractive attribute for functional biomarkers, such as phosphoprotein/pathway biomarkers, or isoforms-specific products of enzymatic activity such as cleavage products (e.g., cleaved PARP), is that the biomarker is very often a direct read-out for a specified biologic process that is being targeted, or is the drug target itself. Monitoring the effect of a kinase inhibitor, say for the mTOR proteins, may be accomplished by a simple assay that measures the phosphorylation of a known mTOR kinase substrate such as p70S6 kinase. In this example, stratification for an mTOR inhibitor trial might be accomplished by measuring the level of phosphorylated p70S6 instead of a multigene signature that is many degrees of separation removed from the underpinning biologic/biochemical process. Pathway-driven biomarker discovery efforts, focused on upfront posttranslational modification enrichment techniques such as phosphoprotein capture followed by high-resolution MS can be an effective way of rationally exploring a biomarker discovery effort based on the known importance of deranged cell signaling pathway activation in tumorigenic processes (39). As discussed, broad-scale pathway activation mapping using techniques such as the RPPM provide a means to quantify the activation/phosphorylation state of hundreds of key signaling proteins and pathways at once from clinical material. Thus, both de novo discovery efforts and focused profiling efforts can be rationally focused, a priori, around biochemical processes that are directly tied to the underpinning biology of the tumor which can greatly assist in translational research efforts and speed validation and clinical implementation of the biomarker(s).

Although the field of -omic-based characterization and analysis of tumorigenesis and progression is set to unlock the mysteries of cancer like never before, there are considerable barriers that must be overcome to truly realize the promise set before us. We have a growing suite of powerful technologies that give us an unmatched view of the molecular landscape of cancer and we can generate huge databases of -omic data in hours. Cancer biomarkers hold the key to effective managed health care and are the gatekeepers of personalized medicine. All of these efforts, however, rest on the shoulders of the weakest links in the entire process. Often times these weakest links are simply ignoring the biologic reality of the tumor microenvironment, which is then manifest in the way the biobanks were designed in the first place, the lack of tumor cell enrichment, the way the tumor and/or body fluid was procured and processed, and missing evidence supporting a link between the biomarker and the cancer. Ignoring the biology will jeopardize and squander the opportunity given to us.

L. Liotta and E. Petricoin III are co-inventors on technologies that are discussed in the manuscript, and which have been assigned to either the U.S. Government and George Mason University. As such, L. Liotta and E. Petricoin III can receive royalties and licensing fee distributions. L. Liotta and E. Petricoin III are equity interest holders and cofounders of Theranostics Health, Inc. and Ceres Nanosciences, Inc., which have exclusive license to some of these technologies. E. Petricoin III has Employment (other than primary affiliation; e.g., consulting) in Theranostics Health, Inc., as the Chair of the Scientific Advisory Board and in Ceres Nanosciences as one of the Board of Directors.