A Revolutionary Approach to Biomarker Discovery

While questions remain as to whether detecting diseases at their earliest stages actually improves health (see "Proteomics: Promise and Problems"), most agree that screening approaches have led to dramatic changes in the outcome of some cancers such as cervical cancer.² We see a future that expands on this promise: one in which diseases can be diagnosed in their earliest stages in hopes of effective intervention, but also where new classes of sensitive imaging technology (e.g., Spiral CT) are made more specific through combination with a blood-borne biomarker, and also where disease relapse is caught earlier through entirely new types of protein biomarkers. New technology now in development in our lab and in others worldwide should bring us closer to such a future, but will require nothing short of a revolution in the approach to clinical diagnostics.

In 2002, a flurry of reports from our laboratory as well as investigators in the Chan laboratory and the Wright/Semmes laboratory described using high-throughput mass spectrometry to interrogate the serum proteome in the search for disease-related information.^3-5Our laboratory emphasized the low-molecular-weight (LMW) range serum proteome, the peptidome, a previously unexplored reservoir. This method was first applied to ovarian cancer, and then later to other cancers, and to nonneoplastic diseases. These early studies revealed an apparent abundance of LMW information that potentially contains disease-specific information.

More recently, investigators have been sequencing and identifying the ions that comprise the key signatures described in the early profiling work. The surprising outcome of these efforts^6-12 is that the disease-specific ions appear to be fragments of large molecules and fall into one of two general categories: 1) fragments of endogenous high-abundant proteins, such as transthyretin; and 2) fragments of lowabundance cellular and tissue proteins, such as BRCA-2.⁷

Consequently, the peptidome is now emerging as a vast uncharted archive of diagnostic information that constitutes a new source of candidates for sensitive and specific biomarkers. It comprises a milieu of protein fragments generated as a consequence of cell death, cell-cell interactions, immune system function, enzymatic cascades, and metabolic processing. Thus, it may contain a "recording" of cellular and extracellular enzymatic events taking place at the level of the diseased tissue microenvironment.

While some dismissed the LMW range as noise, biological trash, or too small and unstable to be biologically relevant,¹³ we and others propose that just the opposite is the case.^3,6-12 LMW-range peptides are important in part because tissue proteins that are normally too large to passively diffuse through the endothelium into the circulation may be represented in the blood stream as fragments of the parent molecule.

The logical extension of this is that peptidome information is archived in at least three dimensions: 1) the identity of the peptide (e.g., the peptide sequence or parent protein from which it was derived), 2) the quantity of the peptide itself, and 3) the state of the modified form (fragment size and cleavage ends, posttranslational glycosylation sites, etc.). Peptide fragments are developed by the disease system and embody an integrated record of the system. Thus, taking a systems biology approach to measure panels of peptidome markers, we believe we can potentially overcome the failures of previous biomarkers to achieve adequate clinical sensitivity and specificity.

Cancer and many other human diseases are a product of the tissue microenvironment^14,15 (see The Peptidome Hypothesis). Interactions between the diseased or infected cells, the surrounding epithelial and stromal cells, vascular channels, the extracellular matrix, and the immune system are the ultimate determinant of the final pathology. Cell-surface antigens and receptors, cell-anchored and secreted enzymes, cytokines, and extracellular matrix molecules are the mode of communication between disease cells and the surrounding microenvironment. The outcome is a complex cascade of molecules available for sampling by the ongoing vascular perfusion.

The peptidome signature shed from the microenvironment is a reflection of the microenvironment as a whole. Thus, with regard to cancer biomarkers, cancer specificity is not derived from proteins secreted exclusively by tumor cells. The early^3-5 and ongoing¹¹ mass spectrometry (MS) profiling work indicated that a higher level of both specificity and sensitivity might be achieved by measuring the combination of markers emanating from both the diseased cells and the reactive cells in the microenvironment. In this way the peptidome can potentially supplant individual single biomarkers and transcend the issues of tumor and population heterogeneity.

Biomarker discovery is moving away from the idealized single, cancer-specific biomarker such as prostate specific antigen. Since 2002, a growing confluence of scientific data and results point to combinations of blood-borne markers using MS profiling techniques^3-5,11 as well as tissue MS profiling strategies,¹⁶ and multiplexed immunoassay¹⁷ providing more superior results than single markers alone. Despite decades of effort, most single biomarkers have not reached the level of cancer specificity and sensitivity required for routine clinical use in early detection and screening purposes. For example, in 2000, kallikrein 6 was described (along with an immunoassay developed) as being elevated in women with ovarian cancer.¹⁸ Yet, despite the potential for this marker, the analyte is still not in use at the clinic.

Biomarker panels could require measuring multiple-size isoforms of the same molecule. Recently, scientists have found that fragments of well-worn markers such as PSA, the most widely used cancer marker today, appear to have even better sensitivity and specificity than the intact molecule. They have proposed measuring these fragments as a multiplex panel.¹⁹ Poor transition of single markers to routine clinical use may be due to patient-to-patient molecular heterogeneity at several levels. Taking a cue from gene arrays, the hope is that panels of tens to hundreds of protein and peptide markers may transcend the heterogeneity to generate a higher level of diagnostic specificity. Combinations of peptide fragment-based markers should achieve a higher specificity and a higher sensitivity for the detection and monitoring of disease. This optimism is in part based on the concept that the biomarkers are derived from a population of cells more voluminous than the small precancerous lesion itself.

Of course, it is logical to ask how such tiny molecules exist at all in the circulation. Kidney filtration parses molecular information based on several factors with LMW molecules, usually containing the by-products of metabolism, being actively filtered out of the circulation. A concept that we initially presented in 2003 was that candidate peptide biomarkers could be amplified and exist in the circulation bound to molecules whose size resides above the cutoff value for glomerular filtration.⁶ Albumin, because of its relative abundance and long half-life, is a prime candidate. In 2004, using a mathematical modeling approach, we extended the concept to analyze ovarian cancer serum, and we found that some of the LMW ions reported in our initial Lancet publication appeared bound to albumin in a native state.¹⁹ In 2005, we generated initial data that gave substance to the reality of the albuminbound circulatory peptidome, whereby sequencing of the albumin-bound peptides found only in patients with earlystage ovarian cancer revealed fragments of low-abundance proteins that were never known to exist in the circulation.⁷

The LMW blood peptidome contains a wide range of protein classes potentially derived from all cells and tissues. In 2003, the Veenstra/Conrads laboratory at the National Cancer Institute (NCI) found a vast repository of LMW information in the blood⁸ by using ultrafiltration under conditions that disrupt peptide-albumin interactions. Other investigators had also made significant contributions to the use of the peptidome as a diagnostic during this time period.²⁰ Later, in 2004 the Veenstra/Conrads laboratory collaborated with us to expand these initial observations to evaluate the LMW information archive complexed with different high-abundance carrier-protein sources such as albumin and immunoglobulin (among others).²¹

These analyses revealed a vast and diverse source of LMW and low-abundance fragments of cellular proteins produced in vivo that were complexed with different carriers. Our own work in 2005,⁶ looking at albumin-bound ovarian cancer-specific peptide markers, identified a large number of candidate ovarian cancer-specific peptides. Sequencing these peptides indicates that fragments of low-abundance molecules such as BRCA-2 (proven by western blot of patient serum with antipeptide antibodies), tyrosine kinases, and signaling molecules, as well as intracellular scaffolding proteins exist bound to albumin in the blood.⁶

Earlier this year, the Tempst laboratory provided preliminary evidence that some components of the peptidome could be produced through ex vivo production of peptide fragments of components of the clotting cascade.¹⁰ Also earlier this year, Zheng et al. provided further evidence that the activity of proteases affected the peptidome.⁹ As the field begins to identify and characterize the full peptidome complement, the promise of the work lies in the creation of information that transcends whether a particular type of measurement technology, such as matrix-assisted laser desorption ionization time of flight (MALDI-TOF) mass spectrometry, is employed.

That said, the effective measurement of peptide-fragment biomarkers will require the development of new types of technology that can measure both the identity and fragment size of the biomarker. Conventional immunoassay platforms, and newer multiplexed technologies such as antibody arrays and suspension bead arrays cannot measure panels of peptide analytes that carry their diagnostic information based in two dimensions of both size and identity. This is because immunoassays, by their very definition, rely on antibody-based capture and detection. Unless a neo-epitope is formed, which is not predictable and not likely, this type of platform cannot readily distinguish the fragmented peptide molecule that contains diagnostic information from the "parent" molecule.

Likewise, while surface-enhanced laser desorption ionization (SELDI-) TOF platforms provided MS approaches that could generate size and identity information, the inability of surface-based retentate chromatography to provide a facile enrichment strategy for low-abundance analytes necessitates the development of new approaches of fragment-based analyte detection. One ideal format would be high-throughput MS technology coupled with true affinity chromatography whereby larger quantities of body fluids could be queried over a flow-through high-capacity surface.

Consequently, the future of peptide-based diagnostics will require the invention and adoption of wholly new technologies that rapidly read both the identity and the exact size of the molecule. Immuno-MS provides a means to do this. Using this technology, a microaffinity antibody column, perhaps in a multiplexed microwell format, is first used to capture all species of molecules that contain the antibody recognition site, regardless of size. MS analysis of eluted peptides provides an extremely accurate mass determination of the entire population of captured peptides. Thus, in only two steps, immuno-MS can rapidly tabulate a panel of peptide fragments derived from a known parent molecule.

We are now working with PerkinElmer to develop such a platform, based on the success of the orthogonal-TOF-configured MALDI instrument (prOTOF) they developed, in partnership with SCIEX, to specifically analyze albumin-bound LMW and other analytes.¹¹ The goal of this alliance between our university laboratory and PerkinElmer will be to develop an immunoassay-based application using a mass spectrometer as the detector rather than a fluorescence detector. Over the next two to three years, we will be assessing this approach in a setting that is College of American Pathologists/Clinical Laboratory Improvement Amendments (CAP/CLIA) compliant. The approach could provide a facile solution to the challenge of multiplexed fragment-based analyte measurements. Other types of configured formats, such as plasmon resonance-based affinity mass spectrometry,²² may also be highly successful in transitioning mass spectrometry to the clinic.

While its richness of information means that the blood peptidome can potentially reflect subtle disease events in a small tissue volume, it also means that the peptidome is constantly fluctuating owing to ongoing daily physiologic events. Epidemiologists and clinical chemists fear that the level of individual blood-borne peptidome biomarkers can be greatly influenced by a variety of epidemiologic factors and normal physiologic conditions. This includes stability of the peptidome in the collected blood sample. In 2005, a consortium of scientists participating within the Human Proteomics Organization reported on the impact of sample handling and storage vagaries, revealing the necessity of standardized and consistent methodologies.²³

Before any work on biomarker discovery begins, analysis research teams must work with an epidemiologist to develop a discovery subset that is matched in every possible epidemiologic and physiologic parameter. This includes age, sex, hormonal status, treatment and hospitalization status, clinic location, and any other variable that can be known. Reduction of the potential bias up front is critical before undertaking the discovery phase of the research.

The next stage is peptidome separation and MS-based identification. At this stage it is essential that iterative and repetitive MS sequencing be conducted on each sample. Candidate peptides identified repetitively over many iterations have a high likelihood of being correct. At the end of this process the researcher ends up with a list of candidate diagnostic markers that are judged to be differentially abundant in the cancer versus the control populations.

The next step is to find or make specific antibodies for each candidate peptide marker. After each antibody is validated for specificity using a reference analyte, the antibody can then be used to validate the existence of the predicted peptide marker in the disease and nondiseased discovery set samples.

The goal is to develop a panel of candidate peptide biomarkers along with measurement reagents that are independent of the analytical technology that will ultimately be used in the clinical lab. Our own efforts in this regard are best illustrated in our research on ovarian cancer biomarkers. In 2005 we published on the identification of a large collection of low-abundant and seemingly differentially displayed peptide fragments found only in patients with stage I cancer, as compared to late-stage cancer patients and high-risk control subjects.⁷

Clinical validation begins with choice and validation of the measurement platform. Once the immuno-MS platform is proven to be reliable and reproducible, then clinical validation could proceed rapidly owing to both the high-throughput attributes of mass spectrometry coupled with the knowledge that immuno-MS is simply an immunoassay that uses mass analysis as the detector. Fortunately, there is a well-worn and straightforward process for FDA approval and CAP/CLIA accreditation for laboratories performing immunoassay-based diagnostics. Challenges for immuno-MS-based approaches will include: automation of the entire process to achieve a cost-effective, high-throughput methodology; the sensitivity of the mass spectrometer and the antibody-capture efficiencies; antibody affinities and their ability to recognize the correct fragment isoform.

The final, and most critical stage of research clinical validation is blinded testing of the biomarker panel. Sensitivity and specificity in an experimental test population won't likely translate to predictive value seen routinely in the clinic. But the overall requirements of sensitivity and specificity for any biomarker (single or multiplex) are predicated on the intended use of the bio-marker.

For this reason ultimate adoption peptidome-based will be strongly dependent on clinical context its use in fact we believe approaches that set out to validate biomarkers detectors of early-stage diseases such cancer as screening the general population while noble are overambitious and a poor test ground for markers or new methodologies for the general population, while noble, are overambitious and a poor test ground for new markers or new methodologies.

We instead advocate an approach that first validates the potential of a given marker in the context of recurrence monitoring, or combining the marker with diagnostic imaging in a high-risk setting or when a suspicious mass is already detected.

While MS technology has undergone the predictable growth from research-based instrumentation to more reproducible formats, the field itself is evolving from patterns of unknown ions to fingerprints of known analytes. Now, armed with the information about LMW peptidomes and the knowledge that they are mainly endogenously bound to carrier proteins, we believe we have established a facile conduit for characterization of this important information archive. As new classes of clinical MS technologies emerge and methodologies such as immuno-MS become evaluated and developed for routine clinical use, we are more optimistic than ever that the emergence and newfound appreciation of the circulatory peptidome will have a positive impact on patient care.