Given the paucity of new clinical tests emerging from proteomics research on the one hand (1) and the large and increasing levels of NIH funding devoted to biomarker research on the other (2), important questions have been raised about our strategy for translating basic research into clinical diagnostics. At a fundamental level, it has been unclear whether the lack of clinical results from numerous biomarker discovery efforts is due to our having exhausted the low-hanging fruit (or all fruit?) in years past, or to defects in the approaches being used in the search for new markers. It seems unlikely that all useful clinical protein tests have been discovered; therefore, discussion has focused on the defects in the “biomarker pipeline” that should be conveying candidate biomarker proteins toward a clinical application. The ultimate importance of this debate is not academic: Improvements in clinical care and in basic health economics increasingly depend on the development of new tests (including companion diagnostics that allow personalization of therapy) and markers addressing the long-standing list of unmet clinical diagnostic needs in such areas as Alzheimer disease, chronic obstructive pulmonary disease, and cancer.
Fortunately, recent publications (3, 4) suggest that progress is being made in implementing biomarker pipeline concepts that work: approaches that link a “discovery” process that uses a small number of particularly favorable samples with one or more “qualification” and/or “verification” steps that focus on testing the resulting candidate biomarkers with independent sets of real clinical samples. The work of Addona et al. (3) serves as a useful benchmark in this context because it is focused on a clinical indication (cardiac damage markers) in which numerous proteins, such as creatine kinase isoenzyme MB, troponins, and B-type natriuretic peptide, have achieved undisputed clinical success. One can thus ask whether such a modern systematic approach can find the previously identified markers and whether it can extend them by finding new markers of similar or potentially better performance.
Addona et al. began their discovery work with especially favorable samples for detecting cardiac damage: blood samples collected through a catheter in the coronary sinus of 3 patients immediately before and at 2 time points after (10 and 60 min) a planned destruction of heart muscle caused by alcohol injection. This planned myocardial infarction is a successful treatment for hypertrophic cardiomyopathy and provides a carefully controlled human model of a spontaneous myocardial infarction—one in which each patient provides his/her own control samples, thereby minimizing the impact of interindividual differences.
In the discovery phase, approximately 1000 different proteins were identified by using a powerful mass spectrometer system for the confident detection of 2 or more constituent peptides in fractionated trypsin-digested plasma. Their relative concentrations in the 9 samples were estimated by label-free mass spectrometric methods. Concentration changes of at least 5-fold among the time points were observed for 121 of these proteins, which included a number of well-known cardiac markers, such as creatine kinase isoenzyme MB, myoglobin, myeloperoxidase, fatty acid–binding protein, and cardiac troponin T (this last marker was barely detected, on the basis of only 1 peptide in 2 of the 3 patients). These results established that the discovery process was capable of seeing known markers with high probability and suggested that the novel candidates were worth further examination.
In a second, “qualification” phase, peripheral plasma samples from 10 new patients with planned myocardial infarction procedures were pooled at each of the 3 time points, and the resulting 3 samples were analyzed with a targeted “accurate inclusion mass screening” strategy that looked for 1904 peptide features representing the 121 protein candidates. Two or more peptides were observed at one or more time points for 83 of the proteins, indicating that they could be detected in peripheral plasma at some point in the time course. Of these proteins, 52 showed a significant correlation between the discovery and qualification profiles across the 3 time points. This filter not only independently confirmed the results of the discovery process with different samples but also demonstrated that the planned myocardial infarction–related proteins observed in coronary sinus blood were still detectable after dilution into the much larger volume of circulating blood.
A third, “verification” phase was undertaken to further confirm the results obtained to this point by the use of semiquantitative mass spectrometric methods (i.e., lacking labeled internal standard peptides of the same sequences). The real challenges of the biomarker pipeline emerged at this stage. Of the 52 candidate proteins that could be pursued, only about 20 could be measured quantitatively by any of 3 methods: 6 with commercial or homemade ELISA assays, 7 by using single antibodies to stain Western blots, and 7 by highly specific multiple reaction monitoring assays carried out with internal standard peptides on triple-quadrupole mass spectrometers. Thus, the selection of the candidates that moved forward from the qualification stage was based primarily on the availability of preexisting antibody reagents, which are limited and of questionable performance in many cases. A relatively small number were pursued with quantitative mass spectrometry. Each group of assays was used in a different small set of samples. Use of this heterogeneous collection of quantitative and semiquantitative assays allowed only about 40% of the candidates to be tested.
Finally, a small subset of 6 candidate proteins (<12% of the 52 qualified candidates) were measured by immunoassay in several somewhat larger clinical cohorts totaling approximately 100 patients, including some undergoing routine cardiac catheterization without myocardial infarction, patients experiencing a spontaneous (real) myocardial infarction, and patients undergoing an exercise stress test (including groups with and without reversible perfusion defects). These clinical studies showed that the concentrations of some biomarker candidates were increased during spontaneous myocardial infarction and exercise ischemia, thus strongly suggesting a clinical association with cardiac muscle damage or stress, whereas others were increased by the catheterization procedure alone and hence were indicative of other processes— critical insights in terms of their potential clinical utility. The initial stages of the effort, carried out efficiently by highly experienced staff, used 2800 h of time on an expensive (approximately $750 000) mass spectrometer funded by an NIH grant that totaled approximately $2.2 million over 3 years.
A similar overall pipeline structure with similar results and cost parameters is discussed in the separate concurrent publication of a study by Whiteaker et al. (4), in which protein biomarkers of breast cancer were evaluated in a mouse model system. In this study, the discovery samples were also carefully chosen (a well-characterized mouse model of breast cancer). A lengthy list of 1144 initial candidates was triaged to 118 candidates by means of accurate inclusion mass screening and semiquantitative mass spectrometry. Finally, 91 of these candidates were examined in larger sample sets with specific assays. In this case, a consistent general approach was applied for constructing specific verification assays. This approach used quantitative mass spectrometry with labeled same-sequence internal standards—either directly (a first set of 57 multiplexed assays) or with sensitivity enhancement by specific immunoaffinity capture (a second set of 31 multiplexed assays)—to yield a homogeneous analytical data set (triplicate determinations across 80 samples). The use of a highly multiplexable assay methodology in this scheme allowed the measurement of many analytes together in small sample aliquots. Using individual assays requiring separate sample aliquots would consume much more total sample, thereby severely limiting the number of candidates that could be subjected to verification with typical sample volumes. Instrument time, instrument types, and project costs were comparable to those reported by Addona et al.(3).
These 2 biomarker pipelines—the first of which focused on a human clinical situation in which known good biomarkers were replicated and extended, and the second of which used an animal model of a disease for which no good clinical biomarkers exist for humans—were both successful in demonstrating novel protein changes that were confirmed in peripheral plasma via multiple analytical approaches. Neither, however, approached the end of the biomarker development process, and thus further work is needed to provide the groundwork for definitive clinical studies. In particular, the work on cardiac markers that Addona et al. have reported thus far did not encompass the development of specific multiplexable assays for its 52 primary candidates—a basic requirement for clinical validation of these candidates in thousands of clinical samples. The publication by Whiteaker et al. (4) on breast cancer did include the development of high-quality mass spectrometry–based assays for 88 of the 118 candidates. Nevertheless, because of the numerous differences in sequence between mouse and human proteins, these candidates remain to be translated into the cognate human assays required to test the candidates as potential biomarkers of human breast cancer.
At this point, we thus have at least 2 broadly similar road maps that lay out effective biomarker pipeline strategies extending well beyond simple ‘omics discovery experiments, with results that justify confidence that biologically real protein differences exist in peripheral plasma. The remaining steps toward clinical validation are still challenging, however. A very crude estimate of the cost to generate and apply high-quality mass spectrometry–based specific assays for approximately 50 to 100 good candidates would be $1.5 million and 1 year in the case of either of these 2 studies, suggesting that a full pipeline exercise would cost roughly $4 million and require about 4 years. That seems a very reasonable cost to reach a true clinical validation threshold for 50 to 100 good candidates, especially considering the much larger amounts of money that have been spent on inconclusive biomarker ‘omics studies to date. Although the challenges of securing grants of this size to attack clinically important biomarker problems are substantial and although the process is slow, the segment of the biomarker pipeline within academic territory will not be complete until investigations like those examined here lead to the development of high-quality multiplexable assays for a complete spectrum of the verified biomarker candidates, ideally in a format that makes them generally available to clinical researchers in a short time frame.
Author Contributions: All authors confirmed they have contributed to the intellectual content of this paper and have met the following 3 requirements: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; and (c) final approval of the published article.
Authors' Disclosures or Potential Conflicts of Interest: Upon manuscript submission, all authors completed the Disclosures of Potential Conflict of Interest form. Potential conflicts of interest:
Employment or Leadership: Leigh Anderson, SISCAPA Assay Technologies and Clinical Chemistry, AACC.
Consultant or Advisory Role: None declared.
Stock Ownership: Leigh Anderson, SISCAPA Assay Technologies.
Honoraria: None declared.
Research Funding: None declared.
Expert Testimony: None declared.
- Received for publication October 10, 2011.
- Accepted for publication October 12, 2011.
- © 2012 The American Association for Clinical Chemistry