Proteomics is about measuring proteins, and ideally, about measuring many of them. Not surprisingly, given the universal importance of proteins in biological systems and the variety of their properties when employed as machines, signals, structural materials, and food, proteomics has enabled real progress in many areas of biology while falling short in others.
Fortunately, we are beginning to understand the differences between these cases, and have begun to adapt the technology and the way we use it to address the harder questions successfully.
For simplicity, I refer here to proteomics in the service of basic biology as Type 1 proteomics, and to population proteomics (mainly clinical applications) as Type 2. My belief, outlined in this article, is that we are at an inflection point in terms of what Type 2 proteomics can deliver in clinical applications.
Type 1 proteomics is making major contributions toward illuminating biological model systems for clues to basic mechanisms. Phosphorylation, glycosylation,
and a host of other mechanistically important post-translational modifications have been revealed and made measurable. Increasingly, complete inventories of proteins are being generated from matched samples (e.g., SILAC comparisons of cells under perturbation or label-free analyses of disease vs. normal clinical samples). These results will provide the foundational data needed for a real systems biology—one that incorporates experimental data exposing the operation of complex cellular regulatory mechanisms.
However, the price that has been paid to achieve deep proteome coverage has been the limited number of samples subjected to analysis, which represents a tradeoff that works well in pursuit of widely conserved, basic biological mechanisms, but necessarily slights the complexities of how such mechanisms behave in nonidentical individuals.
In a sense, the strengths and weaknesses of this approach to basic biology mirror those involved in the use of biological model systems generally: they represent a simplification needed to provide a clear view of some biological mechanism, but don’t necessarily ensure that the mechanism operates similarly “in the wild”.
Proteomics has had less success so far in finding clinical biomarkers, a classical Type 2 application and an area of biology in which population heterogeneity is a key limitation. Essentially, the quest for biomarkers is a search for biological mechanisms that are invariant, or nearly so, across a real population of individuals.
In principle, all mechanistic discoveries from Type 1 work can be considered candidate biomarkers to the extent that the mechanism in question is related to disease, drug treatment, etc.
Published biomarker studies have identified some disease association for almost 25% of the 20,000 human proteins. Unfortunately, none of these prospective “discoveries” has yet been confirmed at the level required to achieve FDA clearance for a clinically useful protein test.
That proteins can be excellent clinical biomarkers is indisputable: 109 proteins are measured by FDA-cleared tests and another 96 by generally available laboratory-developed tests (homebrews) through the efforts of a multibillion dollar in vitro diagnostics (IVD) industry. Cardiac troponin, for example, when measured in the blood, is the clinical definition of a heart attack.
Yet the rate at which new protein analytes are cleared has remained flat at about 1.5 new proteins per year for the last 15 years, much lower than the rate during the initial wave of monoclonal antibody-driven discoveries and far less than the number required to address critical diagnostically underserved indications such as Alzheimer disease, COPD, and stroke.
Why is it so difficult to find biomarkers that work? As we have learned in drug development, most disease mechanisms involve complex but important differences among people, leading to the inconvenient finding that most drugs provide intended levels of benefit to a minority of patients. Hence, the large clinical trials that dominate the cost of late-stage drug development. Biomarker proteomics has now encountered the same “statistical barrier.”
Unfortunately, overcoming this barrier has proven quite difficult using the favored tools of deep coverage (Type 1) proteomics, with their high cost per sample and limited quantitative precision.
What should be the central component of the “biomarker pipeline” is missing: an easily accessible capacity to accurately measure, in large clinical sample sets, the candidate biomarkers emerging from proteomic (or genomic) studies.
Werner Zolg crystallized this requirement by pointing out that good analytical data from at least 1,500 samples is required to support a convincing case for serious, i.e., commercial, interest in any protein biomarker. Of the thousands of papers in biomarker proteomics, I can only think of one that involved more than 1,000 samples. All the rest fall short of the Zolg number.
This means that the biomarkers “discovered” in these studies have not been tested to a level that establishes real clinical utility (often referred to as “verification”). Absence of such data leaves us speculating as to the fraction of published candidates that ultimately ought to find use in medicine, but a persuasive case can be made that the failure rate is greater than that of drug candidates going into Phase I trials, and probably exceeds 95%.
Clinical verification of new protein biomarkers is constrained by several factors, including lack of grant funding available to “confirm the discoveries of others” and, until recently, the lack of a suitable technology base. Immunoassays, the default method of high-throughput protein quantitation, are difficult and expensive to construct and more difficult to multiplex in a reliable fashion as required in large-scale candidate verification.
Mass spectrometry has now emerged as the favored path for development of the targeted assays required for Type 2 research, largely as a result of applying to peptides the multiple reaction monitoring (MRM) technology long used by analytical chemists for quantitation of smaller molecules.
MRM measurements provide near-absolute structural specificity, true internal standardization and flexible multiplexing, none of which is available in conventional immunoassays.
MRM has also overcome one of the long-standing criticisms of proteomics—reproducibility. At one point, in the wake of the SELDI debacle, it was believed, especially in the genomics community, that the methods of proteomics were simply not reliable enough to get the same result in different labs.
Multilaboratory efforts, largely spearheaded by the NCI’s CPTAC program, have now shown that peptide MRM measurements are accurate and consistent across different labs and instrument platforms, as analytical chemists knew they would be.