Casting doubt on the traditional approach of cancer biomarker discovery through proteomics

Abstract

Towards revolutionary biomarkers, a considerable amount of research funds and time have been dedicated to proteomics. Although the discovery of novel biomarkers at the dawn of proteomics was a promising development, only a few identified biomarkers seemed to be beneficial for cancer patients. We may need to approach this issue differently, instead of only extending the conventional approaches that have been used historically. The study of biomarkers is essentially a study of diseases and the biochemistry relating to peptide, protein and post-translational modifications is only a tool. A problem-oriented approach should be needed in biomarker development. Clinician participation in the study of biomarkers will lead to realistic, practical and interesting biomarker candidates, which justify the time and expense involved in validation studies. Although discussion in this article is focused on cancer biomarkers, it can generally be applied to biomarker studies for other diseases.

Keywords::

• biomarker
• cancer
• proteomics

An inconvenient situation for cancer biomarker discovery through proteomics

Cancer is a major cause of disease-related death in developed countries. Early diagnosis, the ability to predict treatment response and evaluation of the prognosis will optimize the therapeutic strategy and improve the clinical outcome of cancer patients. The use of biomarkers is a fundamental tool in achieving this goal. However, existing biomarkers are only beneficial in limited cases, and novel biomarkers have long been desired for almost all cancers in various clinical situations. In anticipation of the emergence of revolutionary biomarkers, a considerable amount of research funds and time have been dedicated to proteomics during the last decade [1]. Biomarker development has aided the progression of proteomic technologies, and, in fact, the most progressive technologies are applied to the study of biomarkers. Therefore, biomarker development has been a driving force in proteomics as well as the benchmark for novel proteomic technologies.

The discovery of novel biomarkers at the dawn of proteomics was a promising development [2], and literally, thousands of papers about the identification of biomarker candidates using proteomics were published. Unfortunately, only a few identified biomarkers seemed to be beneficial for cancer patients [3], and the actual benefits achieved by the biomarkers still require evaluation. Therefore, we have an opportunity to reconsider the strategy of using proteomics to develop biomarkers [4]. We may need to approach this issue differently, instead of only extending the conventional approaches that have been used historically. The problems and possible solutions for proteomic biomarker development were previously reviewed [5–7], and the efforts toward effective biomarker discovery and validation have been organized by Human Proteome Organization [8]. In this article, two unique viewpoints are introduced on the basis of the author's experience.

Casting doubt on the ‘ultimate' solution for biomarker discovery

In 2002, Anderson et al. published a landmark paper, plotting plasma proteins as a function of their possible expression level [9]. The authors indicated that, while the dynamic range of plasma proteins reached up to 11 orders of magnitude, the proteomic modalities uncovered only a small portion of the plasma proteomes; this limitation still exists today. Moreover, the traditional plasma biomarkers, such as alpha fetoprotein, prostate specific antigen and carcinoembryonic antigen, existed below the limits of detection of the available proteomic modalities. As a result, researchers were certain that the major obstacle to biomarker discovery was technical limitations. Achieving higher sensitivity through technology improvements became the major challenge to overcome in discovering biomarkers. However, this approach to the research may be pointless.

I agree that existing plasma biomarker proteins are often present at sub-pg/ml levels, and current proteomic modalities are unable to comprehensively detect proteins at such low levels of expression. However, even if the ultimate proteomic solution, which allows observing all plasma proteins, becomes available, the target biomarkers will not be discovered by proteomics alone. This is supported by the history of research related to mRNA analysis.

In the study of transcriptomes, the comprehensive analysis of mRNA became feasible in the early 21st century with DNA microarray technology, and a global effort to examine the expression of mRNA in many cancers occurred, with an intention to discover biomarkers. While a considerable number of papers were published about the discovery of biomarker candidates through transcriptomics, only a few prognostic biomarkers, such as Oncotype DX [10] and MammaPrint [11,12], were reported in the clinical setting. Although some of these discovery-related issues might be specific to the study of transcriptome, this example suggests that comprehensive analysis through the ‘ultimate' proteomic modalities will not be the solution for biomarker development. We may need to approach to biomarker research not only by expanding the capability of proteomics modalities, but also by considering more essentially important viewpoints.

A problem-oriented & interdisciplinary approach in biomarker development

Limited performance of proteomics modalities was such an obvious obstacle in biomarker discovery that it masked essential points of biomarker study and led to superficial optimistic perspectives. The study of biomarkers is essentially a study of diseases, and the biochemistry relating to peptide, protein and posttranslational modifications is only a tool. A problem-oriented approach should be needed for biomarker development: we have to consider what the requirements for biomarkers are, as determined by the clinicians, and how will they be used to improve the clinical outcomes using the currently available diagnostic modalities and treatments. Comprehensive proteomic observations without concrete clinical aims and backgrounds may not lead to successful biomarkers.

For biomarker discovery, it is critically important to design the study considering adequate stratification of patients according to clinical stage, pathological grading, response to treatments and other disease-specific parameters. An additional consideration is the identification of diseases for differential diagnosis, especially when identifying plasma biomarkers for early detection of disease. For prognostic biomarkers, the effect of a prognostic diagnosis on specific treatments should be understood. Staying up-to-date with recent therapy trends is required when predictive biomarkers are considered. In short, without adequate treatment options related to the results of biomarker tests, the usefulness of these diagnoses is limited. Without demonstrating a concrete application, the benefits of validation studies for biomarker candidates may not be obvious and the validation may not be achieved. Indeed, one of the keys to success for Oncotype DX [10] and MammaPrint [11,12] may be the variety of available risk stratification therapies for breast cancers with possible metastasis. When these factors are taken into consideration in the study design, the biomarkers that are subsequently discovered will hold greater value in the clinical settings. Unfortunately, these essential issues are often missed in biomarker study, and biomarkers discovered by poor study design and unclear aims seem to be unpromising in the clinical settings.

Needless to say, one of the most important steps in biomarker development is extensive validation. Validation studies require the nationwide cooperation of industry, academics and government, and it is critically important that clinicians participate in multi-institutional validation efforts for the biomarker candidates. However, there may be a lack of clinician interest in the biomarkers published using the proteomics approach; the study designs may seem unrealistic, and the utilities of the biomarkers are not clearly stated. For example, in plasma biomarker studies, comparisons are often conducted in a similar number of cancer patients and healthy donors during the validation process. Even if high sensitivity and specificity for the biomarkers are obtained, their practical use remains very questionable because this situation is not representative of the patient distribution in the clinical setting; the number of cancer patients is always considerably smaller than noncancerous persons. The research aim and study design should take into consideration the prevalence of the target disease and the characteristics of the patient population who are meant to benefit from the biomarkers. Moreover, healthy donors are not appropriate controls for discovery and validation. Owing to the age at which cancers are common, other minor physical disorders are often present; these patients only visit the hospital when cancer is suspected and are then subjected to plasma tests. Therefore, the use of age- and sex-matched controls is not sufficiently representative, and people with common benign diseases, especially those with symptoms that are similar to those of cancers, should be included in the validation studies and even potentially in the discovery phase. There are many examples of biomarkers, which do not convince clinicians, and the validation studies for which were not supported.

Obviously, a problem-oriented approach needs strong backgrounds of medicine, which are generally weak subjects of nonmedical proteomics researchers, for both discovery and validation. Clinicians may often be involved in biomarker study in a superficial way, by providing clinical materials. However, clinicians should play a more central role in biomarker discovery by providing clinical experience and idea, not only providing samples. Ideally, the clinicians in the biomarker discovery team should have some backgrounds in clinical or basic research. Hopefully, they are also able to discuss the advantages and limitations of the current global expression study. The issues of bias in the study design and sampling have been discussed previously [6,7]; to address these issues, clinicians who understand them and their solutions should be included. At the same time, basic researchers involved in the collaboration should be enthusiastic and willing to become familiar with clinical oncology. We need to realize that methods of analytical chemistry and modern machines, such as mass spectrometry, which are just tools for biomarker study; they are important but not essential. Mutual efforts to improve communication and work toward a productive collaboration will lead to successful biomarker development.

Examples of successful biomarker developed by interdisciplinary study

During the past 13 years in my laboratory, 25 clinicians have committed themselves to biomarker development. The majority of experiments lasted for 2 or 3 years. Although I have a medical license, my role in biomarker projects is limited to that of basic researcher, owing to my lack of experience in clinical oncology and inability to identify currently most important problems in oncology. Therefore, I have discovered that successful biomarker development occurs through the collaboration of basic researchers and enthusiastic, experienced oncologists.

An example of a biomarker discovered in this research environment is pfetin, a novel prognostic biomarker in gastrointestinal stromal tumors [13]. Prognostic biomarker is required to indicate the most effective adjuvant molecular targeting therapy in gastrointestinal stromal tumors. Our research group identified pfetin by comparing patient groups who were strictly stratified by prognosis and pathological features. The prognostic utility of pfetin was demonstrated using immunohistochemistry in approximately 500 cases located in six different hospitals [14,15]. Pfetin is one of the most successful tissue biomarkers discovered by proteomics and validated by a conventional method, immunohistochemistry, in an independent sample sets. It is noteworthy that pfetin was discovered using 2D-DIGE [16–18], which is based on the classical proteomics method such as 2D-PAGE. Therefore, I do not intend to say that the development of proteomic technologies is meaningless, only that successful biomarker development will be preceded by a clear and convincing aim and an appropriate study design.

Opinions

We should be aware that the development of biomarkers needs a problem-oriented, interdisciplinary process. Comprehensive study does not always lead to promising biomarker by itself. Clinician participation in the study of biomarkers will lead to realistic, practical and interesting biomarker candidates, which justify the time and expense involved in validation studies. Many problems in biomarker development have been pointed out and many of them will be solved by deeply including clinicians to biomarker projects. Although discussion in this article focused on cancer biomarkers, it can generally be applied to biomarker studies for other diseases.

Financial & competing interests disclosure

The author has no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this manuscript.