Sci. Aging Knowl. Environ., 4 January 2006
Vol. 2006, Issue 1, p. pe1
[DOI: 10.1126/sageke.2006.1.pe1]


Biomarkers of Aging: Combinatorial or Systems Model?

Andres Kriete

The author is at the School of Biomedical Engineering, Science, and Health Systems at Drexel University, Philadelphia, PA 19104, USA, and the Coriell Institute for Medical Research, Camden, NJ 08103, USA. E-mail: andres.kriete{at} and akriete{at}

Key Words: combinatorial biomarker • systems biology • gene enrichment • networks • reverse engineering principles • aging model


Like research in many other areas of biology, aging-related research has experienced an increase in the diversity of applicable experimental techniques and an accelerated rate of growth of a large body of data, which has outpaced the rate at which a deeper understanding has emerged. In a way, the situation now appears exacerbated as compared with that of ~15 to 20 years ago, given the range of new "-omics" profiling technologies (see Kaeberlein Perspective, Gafni Perspective, and Kristal Perspective). In addition, the biology of aging was and still is a field in which new hypotheses are frequently postulated, challenging both experimental validation and interpretation of the many phenomena involved. One particular point of discussion, which is often singled out as a hallmark of current problems, is the failure of the aging-related research community to find a suitable biomarker of aging (1, 2) (see Miller Perspective, Dimri Perspective, and "Will We Find Biomarkers of Aging?").

Problems in the Search for Biomarkers of Aging

What a biomarker for aging should be or predict are quite broadly defined. At the minimum, a biomarker should not only (i) reflect some basic property of aging, but also (ii) be reproducible in cross-species comparison, (iii) change independently of the passage of chronological time (so that the biomarker indicates biological rather than chronological age), (iv) be obtainable by nonlethal means, and (v) be measurable during a short interval of life span. One might argue that these expectations are too broadly defined to allow success and that one might be better off by first narrowing the search down to a particular species, organ, or even cell type. Still, particular markers might display different properties between cell types, individuals, or groups. Another problem, which is even more challenging, is that unless we understand how aging "works," we might not be able to define ideal biomarkers at all. A biomarker would have only limited utility without an understanding of the biological reason it is a biomarker. A key to progress in the search for a suitable biomarker of aging might be found in the field of drug development. Failures in drug target identification, drug development, and disease intervention are often caused by heterogeneity in the population and/or a limited understanding of how a drug works on the molecular level. The recognition that many disease processes manifest themselves on a system level, and have to be treated as such, is slowly restructuring the research in this area--and strategies evolving now may serve as a blueprint to improve methodologies in aging-related research.

Combinatorial Biomarkers

Most simply, a biomarker would comprise a single entity, such as a particular serum protein. However, biomarkers can also consist of a panel of multiple genes, proteins, or metabolites, for example, or be combinatorial, in which a variety of different attributes are monitored. The beauty of a combinatorial systems biomarker is that it might not be limited to a particular level of biological organization. It can be a combination of quantifiable features on the level of an organism, organ, cell, protein, or gene. As an example, physiological data could be combined with the concentration of a specific secreted protein and the expression level of a particular gene. The identification of such biomarkers would initially require a quantitative, reproducible assessment of many profiles on an individual basis, ideally in longitudinal studies. If genetic or gene-expression profiling were used, several genes could be used jointly to define a panel of markers. In drug development, typically five or more measures are combined to assess drug metabolism or toxicity, leading to far better sensitivity and specificity than if a single measure had been used. Combined markers would be more likely than individual markers to allow subpopulations to be singled out--a prospect of great value in the "diagnosis" of aging, because a group that develops cardiovascular problems should be differentiated from another group with a weakened immune system, for example.

Indeed, scientists involved in the search for a biomarker of aging have suggested the use of panels (3) and turned away from the homogeneous strains of animals initially used toward heterogeneous stocks, to improve detection of aging-related genetic loci and T cell type subsets predicting longevity (4, 5) (see Rikke Perspective). However, one has to recognize that assessment on one level of biological organization, such as gene expression profiling, may not be sufficient to identify all relevant genes. Identification of the genes that exhibit the largest change in expression in two situations, or for which correlated changes are revealed through biostatistical clustering, might represent reliable properties only in very homogenous populations. However, for the expected heterogeneity caused by differences in genetic makeup and environmental factors, the chief alternative is to identify and sort out genes by correlating them with phenotypical measures such as tissue morphology or clinical chemistry, assuming that those genes or pathways are relevant that show a measurable downstream effect (Fig. 1). This approach has shown success in time-controlled toxicology and animal studies of diabetes (6, 7), using phenotypical measures as statistical covariants in the analysis of gene expression, to enrich for genes that are relevant to the topic under study. At the present time, such a systems-oriented data-mining approach is not widely practiced in aging-related research.

View larger version (36K):
[in this window]
[in a new window]
Fig. 1. The search for a combinatorial biomarker of aging. A systemwide correlation between molecular data and tissue phenotypes derived from samples from heterogeneous populations should enrich for genes that are relevant to aging and reveal reproducible and sensitive biomarkers.

Network Analysis

Combinatorial biomarkers may point to relevant factors in the aging process, yet generally biomarkers are sought without consideration for such larger issues. One has to realize that a marker for age, whether a gene or a gene product, is not regulated by one factor alone, nor are the downstream effects of such genes and products isolated but rather fan out in intricate cellular and organismic networks. Likewise, many observations of the aging process on the molecular, cellular, and organ level are connected in one way or the other. The pitfalls and limitations of current research are not so much in the experimental approaches being taken but in the integrated analysis or bioinformation engineering of the available data. As an understanding of aging emerges on a systems level, another spin of the discussion might therefore leverage consideration of the networks involved, turning the focus from single genes to gene regulatory pathways and from single proteins to protein-protein networks (see Pletcher Perspective). The exploration of such networks is the primary goal of computational systems biology [for reviews, see (8-10)]. Systems biology implements reverse engineering principles to study individual parts and the complex dynamic interactions between all components, so that the resulting mechanistic models can describe complex gene regulatory, metabolic, and signaling networks.

The ability of computer-based, in silico representations to predict how a system in a particular state might react and adjust to perturbations has made systems biology an attractive component for basic research, drug development, and predictive medicine. Clearly, aging-related research has not thoroughly implemented this type of approach; however, Kowald and Kirkwood published a model-based systems approach in 1996 (11) that was widely referred to as the "network theory of aging." In retrospect, this work was the first comprehensive and computational systems approach to aging, in an attempt to combine models involving mitochondrial production of reactive oxygen species, free radicals, and aberrant proteins.


It may turn out that in silico, mechanistic modeling may refresh an otherwise stalled discussion on biomarkers. To be successful, the development of models will first require comprehensive profiling, at the level of the individual genome, proteome, cell, organ, and organism in heterogeneous populations. Models will be at first minimal in nature but will be expected to grow in complexity and scale. Such model-based representations not only will lead to a better mechanistic understanding of aging but also will help to single out those critical components (representing potential biomarkers) that are key to the aging process and relevant for diagnostic purposes. Because in silico representations of a given entity, whether cell or organism, reflect individual rather than average properties of a population, a path to personalized medicine becomes apparent. Additionally, the mechanistic nature of in-silico models makes them distinct from ad hoc defined biomarkers in their ability to predict. A systems model should provide an ability to make a prediction about the rate of the progression of aging, if not the most likely path toward a catastrophic systems-failure endpoint.

January 4, 2006
  1. H. R. Warner, Current status of efforts to measure and modulate the biological rate of aging. J. Gerontol. A Biol. Sci. Med. Sci. 59, B692-696 (2004).[Abstract/Free Full Text]
  2. R. J. Davenport, Will we find biomarkers of aging? Sci. Aging Knowledge Environ. 2005 (26), nf48 (2005).[Abstract/Free Full Text]
  3. R. L. Sprott, Biomarkers of aging. J. Gerontol. A Biol. Sci. Med. Sci. 54, B464-B465 (1999).[Medline]
  4. A. U. Jackson, A. T. Galecki, D. T. Burke, R. A. Miller, Mouse loci associated with life span exhibit sex-specific and epistatic effects. J. Gerontol. A Biol. Sci. Med. Sci. 57, B9-B15 (2002).[Abstract/Free Full Text]
  5. R. A. Miller, Biomarkers of aging: Prediction of longevity by using age-sensitive T-cell subset determinations in a middle-aged, genetically heterogeneous mouse population. J. Gerontol. A Biol. Sci. Med. Sci. 56, B180-B186 (2001).[Abstract/Free Full Text]
  6. K. Boyce, A. Kriete, S. Nagatomi, B. Kelder, K. Coschigano, J. J. Kopchick, Phenotypical enrichment strategies for microarray data analysis applied in a type II diabetes study. Omics 9, 251-265 (2005).[Medline]
  7. A. Kriete, M. K. Anderson, B. Love, J. Freund, J. J. Caffrey, M. B. Young, T. J. Sendera, S. R. Magnuson, J. M. Braughler, Combined histomorphometric and gene-expression profiling applied to toxicology. Genome Biol. 4, R32 (2003).[CrossRef][Medline]
  8. A. Kriete, R. Eils, Computational Systems Biology (Elsevier Academic Press, San Diego, 2005).
  9. L. Hood, Systems biology: Integrating technology, biology, and computation. Mech. Ageing Dev. 124, 9-16 (2003).[CrossRef][Medline]
  10. H. Kitano, Biological robustness. Nat. Rev. Genet. 5, 826-837 (2004).[CrossRef][Medline]
  11. T. B. Kirkwood, A. Kowald, Network theory of aging. Exp. Gerontol. 32, 395-399 (1997).[CrossRef][Medline]
Citation: A. Kriete, Biomarkers of Aging: Combinatorial or Systems Model? Sci. Aging Knowl. Environ. 2006 (1), pe1 (2006).

Science of Aging Knowledge Environment. ISSN 1539-6150