Guss, Daniel. Genomics and Bioinformatics, MB&B452a. "The triumph and oversight of the pharmaceutical use of cDNA microarry technology."
In order to fully understand gene function, one must know when and where the gene is expressed, what affects its expression levels, and in a larger sense, how does this gene’s expression correspond to the expression of other genes that work together to carry out cellular processes. Facilitating this very task, cDNA microarrays allow scientists to quantify the mRNA expression of thousands of genes simultaneously. This has also presented the field of bioinformatics with the challenge of making sense of the immense amount of data that microarray experiments generate.
The intuitive response to this bioinformatic problem is to arrange genes according to correlations in their pattern of gene expression, which has been shown to efficiently group together genes of known similar function in yeast [1]. Viewed within the larger aims of bioinformatics, correlating the expression of a novel gene with that of a gene of known function may subsequently lead to its characterization. Among the most commonly used algorithms for clustering gene expression is the pairwise average-linkage cluster analysis described by Eisen, et. al., that ultimately generates a tree whose branch lengths reflect the degree of gene product similarity.
The beauty of using bioinformatics to analyze microarray data is that in addition to clustering genes within normal cells, with the aim of elucidating function, both known and unknown genes can additionally be clustered according to expression levels in a diseased versus a non-diseased state. Medically speaking, one can diagnostically associate changes in normal gene expression levels with a specific disease. Thus far, changes in gene expression have been correlated with many pathological states including breast cancer [2], multiple sclerosis lesions [3], and renal disease [4]. In addition, host-microbe interactions can be elucidated either by measuring gene expression in the infected host, or by quantifying the actual pathogen’s gene expression [5]. Examples of the latter include the monitoring of gene expression in M. tuberculosis while it infects cultured monocytes [6]. Once a specific gene has been associated with a disease state, its function can subsequently be elucidated specifically within normal and diseased tissue via clustering algorithms that correlate its expression levels with those of characterized genes. Ultimately, such an approach allows selective characterization of genes associated with pathological conditions.
Identifying genes that are medically relevant to specific disease states offers a means of both diagnosing disease and identifying potential drug targets. The former is achieved by correlating specific gene expression patterns with the presence and progression of a disease state, and has been studied in pathologies such as the progression of human breast cancer [7] and in neonatal screening for diseases including sickle cell, alpha-1-antitrypsin deficiency, and Factor V Leiden [8]. Ideally, a specific cDNA microarry pattern could be used to diagnose the presence of the disease and assess its degree of progression. Identifying potential drug targets also stands in the forefront of DNA microarray use, and potential targets are similarly determined by characterizing the function of those genes whose expression correlates with the disease state [9]. Drugs can then be developed that target specific gene expression. Just as importantly, a drug’s effectiveness can subsequently be quantified by using microarrays to definitively assess drug-specific gene expression [5]. Such approaches have recently shown, for example, that superoxide dismutases are effectively targeted by oestrogen derivatives in the selective killing of cancer cells [10]. DNA microarrays thereby mark a new age in medical diagnosis and drug development.
The aforementioned medical and pharmaceutical uses of DNA microarrays are largely the hallmark of recent excitement. But as successful as these strategies have proven to be, there is still an underlying oversight to such approaches. These comparisons usually involve two or more distinct conditions, such as normal cells versus tumor cells or treated versus untreated cells, with the hope of describing the number and nature of differently expressed genes. Beyond these gene-centered studies, however, one must also ask: "What do the expression profiles tell us as a whole? [11]" The nature of this question is outlined in an article by Huang S. Cellular homeostasis is a dynamic process that involves a wide variety of genes whose multiple interactions generate a picture of a dense network of cross-talk between cellular signaling pathways. This homeostasis can take various forms, such as cell division, differentiation or apoptosis, each of which creates a different expression pattern on a DNA microarry. Cellular states can thereby be described by gene expression profiles, also known as gene activity profiles (GAP). Since feedback circuits are contained within the genomic network, multiple discrete and stable states of gene expression are expected as genes regulate the expression of other genes in a web of hierarchal interactions. In the simplest case, individual gene expression can be simplified to be either "on" (=1) or "off" (=0). In turn, for each homeostatic state of ‘n’ genes, there is a GAP of length ‘n’ that is simply a series of 1’s and 0’s that signify each individual gene’s expression. Despite the vast number of genes, however, there is in reality only a small number of GAPs, because due to the regulatory interactions of genes only a fraction of theoretical GAPs account for stable homestatic states that can be ‘realized’. ‘Forbidden’ GAPs end up ‘flowing’ along a chain of other unstable states until a stable GAP is reached, also called an attractor. This boolean network of GAPs is an idealization that enormously simplifies the use of computer models to study the genomic regulatory network. Within this context, there is a degree of ‘GAP inertia’ that inhibits straying from one stable GAP to another, representing the stability of each homeostatic state.
The idea behind pathological states such as tumorigenesis is therefore that mutations affect the wiring architecture of the network, such as by irreversibly inactivating a gene (=fixed activity status 0). The fact that neoplasia doesn’t occur more often can be explained by the robustness of attractor states, in which the switch of one gene usually does not change the attractor. Global changes of states (such as from cell growth to differentiation) must therefore be a result of some master switch that can change the attractor state, such as an activated cell receptor that triggers a signaling pathway. These master switches may need to be active only during the transition between cellular states, during which ‘forbidden’ GAPs flow into a new, ‘stable’ GAP. It is this fact that current pharmaceutical use of DNA microarrays usually fails to consider. By comparing only stable GAPs with DNA microarrays, such as a normal state versus a tumor state, one may overlook the gene ideal for drug targeting - the master switch. In addition to the current categorization of genes according to their conditional expression, functional genomics must therefore also identify attractor states and establish a database of gene expression and activation profiles (GAPs) that represent such attractors. Only by these means can all potential drug targets truly be identified, paving the way for using DNA microarrays to their ultimate potential.
The use of bioinformatics in the context of DNA microarrays is therefore a burgeoning field. As successful as current pharmaceutical and medical uses of these microarrays have proven to be, one must still reanalyze one’s approach to make full use of this new technology. And as with any new technology, it will be years before its full potential is realized.
References: