Synthetic glucocorticoids, corticosteroids, are widely used to suppress inflammatory and immune responses. However, this class of drugs has a low therapeutic index due to a multiplicity of adverse effects on many tissues, including skeletal muscle [1
]. The adverse effects on skeletal muscle derive from the role of this tissue in glucose homeostasis. One aspect of the broad systemic function of glucocorticoids is to increase gluconeogenesis in the liver and kidney [4
]. A primary substrate for gluconeogenesis is amino acid carbon derived from the net degradation of muscle protein. Glucocorticoids also cause muscle to become insulin resistant, thereby preventing the large bulk of the musculature from taking up the glucose produced by the liver and kidney [4
]. When corticosteroids are used therapeutically, their effects on the musculature are accentuated, which results in muscle wasting and insulin-resistant diabetes.
Corticosteroids produce their effects on skeletal muscle by altering the transcription of specific genes. These transcriptional effects take two fundamental forms, a direct and indirect method. Many regulated genes contain glucocorticoid responsive elements (GREs) in their regulatory sequences and their transcription is influenced directly [2
]. However, there are a large number of genes whose transcription is altered indirectly by glucocorticoids. In these cases, glucocorticoids alter the expression or function of other transcription factors, which in turn alter the transcription of other genes [13
]. Although still not entirely understood, the phenomena of muscle wasting and insulin resistance clearly involve temporal cascades of changes in the expression of a multiplicity of genes [2
]. Many genes involved in the phenomena of muscle wasting and insulin resistance have been identified in a piecemeal fashion using diverse in vitro
and in vivo
experimental systems. However, understanding such phenomena requires that the temporal cascade of gene expression events be viewed as a whole.
Previously we have used pharmacokinetic/pharmacodynamic (PK/PD) modeling in studies to describe the relationship between bolus dosing with methylprednisolone (MPL) and the change in the expression of a few genes in liver and skeletal muscle [3
]. For those experiments, a single bolus dose of MPL was given intravenously to groups of adrenalectomized animals. Animals were sacrificed at 16 time points over a 72-h period. The PK/PD models describe the deviations from and return to baseline (defined by vehicle-treated controls) of gene expression responses. The livers and muscles used for both of these studies were derived from the same animals. Data were analyzed as if samples were taken from a single animal. The data for the change in the expression of mRNA for the PK/PD models was generated using quantitative northern hybridization. Although, more recently, we have converted such measurements to quantitative real-time reverse transcriptase polymerase chain reaction (RT-PCR), even this method does not allow the scope of data collection necessary for developing models for the type of polygenic phenomena initiated by corticosteroids. We previously described the availability of data sets developed by using the Affymetrix GeneChips®
Rat Genome (R_U34A) (Affymetrix, Inc., Santa Clara, CA, USA) microarray chip available online, which allows for single gene queries [27
]. Those data sets were developed using the same rich time series employed in our earlier studies. The intent was to use gene arrays as a method of high-throughput data collection in order to obtain the scope of data necessary for applying PK/PD modeling to describe broad polygenic phenomena, such as insulin resistance caused by corticosteroids.
Mining such data sets presents uniquely different problems from those encountered when microarrays are used to distinguish one group from another (e.g., cancerous versus non-cancerous tissues). For those applications, one attempts to define a pattern or fingerprint that distinguishes, with very high probability, one group from another [28
]. In many cases it is the pattern of gene expression rather than the relationship between the genes that is the important focus. In the present application of microarrays, the difficulty lies with sorting through the vast amount of data to identify probe sets with temporal patterns of change in expression, which indicate that the gene is regulated in response to the drug. In this case, the causal relationship between the genes whose expression is changing in response to the drug is of paramount importance. For example, the drug may change the expression of a particular transcription factor, which in turn alters the expression of downstream genes. For this reason the most important aspect of the mining approach is to avoid discarding valuable data. This is of particular importance because each differentially expressed gene becomes the subject of extensive literature searches in order that it can be placed into a temporal context of all other transcriptionally altered genes. The purpose of the endeavor is to use PK/PD modeling to develop a ‘motion picture’ of the polygenic response to the drug.
In the present report we describe a filtering approach to mining the skeletal muscle data set, which is designed to eliminate probe sets that do not meet criteria expected of transcriptionally altered genes. These criteria are based on our extensive prior knowledge of data for individual genes and their use in PK/PD modeling. This report, therefore, details the small percentage of probe sets in the skeletal muscle data set that meet a specific criteria for further and more intense scrutiny. That same skeletal muscle data set was initially described and its online availability has been detailed in a previous report [27