The behavior of multiple genes under different experimental conditions is often analyzed by clustering mRNA expression data from microarrays, followed by computational discovery of regulatory motifs in promoter sequences of clustered genes in a post-processing step. This stepwise approach may lead to the construction of biologically insignificant clusters, and in turn, to errors in motif discovery. The statistical challenge is to simultaneously determine the groupings of genes and subsets of motifs involved in their regulation, when the groupings may vary over time, and a large number of potential regulators may be present. We devise a novel unified Bayesian framework, and a hybrid Monte Carlo methodology to estimate the model parameters under two classes of latent structure, arising due to the unobservable state identity of genes, and the unknown set of covariates influencing the response within a state. The performance of our method is demonstrated though an application to a yeast cell-cycle data set and simulation studies.