Here we have presented a novel approach to integrate experimental gene expression profiles with already available data. We were able to combine our own data from microarray experiments and data from public databases of PPIs, TF, and GO in a rational manner to address the global biological functions of Hsp90 and p23 both individually and, perhaps more interestingly, concomitantly.
Several key features of our working pipeline (Figure ) allowed us to process and to understand certain aspects of our data. First, we worked with genes showing a significant fold change in at least one, often several pairwise comparisons that were analyzed (Figure ). Second, we organized them into a graph establishing connections between nodes according to their co-expression levels and their ability to interact at the protein level (Figure ). Third, we organized the nodes in the graph according to their annotated intracellular localization. Using the Cerebral layout, highly interconnected nodes with the same intracellular localization get grouped in the final individual graphs [27
] (Figure ). And finally, a novel and essential feature of the analysis was to convert all the generated graphs into an animation to facilitate the visualization of similar expression patterns of genes topologically grouped in the individual graphs (Figures and , and Additional File 3
Movie S1). It is important to emphasize that this pattern detection was only possible by this method and that we failed to detect these patterns with more standard clustering algorithms (data not shown). The latter may have discarded or "undervalued" nodes with only marginal changes in gene expression whereas a human subject's pattern recognition would have incorporated some of these as well. While small fold changes may not have any significant value by themselves, as part of a larger animated pattern, they attract the attention of a human observer and they may become corroborative. The interpretation of these patterns matched the information contained in the pairwise comparisons of the microarray data (Figure ), which finally allowed us to uncover the genes that are regulated by both Hsp90 and p23. Thus, the use of an animation to visualize patterns was a key tool to connect the data of the different microarray experiments.
Once the networks had been built, we were able to perform two kinds of analyses, single pairwise ones and an animation-assisted one that takes advantage of all pairwise comparisons. With the single analysis we found that 114 genes were up-regulated upon pharmacological Hsp90 inhibition (Figure , red nodes and Additional File 1
Table S1, genes boxed in blue). Of these, 26 presented an enrichment of GO terms related with stress response (Additional File 1
Table S1, genes highlighted in orange), in accordance with the finding that these genes contained an increased occurrence of STREs in their promoters. The stress response in yeast is regulated by the TFs Hsf1 and Msn2/Msn4, which bind to promoters containing heat-shock response elements and STREs, respectively [34
]. Some of these promoters could also contain both elements [36
]. From our study, it appears that the sole inhibition of Hsp90 is sufficient to up-regulate stress-related genes with STREs in their promoters. This would indicate that Hsp90 influences the Msn2/Msn4 system. Analyzing the yeast PPI networks extracted from public repositories and the literature, we can infer that Hsp90 (and co-chaperones) interacts with kinases known to repress the Msn2/Msn4 system (and the stress response) [38
] (Additional File 5
Table S4). Since many kinases are known Hsp90 clients (see regularly updated list at http://www.picard.ch/downloads/downloads.htm
), we can speculate that Hsp90 and its co-chaperones maintain kinases responsible for Msn2/Msn4 repression in a functional state thereby regulating the stress-response (Additional File 6
Figure S1). Interestingly, with animation-assisted analysis, we noticed that not all stress-related genes belong to the same expression pattern. Most of these genes belong to patterns B and C, indicating that they are at the same time affected by Hsp90 and p23 (Additional File 3
Movie S1). It is also clear that Hsp90 and other co-chaperones (Sti1, Aha1, Cpr6) are up-regulated upon pharmacological Hsp90 inhibition, and yet they do not show any particular pattern. Hsf1 is the only stress-inducible transcriptional regulator of Hsp90 protein expression in yeast [42
]. Conversely, Hsp90 represses gene expression from Hsf1-dependent promoters [44
], establishing a negative feedback loop for its own expression. Conceivably, the genes for Hsp90 and at least some of its co-chaperones are regulated only by Hsf1, while the other stress-related genes exhibit a more complex regulation involving the concurrent action of Hsp90 and p23, potentially through the Msn2/Msn4 system.