In this study, microarray gene expression data obtained during the initial 6-hour period after partial hepatectomy were used to characterize the transcriptional regulatory network that drives the onset and early progression phase of liver regeneration. Following clustering of the gene expression data, PAINT analysis was used to characterize significantly enriched TF binding sites in the different clusters to identify TFs that might have contributed to the temporal profile of gene expression obtained. TF activation could be directly confirmed by analysis of nuclear extracts. Insight into the functional role of the genes regulated by these TFs was obtained from the gene ontology analysis of TF-gene relationships. Not unexpectedly, the analysis suggests that multiple TFs coordinate to control a wide range of functions during the early phase of liver regeneration (Fig. 7). Importantly, functional categories identified by GO analysis often are broad and overlapping and should be interpreted with considerable caution. For that reason, we curated the individual assignments obtained from the GO analysis to optimize the functional associations presented in Figs. 5 and 7. A more detailed discussion of the functional categories identified in this analysis is provided as Supplemental text (see Additional file 4). In agreement with an earlier study on mice , sham-operated animals did not show significant changes in gene expression accompanying the early response to PHx that could not be accounted for by multiple comparison errors or animal-to-animal variability and there was no overlap with the differentially expressed genes detected after PHx.
Several previous studies reported microarray studies of gene expression changes in rodents after partial hepatectomy using a variety of platforms. The majority of these studies presented data on mice, including some that included early time points [17–20, 48]. However, the onset and progression of liver regeneration after PHx is considerably slower in the mouse than in the rat . Reported experimental results vary considerably between studies, both in the number and the nature of genes reported and in the number of replicates, making consistent evaluation of the statistical significance and validation of the resulting changes difficult. Therefore, these studies have not generally resulted in broader insights into the functional processes associated with these changes in gene expression. One previous study used the rat model , starting with the 6 hr time point. However, this study observed significant differences in only 16 (out of 4608) genes. Thus, our study is unique in presenting a robust analysis of the gene expression changes in the rat and, importantly, in using the temporal response profile to obtain information on the transcriptional regulation that drives these responses.
These results demonstrate that relevant functional information on the transcriptional control of the early response to partial hepatectomy can be obtained from the PAINT analysis of clustered microarray data. Each of the six temporally distinct gene expression clusters is characterized by a unique pattern of significantly over-represented binding sites for TFs. Activation of a selection of the candidate TFs was confirmed by oligonucleotide binding assays of nuclear extracts. Notably, there was relatively little overlap in the TFs driving the response in different temporal clusters. This is not to say that transcription factors involved in one cluster didn't play any role in the response in other clusters (e.g. note the broad presence of TF binding sites for NF-κB, HNF-1, or PolyA), but those associations didn't reach statistical significance in our analysis. It is possible that these factors contribute to the fine-regulation of the gene expression responses within clusters, but the number of differentially expressed genes in these studies was too low to identify such combinatorial control by multiple TFs with sufficient statistical weight.
As with any computational approach, it is important to note that our unbiased discovery approach using PAINT, while informative in predicting a role for novel TFs, is subject to false negatives, i.e. not all the currently known TFs in liver regeneration were present in the computational predictions. For example, STAT-3 is known to play role in liver regeneration [28, 46, 47] and analysis of our samples confirmed that STAT-3 activation occurred during the time frame of early responses that we investigated here. However, our PAINT analysis indicates that the differentially expressed genes were not enriched for STAT-3 target genes (based on the results from MATCH/TRANSFAC, only 2 of the 309 genes contained STAT-3 binding sites in their promoters). The STAT-3 binding site is characterized by a position weight matrix of 21 base pair length in the TRANSFAC database (Accession number M00225), with a position weight matrix similarity threshold of 0.934 for minimizing false positives in finding the binding site on genomic sequences. According to the TRANSFAC database, this similarity threshold corresponds to a false negative rate of ~20%. Hence, the unexpectedly low number of predicted STAT-3 binding sites may not be due to our choice of parameters in MATCH, but appear to arise from a combination of the pattern matching algorithm and the STAT-3 binding site description. As the databases and associated computational tools continue to improve, false negatives like these are likely to decrease, although they will probably never be completely eliminated. Nevertheless, our computational approach successfully predicted many known and novel TFs as playing a role in the onset of liver regeneration, several of which were experimentally validated (Figure 6).