In this study, we used ChIP-seq to generate genome-wide profiles of the binding sites of the major adipogenic transcription factors, C/EBPα and PPARγ, in mature human SGBS adipocytes. We identified 52,733 C/EBPα and 23,328 PPARγ binding sites, and consistent with previous studies in mouse adipocytes [15, 16] we found that PPARγ binding sites are highly enriched in the vicinity of genes upregulated during adipogenesis. Furthermore, and also consistent with what has been reported for mouse adipocytes , a high percentage (49.5%) of the PPARγ binding sites overlap with C/EBPα binding sites. Thus, these data indicate conservation of the overall regulatory regime of C/EBPα and PPARγ between mouse and human adipocytes, including their potential direct cooperativity through binding to adjacent sites. However, despite the conservation of the overall regulatory regime and putative target genes between mouse and human adipocytes the retention of mouse binding sites in human is limited, i.e. most sites are species-specific. Similar results demonstrating limited retention of binding sites despite extensive conservation of association with putative functional targets have previously been reported for other transcription factors [20, 22–25], and was recently also reported by Rosen and colleagues when comparing PPARγ binding sites in primary human in vitro differentiated adipocytes with the binding sites identified in murine 3T3-L1 cells .
Despite the fact that most binding sites of PPARγ and C/EBPα are species-specific, there are some sites that we can confidently classify as retained between human and mouse. There are also larger genomic regions, like the PPARG2 locus, where most sites are retained (Figure 3C). This observation prompted us to systematically investigate what biological features determine whether a site is retained or not. Thus, we have systematically investigated several biological features that could affect the retention of mouse C/EBPα and PPARγ binding sites, using two complementary methods with different advantages, i.e. a binary approach where overlaps of sites are counted, and an enrichment approach where the signal strength of human ChIP reads is assessed. The binary approach has the advantage of being conservative (i.e. only strong signals are being labeled as bound sites), and it is easier to interpret the outcomes as sites can be either lost or retained. Conversely, the enrichment method can take weaker sites into account and is not dependent on specific thresholds in peak finding; however, this method will not clearly define what sites are retained and lost, as it imposes no thresholds on the data.
Regardless of method, we find that there is a significantly increased retention as well as turn-over of PPARγ binding sites in the vicinity of genes upregulated during adipogenesis compared with binding sites more distant from such genes. By contrast, retention of C/EBPα binding is only slightly increased in the vicinity of genes that are highly upregulated. The difference indicates that there is a higher evolutionary pressure to maintain specific PPARγ binding sites compared with specific C/EBPα binding sites for the regulation of adipocyte differentiation and function. This may reflect the higher importance of PPARγ over C/EBPα in adipocyte differentiation, as most clearly demonstrated by Spiegelmann and colleagues, who showed that that ectopic expression of PPARγ in C/EBPα -/- mouse embryo fibroblasts (MEF) , but not C/EBPα in PPARγ -/- MEFs , can induce adipogenesis. In addition, since there are more C/EBPα binding sites than PPARγ binding sites there may be less evolutionary constraint on the individual C/EBPα sites. This could reflect the lower binding specificity of C/EBPα, since it is easier to create a new site if the factor can bind to many different sequences.
We also observed that mouse sites that are strongly bound by either factor are more likely to be retained, but the causality is unclear. Stronger binding sites could be under higher selective pressure, since they are more likely to have a vital function. In addition, stronger binding sites could be harder to erode over time, as these are likely to represent sites where multiple factors such as co-binding transcription factors and chromatin accessibility contribute to a high occupancy of the site. Thus, strongly bound sites in a common ancestor between two species might require more evolutionary changes to be reduced to a fully inactive site in one species.
The remarkable overlap between PPARγ and C/EBPα binding reported previously [15, 16] indicates cooperativity between the two transcription factors. In support of this, we find that retention of both C/EBPα and PPARγ binding is increased, when the binding sites overlap with a binding site of the other factor, suggesting that a potential synergy between these two factors is of such significance that selective pressure on these overlapping sites is increased. Interestingly, while retention of C/EBPα-only sites is not affected by vicinity to regulated genes, the C/EBPα sites that overlap with PPARγ sites display increased retention in the vicinity of upregulated genes, further indicating an importance of C/EBPα co-binding with PPARγ in adipocyte gene regulation.
We also analyzed the sequence of human genomic regions corresponding to overlapping PPARγ and C/EBPα binding sites in mouse adipocytes. These analyses revealed that presence of the C/EBP consensus sequence in the human sequence not only predicts the retention of C/EBPα binding but, surprisingly, also retention of PPARγ binding at sites that overlap with C/EBPα binding sites in mouse. By contrast, the PPAR consensus sequence in the human sequence only predicts retention of PPARγ binding, not retention of C/EBPα binding at sites overlapping with PPARγ binding sites. This indicates that at these overlapping sites, the C/EBP consensus sequence in the binding region is being selected for over evolution to a higher degree than the PPAR consensus site, suggesting that PPARγ binding is directed by C/EBPα binding. By analogy, differential binding of nuclear factor κB (NFκB) in 10 human individuals correlates with changes in a consensus sites of a reported cooperating transcription factor, signal transducer and activator of transcription 1 (Stat 1) , and similar results have been found in studies in yeast  and fungi .
Possible mechanisms explaining the interdependence between observed C/EBP consensus site conservation and PPARγ retention include C/EBP-assisted binding of PPARγ to an adjacent consensus site as well as indirect binding of PPARγ to DNA by tethering to C/EBPα or by long range intrachromosomal loops [34, 35]. Lazar and colleagues have previously investigated the interdependence of C/EBPα and PPARγ binding in mature 3T3-L1 adipocytes and failed to see an effect of C/EBPα knockdown on PPARγ binding to selected target sites in mature 3T3-L1 adipocytes . However, since only a few sites were investigated it is difficult to conclude on the general importance. Furthermore, it is possible that other members of the C/EBP family, which all share the same consensus sequence, play a role in the establishment of the PPARγ transcriptional complex during adipogenesis and a more limited role once the complex is established. Intriguingly, recent data from our laboratory indicate that C/EBPβ, which is expressed early in differentiation, may play a role in early chromatin remodeling of PPARγ binding sites (Siersbæk, Nielsen, John, Sung, Baek, Loft, Hager, and Mandrup, EMBO Journal, in press)).
Odom and colleagues have previously shown that the transcription factor binding regions that were retained between multiple species had higher sequence constraint than species specific binding regions , however the interdependence between conservation of the region around the bound site and the retention of the consensus site was not addressed. Interestingly, we show here that the high identity quartile of binding sites (i.e. the binding regions with the highest sequence conservation) displayed increased retention of C/EBPα and PPARγ binding independent of conservation of their respective consensus sites. This strongly indicates that additional sequence patterns are important for C/EBPα and PPARγ binding. Such sequence patterns could among others be directing binding of cooperating transcription factors.