Here, we have analyzed binding patterns of the common hematopoietic transcription factor PU.1 to reveal novel insights into prerequisites for DNA binding of this master regulator: overall, we can distinguish three major categories of consensus binding sites (summarized in ): (i) non-bound sites that mainly show low-binding affinity and reside in inactive chromatin. (ii) PU.1-bound sites that are DNase I inaccessible and represent ‘autonomous’ binding events preferentially at high-affinity sites. (iii) PU.1-bound mostly intermediate- and low-affinity sites that are DNase I accessible, and where binding is likely stabilized by cooperativeness with neighboring transcription factor binding sites. Increasing PU.1 concentration, which has previously been implicated in lineage-specific functions of PU.1 (16
) reduces the binding affinity threshold, leading to a marked increase in autonomous binding sites and to a lower extent in cell type-specific sites.
Schematic depicting the three classes of PU.1 consensus motifs.
Expression of the master regulator PU.1 is induced in early progenitor cells during hematopoiesis and retained at high levels in human monocytes/macrophages, myeloid dendritic cells, as well as granulocytes and to a lower level in early and pro-B-cell stages, erythrocyte–megakaryocyte progenitors or mast cells (40
). Although PU.1 binds common sets of targets in these cell types, previous work identified a significant share of cell type-specific–binding events. The observed binding patterns were either attributed to cell type-specific transcription factor cooperativeness (7
) or different PU.1 expression levels (16
), which, however, explain only parts of the binding patterns. In fact, only a strikingly small fraction of all putative binding sites across the genome (<1%) is actually occupied by PU.1 in any of the cell types, and it is unclear how PU.1 is sequestered to the small and often cell stage-specific proportion of its binding sites. The DNA sequence represents a proximate determinant of transcription factor binding, and we initially addressed the aforementioned question by systematically analyzing sequence features of PU.1-bound sites. The PU.1 consensus PWM comprised >2500 different 12mers, which were found to bind PU.1 at variable frequencies. For example, 80% of all AAAGAGGAAGTG 12mers (809 instances) overlap with PU.1 ChIP-seq peaks within the non-repetitive genome, whereas only 1% of TAACTGGAAGTG 12mers (three instances) were considered occupied. The in vivo
binding preference is reflected by the PWM motif log-odds score, which is a measure for the similarity of a given motif to the consensus PWM. Using microscale thermophoresis (21
), we could demonstrate that the PU.1 PWM log-odds scores represent a surrogate for PU.1 motif-binding affinity. The comparison of motif features in different log-odds score (affinity) classes revealed a number of notable correlations. Most strikingly, we observed an inverse correlation between motif affinity and local DNA accessibility (as measured by DNase I cleavage frequency). High-affinity motifs were more often located in less accessible regions that also showed the lowest average vertebrate conservation. In contrast, highly accessible and more conserved regulatory modules were mostly populated by intermediate/low affinity motifs. Although we cannot directly infer functionality from this data, bound intermediate/low affinity motifs more likely reside in functionally important regions, as they are preferentially found in motif clusters that are also associated with ‘active’ epigenetic marks like H3K27ac and H3K4me1.
Even high-affinity motifs were not always associated with ChIP-seq tags, suggesting that some sites offer no or little opportunity for PU.1 to bind. DNA methylation, a principal mechanism of binding site selection for several other transcription factors, is not a major determinant of PU.1 binding—PU.1 binding was generally associated with local DNA demethylation, both in monocytes and macrophages, suggesting that PU.1 may participate in recruiting the DNA demethylation machinery to its binding sites. As a ‘master regulator’, PU.1 might actually be required to access relevant genomic sites, including cell type-specific enhancers, which are ‘silenced’ by DNA methylation in progenitor cells. This observation concurs with a recent study demonstrating the ability of transcription factors to induce local DNA demethylation (25
). In line with such a role, the consensus sequence of PU.1 does not contain CpG dinucleotides within its core sequence, which may avoid a direct steric effect of DNA methylation on PU.1 binding. Insights from our comparison of bound and non-bound PU.1 elements across the genome, however, point to a role for higher-order chromatin structures in regulating PU.1 binding, regardless of the cell stage. Non-bound PU.1 elements were enriched in gene deserts or chromatin domains (defined as being flanked by the boundary transcription factor CTCF) that lack domain-wide monomethylation of histone H3 at lysine 4, a modification that correlates with transcriptional activity (41
). Thus, a large proportion of non-bound PU.1 consensus motifs is located in inactive chromatin domains that are likely not or only partially accessible to PU.1. Exceptions most often include autonomous high-affinity motifs, which frequently show some degree of binding even in inactive chromatin domains. Interestingly, we also note that motif co-association patterns (the presence of PU.1 recognition sites close to other predicted binding sites) distinguish active and inactive domains, suggesting that the activity of chromatin domains is at least partially pre-determined by the underlying sequence context.
In contrast to factors like the glucocorticoid receptor (9
), PU.1 binding is clearly not pre-determined by baseline chromatin accessibility patterns. Although DNase I non-accessible, autonomously bound motifs generally show little conservation (implying a lack of function), the ability of PU.1 to efficiently compete with nucleosomes at high-affinity sites may still be an important feature of its master regulator function. In this context, it will be interesting to clarify whether PU.1 binding simply relies on opportunity (on a stochastic basis) to access freely available DNA, or whether it can also bind or induce remodeling of nucleosome-associated DNA.
In conclusion, our analysis supports a hierarchical model for the regulation of PU.1 binding. The fact that some predicted high-affinity sites are not bound suggests that (higher order) chromatin structures can provide a first level of restriction. Motif-binding affinity generally presents the second layer of binding control that is fine-tuned by PU.1 expression levels and local binding site cooperativeness, which both lower the binding affinity thresholds.