|Home | About | Journals | Submit | Contact Us | Français|
Binding of transcription factors to DNA is a key regulatory step in the control of gene expression. DNA sequences with high affinity for transcription factors occur more frequently in the genome than instances of genes bound or regulated by these factors. Although several mechanisms have been identified that influence the specificity of transcriptional regulation, it is not known if these can explain the observed genome-wide pattern of binding or regulation for a given transcription factor. We used genome-wide approaches to study how trans influences shape the binding and regulatory landscape of Pho4, a budding yeast transcription factor that activates gene expression in response to phosphate limitation. We find that nucleosomes significantly restrict the sites to which Pho4 binds. At nucleosome-depleted sites, competition between Pho4 and another transcription factor, Cbf1, determines Pho4 occupancy, raising the threshold for transcriptional activation by Pho4 in phosphate replete conditions and preventing Pho4 activation of genes outside the phosphate regulon during phosphate starvation. Pho4 binding is not sufficient for transcriptional activation – a cooperative interaction between the transcription factor Pho2 and Pho4 occurs specifically at genes that are activated. Combining these experimental observations, we are able to globally predict Pho4 binding and its functionality. Our study provides insights into the mechanisms of global control by sequence-specific transcription factors.
Specific transcriptional regulation is essential for precise control of many biological processes, including tissue development and responses to environmental stimuli. Alterations in gene expression are commonly associated with human diseases and disorders (Jimenez-Sanchez et al., 2001; Vaquerizas et al., 2009). A key regulatory step in the control of gene expression is the binding of transcription factors to DNA to activate gene transcription (Hochheimer and Tjian, 2003; Ptashne and Gann, 1997). Both cis and trans factors can influence transcription factor binding. In cis, binding typically requires recognition of a specific DNA sequence (Jacob and Monod, 1961); sequence variation in DNA binding motifs affects transcription factor binding occupancy (Kasowski et al., 2010; Schmidt et al., 2010; Zheng et al., 2010). However, high affinity binding sites occur more frequently than experimentally detected binding events, even for transcription factors in organisms with a relatively small genome such as Saccharomyces cerevisiae (Harbison et al., 2004; MacIsaac et al., 2006). Among vertebrates, binding events display a species-specific pattern despite similarity between the consensus binding motifs of transcription factors (Schmidt et al., 2010). Among human individuals, the majority of transcription factor binding variance cannot be explained by genetic differences in binding sites (Kasowski et al., 2010), suggesting that trans factors also influence transcription factor binding. In trans, chromatin structure restricts DNA accessibility (Liu et al., 2006; Wasson and Hartemink, 2009), but the effect of nucleosome occupancy on the sequence-specific binding of transcription factors binding has not yet been systematically explored. Other trans factors, such as cooperating and competing factors (Pan et al., 2010; Pierce et al., 2003), have been studied primarily at the level of individual genes for their influence on transcriptional control, but not evaluated on a global scale. Moreover, binding of a transcription factor per se is often not sufficient for transcriptional regulation (Birney et al., 2007; Farnham, 2009; Harbison et al., 2004; MacIsaac et al., 2006). Thus, it remains unclear what determines the genomic locations to which a transcription factor binds and whether this binding is able to influence the transcription of a gene (Farnham, 2009).
To further complicate matters, transcription factors of the same family contain structurally conserved DNA binding domains and usually recognize similar short DNA motifs (Badis et al., 2009; Berger et al., 2008; Wei et al., 2010). However, these factors frequently regulate distinct biological responses (Robinson and Lopes, 2000; Sharrocks, 2001). The conflict between similarity in binding and divergence in regulation raises two questions: How are distinct patterns of transcriptional regulation achieved? How is the regulation by a given transcription factor influenced by others that recognize very similar DNA motifs? To answer these questions, it is necessary to systematically interrogate, on a genome-wide scale, the factors that may contribute to the specificity of transcription factor binding and regulation.
We used gene regulation by the phosphate-responsive (PHO) signaling pathway as a model system to explore the determinants for transcription factor binding and function. Pho4, a basic-helix-loop-helix (bHLH) transcriptional activator in S. cerevisiae, is regulated in response to environmental inorganic phosphate (Pi) availability and activates a transcriptional program together with the transcription factor Pho2 when cells are limited for Pi (Ogawa et al., 2000). In Pi rich conditions, Pho4 is phosphorylated and transported from the nucleus into the cytoplasm; in phosphate-limited conditions, Pho4 is de-phosphorylated and transported from the cytoplasm into the nucleus (O'Neill et al., 1996; Schneider et al., 1994). Pho4 binds with high affinity to ‘CACGTG’ motifs in vivo (Harbison et al., 2004) and in vitro (Badis et al., 2008; Maerkl and Quake, 2007; Zhu et al., 2009). In this study, we found that Pho4 is bound to only ~14% of its consensus binding sites, and only a quarter of those binding events result in activation of gene transcription. We demonstrated that this genome-wide binding and regulatory pattern is determined by a combination of chromatin restriction, competitive binding from the factor Cbf1 that recognizes the same motif as Pho4, and cooperation with Pho2. We also discovered that Cbf1 competition raises the threshold of gene activation and helps to define the specificity of PHO gene regulation. Combining our experimental observations, we can predict transcription factor binding and function at a whole genome level.
We applied biotin-tagging chromatin immunoprecipitation (Kolodziej et al., 2009; van Werven and Timmers, 2006) combined with high throughput sequencing (Bio-ChIP-Seq; See Experimental Procedures) to identify Pho4 binding events in vivo in phosphate starvation conditions (Figure 1A and Figure S1A). Although there are 843 ‘CACGTG’ consensus binding sites (all ‘NCACGTGN’ motifs except ‘TCACGTGA’ are considered to be consensus binding sites; see Experimental Procedures for more details) in the genome, we observed Pho4 binding to only 115 (~14%) of these sites (Experimental Procedures), implying that factors other than DNA binding specificity influence Pho4 binding in vivo. Since chromatin restricts the access of transcription factors to their potential binding sites (Khorasanizadeh, 2004; Kornberg and Lorch, 1999; Narlikar et al., 2002), a Pho4 ‘CACGTG’ consensus binding site could exist in one of two possible states: an inaccessible state, occluded by nucleosomes; or an accessible state, exposed in a nucleosome-depleted region (nucleosome-free or nucleosome linker region). To determine whether local chromatin structure influences transcription factor binding, we mapped nucleosome occupancy in no Pi conditions with micrococcal nuclease digestion followed by paired-end deep sequencing (Figure 1B, Figure S1B–E and Experimental Procedures). As expected, Pho4 is not bound to the binding sites that are inaccessible and most occluded by nucleosomes (See Experimental Procedure for details) (209 of 216 sites, 97%; Figure 2A and Figure S2A). However, using the same threshold, Pho4 is also not bound to two thirds of the most accessible sites (172 of 248 sites) (Figure 2A and Figure S2A). We conclude that chromatin structure inhibits transcription factor binding, but open chromatin structure is not sufficient for Pho4 binding at its ‘CACGTG’ consensus sites.
It is possible that proteins with similar specificity compete with Pho4 for binding to accessible ‘CACGTG’ consensus sites. Cbf1, another member of the bHLH transcription factor family, is present in the nucleus at high concentration (Ghaemmaghami et al., 2003), is not known to interact with Pho4 (Graumann et al., 2004), and binds with high affinity to the same consensus binding motif ‘CACGTG’ in vitro and in vivo as does Pho4 (Harbison et al., 2004; MacIsaac et al., 2006; Maerkl and Quake, 2007; Zhu et al., 2009). To test if Cbf1 competes with Pho4 for binding, we identified in vivo binding sites for Cbf1 in high and no Pi conditions. In no Pi conditions, 77% (132 of 172) of the accessible consensus ‘CACGTG’ binding sites not bound by Pho4 are occupied by Cbf1 (Figure 2B). Intriguingly, most of the accessible sites bound by Pho4 (72 of 76, 95%) are also bound by Cbf1. Thus, the accessible, high affinity ‘CACGTG’ sites mainly fall into two classes: those where Cbf1 competes with Pho4 most effectively, resulting in detectable binding of Cbf1 but not Pho4; and a second class where Cbf1 competes less effectively, resulting in significant occupancy of both Pho4 and Cbf1. In high Pi conditions, Cbf1 is bound to both of these classes of sites (119 of 132 and 67 of 72, respectively; Figure S2B).
Pho4 and Cbf1 have different preferences for bases flanking the ‘CACGTG’ consensus binding site in vitro (Maerkl and Quake, 2007); these sequence features might explain differences in Cbf1 and Pho4 occupancy in vivo. In accord with the observed in vitro sequence preferences, we find that accessible sites with less Pho4 binding have a single 5’ ‘T’ base flanking the ‘CACGTG’ (Figure 2C), suggesting that Cbf1 can compete most effectively at these sites (Figure 2D).
What is the physiological role of the interplay between Pho4 and Cbf1 at consensus sites? One idea is that Cbf1 is required in high Pi conditions to keep nucleosomes properly positioned, and therefore keep consensus binding sites nucleosome free (Figure S3A; “Cbf1 priming model”). Although there is precedent for Cbf1 positioning nucleosomes in the promoters of some sulfur metabolism genes (Kent et al., 2004; Kent et al., 1994), when we analyzed the cbf1Δ strain in high Pi conditions we observed no change in nucleosome occupancy at Pho4 binding sites in Pho4-regulated genes (r = 0.943 between cbf1Δ strain and wild-type, r = 0.953 between wild-type replicates; Figure S3B); in contrast, in the cbf1Δ strain nucleosome occupancy increases and nucleosome position shifts at binding sites in the regulatory regions of sulfur metabolism genes (Lee et al., 2010) (Figure S3C).
An alternative model is that competition from Cbf1 prevents spurious activation of phosphate-responsive genes in high phosphate conditions, and of other ‘CACGTG’ containing genes during phosphate starvation (Figure S3D; “Cbf1 blocking model”). In the absence of Cbf1, most Pho4-regulated genes (see definition in Figure 4A) showed significantly increased expression in high Pi medium (20 of 28, p ≤ 0.05; Figure 3A, column 2) – conditions where Pho4 is less active and localized primarily to the cytoplasm (Komeili and O'Shea, 1999). This aberrant expression is Pho4-dependent (compare Figure 3A, columns 2 and 3) and is not the indirect result of an upstream signaling defect caused by deletion of Cbf1 (Figure S3E). Moreover, deletion of Rtg3 and Tye7, two other members of the bHLH family that bind the ‘CACGTG’ consensus site (Zhu et al., 2009) did not result in spurious activation (Figure 3A, column 4). Since deletion of Cbf1 causes a growth defect and impaired activation of the PHO pathway in no Pi conditions (Figure S3E), we evaluated the consequences of Cbf1 competition when the PHO pathway is fully activated using a strain lacking the cyclin Pho80 grown in high Pi medium (O'Neill et al., 1996). In the cbf1Δ pho80Δ strain we observed Pho4 binding to sites bound by Cbf1 in the wild-type strain (Figure 3D), particularly at the ‘T-CACGTG’ motifs (Figure 3C, Figure S3F). Indeed, Pho4 is bound to and activates transcription of 13 genes that were bound by Cbf1 and not regulated by Pho4 in the wild-type strain (Figure 3B). Our observations support two roles for Pho4-Cbf1 competition: in high Pi conditions Cbf1 prevents spurious activation of the PHO genes induced by a low level of nuclear Pho4, ensuring that phosphate-responsive genes are turned off when Pi is available; in no Pi conditions, Cbf1 prevents Pho4 from inappropriately activating genes containing a ‘CACGTG’ motif that are not part of the phosphate regulon, ensuring that only genes needed for the response to phosphate limitation are turned on.
Pho4 binds to the ‘CACGTG’ motif in the promoters of over 80 genes, but only ~10–20 genes are regulated by Pho4 in response to Pi starvation (Ogawa et al., 2000; Springer et al., 2003). Moreover, these regulated genes have different dependence on Pho4 and on Pho2, a homeodomain factor that interacts with Pho4 and regulates the phosphate starvation response (Vogel et al., 1989). We used epistasis expression analysis (mutant cycle analysis (Capaldi et al., 2008)) to dissect and quantify the contribution of the regulatory interactions between Pho2 and Pho4 to transcriptional activation (Figure S4) in terms of three “expression components”: the contribution of Pho2 acting alone (Pho2), the contribution of Pho4 acting alone (Pho4), as well as the contribution of Pho2 and Pho4 acting together (which we refer to as a “cooperative” component). To determine the values of these expression components, we directly compared gene expression between all possible pairs of wild-type, pho2Δ, pho4Δ and pho2Δ pho4Δ strains in no and high Pi conditions, where the measured gene expression difference between each pair of strains consists of a linear sum of the expression components relevant for that pairwise comparison (Figure S4). We calculate the expression components for each gene through regression methods with equations describing the expression components measured in each microarray (Supplemental Experimental Procedures). We observe only three regulatory interactions operating at phosphate-responsive genes: gene activation by Pho4 acting alone (Pho4 only), genes activated by the interaction between Pho2 and Pho4 (Cooperative, Co), and mixed regulation (genes partially activated by Pho4 alone and partially dependent on the interaction between Pho2 and Pho4, Pho4 + Co) (Figure 4A). Notably, almost all Pho4 regulated genes contain a cooperative component, suggesting that the interaction between Pho2 and Pho4 is necessary for gene activation.
Since Pho2 and Pho4 interact and bind cooperatively to the PHO5 promoter (Barbaric et al., 1998; Barbaric et al., 1996; Vogel et al., 1989), we hypothesized that this cooperative binding might be correlated with the functionality of Pho4 binding events – the ability to trigger gene activation. We observed a strong correlation between Pho2 and Pho4 binding events in no Pi conditions (r = 0.926, p < 10−49; Figure S5A). Only some of these coincident binding events appear to be instances of cooperative binding (See Experimental Procedures), where both Pho2 and Pho4 occupancy increases in response to phosphate starvation – it is these cooperative binding events that correlate with gene activation (‘regulated’ class in Figure 4B, p = 8.1 × 10−16, Fisher’s exact test). Recruitment of Pho4 to cooperative, regulated sites is entirely dependent on Pho2, whereas Pho4 binding to the non-cooperative, non-regulated sites is largely unaffected by deletion of Pho2 (Figure 4C). The reduction of Pho4 occupancy in pho2Δ strains in no Pi conditions is correlated with the transcriptional activation contributed by the cooperative interaction between Pho2 and Pho4 (Co component) for Pho4 regulated genes (R = 0.63, p = 0.0016). For the cooperative Pho4 binding events, the Pho2 and Pho4 ChIP signals are overlapping (Figure 4D) and the predicted Pho2 binding sites are enriched at a distance of 15 bp from the consensus Pho4 binding sites (Figure 4E). In contrast, when only one of the factors is significantly recruited in response to Pi limitation, the gene is not activated (‘non-regulated’ class in Figure 4B, p = 0.99, binomial test), there is no juxtaposition of Pho2 and Pho4 ChIP signals (Figure 4D), and predicted Pho2 binding sites do not exhibit a consistent spatial relationship with Pho4 binding sites (Figure 4E). Thus, the spatial organization of Pho2 and Pho4 binding motifs may promote cooperative binding of Pho2 and Pho4 and the ability to activate transcription.
Prediction of transcription factor binding and functional targets has been a challenging task. Here we provide an integrated mechanistic view of the determinants that influence Pho4 binding and regulation, taking into account all ‘CACGTG’ sites independent of evolutionary conservation, clustering of motifs, and relative positioning in the promoter. Incorporating the influence of trans effects into an equilibrium binding model (see Experimental Procedures), 43 of 50 (86%) (Figure 4F) of the top predicted binding sites are indeed bound by Pho4 (AUC-ROC = 0.87). We find that all trans effects contribute significantly to the pattern of Pho4 binding in the genome (Figure S5C). Either nucleosome occupancy or flanking sequences predicts Pho4 binding to high affinity consensus sites (top panel; AUC-ROC). However, if we consider only the group of top predicted targets (bottom two panels), a prediction based on nucleosome occupancy or Cbf1 competition alone has poor accuracy, whereas the synergy between Cbf1 competition and nucleosome occlusion more accurately predicts binding to this group of target sites (Figure S5C and Supplemental Experimental Procedures). Of the 115 experimentally-determined Pho4 binding sites, fewer than 25% are able to promote activation of gene transcription in no Pi conditions. Considering the cooperative interaction between Pho2 and Pho4, 23 of 28 (82%) (Figure 4F) binding sites predicted to be functional are actually associated with transcriptional activation (AUC-ROC = 0.992). From this prediction, we identified three non-promoter binding events that are associated with Pi dependent anti-sense transcripts (B. Zid, et al., data not shown) – two of them have not been previously identified and are potentially linked to phosphate signaling. We find that the presence of a Pho2 binding motif 15 bp from the Pho4 binding site is highly predictive of the functionality of the Pho4 binding events (Figure S5D), indicating that the spatial arrangement of Pho2 and Pho4 binding sites is critical in activating gene transcription.
Sequence-specific transcriptional factors recognize short cis elements in promoter regions to regulate gene transcription (Farnham, 2009; Jacob and Monod, 1961). However, these elements exist tens of times more frequently than the number of in vivo binding events. How can transcription factors bind to and regulate a specific set of genes when many other genes carry the same binding sites? We used a genome-wide approach to systematically dissect the determinants of transcription factor binding and regulation, and investigated whether these determinants can explain transcriptional specificity. We provide evidence that competition from chromatin, competition from another transcription factor that recognizes Pho4 consensus binding motifs, and cooperativity with a partner protein determine the specificity of Pho4 binding and regulation. We also discovered novel regulatory functions for Cbf1 competition and provide a general view of the specific regulation of the PHO regulon.
In high Pi conditions, Pho4 is phosphorylated and exported to the cytoplasm (O'Neill et al., 1996; Schneider et al., 1994). Competition from Cbf1 increases the threshold of transcriptional activation by Pho4, preventing spurious activation by low levels of nuclear Pho4 (Fig. 2D). In no Pi conditions, phosphorylation of Pho4 is inhibited (O'Neill et al., 1996; Schneider et al., 1994) and the increase in the concentration of unphosphorylated nuclear Pho4 enables it to compete effectively with Cbf1 at sites where Cbf1 is weakly bound – those with certain flanking sequences found in genes within the phosphate regulon; inappropriate binding and activation of ‘CACGTG’ containing genes outside the phosphate regulon is prevented. The dynamic range of binding and gene activation is increased as a result of the reduction in Pho4 binding due to Cbf1 competition in high Pi conditions and its condition-dependent cooperative interaction with Pho2 in no Pi conditions (Pho2 interacts only with unphosphorylated Pho4 during phosphate limitation (Komeili and O'Shea, 1999)). Other stress- and nutrient response transcriptional programs may employ similar strategies to trigger high levels of induction of specific genes that were tightly repressed in the absence of the perturbation.
In mammalian cells, transcription factors within the same structural family commonly recognize similar DNA motifs but regulate diverse biological processes (Badis et al., 2009; Berger et al., 2008; Wei et al., 2010). It is unclear how faithful and distinct regulation is achieved with transcription factors of overlapping specificity. Our results suggest that chromatin structure influences the accessibility of potential binding sites and competition among transcription factors can facilitate discrimination of subtle differences in DNA binding specificity across the genome. Cooperation with conditionally, spatially or temporally expressed co-activators (Sharrocks, 2001) may provide additional selection for binding events to elicit specific transcriptional outcomes. Moreover, competition among transcription factors may have significant effects on transcriptional regulation, controlling the threshold for gene activation and limiting potential crosstalk between different signaling pathways. Overall, our findings suggest that the specificity of transcriptional regulation is a composite of the DNA binding specificity, nuclear abundance and functional interactions of each transcription factor in the context of eukaryotic genome architecture. Understanding the grammar of specific regulation of individual transcription factors will provide insights into decoding the complex regulatory network of the eukaryotic genomes.
Media and growth conditions are described in Supplemental Experimental Procedures.
High affinity Pho4 binding motifs were determined using the position specific scoring matrix (PSSM) described in (Lam et al., 2008), which was derived from in vitro measurement of Pho4 DNA binding affinities (Maerkl and Quake, 2007). We selected the most stringent threshold (0.0075) to recapitulate in vivo validated high affinity binding sites as the threshold of Pho4 high affinity binding motifs (Lam et al., 2008). Since all determined high affinity binding motifs at this threshold contain ‘CACGTG’ as core sequence, we define these motifs as the consensus ‘CACGTG’ binding motif for Pho4. All ‘NCACGTGN’ motifs except ‘TCACGTGA’ in S. cerevisiae genome meet the threshold.
The procedure for biotin-tagging immuno-precipitation is described in detail in Supplemental Experimental Procedures. In summary, we performed Bio-ChIP-seq experiments on Pho2, Pho4, Cbf1 and mock samples in both high and no Pi conditions (60 minutes after Pi starvation), and on Pho4 in a pho2Δ strain grown in no Pi conditions, and in pho80Δ and cbf1Δ pho80Δ strains grown in high Pi conditions. Sequencing libraries were prepared for both ChIP DNA and input DNA (from the supernatant of total cell lysate) following the Illumina protocol. Libraries with size between 200 to 300 bp were selected for PCR amplification and sequenced with an Illumina Genome analyzer II. 36 base sequence tags were aligned to the S. cerevisiae genome with ELAND. On average, 2.9 and 12.5 million uniquely aligned reads were obtained for ChIP and input samples, respectively. Bio-ChIP-seq results of all Pho4 consensus binding sites are presented in Table S2.
Libraries from mono-nucleosomal DNA were sequenced from both ends with an Illumina genome analyzer II; details are described in Supplemental Experimental Procedures. We assumed that the center of each sequenced DNA fragment is the nucleosome dyad and extended 73 bp on both sides to generate mono-nucleosome coverage (Figure S1C–E). We obtained nucleosome maps for the wild-type strain in high and no Pi conditions (40 minutes after Pi starvation) and for the cbf1Δ strain in high Pi conditions. For each sample, 8–10 million uniquely aligned nucleosomal DNA sequencing reads were obtained. Nucleosome occupancy of all Pho4 consensus binding sites is presented in Supplemental Table 2.
Details of yeast cell collection, RNA isolation, cDNA synthesis and labeling, and microarray hybridization are described in Supplemental Experimental Procedures. Whole genome expression profiling was performed with competitive hybridization to Agilent 8×15K S. cerevisiae two-color expression microarrays (G2509F, AMADID #019838). Microarrays were scanned using an Axon 4000B scanner and gene features were extracted using GenePix 5.1 software. Lowess and quantile normalization were performed using the MATLAB bioinformatics toolbox before further analysis.
Wild-type (WT) no Pi vs WT high Pi, cbf1Δ high Pi vs WT high Pi, cbf1Δ pho4Δ high Pi vs WT high Pi, and tye7Δ rtg3Δ high Pi vs WT high Pi microarrays were performed with dye-swaps to eliminate dye labeling bias and analyzed in four biological replicates (Churchill, 2002; Yang and Speed, 2002). pho80Δ vs pho80Δ pho4Δ and pho80Δ cbf1Δ vs pho80Δ cbf1Δ pho4Δ were performed in high Pi conditions and analyzed in two biological replicates. Mutant cycle analysis was constructed with a cyclic comparison so that the expression components could be directly inferred and dye labeling bias would be cancelled in the analysis (Churchill, 2002; Quackenbush, 2002; Yang and Speed, 2002) (Figure S4). The mutant cycle was repeated with three biological replicates.
Details of the design and analysis of the mutant cycle (Figure S4) are described in Supplemental and Experimental Procedures. 80 genes were selected as Pi starvation responsive genes by directly comparing the expression of wild-type in no and high Pi conditions (activated more than 1.8-fold and significantly induced with a null hypothesis of < 1.5 fold, p ≤ 0.01). For each of those genes, linear regression was performed and genes with significant Pho2, Pho4, or Co components (significantly induced with a null hypothesis of < 1.4 fold, p ≤ 0.05, (Capaldi et al., 2008)) were shown in Figure 4A. None of the genes induced after Pi starvation contains a significant Pho2 only component, indicating that Pho4 is the primary activator and Pho2 functions only as a cooperating factor (Komeili and O'Shea, 1999; Springer et al., 2003). Full mutant cycle analysis results are summarized in Table S3.
We calculate the average nucleosome occupancy of a 20 bp window centered on all consensus ‘CACGTG’ sites to determine local nucleosome occupancy. ~80% of the S. cerevisiae genome is estimated to be covered with nucleosomes (Lee et al., 2007); we thus say a transcription factor binding site is in the accessible state if the average nucleosome occupancy on the site is in the lower quartile of the average genome nucleosome occupancy. Symmetrically, we say the site is inaccessible if the average nucleosome occupancy is in the upper quartile of the average genome nucleosome occupancy. The average genome nucleosome occupancy distribution was determined by randomly sampling the 20 bp window average for 10,000 genome locations and repeating this 10 times. The average of the upper and lower quartile cut-off for 10 times sampling of the average genome nucleosome occupancy was used to define the accessible and inaccessible states of binding sites. Varying the percentile threshold over a wide range (15 % – 30%) does not affect the outcome (Figure S2A).
Methods to determine the binding at Pho4 consensus binding sites are described in Supplemental Experimental Procedures.
The increase in Pho2 and Pho4 binding after Pi starvation (recruitment) is calculated using the following equation: taking Pho4 as an example, EnrichmentNo/High = (Pho4ChIP No / Pho4Input No) / (Pho4ChIP High / Pho4Input High). To estimate a threshold to define transcription factor recruitment, we calculated the mean and standard deviation of Pho2 and Pho4 recruitment at all consensus ‘CACGTG’ sites, excluding the sites at regulatory regions of Pho4 regulated genes (functional binding sites). We used mean + 2 standard deviations (s.d.) as the threshold to identify the sites showing the most significant recruitment (Figure 4B). 23 of 28 sites showing recruitment of both Pho2 and Pho4 are regulated by both factors (p = 8.1 × 10−16, Fisher’s exact test), and none of the Pho4-only recruited sites is regulated by Pho4 (p = 0.003, Fisher’s exact test).
Information about Pho4 binding at its non-consensus binding sites is described in Supplemental Experimental Procedures.
The procedures for fluorescence microscopy are described in Supplemental Experimental Procedures.
To predict Pho4 binding at its consensus binding motifs, we integrate information about Pho4 DNA binding preference, local nucleosome occupancy, and competition from Cbf1 into an equilibrium model (Granek and Clarke, 2005). At the equilibrium state, the probability of Pho4 not binding to an 8-mer sequence at position i can be expressed as:
Where WNuc, i, WCbf1, i, Ka, i, adj and [Pho4] represent occlusion from nucleosomes, competition from Cbf1, equilibrium association constant of Pho4 with the 8-mer sequence, and the nuclear Pho4 concentration.
If there are n potential Pho4 binding sites near position x on the genome, the binding probability at x is finally calculated as:
Derivation of equations and details about the calculation are described in Supplemental Experimental Procedures.
We thank C. Daly and J. Zhang for help with Illumina sequencing; X. Zhang for assistance with microscopy; former and present O’Shea lab members for discussion and commentary; and V. Denic, N. Friedman, A. Regev, B. Stern for critical reading of the manuscript. This work was supported by National Institutes of Health grant R01 GM051377 and the Howard Hughes Medical Institute.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Agilent custom 8×15K S. cerevisiae two-color whole genome expression microarray data are deposited in the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE23580. The data can be accessed now through the link (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=fjmxtouwogmqozk&acc=GSE23580). The nucleosome and Bio-ChIP-seq occupancy maps can be accessed at (link). All raw sequencing data will be deposited in the GEO database under accession number XXXX.
Supplemental data include three supplemental tables, 5 supplemental figures, Supplemental Experimental Procedures, and Supplemental References and can be found with this article online at (link).