|Home | About | Journals | Submit | Contact Us | Français|
The estrogen receptor (ER), glucocorticoid receptor (GR), and forkhead box protein 1 (FoxA1) are significant factors in breast cancer progression. FoxA1 has been implicated in establishing ER binding patterns though its unique ability to serve as a pioneer factor. However, the molecular interplay between ER, GR, and FoxA1 requires further investigation. Here we show that ER and GR both have the ability to alter the genomic distribution of the FoxA1 pioneer factor. Single-molecule tracking experiments in live cells reveal a highly dynamic interaction of FoxA1 with chromatin in vivo. Furthermore, the FoxA1 factor is not associated with detectable footprints at its binding sites throughout the genome. These findings support a model wherein interactions between transcription factors and pioneer factors are highly dynamic. Furthermore, at a subset of genomic sites, the role of pioneer can be reversed, with the steroid receptors serving to enhance binding of FoxA1.
Pioneer factors (PFs) have been described as a class of proteins that penetrate closed chromatin to create accessible binding sites for general transcription factors (TFs) during development (Zaret and Carroll, 2011). The forkhead box protein 1 (FoxA1) has been shown to interact with compact chromatin, modulating chromatin structure as an early event. Upon chromatin binding, FoxA1 is thought to initiate nucleosome binding via the winged-helix domain that it shares with the H1 linker histone, and induce nucleosomal rearrangements by a mechanism independent of ATP-dependent remodeling complexes. These transitions in turn result in an increase in the accessibility of DNA binding elements (Bernardo and Keri, 2012, Cirillo et al., 1998, Cirillo et al., 2002). This mechanism has been widely implicated for the recruitment of steroid receptors (SRs) (Bernardo and Keri, 2012, Eeckhoute et al., 2006, He et al., 2012, Hurtado et al., 2011) specifically for the estrogen receptor (ER) and androgen receptor (AR) in breast and prostate cancer cells, respectively. Early studies reported that FoxA1 binding sites overlap with ~50% of ER binding sites (Carroll et al., 2005), and that FoxA1 is required for at least half of all ER binding events in MCF-7 breast cancer cells (Carroll et al., 2005, Laganiere et al., 2005). Later findings via genome-wide analysis (Carroll et al., 2006, Lupien et al., 2008, Hurtado et al., 2011) have been interpreted in support of this general model. In addition, it has been reported inhibition of ER produces no change in FoxA1 genomic binding patterns (Lupien et al., 2008, Hurtado et al., 2011). These investigations have either focused on a small number of binding locations, or have compared FoxA1 binding only between unstimulated cells and cells treated with an ER antagonist (Hurtado et al., 2011, Lupien et al., 2008). Contrary to these findings, an independent study reported that upon knockdown of ER, FoxA1 binding is lost at many unstimulated ER binding sites (Caizzi et al., 2014), evidence that ER may in fact regulate binding of FoxA1.
More recently it was demonstrated that multiple TFs can modulate each other’s binding patterns through a mechanism termed dynamic assisted loading. In this model, one factor can induce accessibility for another through the recruitment of ATP-dependent remodeling complexes that create transient open chromatin states (Biddie et al., 2011, Grontved et al., 2013, Miranda et al., 2013, Voss et al., 2011), allowing the secondary factor to bind. This model is distinguished from the classic pioneering concept by three important parameters; (1) the initiating and secondary binding factors can reverse roles, depending on the local chromatin environment, (2) residence times for the binding factors are quite short, measured in seconds (sec), and (3) a central role for ATP-dependent remodeling proteins is proposed (Voss and Hager, 2014, Voss et al., 2011).
Here, we show that activation of either ER or the glucocorticoid receptor (GR) induces the reprograming of the chromatin landscape in breast cancer cells, and results in the recruitment of FoxA1 to a subset of sites that were previous ly inaccessible. In addition, we find no evidence of FoxA1, ER, or GR footprints within DNaseI hypersensitive (DHS) sites in multiple breast cancer cells. Finally, as measured by single-molecule tracking (SMT), FoxA1 manifests a highly dynamic behavior in the nucleus (comparable to that of ER and GR), with relatively fast dwell times. Together, these results suggest that these factors interact dynamically with chromatin through a symmetric mechanism. That is, the SRs can induce loading of FoxA1, or FoxA1 can reorganize nucleoprotein states consistent with receptor binding.
We reported recently that ER and GR can facilitate the binding of one another at a subset of binding sites, suggesting that ER and GR have the ability to function as initiating factors for each other (Miranda et al., 2013). To further investigate the “pioneering mechanism” of FoxA1, ER, and GR we mapped the binding profiles for FoxA1 following activation of ER or GR in MCF-7, ZR-75-1, and T-47D breast cancer cell lines, three of the most commonly studied, estrogen responsive breast cancer models (Lacroix and Leclercq, 2004). Analysis of the direct overlap between FoxA1 and ER binding patterns and FoxA1 and GR binding patterns revealed a number of unique clusters (denoted by lower case letters). For FoxA1 binding sites identified in unstimulated and 17β-estradiol (E2) treated MCF-7 cells, a large proportion (~80%) appear to have no close range proximity to ER binding sites (Figure 1A, cluster “a”, “b”, “c”). For all ER sites stimulated by E2 treatment, 44% overlap with FoxA1 sites active in unstimulated cells and E2 treated cells (cluster “e”). A lesser yet significant proportion (~14%) of ER binding sites overlap with FoxA1 sites functional only in E2 stimulated cells (cluster “f”) with only ~0.8% of ER sites overlapping with unstimulated FoxA1 sites in the absence of E2 (cluster “g”). The individual binding intensities of the FoxA1 and ER peaks have been compared via heat map analysis, indicating the presence or absence of FoxA1 and ER binding at each binding group described by the Venn diagram [denoted by lower case letters (Figure 1C)]. The locations of the FoxA1 and ER peaks and the average tag count have been presented for each of the three significant binding groups by aggregate plots, confirming the uniqueness of each binding group (Figure 1E). These findings indicate that the relationship between FoxA1 and ER is more complex than previously reported.
The interaction between FoxA1 and GR in breast cancer cells has not been comprehensively explored. In fact, ~69% of dexamethasone (Dex) inducible GR sites overlap with FoxA1 sites present in unstimulated and Dex treated cells (Figure 1B, cluster “l”), suggesting that FoxA1 plays a significant role in GR recruitment. A large fraction of FoxA1 sites (~77%) are unrelated to GR binding (cluster “h”, “j”, “i”). Seventeen percent of GR sites have no overlap with FoxA1 binding (cluster “k”) and a smaller number (~12%) overlap with FoxA1 that is bound only in Dex stimulated cells (Figure 1B, (cluster “m”)). Heat map analysis (Figure 1D) presents individual intensities for FoxA1 and GR for the binding groups denoted by lower case letters. The average tag count and binding location of GR and FoxA1 is shown via histogram (Figure 1F). Thus the hormone activated differences in FoxA1 genome-wide occupancy are similar between GR and ER. Analysis of FoxA1, ER, and GR binding in ZR-75-1 (Figure S1A–B) and T-47D (Figure S1C–D) cells presents similar findings, supporting the results identified in MCF-7 cells.
Of particular interest is the subset of ER and GR binding sites that overlap with FoxA1 only after hormone stimulation ( “f” and “m”), as well as the FoxA1 population of binding sites altered by hormone treatment, yet not overlapping with a SR site (“a” and “h”). These classes represent sites where ER and GR serve as initiating factors for FoxA1. To investigate this observation further we first determined, genome-wide, the binding locations and distances between all FoxA1 and ER peaks (Figure 2A–2B) and all FoxA1 and GR peaks (Figure 2C–2D). We combined all FoxA1 binding sites identified in untreated and E2 samples and untreated and Dex samples. The cumulative distribution of FoxA1 peaks to the closest ER binding event (broken line) and closest estrogen response element (ERE) motif (solid line) has been determined at a range of 0–500 base pairs (bp) (Figure 2A) and 0–10,000 bp (Figure 2B). This distribution was also determined for all FoxA1 peaks in relation to closest GR binding event (broken line) and the closest glucocorticoid response element (GRE) motif (solid line) at a range of 0–500 bp (Figure 2C) and 0–10,000 bp (Figure 2D). Less than 5% of all FoxA1 peaks are located within 100 bp of an ERE motif and ~15% of an ER peak (Figure 2A). Furthermore, only 2.5% of FoxA1 peaks are within 100 bp of a GRE motif and ~20% of a GR peak (Figure 2C). These close range binding classes represent events where potential localized nucleosome reorganization by FoxA1 could lead directly to SR binding. However at longer ranges, only ~30% and ~25% of FoxA1 peaks are located within 10,000 bp of an ER peak (Figure 2B), or a GR peak (Figure 2D), respectively. In addition, ~80% of FoxA1 peaks are located within 10,000 bp of the closest ERE motif (Figure 2B), and ~55% within 10,000 bp of the closest GRE motif (Figure 2D), indicating there are many unbound ERE and GRE motifs within 10,000 bp of a FoxA1 peak. These findings are quite unexpected, given the widely-described behavior of FoxA1 as a PF for SR function.
A large number of ER and GR sites pioneered by FoxA1 lack a binding interaction and SR motif within 100bp. To look more closely at this observation we examined the FoxA1 peaks identified in the Venn diagrams in Figure 1. Specifically we used the ER and GR peaks that overlap with FoxA1 peaks present only in hormone treated cells (“f” and “m”), ER and GR peaks that overlap with FoxA1 in unstimulated and hormone treated cells [classically considered to be ER and GR peaks pioneered by FoxA1 (“e” and “l”)] and FoxA1 peaks unique to hormone treatment in the absence of an ER or GR peak (“a” and “”h”) (Figure 2E and 2F). Of the classical pioneer FoxA1 sites for ER (“e”), ~90% are located within 100 bp of an ER peak (Figure 2E) and for classical pioneer FoxA1 sites for GR (“l”), ~93% located within 100 bp of a GR peak (Figure 2F). Therefore of the small number of FoxA1 sites (~15–20%) that function as a PF for either ER or GR ~90% of those are located within 100bp of an ER or GR peak. However, an important observation is that FoxA1 peaks overlapping ER and GR in the hormone treated groups only (“f” and “m”) have a higher percentage (95% and 97% respectively) located within 100 bp of an ER or GR peak (Figure 2E and 2F). Further, approximately 60% of these peaks are within 10 bp of an ER or GR peak compared to only ~35% of the observed classical FoxA1 peaks known to pioneer for ER or GR (“e” and “l”) (insert of Figure 2E and Figure 2F). These findings suggest there is populations of FoxA1 peaks regulated by hormone in close proximity of the regulating TF, signifying that ER or GR have the potential to recruit FoxA1 to binding locations.
To investigate the potential hormone regulation of FoxA1 binding pattern we further analyzed the FoxA1, ER, and GR chromatin immunoprecipitation (ChIP)-sequencing (seq) data to determine the differential hormone regulation of FoxA1. This analysis of FoxA1 binding sites show that ER and GR can also recruit FoxA1 to a subset of specific sites in MCF-7, ZR-75-1, and T-47D cells (ZR-75-1 and T-47D Figure S2A–B). In MCF-7 cells there are 19,068 FoxA1 binding sites found to be in common with untreated and E2 treated cells and 1,219 binding sites that are gained or lost upon E2 stimulation (Figure 2G). Of the total FoxA1 binding sites identified in cells treated with Dex or left untreated, 18,142 are found to occur in both treatment conditions, with 571 FoxA1 sites gained and 72 lost with Dex treatment (Figure 2H). This demonstrates that activated ER and GR have the ability to re-distribute the binding patterns of FoxA1, consistent with a previous report whereby the dual activation of ER and GR reprograms the binding landscape through the gain and loss of binding sites (Miranda et al., 2013). Together these results suggest that FoxA1, ER, and GR in breast cancer cell lines each have the capability of altering a subset of binding sites, reinforcing the notion that multiple factors can function in a “pioneering” mode.
To further define the specific role of each receptor in redistributing FoxA1 binding across the genome, supervised clustering analysis of FoxA1, ER, and GR ChIP-seq data in each cell line has been performed to extract specific binding modules. In MCF-7 cells three unique binding clusters have been identified for FoxA1 and ER sites induced and lost by E2 (Figure 3A–B). Cluster 1 (122 peaks) represents FoxA1 sites that are lost upon E2 treatment and do not overlap with ER binding. Cluster 2 (625 peaks) includes FoxA1 sites that are gained by E2 treatment, but do not overlap with ER binding sites. The lack of overlap with ER at cluster 2 suggests that FoxA1 binding at these sites is either through a tethering interaction or is a representation of a hit and run event (McNally et al., 2000, Hager et al., 2002) whereby ER is binding with such a short residence time that a ChIP signal is not detected. Lastly, cluster 3 represents unique sites whereby FoxA1 binding is gained with E2 treatment only and overlaps with ER binding sites (470 peaks) suggesting ER activation may result in ER-induced FoxA1 sites through assisted loading.
In addition, four FoxA1 and GR binding clusters have been identified (Figure 3D–E). Specifically, cluster 1 (236 peaks) represents FoxA1 sites gained with Dex treatment and overlap with GR binding sites. This cluster also demonstrates that GR activation may recruit FoxA1 to specific sites through a GR-induced mechanism (GR-induced FoxA1 sites). Cluster 2 (335 peaks) represents FoxA1 binding sites gained with Dex treatment that do not overlap with GR binding sites, again suggesting a potential hit and run event for FoxA1 recruitment. Cluster 3 (16 peaks) represents FoxA1 sites lost upon Dex treatment and overlap with GR sites. Cluster 4 also contains FoxA1 lost sites (56 peaks); however, these peaks do not overlap with GR binding sites. Supervised clustering analysis of ZR-75-1 cells (Figure S3A–B) and T-47D cells (Figure S3C–D) for ER, GR, and FoxA1 reveal very similar binding pattern.
Interestingly, the specific ER and GR-induced FoxA1 sites identified in the three cell lines reveal very little overlap, suggesting that while the mechanism of assisted loading is active in breast cancer cells, the binding sites are cell line specific (Figure S3E–F). This indicates that different types of breast cancer may have an altered assisted loading binding pattern. It has previously been demonstrated that a number of FoxA1 and ER binding sites are cell line specific in MCF-7, ZR-75-1, and T-47D cells (Hurtado et al., 2011). This supports the concept that multiple TF binding sites can be largely unique to individual cell lines.
To further examine the direct role of ER and GR on dictating FoxA1 binding at the induced sites identified in MCF-7 cells, de novo motif analysis has been performed. At ER-induced FoxA1 sites there is a strong ERE and a weaker FoxA1 motif. In contrast, FoxA1 classical binding sites observed in both the untreated or E2 treated cells, the FoxA1 binding element is the most highly enriched motif (Figure 3C). At the GR-induced FoxA1 sites a GRE element is the most highly enriched motif and there is a weak FoxA1 motif (comparable to the ER-induced FoxA1 sites). At FoxA1 classical sites observed in the untreated and Dex treated cells, the FoxA1 motif is highly prevalent (Figure 3F). Thus, while FoxA1 binding can be modulated by activation of SRs through an assisted loading mechanism, FoxA1 also dictates ER and GR binding at a small set of FoxA1 specific sites. In addition, there is an AP-1 motif identified at the ER and GR-induced FoxA1binding sites suggesting AP-1 may be playing a functional role with SRs to assist the loading of FoxA1. This supports previous findings where AP-1 is a required component of GR binding at 40% of binding sites (Biddie et al., 2011). Further, it has also been identified that AP-1 is required for GR induced binding of ER sites (Miranda et al., 2013). Together these findings demonstrate that ER and GR have the ability to reshape the FoxA1 binding landscape, in collaboration with factors like AP-1, frequently acting as initiating factors for FoxA1.
To further understand the chromatin landscape of sites identified at the SR-induced FoxA1 sites, we performed DHS-seq analysis in MCF-7 and ZR-75-1 cells under E2 or Dex treatment, and characterized the change in DHS at all identified clusters in both cell lines (Figure 4 and Figure S4). In MCF-7 cells at ER-induced FoxA1 sites (cluster 3), there is an increase in DHS upon E2 treatment (Figure 4A). The same pattern is observed in chromatin accessibility at GR-induced FoxA1 sites (Figure 4B). These same results are also observed at the clusters identified in ZR-75-1 cells (Figure S4A–B). Of particular note, FoxA1 sites activated with E2 or Dex that do not overlap with SR binding (cluster 2) are also remodeled, indicating the receptor is involved in chromatin remodeling despite the absence in binding, either by hit and run, or by tethering. Examples of specific binding locations are shown as UCSC browser shots for ER-induced FoxA1 sites (Figure 4A) and GR-induced FoxA1 sites (Figure 4B).
Our results from multiple cell lines demonstrate that while the specific binding sites have unique cell specificity, the assisted loading mechanism is functional in breast cancer cells through ER and GR modulation of FoxA1 binding patterns. Thus ER and GR, much like FoxA1 (Hurtado et al., 2011), can cause changes in chromatin accessibility allowing for the recruitment of other TFs to specific response elements.
When occupied by their cognate factor, TF binding sites on DNA are resistant to nuclease digestion, and can be monitored by the generation of a TF footprint (Galas and Schmitz, 1978). Similar patterns of protection have been identified in nuclease digested nuclei (Church et al., 1985), and are often inferred to represent stable binding of a factor at the recognition site. This interpretation of a footprint profile fundamentally conflicts with many studies showing very brief residence times for DNA binding proteins in living cells (Voss and Hager, 2014). Furthermore, it has been recently reported that factors with robust ChIP-seq peaks but short DNA residence times in live cells manifest either a minimal, or complete lack of, a footprint in vivo (Sung et al., 2014, He et al., 2014, Grontved et al., 2015).
FoxA1 has been considered a master PF and has been argued to present stable DNA interactions with low mobility (Caravaca et al., 2013, Sekiya et al., 2009), in contrast with the dynamic function of a number of TFs. To characterize FoxA1, ER, and GR intranuclear dynamics, we performed SMT for these factors (Movie S1–S6). Using HaloTag and Janelia Fluor 549 (JF549) HaloTag ligand versions of FoxA1, ER, and GR expressed in MCF-7 cells, we identified and tracked single molecules using highly inclined laminated optical sheet (HILO) illumination (Tokunaga et al., 2008). We observed three dwell time classes for FoxA1, ER, and GR molecules (Figure 5), unbound, fast, and slow. Quantitative analysis of large track sets (Figure S5A–F) shows that the dwell time distributions fit a double-component exponential decay model for all three TFs (Figure S5G–L), indicating the presence of two distinct bound-populations. The lifetimes of bound factor molecules resolve into short and longer lived populations, referred to as fast short-lived (Tns, fast) or slow long-lived (Ts, slow) fractions, respectively. Previous reports suggest that the slow long-lived population of molecules represents specific binding events associated with enhancers or promoters, while the fast short-lived class describes non-specific binding to chromatin (Morisaki et al., 2014). Interestingly, FoxA1 presents characteristic residence times of 1.78±0.03 sec at the fast short-lived fraction, and 10.8±0.5 sec at the slow long-lived fraction in the untreated state (Figure 5A, Figure S5A). This relatively fast dwell time of FoxA1, especially at specific binding sites, is inconsistent with a stable binding model for the FoxA1 factor. ER or GR activation leads to a modest yet significant modulation of the FoxA1 slow long-lived fractions, to 8.4±0.3 sec and 8.8±0.6 sec, respectively (Figure 5B–C Figure S5B–C). Thus activation of either SR slightly decreases the stability, on average, of FoxA1 binding to chromatin. The mean residence times that we measure reflect the average of residence times for FoxA1 at all endogenous response elements, including FoxA1 sites not associated with ER or GR assisted loading.
The residence times of both ER and GR were found to be comparable to that of FoxA1. Without hormone stimulation, ER presents two distinct bound populations of molecules with a fast short-lived and slow long-live of 0.81±0.01 sec and 4.4±0.2 sec respectively (Figure 5D, Figure S5D), suggesting some specific binding events to chromatin in a ligand-independent manner. This finding is consistent with studies indicating there are subsets of functional ER binding site in the unligand state in MCF-7 cells (Caizzi et al., 2014, Kong et al., 2011). Upon activation of ER, the residence time at the slow long-lived fraction is increased significantly averaging 11.7±0.6 sec (Figure 5E, Figure S5E). In addition, the percentage of ER molecules in slow long-lived fraction is markedly increased (Figure 5D and 5E), consistent with increased site specific binding for the receptor after ligand activation. GR manifests residence times of 1.24±0.02 sec and 7.4±0.4 sec in the Dex-treated state (Figure 5F, Figure S5F), similar to previous reports (Morisaki et al., 2014). Interestingly, while FoxA1 is significantly slower than GR+Dex complexes, ER+E2 is significantly slower than FoxA1 (Figure 5G). Furthermore, comparison of FoxA1, ER, and GR indicates that activated ER shows the slowest residence time of the three factors. Thus, the SMT data does not support a stable model of FoxA1 DNA binding, but indicates rather that FoxA1, like ER and GR (Figure 5G), is highly dynamic with relatively fast DNA residence times.
The question is now raised as to whether the binding events of FoxA1 can provide protection for its binding sites as detected by footprint analysis. An example of a well described footprint is shown for the CTCF protein, a transcriptional repressor, in MCF-7 untreated cells (Figure S6A). As previously described, this deep footprint correlates with a relatively slow DNA residence time for the CTCF factor (Boyle et al., 2011, Nakahashi et al., 2013, Siersbaek et al., 2014, Sung et al., 2014). On the contrary, in MCF-7 cells there is no genome-wide FoxA1 footprint detected in the untreated, E2, or Dex treated samples (Figure 6A–C). This finding is also observed in ZR-75-1 cells (Figure S6B–D). Furthermore, we observe the lack of a detectable ER or GR footprint in MCF-7 and ZR-75-1 cells, in E2 and Dex treated samples respectively (Figure 6D–E, Figure S6E–F), which supports previous studies (Sung et al., 2014). Furthermore, the ability to detect a FoxA1 footprint is not changed subject to protein binding intensity with the bottom 2000 and top 500 FoxA1 peaks not altering the outcome of footprint detection (Figure S6G–H). The footprint data, in conjunction with the SMT findings, indicate that FoxA1, like ER and GR, is a highly dynamic TF. This contrasts with previous models for FoxA1 interactions with chromatin, wherein FoxA1 is hypothesized to penetrate inaccessible chromatin in conjunction with nucleosomes and function as a PF through stable interactions with chromatin. These dynamic properties of FoxA1 indicate that the factor is not stably bound to chromatin with a slow mobility, and support a model wherein FoxA1, like GR and ER, is reprogrammable in the chromatin context.
FoxA1 has been widely discussed as a PF for SR recruitment to specific sites across the genome (Bernardo and Keri, 2012). Importantly, the interaction between FoxA1 and ER has been shown to be a prominent factor in breast cancer development (Yamaguchi et al., 2008). Several studies have argued that FoxA1 is a vital component for ER recruitment at the majority of sites in breast cancer cells (Carroll et al., 2006, Hurtado et al., 2011, Lupien et al., 2008). It is also been reported that upon inhibition of ER, no change in FoxA1 binding is observed (Lupien et al., 2008, Hurtado et al., 2011). These models envisage FoxA1 as an asymmetric PF that establishes the binding landscape for ER, while ER plays no role in FoxA1 binding and recruitment to chromatin. However, other studies, inconsistent with these findings, reported that FoxA1 recruitment is dependent on stimulation of cells with E2 at 29% of sites where both ER and FoxA1 bind, although no mechanism was determined (Kong et al., 2011). In addition, it was reported that upon knockdown of ER, FoxA1 binding is lost at unstimulated ER binding sites (Caizzi et al., 2014).
SRs have also been shown to modify the binding landscape for each other. As an example, ER and GR can each act as initiating factors for binding of the other in mammary cells (Miranda et al., 2013). Furthermore, an early report (Rigaud et al., 1991) found that GR could modulate FoxA1 binding at a specific GRE within the GR responsive unit of the rat tyrosine aminotransferase gene. This enhanced binding of FoxA1 was associated with GR dependent disruption of nucleosomal structure, and suggests that the role of SRs in the regulation of FoxA1 responses is not well understood.
Alternative mechanisms have been suggested for the “pioneering” action of TFs. AP-1 was shown to be necessary for opening chromatin at 40% of GR binding in mammary cells (Biddie et al., 2011). However, at a smaller number of sites in the same cells, GR serves as an initiating factor for AP-1 binding. Finally, at some sites, the initiating factor cannot be co-resident with the secondary loaded factor (Voss et al., 2011, Rigaud et al., 1991). Taken together, these findings suggest a more dynamic and symmetric model for the pioneering action of these factors, a mechanism that has been termed “dynamic assisted loading” (Voss et al., 2011).
The confusing and contradictory findings led us to examine the pioneering roles of these proteins in more detail. Interactions between FoxA1 and SRs were characterized here with multiple methodologies. Global ChIP-seq analysis reveals that a significant fraction of FoxA1 binding events are dependent on either GR or ER, in agreement with Kong et al. and Caizzi et al. Furthermore, each of the receptors was found to initiate chromatin opening, as determined by DNaseI hypersensitivity, at sites where FoxA1 loading was dependent on the SR. Furthermore, these sites tend to contain a weaker FoxA1 motif suggesting after ER or GR induction, FoxA1 binding intensity changes reflecting transient FoxA1 and DNA interactions. These results indicate that the SRs have the potential to bind chromatin and recruit remodelers to make the sites more accessible, allowing for the secondary loading of FoxA1 at these sites.
If FoxA1 is bound stably to chromatin in the classic wedging model, it would be expected to produce a footprint of protection at pioneering sites. However, TFs that are highly dynamic, with rapid chromatin exchange rates, have been shown to lack a nuclease resistant footprint (He et al., 2014, Sung et al., 2014). ER and GR each display highly dynamic binding events with a short residence time on chromatin (McNally et al., 2000, Stenoien et al., 2001). As shown here, FoxA1 also fails to manifest a detectable footprint at binding sites in the genome, either SR-dependent or independent. The relatively fast residence times measured by the SMT experiments indicate that the binding of FoxA1 is also a highly dynamic process, with rapid FoxA1 exchange at sites in chromatin. Another PF, SOX2, was recently shown to present a transient interaction with chromatin, on the order of 12–14 sec (Chen et al., 2014). These findings again support the concept of PFs as highly mobile during functional enhancer interactions.
Finally, we note that the discussion of pioneer protein function has centered almost exclusively on potential interactions between TFs and nucleosomes. In fact, where examined in detail, localized chromatin transitions are often associated with the action of ATP-dependent remodeling complexes (Mellor, 2006, Yoo and Crabtree, 2009). Several reports have shown that factors can be mobile during these remodeling processes (Kassabov et al., 2002, McKnight et al., 2011, Nagaich et al., 2004). A recent study by Wang and colleagues (Li et al., 2015) elegantly demonstrated that the Swi/Snf complex can directly displace a bound TF, supporting previous suggestions for factor mobility during remodeling. As it is well established that SRs recruit remodeling proteins, the action of these complexes in the symmetric recruitment described here requires further attention. It is possible ATP-dependent remodeling systems are more centrally involved in pioneering factor action than previously appreciated. To seek further clarification of the direct role of ER and GR in the assisted loading mechanism future studies involving mutation of the SR DNA binding domain or direct mutation of the FoxA1 binding site could be performed. This would determine whether the SRs bind to their cognate site directly facilitating FoxA1 binding. In conclusion, the results presented here support a symmetric model wherein multiple TFs can serve a pioneering function, largely dependent on the local chromatin structure at a given site. In addition, PFs FoxA1 and SOX2 each manifest highly dynamic interactions with chromatin, with exchange rates comparable to those for SRs and other highly mobile factors.
The MCF-7, ZR-75-1, and T-47D breast cancer cell lines have been utilized in this study. Additional description of cell culture conditions is provided in supplemental procedures.
ChIP experiments have been performed as per standard protocols (Kimura et al., 2008, Miranda et al., 2013). Briefly, cells are treated with either 100nM E2, 100nM Dex, or left untreated for 30 minutes (m). The following antibodies have been utilized: ER cocktail (Ab-10, Labvision; sc-543, Santa Cruz Biotechnology), GR (sc-1003, Santa Cruz Biotechnology), and FoxA1 (ab23738 Abcam). Two biological replicates are pooled resulting in one technical replicate. Two technical replicates per treatment group have been sequenced.
The DHS assay has been performed as previously described (John et al., 2011, Hesselberth et al., 2009). MCF-7 and ZR-75-1 cells are treated with either 100nM E2, 100nM Dex, or left untreated for 30 m. Two technical replicates per treatment group have been sequenced. Additional description of DHS experimental procedure is provided in supplemental procedures.
Sequence reads have been generated and the unique tags are aligned to the human reference genome (UCSC hg19 assembly) for all ChIP- and DHS-seq data from MCF-7, ZR-75-1, and T-47D breast cancer cell lines. All sequenced data has been distributed in the GEO under GSE72252 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=mhkzgqcwzjqtbmj&acc=GSE72252.) Further description of the method can be obtained from the supplemental procedures.
Ultra deep sequencing of DHS libraries has been performed in MCF-7 and ZR-75-1 cells treated with either 100nM E2, 100nM Dex, or left untreated for 30 m. DHS cut count libraries have been generated as previously described (Siersbaek et al., 2014). The average DNaseI cleavage of observed footprints, expected footprint, and log ratio of observed and expect has been plotted at a bp resolution centered on the TF motif. Further description of DHS footprint analysis procedure can be obtained from the supplemental procedures.
The MCF-7 cells have been transiently transfected with HaloTag-FoxA1, HaloTag-ER, or HaloTag-GR respectively to achieve appropriate protein levels for single molecule visualization. Cells are treated with 5nM JF549 HaloTag ligand (Grimm et al., 2015). MCF-7 cells are treated with either 100nM E2, 100nM Dex, or left untreated for 30 m and cells are then imaged for ~ 2 hours at 37 °C in 5% CO2 using a custom built microscope. Specific details and description of SMT analysis including information on plasmid constructs can be obtained in supplemental procedures.
The authors thank the National Cancer Institute Advanced Technology Program Sequencing Facility for sequencing services. The research was supported, in part, by the Intramural Research Program of the NIH, NCI, Center for Cancer Research and the Howard Hughes Medical Institute. E.E. Swinstead was supported, in part, by an Australian Postgraduate Award. T.B. Miranda was supported, in part, by a National Institute of General Medical Sciences Pharmacological Research and Training Fellowship. V. Paakinaho was supported, in part, by the Sigrid Jusélius Foundation. D. Mazza was supported, in part, by a Marie Curie International Incoming Fellowship (GA: 27432). T. Morisaki was supported, in part, by the Japan Society for the Promotion of Science. L. Grøntved was supported by the Danish Research Council and SDU2020. The authors declare the following competing financial interests: J.B.G. and L.D.L. have filed patent applications on the Janelia Fluor dyes, such as JF549, whose value may be affected by this publication.
Author ContributionsE.E.S. and T.B.M. performed all genome-wide experiments, and prepared the manuscript. G.L.H. initiated and directed the project. V.P. and D.M.P. carried out SMT. I.G. participated in manuscript and figure preparation. T.M. and D.M. constructed the HILO microscope. T.S.K. and D.B. provided imaging instrumentation and programming support. L.G. and S.B. carried out the footprinting analysis. M.H. supported the ChIP-seq experiments. L.L. and J.B.G provided SMT chromophores.
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.