Evidence That Partial Unwrapping of DNA from Nucleosomes Facilitates the Binding of Heat Shock Factor following DNA Replication in Yeast*

In the yeast Saccharomyces cerevisiae, heat shock transcription factor (HSF) binds heat shock element (HSE) DNA shortly after DNA replication, independently of its activation by heat shock. To determine if HSF binding occurs before newly replicated DNA is packaged into nucleosomes, we inserted an HSE into a DNA segment that normally forms a positioned nucleosome in vivo. Transcription from constructs designed to create steric competition between binding of HSF and histone H2A-H2B dimers was generally poor, suggesting that nucleosome assembly precedes and inhibits HSF binding. However, one such construct was as transcriptionally active as a nucleosome-free control. Structural analyses suggested that ∼40 base pairs of DNA, including the HSE, had unwrapped from the 3′ edge of the histone octamer, allowing HSF to bind; ∼100 base pairs remained in association with the histone octamer, with the same translational and rotational orientation as was seen for the poorly transcribed constructs. Modeling studies suggest that the active and inactive constructs differ from one another in the ease with which the HSE and flanking sequences can adopt the curvature needed to form a stable nucleosome. These differences may influence the probability of DNA unwrapping from already assembled nucleosomes and the subsequent binding of HSF.

In the yeast Saccharomyces cerevisiae, heat shock transcription factor (HSF) binds heat shock element (HSE) DNA shortly after DNA replication, independently of its activation by heat shock. To determine if HSF binding occurs before newly replicated DNA is packaged into nucleosomes, we inserted an HSE into a DNA segment that normally forms a positioned nucleosome in vivo. Transcription from constructs designed to create steric competition between binding of HSF and histone H2A-H2B dimers was generally poor, suggesting that nucleosome assembly precedes and inhibits HSF binding. However, one such construct was as transcriptionally active as a nucleosome-free control. Structural analyses suggested that ϳ ϳ40 base pairs of DNA, including the HSE, had unwrapped from the 3 edge of the histone octamer, allowing HSF to bind; ϳ ϳ100 base pairs remained in association with the histone octamer, with the same translational and rotational orientation as was seen for the poorly transcribed constructs. Modeling studies suggest that the active and inactive constructs differ from one another in the ease with which the HSE and flanking sequences can adopt the curvature needed to form a stable nucleosome. These differences may influence the probability of DNA unwrapping from already assembled nucleosomes and the subsequent binding of HSF.
Transcription factors in eukaryotic cells must contend with DNA that is packaged in nucleosomes. Nucleosomes contain approximately 146 bp 1 of DNA wrapped around an octamer of histone proteins and are separated from one another by linker DNA segments of varying length (1, 2). Nucleosomes that package promoter DNA interfere with the binding of some transcription factors and thereby prevent the assembly or activa-tion of transcription initiation complexes (3,4). For certain genes and chromosomal domains, transcription is further suppressed by DNA sequence-specific binding factors that appear to position or stabilize nearby nucleosomes (e.g. Refs. 5 and 6 and references therein). Such selective, nucleosome-mediated repression is established shortly after DNA replication (e.g. Refs. 7-10).
Cells use a variety of mechanisms to overcome nucleosomemediated repression of transcription (11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23). First, replication partially disrupts preexisting nucleosomes and creates new DNA that also must be packaged into chromatin (24 -27). In principle, this provides an opportunity for de novo binding of transcription factors prior to the assembly of new nucleosomes (e.g. Refs. 28 and 29). Second, DNA in many nucleosomes adopts a preferred rotational orientation with respect to the histone octamer, due to a nonrandom distribution of short GC-rich DNA segments in the eukaryotic genome (30). GC-rich DNA segments do not readily tolerate minor groove compression associated with the wrapping of DNA about a histone octamer and thus tend to face outward from the histone octamer (31). As a result, some transcription factors can bind to preexisting nucleosomes, in part because their targets are correctly oriented with respect to the histone octamer (e.g. Refs. 32 and 33). Other factors are unable to bind nucleosome-packaged targets, regardless of their rotational orientation (e.g. Ref. 34). Third, DNA in some nucleosomes adopts a preferred translational position with respect to the histone octamer, such that cis-acting regulatory elements occur in more accessible linker regions (e.g. Ref. 35 and references therein). Finally, cells contain accessory factors that modify histones to facilitate transcription factor access as well as factors that alter or disrupt preexisting nucleosomes. These accessory factors enable many transcription factors to bind their DNA targets in a replicationindependent manner.
Heat shock factor (HSF) activates specific genes in response to stresses such as hyperthermia and anoxia (36,37). Early in vitro reconstitution studies indicated that HSF is unable to bind to nucleosomal DNA unless TFIID is allowed to bind DNA prior to nucleosome assembly (38,39). Wu and colleagues later isolated from Drosophila extracts a nucleosome remodeling factor that enables HSF to bind to chromatin templates in vitro (Ref. 40 and references therein). Meanwhile, in vivo competition experiments in yeast indicated that HSF could bind to a limited number of target sites within but near the edge of nucleosomes (41). It is possible that the observed in vivo binding of HSF to nucleosomal DNA was dependent on the prior binding of TFIID, since later studies indicated that TFIID binds to the promoter used in those experiments independently of upstream regulatory factors (42). This seems unlikely, however, since overexpression of HSF enables it to bind nucleosomal DNA in vivo in the absence of TFIID (43). It is also possible that HSF binding to nucleosomal DNA in yeast depends on nucleosome remodeling factors, although the observed HSF binding did not detectably disrupt the HSE-containing nucleosome (41). Finally, the in vivo binding capacities of HSF could stem from the fact that, in yeast, HSF binds HSE DNA independently of its activation by heat shock (41, 42, 44 -49). This implies that HSF binds soon after replication of HSE DNA. Consequently, the outcome of competition between binding of HSF and nucleosome formation may depend on the time of HSF binding relative to that of nucleosome assembly. To test this possibility, we set out to determine whether HSF binds newly replicated DNA before or after it is packaged into nucleosomes.
Nucleosomes assemble with newly replicated DNA in discrete steps, beginning with the rapid binding of H3-H4 tetramers; H2A-H2B dimers bind in a second independent step (22,50). We previously conducted in vivo competition experiments to assess the ability of HSF to bind to DNA in a region of the nucleosome that is organized primarily by histone H2A-H2B dimers (51,52). Modeling studies suggested that HSF is able to bind if critical determinants in the HSE face outward from the histone octamer (41). These studies suggested that we could create competition between the binding of HSF and a histone H2A-H2B dimer by rotating the HSE determinants so that they now faced into the histone octamer. Failure of HSF to bind such a construct would suggest that HSF binds after the addition of H2A-H2B dimers to nascent nucleosomes (and vice versa), as illustrated in Fig. 1. In this paper, we have systematically rotated the HSE relative to the preferred orientation of the histone octamer, in 1-bp steps. We find that nucleosome formation interferes with efficient in vivo binding of HSF to most of the resulting constructs, suggesting that H2A-H2B dimers bind prior to HSF. However, HSF binds efficiently to one construct also predicted to present steric conflict between HSF and histone H2A-H2B dimers. Analysis of this construct suggested that HSE DNA had unwrapped from the histone octamer, allowing HSF to bind without complete disruption of the underlying nucleosome. Modeling studies suggest that the HSE and surrounding DNA in this particular construct is less able to adopt the curvature needed to form a stable nucleosome than are most of the other rotational constructs. This may promote transient unwrapping of DNA from the edge of an already assembled nucleosome, thus permitting HSF to bind.

EXPERIMENTAL PROCEDURES
Plasmid Construction, Transfection, Chromosomal Integration, and Assay of Transcriptional Activity-Plasmids used in this study are depicted in Fig. 2. p91.3 through p100.3 constitute a series in which the helical orientation of a HSF DNA-binding element (heat shock element; HSE) changes 1 bp at a time relative to the underlying nucleosome positioning determinants. These and the control plasmid p41.3 were derived, using standard methods (53), from the lacZ expression plasmids described by Pederson and Fidrych (41). For the present study, we restricted HSF binding to a unique orientation relative to the histone octamer by replacing the four-part HSE in plasmids pZJd41, pZJd91, and pZJd100 with a three-part HSE (5Ј-GCTAGCTTCTAGAAGCTTCT-GTCGAC-3Ј), forming plasmids p41.3, p91.3, and p100.3, respectively. p91.3 then was used as a template for PCR-mediated synthesis of eight BglII-NheI fragments, each containing nucleosome positioning determinants (cf. Ref. 41) but lacking 1-8 bp of DNA immediately 5Ј of the NheI site. By replacing the BglII-NheI fragment in p91.3 with these PCR-generated fragments, we created plasmids p92.3 through p99.3. All substitutions were verified by DNA sequencing. These manipulations did not create any new HSEs, nor did they alter major positioning determinants in the nucleosome positioning element (NPE) (see Figs. 5 and 6). The single HSE in each of these plasmids was linked to the proximal promoter and translation start site of the CYC1 gene, and to the coding region of the Escherichia coli lacZ gene. The sequence and distance between the HSE and the downstream promoter elements are identical for all plasmids. Each construct was transfected into Saccharomyces cerevisiae strain TD4 (MATa ura3-52 his4 leu2 trp1 gal2 [cir ϩ ]; from G. Fink, MIT), and the heat shock-inducible transcriptional activ-ity of six independently isolated transformants was assayed as described (41). To minimize the variation commonly observed with this reporter gene, all measurements were made using transformants grown in an identical manner to nearly identical cell densities. Transformants with expression levels closest to the mean were used for chromatin structure analyses. To integrate the constructs, a 7.0-kilobase pair PstI fragment, containing URA3 and the E. coli lacZ reporter gene with its attached NPE-HSE promoter, was gel-purified and ligated to a 2.4kilobase pair PstI fragment from the yeast integrating vector YIp5. The resulting plasmid was linearized by ApaI cleavage within URA3 and transformed into strain TD4. URA ϩ transformants were analyzed by Southern blotting to ensure that each contained a single copy of the desired HSE-NPE construct, integrated at the ura3-52 locus. Heat shock-inducible ␤-galactosidase activities were assayed as described above.
Chromatin Structure Analyses-Procedures for cell growth, preparation of metabolically active spheroplasts, heat shock of spheroplasts, and isolation of nuclei were exactly as described previously (41). Nuclei isolated by these methods support transcription elongation reactions and contain intact DNA, histones, and HSF (54). 2 Nuclei were treated at 25°C with varying amounts of micrococcal nuclease (MNase) or DNase I, and DNA was isolated as described previously (41). Double strand nuclease cleavage sites were mapped relative to both upstream and downstream EcoRV sites by indirect end labeling, as described previously (55). Single strand nuclease cleavage sites were mapped at single base pair resolution, using PCR-mediated reiterative primer extension reactions, as described previously (41), except that 0.2-ml rather than 0.5-ml tubes were used, eliminating the need for a mineral oil overlay, and primer annealing temperatures were increased from 45 to 50°C. As before, PCR templates were cleaved at nearby restriction sites to ensure efficient denaturation of templates and annealing of oligonucleotide primers. Use of linearized templates also results in a parental band whose intensity reflects the fraction of molecules cleaved between the primer and the restriction site. Quantification of this band provided assurance that comparisons made in this paper were between samples digested to equivalent extents and based largely on single hit nuclease data. This reduces the possibility of artifacts due to chromatin disruption over the course of digestion. Using these procedures, we have shown that mutations that affect the binding of factors to defined genetic elements in vitro and the function of these factors in vivo also alter our genomic footprints (42,47). Thus, our genomic footprinting methods faithfully reflect protein-DNA interactions that occur in cells.
Nuclei to be treated with restriction enzymes were washed once in the supplier's recommended restriction enzyme buffer, suspended in the same buffer at approximately 5 ϫ 10 8 nuclei per ml, and treated with varying amounts of enzyme for 30 -60 min at 30°C. Digestions were stopped, and DNA was isolated as described previously (41). The extent of cleavage at specific restriction sites was measured either by reiterative extension of single end-labeled primers, in 20-cycle PCR reactions containing 250 ng of template, or by exponential, 16-cycle PCR reactions containing 20 ng of template and end-labeled primers that flanked the restriction site of interest (cf. Fig. 4). These reaction parameters gave a uniform amplification per reaction cycle, such that the amount of product was directly proportional to the amount of intact template. Parallel reactions using primers that do not flank cleavage sites controlled for PCR efficiency. PCR products were fractionated on 5% sequencing gels and quantified by phosphor imagery.

RESULTS
Experimental Design-In S. cerevisiae, HSF binds DNA shortly after replication, regardless of whether it has been activated by heat shock. (Activation of HSF does, however, appear to stabilize its binding to DNA in vivo (41,48)). This property made HSF a good model for investigating the relative order of binding of transcription factors and histone 2A-2B dimers to newly replicated DNA in vivo. HSF binds as a homotrimer to HSEs. The consensus HSE sequence is 5Ј-NTTC-NNGAANNTTCN-3Ј (36,37). Methylation interference and structural analyses indicate that HSF contacts DNA determinants in the major groove and that each HSF monomer interacts with a single 5Ј-NGAAN-3Ј repeat (56 -58). The two 5Ј-NGAAN-3Ј repeats spaced 10 bp apart lie approximately on the same face of the DNA helix and are separated from the third 5Ј-NGAAN-3Ј repeat by 7 or 3 bp (ϳ108°). This suggests that HSF binding determinants occur largely on one side of the DNA helix, consistent with our earlier finding that HSF can bind to a limited number of sites in nucleosomes in vivo (41). Those studies also suggested that we could create competition between the binding of HSF and a histone H2A-H2B dimer by rotating the HSF binding determinants so that they face into the histone octamer.
To systematically rotate the HSE relative to the preferred orientation of DNA in a nucleosome, we inserted a synthetic HSE into a NPE, a naturally occurring DNA segment known to form a positioned nucleosome in vivo (41,59,60). Previous studies indicated that insertion of an HSE within but near the edge of this NPE would not interfere with the positioning determinants (Ref. 41 and Figs. 5 and 6). The NPE-HSE hybrid DNA then was inserted upstream of the CYC1 proximal promoter linked to a lacZ reporter gene. By deleting DNA between the HSE and the upstream, presumptive nucleosome dyad, 1 bp at a time, we formed a complete rotational series, consisting of plasmids p91.3 though p100.3 (Fig. 2). The HSE in the control plasmid p41.3 is located between the NPE and TATA boxes in a nucleosome-free segment (Ref. 41 and Fig. 5). It is important to note at the outset that the free energy of intrinsic positioning for any nucleosome is relatively modest. For example, GC-rich segments in the NPE, spaced at integral multiples of the DNA helical repeat, probably contribute to its rotational positioning. High salt partitioning experiments suggest that segments such as these contribute at best only a few kcal/mol to the free energy of nucleosome formation (61). Likewise, nucleosomes that have a single predominant translational position sometimes occupy minor translational positions that differ from the major position by integral multiples of the DNA helical repeat (32,62). This implies that the free energy of translational positioning also is modest. If the free energy of positioning of the NPE nucleosome is comparable with that of other well positioned nucleosomes, the predominant nucleosome configuration for a particular rotational construct may occur in, for example, only ϳ75% of all molecules. If we imagine that this configuration inhibited HSF binding 9-fold but that HSF were able to bind efficiently to the remaining 25% of the molecules, we would observe an overall inhibition of 3-fold. Thus, the apparent differences among the rotational variants in the effect of nucleosome formation on HSF binding would be less than the actual differences but would still accurately reflect the effect of nucleosome formation on HSF binding.
Three experimental parameters in this study made it possible to use heat-inducible transcriptional activity to measure the effect of nucleosome formation on HSF binding. First, in all of the constructs used in this study, the HSE itself and the DNA 3Ј of the HSE were identical. Therefore, the efficiency with which HSE-bound HSF interacts with its downstream targets was identical in all constructs. Second, more than one HSE is required to achieve maximal heat shock induction (47). Consequently, the activity of promoters used in this study was HSF-limited. Third, the extent to which HSF activates transcription is proportional to its DNA binding affinity (44). Finally, the copy number of the various transfected DNAs was uniform (8 -10 copies/haploid cell; data not shown) and far lower than is required to titrate endogenous HSF (63,64). To be sure of this point, we also integrated each construct, thus enforcing a uniform copy number of one per haploid genome (see "Experimental Procedures"). As predicted, transcription from the integrated constructs averaged 8 -10-fold less than observed for the corresponding plasmid constructs (Fig. 2). Given these parameters, any differences in heat shock-induc-ible transcription could be attributed to the influence of the NPE nucleosome on HSF binding.

Transcription Measurements Suggest That Nucleosome Formation Precedes and Interferes with HSF Binding in Vivo but to
Varying Extents- Fig. 1 illustrates possible outcomes of competition between nucleosome formation and HSF binding to newly replicated DNA. If HSF were to bind prior to the packaging of its target by histone 2A-2B dimers, then nucleosome maturation would be inhibited ( Fig. 1, Pathway A). In this case, all constructs should be uniformly active, with transcription levels similar to that of the nucleosome-free p41.3 control. On the other hand, if nucleosomes assemble first, they would block HSF binding, except in a few instances where critical HSF binding determinants faced outward from the histone octamer ( Fig. 1, Pathway B). Based on our earlier studies (41), we predicted that, if the NPE nucleosomes formed, the HSE would be accessible to HSF in p100.3 and p99.3 but largely inaccessible in all other rotational variants. (Modeling studies suggested that the adjacent DNA gyre in the NPE nucleosome of p91.3 would interfere with HSF binding more than in p99.3 or p100.3. While the relatively high levels of transcription in the episomal versions of p99.3 and p100.3 are consistent with these predictions, transcription from the integrated versions of p99.3 and p100.3 was lower than predicted (Fig. 2). We did not investigate the basis for this difference since, for reasons outlined below, the focus of the present study is on constructs p41.3, p95.3, and p97.3.) With one notable exception (p95.3; see below), the rotational constructs designed to create competition between binding of HSF and histone H2A-H2B dimers were 2-3-fold less active than the nucleosome-free p41.3 control (Fig.  2). As discussed above, we expected transcription differences among the rotational variants to be small, partly because of imperfect nucleosome positioning. (Another likely reason will become more apparent below.) Nevertheless, these transcriptional differences were highly reproducible, evident in both the integrated and episomal constructs, and statistically signifi-FIG. 1. In vivo competition between HSF binding and nucleosome assembly following DNA replication in yeast. NPE-HSE DNA consists of a binding site for the transcription factor HSF embedded within, but near the edge of, a DNA segment that normally forms a positioned nucleosome in vivo (see "Results" for details). The figure depicts the rapid association of newly replicated NPE-HSE DNA with a histone H3-H4 tetramer. This step ordinarily would be followed by the binding of histone H2A-H2B dimers to form a complete nucleosome. However, if HSF binds in advance of histone H2A-H2B dimers, it would inhibit nucleosome maturation, as depicted by pathway A. Alternatively, if nucleosomes assemble prior to HSF binding, as depicted by pathway B, HSF would be unable to bind (unless critical determinants in the HSF binding site faced outward from the histone octamer). Pathway C depicts partial unwrapping of DNA from the histone octamer, enabling HSF to bind after nucleosomes form. As discussed under "Discussion," HSF binding via pathway C may be facilitated by chromatin remodeling factors and, in some instances, may be followed by complete disruption of the nucleosome. cant, given average S.D. values of Ϯ10% for the integrated constructs and Ϯ26% for the episomal constructs. In addition, structural analyses presented below reveal differences in overall nuclease sensitivity and packaging of these constructs that are in accord with differences in their transcriptional activity. Therefore, the poor transcription of constructs designed to create competition between binding of HSF and histone H2A-H2B dimers suggested that nucleosome assembly occurs prior to HSF binding (Fig. 1, Pathway B).
If nucleosomes assemble prior to HSF binding, we expected p95.3 to be significantly less active than p41.3. Instead, p95.3 was highly transcriptionally active, in both the episomal and integrated versions. There are several possible explanations for this result. For example, the NPE nucleosome in p95.3 might form with a different translational or rotational position that permits HSF to bind. Alternatively, if the NPE nucleosome in p95.3 were less stable than in the other constructs, DNA unwrapping from the nucleosome edge might permit HSF to bind, as depicted in Fig. 1, Pathway C. Structural analyses described below support the latter alternative. (Specifically, the HSE and ϳ20 bp of DNA immediately 5Ј of the HSE appear to have unwrapped from the histone octamer, permitting HSF to bind; ϳ100 bp of NPE DNA appear to have remained associated with the histone octamer.) Consequently, below we sometimes refer to this partially unwrapped nucleosome in p95.3 as a "nucleosome," using quotation marks to help distinguish it from a canonical nucleosome.
Chromatin Structure Analyses Suggest That the NPE in p95.3 Is Packaged in a "Nucleosome" Despite a Predicted Steric Conflict between Nucleosome Formation and HSF Binding-To distinguish among the possibilities depicted in Fig. 1, we compared the chromatin structure of p95.3 with that of p97.3, one of the least active of the rotational variants, and the p41.3 control, a construct in which the HSE is well separated from the NPE nucleosome. Nuclei containing p41.3, p95.3, or p97.3 were treated with varying amounts of MNase, and the resulting DNA samples were purified and size-fractionated (Fig. 3). Ethidium bromide staining revealed the expected oligomeric ladder, reflecting MNase cleavage of linker DNA within nucleosome arrays. Southern blotting and hybridization with a short NPE probe revealed an NPE-associated nucleosomal array in all three constructs. This result suggested that even in the highly active p95.3 NPE-HSE DNA was associated with a nucleosome (but did rule out the possibility that it differed slightly in position relative to other NPE nucleosomes). Fig. 3 also shows that, at equivalent overall digestion levels (judged by the ethidium bromide pattern), the NPE region was more extensively digested in p95.3 than in the less active p97.3. This   FIG. 2. Organization and transcriptional activity of rotational variants. In the control construct p41.3, the HSF binding site (HSE) lies in a nucleosome-free region well separated from the NPE-directed nucleosome. To create competition between HSF binding and nucleosome formation, 50 -59 bp of DNA between the HSE and the presumptive dyad axis of the NPE nucleosome (vertical dashed line) was deleted such that the HSE would now be incorporated into the NPE nucleosome, should it form. The HSE itself, the proximal promoter, and the reporter gene are identical in all constructs. The resulting rotational variants were transfected into yeast as episomes and also integrated into the yeast genome. For each construct, transcriptional activity was measured in multiple independently isolated transformants, as described (41). In all cases, non-heat-shocked transcription values were similar to one another and less than 0.5 Miller units (not shown). Average heat shock-induced transcription values and S.D. values shown are based on n measurements shown in parentheses. Except for p99.3 and p100.3 (see text for discussion), relative differences in activity of the various episomal constructs were similar to the differences evident in the integrated constructs.
FIG. 3. The NPE in p95.3 appears to be packaged in a nucleosome despite a predicted steric conflict between nucleosome formation and HSF binding. p41.3-, p95.3-, and p97.3-containing nuclei were isolated from heat-shocked cells and treated with MNase. DNA was then purified, fractionated, and stained with ethidium bromide to visualize the bulk nucleosomal ladders and verify that digestions had proceeded to similar extents. Size markers (M) were *X174 DNA cut with HaeIII. Samples were blotted, and NPE sequences were visualized by hybridization with a 32 P-labeled, 185-bp BglII-NheI fragment. A mononucleosome-sized band is evident in all rotational variants examined. The intensity of this band among rotational variants digested to similar extents was relatively uniform, suggesting that the NPE nucleosome forms with similar efficiency regardless of differences in the efficiency of HSF binding. Similar results were observed using cells not subjected to heat shock (not shown).
increased level of nuclease sensitivity is a hallmark of highly active genes and was also evident in the genomic footprinting studies described below.
While the MNase digestion assay in Fig. 3 suggested that NPE DNA in p41.3, p95.3, and p97.3 is associated with a nucleosome, this assay is not quantitative. Thus, HSF might have bound a subset of p95.3 molecules not packaged in an NPE nucleosome. This possibility seemed unlikely, given that p95.3 was as transcriptionally active as the nucleosome-free p41.3 control. Nevertheless, we asked whether a naturally occurring RsaI restriction site within the NPE was more accessible in p95.3 than in p41.3, as might be expected if HSF binding interfered with nucleosome formation. The extent of RsaI cleavage in nuclei was measured in quantitative PCR reactions, using primers that flanked the restriction site of interest, as illustrated in Fig. 4. Only about 7% of the RsaI sites in the NPE of p41.3 were cleaved in nuclei isolated from heatshocked cells, whereas an upstream control site was cleaved in about 72% of the molecules. Similarly, there was virtually no cleavage at the NPE RsaI site in p95.3. We also measured the accessibility of AciI sites within and outside the NPE for p41.3, p95.3, p97.3, and p100.3 (not shown). In all cases, the AciI site within the NPE nucleosome was cleaved in only about 10% of the molecules, whereas an AciI site 3Ј of the HSE was cleaved in approximately 50% of the molecules. These results argue that packaging of NPE DNA into nucleosomes was uniform and highly efficient and that HSF was not simply bound to a nucleosome-free subset of NPE molecules in p95. 3.

High Resolution Genomic Footprinting Indicates That the NPE "Nucleosome" in p95.3 Has the Same Translational and Rotational Position as Occurs in
Other NPE-containing Constructs-Given the predicted steric conflict between nucleosome formation and HSF binding, the apparent packaging of the highly active p95.3 NPE into a nucleosome was paradoxical. However, it was possible that HSF binding had altered the normal helical orientation of NPE DNA relative to the histone octamer or shifted the NPE nucleosome to a position further upstream. Indirect end label mapping of the p95.3 NPE nucleosome suggested that its translational position was identical to that of the NPE nucleosome in p41.3 (not shown). However, the resolution of these analyses was not high enough to rule out a translational shift of up to 20 bp. Such a shift would leave the AciI and RsaI sites within the nucleosome but partially expose the HSE. We therefore used genomic footprinting to map the NPE nucleosome in p41.3, p95.3, and p97.3 at 1-bp resolution (Figs. 5 and 6). Nuclei containing p41.3, p95.3, or p97.3 were treated with varying amounts of MNase, and DNA was purified. As a control, intact DNA from the same nuclei was purified and then treated with MNase. Cleavage sites were visualized by PCR-mediated reiterative extension of an end-labeled primer and mapped relative to the translation start site of the reporter gene using DNA sequencing standards. PCR control reactions using undigested templates (e.g. the two leftmost lanes in the left part of Fig. 5A) demonstrated that bands in digested samples reflect nuclease cleavages rather than pausing of Taq polymerase. Inspection of Fig. 5A shows a large, nuclease-protected region in p41.3, p95.3, and p97.3 chromatin. The upstream edge of this protected region is marked by a prominent, chromatin-specific MNase cleavage site on the coding (i.e. bottom) strand of DNA (Fig. 5A, open arrowhead in each part). On the noncoding strand of DNA, this same upstream edge is marked by two prominent, chromatin-specific MNase cleavage sites (Fig. 5B, open arrowheads) and by three, closely spaced (i.e. nonnucleosomal) DNase I cleavage sites (Fig. 6). These cleavages sites are identical for p41.3, p95.3, and p97.3, indicating that the upstream boundary of the NPE nucleosome is the same in all three constructs. If HSF binding had induced only a fraction of the NPE nucleosomes in p95.3 to shift upstream, the intensity of these cleavages would be significantly lower in p95.3 than in p41.3 or p97.3. Instead, the cleavage intensities were approximately the same for each construct, indicating that the p95.3 "nucleosome" was not shifted upstream by HSF binding.
The MNase cleavage patterns also argue against a possible downstream shift in the position of the NPE nucleosome. For example, even a minor (e.g. 10-bp) downstream shift would have exposed naked DNA cleavage sites near the upstream edge of the NPE nucleosome; Fig. 5A shows no evidence of this having occurred. Fig. 5B shows chromatin-specific, MNase cleavages within the NPE nucleosome on the noncoding strand, spaced at approximately 10-bp intervals, as expected for nucleosomal DNA (65). These cleavages occur at the same sequences for all three constructs, providing additional evidence for a single common translational position and suggesting as well that the various NPE nucleosomes share the same rotational position. Finally, a major downstream shift of the NPE nucleosome would have suppressed cleavage at a prominent naked DNA cleavage site 3Ј of the HSE (Fig. 5C, open circles).
To further define the translational and rotational position of the NPE nucleosomes in p41.3, p95.3, and p97.3, we next

FIG. 4. Restriction enzyme analyses indicate that virtually all p95.3 NPE molecules are packaged in a "nucleosome" despite a predicted steric conflict between nucleosome formation and HSF binding.
To estimate the fraction of NPE molecules packaged in nucleosomes, p41.3-and p95.3-containing nuclei were isolated from heat-shocked (HS) and nonshocked (NHS) cells and treated with varying amounts of RsaI. After DNA purification, the extent of cleavage was measured as depicted in the diagram. RsaI cleavage within the p41.3 NPE (at Ϫ388) will preclude amplification of a 327-bp PCR fragment using oligonucleotide primers 52 and 30; the equivalent fragment for p95.3 is 273 bp. This is evident in control digestions of naked DNA (lanes 7-9, left part; lanes 7 and 8, right part). Amplification of a 127-bp fragment using primers 52 and 55 is unaffected by RsaI cleavage and thus provides an internal control for the efficiency of the PCR reaction. The primer common to both reactions (52 in this example) is endlabeled, allowing visualization of PCR products by autoradiography and quantification by phosphor imagery. For each digestion point, the RsaIresistant fraction was calculated as the ratio of the 327-bp fragment to 127-bp fragment for RsaI-treated DNA and was normalized to the ratio calculated for untreated DNA. Analogous reactions were carried out to quantify RsaI cleavage at a control site at Ϫ584 (lanes 1-3, left part). Markers were pBR322 DNA cleaved with MspI and 32 P-end-labeled using Klenow enzyme. mapped DNase I cleavage sites for these three constructs (Fig.  6). DNase I cleaves across the minor groove of nucleosomal DNA wherever it is maximally accessible. Dots in Fig. 6 mark DNase I cleavage sites in nuclei that are spaced at 10-bp intervals in all three constructs, indicating that in each construct the NPE nucleosome forms with a single predominant helical orientation. These cleavage sites occur in precisely the same sequences, indicating that all three constructs share the same helical orientation. As noted earlier, the upstream edge of the NPE nucleosome is marked by three closely spaced, chromatin-specific DNase I cleavage sites that coincide with MNase cleavages that also mark the nucleosome edge. Together, these results provide strong evidence that the NPE "nucleosome" in p95.3 shares the same translational and rotational position as occurs in p41.3 and p97.3. However, careful inspection of Figs. 5 and 6 reveal reproducible differences in the MNase and DNase I cleavage patterns just 5Ј of the HSE in p95.3, as compared with the same DNA segments in p41.3 and p97.3 (indicated by solid vertical bars in all three parts of Figs. 5B and 6). These differences are discussed below and suggest a mechanism that enables efficient binding of HSF to the NPE "nucleosome" in p95. 3.
Evidence for Unwrapping of HSE DNA from the Histone Octamer in p95.3-Having eliminated the possibility that the translational or rotational position of the p95.3 "nucleosome" differed from that of the transcriptionally less active constructs, we next asked whether the p95.3 promoter was packaged differently from that of the other constructs. This possibility was highly unlikely, since the HSE and all sequences 3Ј of the HSE are identical in all constructs. Nevertheless, we compared the p95.3 promoter with that of the p41.3 control and the less active p97.3 rotational construct. Fig. 5A shows that a prominent MNase cleavage site evident in naked HSE DNA was partially protected from cleavage in chromatin in all three constructs (asterisks in Fig. 5A). Use of a primer that anneals to a site closer to the HSE shows the same protected site evident in Fig. 5A (marked with an asterisk in Fig. 5C) and partial protection of other naked DNA cleavage sites in the HSE as well. On the opposite strand, we observed substantial protection of one naked DNA cleavage site within the HSE and induction of a hypersensitive site at the 5Ј edge of the HSE (Fig.  5B, asterisks and open arrowheads, respectively). These results are consistent with partial rather than full occupancy of the NPE nucleosome is not well-marked by MNase cleavages, it is evident in the DNase I cleavage profile for p41.3 and p97.3, but not p95.3. Dashed lines in the diagram to the right of p95.3 depict likely unwrapping of DNA near the 3Ј edge of the NPE nucleosome, inferred from the cleavage profiles in Figs. 5B and 6 (see "Results" for discussion). B, nuclei containing p41.3, p95.3, and p97.3 were isolated from heatshocked (ϩ lanes) and nonshocked (Ϫ lanes) cells and treated with varying amounts of MNase. Single strand MNase cleavage sites in the noncoding (upper) strands of p41.3, p95.3, and p97.3 were mapped as above. Asterisks indicate naked DNA cleavage sites that are protected in nuclei, and open arrowheads indicate nuclease-hypersensitive sites within the HSE and immediately 5Ј of the NPE nucleosome. The two cleavages 5Ј of the NPE nucleosome occur at identical sequences in all three constructs examined and coincide with other MNase and DNase I cleavages that also mark the upstream edge of the NPE nucleosome (cf. Figs. 5A and 6). A solid vertical bar (between the third and fourth lanes in all three parts) marks a DNA segment within the NPE nucleosome that is cleaved identically in p41.3 and p97.3 but differently in p95.3. This local difference in cleavage pattern may reflect the partial unwrapping of NPE DNA from the histone octamer in p95.3 (see "Results" for discussion). C, single strand MNase cleavage sites in the coding (lower) strands of p41.3, p95.3, and p97.3 were mapped using a probe that anneals to sequences within the NPE. Asterisks mark cleavage sites within the HSE and TATA box 1 (open rectangle T1) that are protected in chromatin. Open circles point to cleavage sites evident in both chromatin and naked DNA, indicating this region of the promoter is nucleosome-free.  G and T). In the diagram to the right of each part, numbers refer to positions relative to the translation start site in the lacZ reporter gene; the filled rectangle depicts the HSE; and the asterisk indicates a naked DNA cleavage site in the HSE that is protected in nuclei. The open rectangle depicts the NPE nucleosome, whose upstream (5Ј) boundary is marked by a strong, chromatin-specific cleavage site (open arrowhead); although the downstream (3Ј) boundary of the HSE by HSF. Nonsaturated genomic footprints commonly occur due to nonspecific binding of factors to other chromosomal sequences (cf. Ref. 66). Apparent nonquantitative binding of HSF in yeast may also reflect its transient displacement during DNA replication and during mitosis, 3 as occurs in metazoans (67). Finally, HSF binds more tightly to HSEs consisting of 5Ј-AGAAC-3Ј repeats than to the 5Ј-AGAAG-3Ј repeats used in this study (68). Because it is difficult to use nonsaturated footprints to quantify HSF binding, we relied on transcriptional activity as our best measure of the efficiency of HSF binding, as discussed earlier. Fig. 5C also shows a similar cleavage pattern associated with the proximal promoter in all three constructs examined. For example, naked DNA cleavage sites on the lower strand within and 3Ј of TATA1 are uniformly protected from cleavage in nuclei, consistent with our earlier finding that efficient binding of TFIID to this promoter occurs independently of upstream regulatory elements (42). The second of the two functional TATA sites in this promoter, TATA2, was less well protected, suggesting that in these constructs TATA1 is the more frequently used of the two sites. These observations indicated that the promoters of the three constructs are identically packaged.
Having ruled out all other likely explanations for why HSF might bind efficiently to p95.3 but not p97.3, we considered the possibility that partial unwrapping of DNA from the NPE nucleosome in p95.3 enabled HSF to bind a site that would otherwise be sterically inaccessible. Studies of nucleosome melting (1), thermally induced changes in the twist of nucleosomal DNA (69), binding of factors to DNA near the edges of nucleosomes (70 -72), and the invasion of nucleosomes by RNA polymerase and exonuclease III (73,74) are all consistent with the idea that histone contacts with DNA near the edges of nucleosomes are more labile than those closer to the dyad axis.  T). In all three constructs the 5Ј nucleosomal boundary is marked by three DNase I cleavage sites spaced at approximately 6-bp intervals. Prominent cleavage sites within the NPE (indicated by dots) are spaced at 10-bp intervals, indicative of rotationally positioned nucleosomal DNA. The cleavage site at the dyad axis is additionally marked by an asterisk. A cleavage site within the HSE (depicted in the diagrams by a filled rectangle) that is hypersensitive in chromatin relative to naked DNA is marked by a small arrowhead, while a cleavage site within the HSE that is protected in chromatin is marked by an asterisk. Note that nucleosomal cleavage sites corresponding to sites Ϫ7 to 0 to ϩ7 are evident in p41.3 and in p97.3, whereas only 10 cleavage sites (corresponding to sites Ϫ7 to ϩ2) are evident in p95.3. As in Fig. 5B, a solid vertical bar in each part marks a DNA segment within the NPE nucleosome that is cleaved identically in p41.3 and p97.3 but differently in p95.3. This local difference in cleavage pattern may reflect the partial unwrapping of DNA from the NPE nucleosome in p95.3; as in Fig Fig. 6 and in an earlier study (41) and thus indicate sequences whose minor groove faces outward from the histone octamer. Each CG bp was assigned a value of 1, and each AT bp was assigned a value of 0. The actual GC index for each base pair was calculated using a 5-bp sliding window. This might permit transient, partial unwrapping of DNA, enabling HSF to bind DNA in this region even after nucleosomes form. As noted above, careful examination of the MNase and DNase I cleavage patterns within the NPE but near the HSE (marked by solid vertical bars in Figs. 5B and 6) indicates that the cleavages are virtually identical for the p41.3 control and the poorly transcribed p97.3 but different from those in p95.3. For example, the 10-bp DNase I cleavage pattern evident throughout the NPE nucleosome in p41.3 and p97.3 is evident only in the central and HSE-distal portion of the NPE in p95.3. Closer to the HSE, cleavages in p95.3 occur more frequently and with less regularity. These observations are consistent with the partial unwrapping of DNA from the histone octamer in p95.3, enabling HSF to bind without causing complete disruption of the NPE "nucleosome."

DISCUSSION
In this paper, we have investigated the result of competition between nucleosome formation and HSF binding following DNA replication in yeast. With one notable exception, constructs designed to create competition between binding of HSF and histone H2A-H2B dimers were significantly less active than p41.3, a nucleosome-free control. These results suggested that nucleosome formation precedes and can inhibit HSF binding, as depicted by assembly pathway B in Fig. 1. Structural analyses of the HSE and surrounding sequences in one such construct, p97.3, confirmed the prediction that the HSE was packaged in a nucleosome (Fig. 3), with a single predominant translational and rotational position (Figs. 5 and 6). However, another construct also predicted to create steric conflict between binding of HSF and histone H2A-H2B dimers, p95.3, was highly transcriptionally active. Structural analyses indicated that the p95.3 NPE was partly packaged in a "nucleosome" that shared the same translational and rotational positions as was seen for the poorly transcribed constructs. This finding led us to consider the possibility that the HSE and surrounding DNA had unwrapped from the histone octamer in p95.3, allowing HSF to bind after nucleosome assembly, as depicted by assembly pathway C in Fig. 1. Differences between p95.3 and p97.3 in the nuclease susceptibility of DNA between the HSE and the nucleosome dyad support this hypothesis. As noted earlier, the partial unwrapping of DNA from the edges of already formed nucleosomes is consistent with several previous studies indicating that histone contacts with DNA near the edges of nucleosomes are more labile than those closer to the dyad axis. Additionally, unwrapping of DNA from the edge of nucleosomes in vitro can occur on a time scale of seconds (74). Thus, it probably can account for the rapid binding of HSF to newly replicated DNA in vivo.
DNA Sequence-related Differences in the Predicted Stability of the NPE Nucleosomes May Account for Differences in HSF Binding to the Rotational Variants-The partial unwrapping of DNA from the histone octamer that can account for the high transcriptional activity of p95.3 probably also occurs to some extent for all the other NPE constructs used in this study. This would help explain why transcriptional differences among the rotational variants are modest (albeit statistically significant). Nevertheless, if HSF is able to bind the HSE in p95.3 because of partial unwrapping of nucleosomal DNA, one must ask why unwrapping occurs more readily for p95.3 than for the transcriptionally less active p97. 3. A likely answer lies in the fact that the deletions required to rotate the HSE relative to the histone octamer also altered the spacing of GC-rich segments near the HSE. As noted earlier, GC-rich DNA segments tend to face outward from the histone octamer (30), probably because such sequences resist the minor groove compression associated with DNA wrapping about the histone octamer. DNA fragments that contain GC-rich DNA spaced at 10-bp intervals form stable nucleosomes in vitro (61). Although fragments such as these do not possess all of the determinants required to form positioned nucleosomes in vivo (75)(76)(77), it seems reasonable to suppose that GC-rich segments whose minor grooves do not face outward would destabilize nucleosomes. To determine if "improperly" positioned GC-rich segments could explain differences in the efficiency of HSF binding to our rotational variants, we catalogued potentially destabilizing dinucleotides in a 30-bp DNA segment containing the HSE and adjacent NPE sequences, using the statistical data base generated by Travers and colleagues (78). In the efficiently transcribed p95.3, the dinucleotides in the "valleys" between DNase I cleavage maxima 5 and 6 and maxima 6 and 7 are, respectively, AG and GC, and AG and GC (Fig. 7A). The minor grooves of GC dinucleotides show a strong tendency to face away from the histone octamer, and the minor grooves of AG dinucleotides show a modest tendency to face away from the histone octamer (78). Since the minor grooves of these dinucleotides are compressed in the p95.3 NPE nucleosome, they are likely to be destabilizing. None of the equivalent dinucleotides in p97.3 show a tendency to face away from the histone octamer (78). Overall, the poorly transcribed rotational variants contain fewer destabilizing dinucleotides than does p95.3. The same conclusion holds if one analyzes potentially destabilizing trinucleotides in the HSE-containing portion of the NPE (not shown).
In an attempt to quantify the destabilizing influence in improperly spaced GC-rich segments, we analyzed variation in GC content across the HSE-NPE sequence in each of the rotational variants as follows. As shown in Fig. 7A, each AT base pair was assigned a value of 0, and each GC base pair was assigned a value of 1. The resulting values were summed across the NPE using a 5-bp sliding window. This produced a local "GC index" for each base pair ranging from 0 to 5 and revealed GC-rich segments spaced at integral multiples of the helical repeat. (Use of either 3or 4-bp sliding windows to calculate GC richness revealed the same general tendency but with more scatter.) Several of these GC-rich segments coincide with DNase I cleavage sites (see vertical lines in Fig. 7A), indicating that their minor groove faces away from the histone octamer, as might be expected for a rotationally positioned nucleosome. Fig. 7A also shows that GC-rich segments near the HSE-containing portion of the NPE sometimes occur between rather than at the DNase I cleavage sites. The destabilizing potential of these improperly phased GC-rich sequences was estimated by calculating a local "mismatch" index for each bp in the HSE-containing portion of the NPE. The GC-rich segments are denoted by a GC index of 3-5, while AT-rich segments are denoted by a GC index of 0 -2. Vertical lines indicate GC-rich segments that coincide with DNase I cleavage maxima and thus face outward from the histone octamer; such segments may contribute to rotational positioning. The actual GC index for each construct was compared with an optimal GC index, in which GC-rich segments alternate with AT-rich segments, once per helical repeat. These comparisons reveal potentially destabilizing GC-rich segments between the DNase I cleavage maxima. Potentially destabilizing GC-rich segments were assigned a mismatch value, as described under "Discussion." The total mismatch is the sum of individual mismatch values for individual bp between DNase I cleavage maxima 5 and 7. B, the total mismatch for each rotational variant was plotted against transcriptional activity measured for both episomal and integrated constructs. Likely reasons for the higher than predicted level of transcription for p99.3 and p100.3 are discussed under "Discussion." Aside from p99.3 and p100.3, there is a good correlation between high mismatch and high transcriptional activity, consistent with the hypothesis that "improperly phased" GC-rich segments can lead to partial unwrapping of nucleosomal DNA, enabling HSF to bind. mismatch value was defined as the extent to which any base pair with a GC index of greater than 2 exceeds the GC index at the equivalent position in a nucleosomal fragment of theoretical optimal helical stability (Fig. 7A). For example, a bp with a GC index of 4 that occurs at a site where the optimal GC richness is 2 was assigned a mismatch value of 2. Analysis of the synthetic nucleosome positioning sequence used by Shrader and Crothers (61) gives a total mismatch value of 0. By comparison, the total mismatch in the HSE-containing portion of each NPE rotational variant tends to increase with increasing transcriptional activity (Fig. 7B). While there is a relatively good correlation between these variables, p99.3 and p100.3 in particular were more active in the episomal constructs than the mismatch-driven DNA unwrapping model would predict (Fig. 7B). Interestingly, p99.3 and p100.3 are the two constructs predicted to form nucleosomes in which the HSE is still accessible to HSF (see "Results"). Thus, steric access may also influence the efficiency of HSF binding to nucleosomal DNA, although it is unclear why p99.3 and p100.3 were not proportionately as active when integrated into the chromosome. p99.3 and p100.3 aside, the analysis in Fig. 7B suggests that local GC-rich, improperly phased sequences account, at least in part, for partial DNA unwrapping from nucleosomes and the subsequent binding of HSF.
Fate of the HSF-bound Nucleosome-In cases where HSF binds to nucleosomal DNA, the fate of the nucleosome may vary depending on its overall stability. The NPE nucleosome used in this study appears exceptionally stable, and it persists even in a partially unwrapped state. Our discovery of a partially-unwrapped "nucleosome" in p95.3 is the best evidence to date that DNA unwrapping shown to occur in vitro also occurs in vivo. More commonly, partially unwrapped nucleosomes may occur as unstable intermediates in a process that culminates in nucleosome disruption (perhaps with the aid of nucleosome remodeling factors). Thus, pathway C (Fig. 1) would often be followed by nucleosome disruption. This reasonable assumption allows us to accommodate the observation of Gross et al. (43) that HSF, when overexpressed, was able to bind a degenerate HSE that otherwise was packaged in a nucleosome. In their study, the HSE may have been packaged in a nucleosome less stable than the NPE nucleosome, such that HSF binding led to its dissolution. Finally, it should be emphasized that our results are based on studies with wild type cells and may thus depend on the local chromatin modification state or nucleosome remodeling activities. Indeed, given our evidence that HSF binding follows rather than precedes nucleosome formation, it is likely that the same mechanisms that facilitate inductiondependent binding of transcription factors (e.g. Gal4p and Pho5p) also act during the assembly of initiation complexes associated both with transcriptionally poised heat shock promoters and constitutively active housekeeping genes.