Other environment-responsive TFs conditionally bind X elements

To determine whether this X element-binding property of Oaf1p is shared with other environment-responsive factors, a comprehensive ChIP-chip data set of 203 yeast TFs (Harbison et al, 2004) was analyzed. Factors that were found to interact with at least one X element-containing probe on the microarrays are shown in Figure 2. The top two graphs (Figure 2A and B) show factors that were analyzed in the absence and presence of a stress stimulus, while the bottom graph (Figure 2C) shows factors tested only in the absence of stress. Bars marked with asterisks correspond to factors that significantly enriched at X elements. This analysis identified 19 factors that appeared to (conditionally) concentrate at X elements, including known X element-binding proteins Rap1p (Zhu and Gustafsson, 2009) and Reb1p (Chasman et al, 1990).

Open in a separate window

Figure 2

Many environment-responsive transcription factors (TFs) bind X elements. ChIP-chip data of 203 DNA-binding proteins were analyzed for factors that target X elements. Factors that bound at least one of the 25 probes on the arrays that overlap with X elements are shown. The top graph shows analysis of factors from cells grown in rich medium, and the middle graph shows the same factors from cells grown under other conditions. The bottom graph shows factors that were analyzed only in rich medium. Factors/conditions marked with a white dot on the bar are those shown to have enriched binding in subtelomeric regions (Mak et al, 2009). Underlined factors have been previously shown to bind X elements. The 19 factors with astrisks significantly enrich at X elements (i.e., they have hypergeometric distribution P<0.001 after multiple test correction, and bind to more than eight X element probes).

Some of the TFs that enriched at X elements have previously been shown to enrich in promoter regions of genes within subtelomeres (marked with a white circle on or above the bar) (Mak et al, 2009). Therefore, we determined the subtelomeric-binding profiles of the factors to determine whether they specifically enrich at X elements compared with the rest of the subtelomeric region (Figure 3). Binding profiles correlated with the position of X elements with Pearson's correlation coefficients >0.8 and P-values <1 × 10⁻⁶, with the exception of Phd1p, which had a coefficient of 0.51 and P=0.0045. This correlation of TF binding with X elements was not detected by a previous analysis (Mak et al, 2009) possibly because of the lower resolution of the previous study (which tested genomic fragments of ~390 kb compared with 10 kb used here). In addition, the previous study analyzed the positions of all X elements for correlation, whereas here, X elements that were not present on the microarrays were excluded from the analysis.

Open in a separate window

Figure 3

Binding profiles of transcription factors (TFs) correlate with X-element positions. Subtelomeric-binding profiles are shown for factors found to enriched at X elements in Figure 2. The densities of TF binding and X elements present on the microarrays are represented as heat maps. Binding conditions other than rich medium are marked with colored dots as in Figure 2. With the exception of Phd1p, all factors had maximal binding 5–7 kb from telomeres, matching the position of X element enrichment. Pearson's correlation coefficients (ρ) comparing each binding profile to the positions of X elements are shown at the right.

Analysis of X element binding with high resolution tiled arrays

To further characterize the interaction of TFs with X elements, an in depth analysis of Oaf1p binding was conducted by repeating the ChIP of Oaf1p in the absence or presence of medium containing fatty acids, and analyzing the targets on tiled microarrays of the entire yeast genome (see Materials and methods). Subtelomeric-binding profiles under each condition were determined and are shown as histograms of the mean number of peaks per chromosome arm (Figure 4A, top two panels). These data revealed that Oaf1p targets in the presence of fatty acids were not only enriched 5–7 kb from telomeres, but were also enriched within 1 kb of telomeres, whereas in the absence of fatty acids the profile was similar to background (compare blue bars to red vector of mean binding frequency per kb in the entire genome). The Oaf1p-binding profile was strongly correlated with the positions of X elements (Figure 4A, third panel) in the presence of fatty acids, with a Pearson's correlation coefficient (ρ) of 0.98 (P<0.01), but not in the absence of fatty acids (ρ= −0.15). This strong correlation is supported by the fact that Oaf1p targeted almost all X elements in the genome with a false discovery rate (FDR) <0.001 in the presence of fatty acids (Figure 4B).

Open in a separate window

Figure 4

Oaf1p conditionally binds subtelomeric X elements. (A) Subtelomeric-binding profiles of Oaf1p in the absence and presence of fatty acids (from ChIP-chip using whole-genome tiling arrays with peak detection false discovery rate (FDR; <0.001). In the presence of fatty acids, binding of Oaf1p is highly enriched within 1 kb of telomeres and more subtly enriched 5–7 kb from telomeres, and matches the profile of X-element positions. Red vectors mark mean binding frequency per kb across the entire genome for the experiments. Error bars represent s.e. of three biological replicates of each experiment. (B) Visualization of fatty acid-activated subtelomeres using Circos visualization software (Krzywinski et al, 2009). In all, 32 subtelomeric regions in yeast are shown (peripheral ring) including X elements (marked with X and spokes extending toward center) and 25 kb of centromere-proximal sequence (green bars). Binding positions of Oaf1p as determined by ChIP-chip (FDR<0.001) are shown as colored circles inside of the peripheral ring. Red and blue bars show positions of significantly up and downregulated genes, respectively, in response to fatty acid exposure at 0.5, 1, 3, 6 and 9 h (from outer to inner ring) (Smith et al, 2002). Significantly downregulated genes are enriched in subtelomeric regions. See also Supplementary Figures S1 and S2.

The high-resolution ChIP-chip data set enabled analysis of specificity of Oaf1p binding at individual X elements (Supplementary Figure S1), which revealed that Oaf1p specifically binds to unique probes in most X elements in response to fatty acids. The data set also enabled identification of a putative DNA-recognition sequence for Oaf1p in the context of negative regulation (see Materials and methods). This sequence (G[not A]AGGGTAANNNNN[not C][not C]) is similar to the predicted binding motif of GRF, Reb1p (RTTACCCK) (Badis et al, 2008) and for the reasons outlined below, was termed subtelomeric Reb1-binding (SRB) motif. This motif occurs 238 times in the entire genome, and 65% of the motifs are within 100 bp of an Oaf1p-binding position (e.g., see Figure 6A). CPA analysis of the motif showed that it strongly enriches in subtelomeric regions at a frequency of >1 per arm, and colocalizes with X elements (Supplementary Figure S2). To characterize the interaction of Oaf1p with the SRB motif, an electromobility shift assay (EMSA) was performed (Supplementary Figure S2). Whereas the DNA-biding domain (DBD) of Reb1p bound the SRB motif, the DBD of Oaf1p did not, suggesting that Oaf1p does not interact directly with X elements and that this interaction might be mediated by Reb1p.

Open in a separate window

Figure 6

Oaf1p negatively regulates distally located subtelomeric genes. (A) Two subtelomeric regions (13 and 14 L) are shown. First row, positions of X and Y′ elements, with X element repeat regions (red) and cores (green) shown separately. Second row, log₂ expression ratios (IP versus WCE) for each 50-mer probe for one replicate of Oaf1p ChIP-chip in the presence of the fatty acid oleate. Third row, Oaf1p-binding positions (false discovery rate; FDR <0.05) predicted from this data (red bars). Fourth row, positions of OREs, motifs that bind Oaf1p–Pip2p heterodimers that activate transcription (Rottensteiner et al, 2003), and subtelomeric Reb1-binding (SRB) motifs, predicted to bind Oaf1p without Pip2p to repress transcription (Supplementary Figure S2). Oaf1p binding at X elements coincide with SRB motifs, whereas other peaks coincide with OREs. Fifth row, ORFs with start sites in green. Genes analyzed in B are marked with asterisks. (B) FACS data showing abundance of subtelomerically encoded proteins fused to GFP in the presence and absence of fatty acids in wild-type (WT) and Δoaf1 strains. Three biological replicates are shown for each condition. All changes in levels between condition pairs are statistically significant with Student's t-test P-values <0.01 and mean fold changes are shown for each. Genes centromere-proximal to X elements appear to be negatively regulated by Oaf1p in the presence of oleate. An exception to this is PEX6, which is positively regulated by Oaf1p in oleate and also has an upstream ORE bound by Oaf1p (A). ORE, oleate response element.

X element-binding TFs conditionally regulate proto-silencing

Conditional binding of Oaf1p to X elements in the presence of fatty acids appeared to correlate with an overall repression of subtelomeric gene expression (compare ratio of down versus upregulated genes (blue versus red bars) in Figure 4B). This led us to hypothesize that X element-binding TFs regulate X element-mediated proto-silencing in response to environmental stimuli. To test this, expression profiles of subtelomeric genes centromere-proximal to X elements (within 20 kb) were analyzed during three responses that correspond to dynamic repositioning of factors at X elements in Figures 2 and ​and44 (Figure 5). The region was enriched for genes that significantly decreased in expression in response to fatty acid stress, but increased in expression in response to H₂O₂ or butanol stress (compared with the percentage of significantly up and downregulated genes in the entire genome, represented as green and red shading, respectively). Expression profiles that were determined to be significantly different from the global responses using a Student's t-test are marked with asterisks. Y′ element genes that are telomere-proximal to X elements were analyzed in the same way, and their responses were not significantly different from the global responses to the stimuli, except for a subtle enrichment of upregulated genes at 2 h of H₂O₂ exposure (3 versus 2.5% in the genome) (data not shown). These data suggest environmental regulation of X element-mediated proto-silencing.

Open in a separate window

Figure 5

Regulation of proto-silencing by X element binding TFs in response to stress. Microarray expression data were analyzed to determine the effects of environmental stresses on the expression of subtelomeric genes that are centromere-proximal to X elements, and the roles of X element-binding factors in this regulation. Three stress conditions resulting in differential binding of transcription factors (TFs) to X elements in Figures 1, ​,22 and ​and33 were analyzed including exposure to 0.15% fatty acids (A), 0.32 mM H₂O₂ (B) and 1% butanol (C). For A and B, and C, there were a total of 227 and 230 genes, respectively, in the subtelomeric region of interest. The percentage of these genes that significantly increased (green bars) and decreased (red bars) in expression are shown, as well as the percentage of genes with significant changes in the entire genome (pink and green shading). Bars marked with asterisks have significantly different mean expression profiles than the entire genome (with Student's t-test P-values <0.001). The left panels show binding patterns of TFs that dynamically enrich at X elements in response to the stresses. The right panels show analyses of wild-type and TF deletion strains under the same conditions. OAF1 appears to conditionally enhance proto-silencing whereas ROX1 and PHD1 have the opposite effect. See also Supplementary Figure S3.

Next, the roles of TFs in proto-silencing were investigated by analyzing microarray expression profiles of subtelomeric genes in TF deletion strains (Figure 5). Growth in the presence of fatty acids resulted in decreased expression of subtelomeric genes centromere-proximal to X elements. Under the same condition, deletion of OAF1 resulted in increased expression of these genes compared with the wild-type strain, suggesting that Oaf1p is an enhancer of X element-mediated proto-silencing during this response. However, deletion of ROX1 or PHD1 resulted in enrichment of genes with significantly decreased expression in the presence of H₂O₂ or butanol, respectively, suggesting that Rox1p and Phd1p conditionally antagonize proto-silencing. To evaluate whether these activities were due to direct activation or repression by the TFs rather than regulation of proto-silencing at X elements, the analyses were repeated with direct targets of factors implicated in the responses excluded, and the results were very similar (Supplementary Figure S3). Together, these data suggest that conditional interaction of TFs with X elements regulate proto-silencing activity in response to the environment with long-range effects on genes that are not immediately adjacent to X elements.

To investigate these effects on distal genes, we analyzed the impact of Oaf1p regulation on specific subtelomeric genes (Figure 6). Panel A shows the binding patterns of Oaf1p in subtelomeric regions on chromosome arms 13 and 14 L. On both arms, X elements were detected (top row), bound by Oaf1p in the presence of fatty acids (second and third rows), and overlapped with SRB motifs, but not oleate response elements (OREs), which bind Oaf1p–Pip2p heterodimers that activate transcription (Rottensteiner et al, 2003) (fourth row). GFP fusion proteins encoded by genes not adjacent to X elements (open reading frames; ORFs marked with asterisks in fifth row) were monitored by fluorescence-activated cell sorting (FACS) in the absence or presence of fatty acids in wild-type and Δoaf1 cells (panel B). With the notable exception of Pex6–GFP (see below) the majority of proteins tested had reduced levels in the presence of fatty acid compared with glycerol (left), and this effect was dependent on the presence of Oaf1p (right). Negative regulation by Oaf1p was specific to the fatty acid response and was not apparent in glycerol-grown cells (data not shown). Altogether, 10 genes centromere-proximal to X elements were analyzed by FACS, 3 of which had significant decreases in expression in an OAF1 deletion strain by microarray analysis. In all, 7 of the 10 corresponding fusion proteins had significantly decreased abundance in fatty acid medium, and 6 had significant fatty acid-specific negative regulation by OAF1. These data support our interpretation that the binding of Oaf1p to X elements has a conditional role in controlling subtelomeric silencing.

A notable exception to the trend is PEX6 (shown in Panels A and B), which had increased protein levels in response to oleate and positive regulation by Oaf1p. A possible mechanism of upregulation is that, although Oaf1p is a proto-silencer at the X element on this chromosome arm, it also directly activates PEX6 by binding the upstream ORE as a heterodimer with Pip2p (Smith et al, 2007) (Figure 6A). These results are consistent with previous studies showing that PEX6 is upregulated by fatty acid exposure (Smith et al, 2002), and that transcriptional activators have access to promoters within otherwise silent chromatin (Xu et al, 2006; Mak et al, 2009). This phenomenon offers an explanation for why not all subtelomeric genes centromere-proximal to X elements were observed to be affected by Oaf1p-mediated silencing.

Oaf1p-mediated silencing is dependent on SIR2

To support the conclusion that Oaf1p interacts with X elements and mediates its effect on subtelomeric silencing from this position, we analyzed the expression of a reporter gene of subtelomeric silencing in the strains FEP318-23 and FEP318-19 (Loney et al, 2009). In these strains, a gene encoding Ura3–GFP is positioned in a subtelomeric region of chromosome end III-R or XI-L adjacent and centromere-proximal to the endogenous X element (Figure 7). The effects of fatty acid exposure and OAF1 deletion on this reporter were analyzed by FACS (Figure 7A). The data suggest that Oaf1p conditionally enhances silencing of the reporter gene in the presence of fatty acids. The reporter gene was also used to elucidate which histone deacetylase is involved in Oaf1p-mediated regulation of silencing. We tested for genetic interactions between OAF1 and two genes encoding histone deacetylases, SIR2, which is known to be required for X element-mediated proto-silencing (Pryde and Louis, 1999), and HDA1, which is involved in another subtelomeric silencing mechanism (Robyr et al, 2002) (Figure 7B). FACS analysis of reporter gene output in various genetic backgrounds suggests that Oaf1p-mediated silencing is dependent on Sir2p and not Hda1p. These data support the contention that environment-responsive TFs conditionally regulate Sir-mediated silencing from X elements and that this function is distinct from their roles linked to Hda1p at other subtelomeric loci (Mak et al, 2009).

Open in a separate window

Figure 7

OAF1 conditionally regulates subtelomeric reporter gene expression and is dependent on SIR2. Strains with a subtelomeric gene encoding Ura3–GFP (Loney et al, 2009) were used to test the effects of Oaf1p on subtelomeric silencing by FACS analysis. Bars represent average geometric mean of fluorescence intensity of three to four biological replicates for each strain and error bars show s.d. values of the means. (A) Exposure of cells to the fatty acid oleate for 16 h results in a subtle but significant reduction of the levels of Ura3–GFP encoded in the subtelomeric region of arm 11L or 3R (top graph). Student's t-test P-values of the effects are 0.044 and 0.046 for arms 11L and 3R, respectively. In the presence of oleate, deletion of OAF1 results in a subtle, but significant increase of Ura3–GFP levels encoded on arm 11L or 3R, with t-test P-values of 0.002 and 0.009, respectively (bottom graph). (B) Comparison of a wild-type (WT) strain to isogenic deletion strains after growth in fatty acid medium for 16 h. Δsir2 appears to be epistatic to Δoaf1 (the effect of Δoaf1/Δsir2 double deletion is not significantly different from the effect of Δsir2 alone by Student's t-test). Masking of the Δoaf1 phenotype by Δhda1 was not detected (the effect of the two mutations together were approximately additive).

The reporter gene used here can also be used in a 5FOA viability assay that can measure levels of silencing of a subtelomeric URA3 gene, which directly correlates with cell viability in the presence of 5FOA. This assay was used to test the effects of five TFs on the expression of Ura3–GFP under conditions of X element binding as determined by chromosome position analysis (Figure 2) using previously published ChIP-chip data (Harbison et al, 2004). The results of this approach suggest a role for Gzf3p in reducing X element-mediated proto-silencing in the presence of rapamycin (Supplementary Figure S4). This positive influence on transcription is in contrast to its known role as a negative regulator of nitrogen catabolic genes (Soussi-Boudekou et al, 1997). No effect on silencing was detected for the other four factors tested in the assay (Xbp1p, Yap6p, Rox1p and Cha4p) (data not shown) possibly because of the sensitivity of the assay, or the fact that the assay requires modification of the growth conditions used for chromosome position analysis in Figure 2.

Proto-silencing activities of stress-responsive TFs

The spread of silencing in yeast subtelomeres is not continuous. Subtelomeric X elements interact with telomeric silencing molecules and relay Sir2p-mediated silencing to centromere-proximal genes. This proto-silencing activity is known to be mediated by the GRF, Abf1p (Pryde and Louis, 1999). The data presented here suggest that proto-silencing mediated by X elements is regulated in response to the cellular environment by multiple TFs that conditionally interact with X elements. In particular, Oaf1p binds X elements and enhances this silencing in response to fatty acid exposure, and Rox1p, Gzf3p and Phd1p bind and conditionally antagonize silencing in response to exposure to H₂O₂, rapamycin and butanol, respectively (Figure 5 and Supplementary Figure S4). Analysis of conditional Oaf1p-mediated silencing showed that it is dependent on SIR2 but not HDA1, consistent with a novel role for TFs in X element-mediated silencing that is distinct from their roles related to Hda1p at other subtelomeric loci (Mak et al, 2009).

Although the coregulatory mechanism identified here conditionally regulates a group of positionally and functionally related genes in subtelomeric regions that make up ~5% of the yeast genome, the consequences of this regulation remain elusive. This is consistent with our previous study showing that genes that transcriptionally respond to fatty acid exposure are largely not (measurably) required for fitness when grown on this carbon source (Smith et al, 2006). Some of these responses likely reflect the many fundamental and complex relationships between cells and their environment that enable long-term survival of the species that are not easily measurable in the laboratory. These responses might result in a subtle growth advantage or other desirable property, which can influence the evolution of the organism. It will be interesting to learn to what extent this regulatory mechanism impacts aging, evolution and other ‘broad-scale’ cellular processes.

Anti-silencing activities of stress-responsive TFs

Y′ elements are known to be protected from silencing by telomere-directed anti-silencing activity mediated by GRFs bound to X elements including Tbf1p (Fourel et al, 1999, 2001), Rap1p and Abf1p (Fourel et al, 2002). Y′ element gene expression is responsive to meiosis (Burns et al, 1994), but little is known about the regulation of Y′ element gene expression in response to environmental cues. In this study, we show that a large number of TFs interact with X elements (Figure 2) and appear to positively regulate Y′ element genes in rich medium (Figure 8A). These genes are further upregulated after the cells are switched from glucose (a fermentative carbon source that does not require the Kreb's cycle for metabolism) to non-fermentative growth conditions (with higher metabolic oxidation from active respiration) (Schuller, 2003) (Figure 8B). Our data suggest that Adr1p conditionally binds X elements (Figure 1) and is involved in activating these genes during non-fermentative growth (Figure 8B). Importantly, these data are consistent with a previous study showing that transcriptional activation domains of several TFs, when tethered to tandem DNA sites in X elements, can act as insulators of telomere-proximal genes and that this mechanism of regulation is distinct from transcriptional activation (Fourel et al, 2001). Therefore, we propose that X element-bound TFs regulate Y′ element gene expression by increasing insulation from silencing propagating from X elements.

The role of metabolic control of Y′ element gene expression is not known. Y′ elements have been implicated in an alternative lengthening of telomeres (ALT) mechanism that is active in the absence of functional telomerase, as survivors of telomerase mutants have been found to have amplification of these repeats, which may physically buffer chromatin degradation (Lundblad and Blackburn, 1993). This amplification coincides with increased expression of Y′ element genes, which encode helicases (Yamada et al, 1998) and involves active Ty1 transposable elements which also lead to gene duplications elsewhere in the genome. Therefore, it has been suggested that survivors of telomerase mutants with Y′ element amplification, likely also have increased genetic variation and are more suited to the stressful environment (Maxwell et al, 2004). This mechanism may be evolutionarily conserved, as recent data suggest that some cancer cells use a similar survival strategy to amplify subtelomeric repeat elements to lengthen telomeres (Marciniak et al, 2005). Considering these data, it will be interesting to determine whether conditional regulation of Y′ element gene expression by X element-binding TFs is linked to adaptation.

Chromosome position analysis (CPA)

Unless otherwise stated, all mapping of features to chromosome positions was done using Map Peak tool of NimbleScan software version 2.4. (Roche NimbleGen, Madison, WI). Chromosome ends were the first and last base pair of each chromosome sequence, which does not include telomeric repeats (from the Saccharomyces genome database (SGD) website (www.yeastgenome.org) on 9 September 2008). The positions of ORFs and start sites were from the S. cerevisiae annotation file from NimbleGen (created on 2 February 2007). The positions of X and Y′ elements and other genomic features were from SGD (31 December 2008). For generation of CPA histograms, features spanning more than one bin were counted only once and placed in the bin closest to the telomere. When CPA profiles were compared, Pearson's product moment correlation function of R 2.3.0 was used.

CPA of targets in a fatty acid-responsive network in Figure 1

ChIP-chip analysis of four TFs from cells grown in low glucose medium (SCIM-O), or 5 h after transferring the cells to medium containing fatty acid (SCIM-D) using intergenic microarrays has been described (Smith et al, 2007). Conditional protein–DNA interaction networks were clustered based on network topologies to identify groups of intergenic regions targeted by the same subset of TFs (Smith et al, 2007). For CPA, targets in each cluster were binned by their distance to closest telomere using the Dcount database function of Microsoft Excel. To obtain a background profile, the analysis was also done for all intergenic regions on the microarrays. The expected frequency of probes bound by Adr1p, Oaf1p and Oaf3p (but not Pip2p) was the total number of intergenic regions with this network topology/total number of intergenic regions on the array. To determine the representation of each subtelomeric region on the microarrays, individual 1 kb segments were scored as represented if there was any overlap with one or more intergenic region on the array. In all, 12.5% of chromosomes within 1 kb of telomeres were represented on the arrays; the number rose to ~75% between 5 and 10 kb from telomeres.

CPA of ChIP-chip data in Figures 2 and ​and33

ChIP-chip data of 204 DNA-binding factors (Harbison et al, 2004) was from the Pvalbyintergenic_9.2_forpaper file at http://jura.wi.mit.edu/young_public/regulatory_code/files_for_paper.zip. It was determined by CPA that 25 probes in the data set overlap with X elements. For each set of target probes (with interaction P-values <0.01), the number of X element-containing probes were determined. Significant enrichment of X element probes for each experiment was determined by calculating hypergeometric distribution P-values using R and applying a Bonferroni's multiple test correction of 352, corresponding to the number of experiments in the study. The heat maps of the binding profiles and X-element positions shown in Figure 3 were generated using MeV 4.5 (Saeed et al, 2006).

ChIP-chip analysis using tiled microarrays

ChIP-chip analysis of Oaf1p was performed in the presence (SCIM-D) and absence of fatty acids (SCIM-O) as previously described (Smith et al, 2007) with the changes described below. Three biological replicates of each experiment were performed. For each replicate, linkers were annealed to DNA ends in whole-cell extract and IP fractions, and fragments were amplified by ligation-mediated PCR with high-fidelity Taq polymerase, and shipped to NimbleGen Systems of Iceland for labeling, hybridization, scanning and preliminary analysis as described below: Equal amounts of DNA in whole-cell extracts and IP fraction were labeled with Cy3 and Cy5, respectively, combined and co-hybridized at 42°C to microarrays of 50-mer DNA probes that span both strands of the entire genome positioned every 64 bp (resulting in 14 bases of DNA between probes). Data was extracted using NimbleScan software and the log₂ ratios of IP versus whole-cell extract for each probe was determined. Log₂ ratios of adjacent probes were analyzed together to find peak regions of TF binding along with FDR using Find Peaks function of NimbleScan software. The chromatin localization data have been submitted to Gene Expression Omnibus database under accession number GSE21852.

CPA of tiling array ChIP-chip data in Figure 4

CPA was performed on peak regions of TF binding (FDRs <0.001). Histograms were generated, showing the mean number of peaks per chromosome arm in 1 kb segments of subtelomeric regions. This was calculated by dividing the number of peaks per segment by the number of chromosome arms (32) and by the number of strands analyzed (2). For each experiment, three biological replicates were analyzed separately and used to determine average number of peaks at each position and s.e.

Gene expression microarrays

The time course data set comparing cells grown in glycerol medium (YPBG) to those switched to oleate medium (YPBO) (Figure 5A) have been described (Smith et al, 2002). Comparison of TF deletion strains to a wild-type strain after 5 h of growth in medium containing fatty acids (SCIM-D) (Figures 5A and ​and7B)7B) have been described (Smith et al, 2007). The time course data set of 0.3 mM H₂O₂ exposure (Figure 5B) has been described (Gasch et al, 2000). Comparisons of TF deletion strains to a wild-type strain during growth in rich medium (YPD) (Figure 8A) have been described (Hu et al, 2007). Comparisons of cells grown in glucose (YPBD), glycerol (YPBG), or oleate (YPBO) (Figure 8B) have been described (Smith et al, 2002).

For the TF deletion analyses in Figure 5B and C, TF deletion strains were compared with an isogenic wild-type strain by microarray analysis under conditions used for ChIP-chip analysis of the factors (Harbison et al, 2004). All mutant and wild-type strains were grown in YPD to a density of 1 × 10⁷ cells/ml, treated with stimulus, and compared. For Δrox1, Δxbp1 and Δphd1 analyses, cells were exposed to 4 mM H₂O₂ for 30 min, 0.4 mM H₂O₂ for 20 min, or 1% butanol for 90 min, respectively. To prepare samples for microarray analyses, total RNA was isolated by hot acid phenol extraction and then purified on Qiagen RNeasy columns according to the manufacturer's instructions. Next, cDNA was synthesized and labeled using a Superscript indirect cDNA labeling system (Invitrogen). Equal amounts of differently labeled cDNA samples were mixed and hybridized to S.cerevisiae oligonucleotide gene expression arrays (Agilent). All experiments were performed with duplicate biological and technical replicates of each condition, and included replicates in each labeling orientation. Replicates were merged and analyzed for significantly differentially expressed genes using maximum-likelihood analysis (Ideker et al, 2000). Significantly differentially expressed genes were those with a lambda value above 20.74, which was determined empirically to correspond to a false positive rate of 0.001. These gene expression data have been submitted to Gene Expression Omnibus database under accession number GSE21926.

We thank Dr David Dilworth for insightful suggestions. We thank Dr Edward Louis for the kind gift of URA3–GFP reporter strains. This work was funded by grants NIH/NIGMS R01 GM075152, NIH P50 GM076547 and NIH U54 RR022220. We also thank the Luxembourg Centre for Systems Biomedicine and the University of Luxembourg for support. Author contributions:JS and JA conceived initial concepts and designed the research. JS, LM and LV conducted experiments. JS performed data analysis. YW contributed unpublished data. JS, RK and LM processed and displayed data for presentation. JS and JA wrote the paper.