Transcriptional response to stress in the dynamicchromatin environment of cycling and mitotic cellsAnniina Vihervaaraa,b, Christian Sergeliusa, Jenni Vasaraa,b, Malin A. H. Bloma,b, Alexandra N. Elsinga,b,Pia Roos-Mattjusa, and Lea Sistonena,b,1

aDepartment of Biosciences, Åbo Akademi University, 20520 Turku, Finland; and bTurku Centre for Biotechnology, University of Turku and Åbo AkademiUniversity, 20520 Turku, Finland

Edited by Susan Lindquist, Whitehead Institute for Biomedical Research, Cambridge, MA, and approved July 18, 2013 (received for review March 23, 2013)

Heat shock factors (HSFs) are the master regulators of transcrip-tion under protein-damaging conditions, acting in an environmentwhere the overall transcription is silenced. We determined thegenomewide transcriptional program that is rapidly provoked byHSF1 and HSF2 under acute stress in human cells. Our resultsrevealed the molecular mechanisms that maintain cellular homeo-stasis, including HSF1-driven induction of polyubiquitin genes, aswell as HSF1- and HSF2-mediated expression patterns of cocha-perones, transcriptional regulators, and signaling molecules. Wecharacterized the genomewide transcriptional response to stressalso in mitotic cells where the chromatin is tightly compacted. Wefound a radically limited binding and transactivating capacity ofHSF1, leaving mitotic cells highly susceptible to proteotoxicity. Incontrast, HSF2 occupied hundreds of loci in the mitotic cells andlocalized to the condensed chromatin also in meiosis. These resultshighlight the importance of the cell cycle phase in transcriptionalresponses and identify the specific mechanisms for HSF1 and HSF2in transcriptional orchestration. Moreover, we propose that HSF2is an epigenetic regulator directing transcription throughout cellcycle progression.

ChIP-seq | ENCODE | human genome | proteostasis

Cells exposed to proteotoxic conditions provoke a rapid andtransient response to maintain homeostasis. The stress re-

sponse induces profound cellular adaptation as cytoskeleton andmembranes are reorganized, cell cycle progression is stalled, andthe global transcription and translation are silenced (1, 2). De-spite the silenced chromatin environment, the stressed cellmounts a transcriptional program that involves induction ofgenes coding for heat shock proteins (HSPs). HSPs are molec-ular chaperones and proteases that assist in protein folding andmaintain cellular structures and molecular functions (3).Heat shock factor 1 (HSF1) is an evolutionarily well-conserved

transcription factor that is rapidly activated by stress and abso-lutely required for the stress-induced HSP expression (4). Ab-errant HSF1 levels are associated with stress sensitivity, aging,neurodegenerative diseases, and cancer (5–9). Instead of a singleHSF in yeasts and invertebrates, vertebrates contain a family offour members, HSF1–4. HSF2 and HSF4 are involved in corti-cogenesis, spermatogenesis, and formation of sensory epithe-lium, and they have primarily been considered as developmentalfactors (10–14). HSF1 and HSF2 share high sequence homologyof the DNA-binding and oligomerization domains and are ableto form heterotrimers at the chromatin (15, 16). Moreover,HSF2 participates in the regulation of stress-responsive genesand is required for proper protein clearance also at febriletemperatures (17, 18). Although HSF1 and HSF2 have beenshown to interplay on the heat shock elements (HSEs) of thetarget loci, their impacts on transcription of chaperone genes areremarkably different; HSF1 is a potent inducer of transcription,whereas HSF2 is a poor transactivator of HSPs on heat stress(19–21). HSF1 and HSF2 are also subjected to distinct regulatorymechanisms, because HSF1 is a stable protein that undergoes

rapid posttranslational modifications (PTMs), and HSF2 is pre-dominantly regulated at the level of expression (22, 23).The rapid and robust chaperone expression has served as a

model for inducible transcriptional responses (24). However, theprevious studies have almost exclusively concentrated on theexpression of a handful of HSP genes in unsynchronized cellpopulations (17, 25–28). Currently, comprehensive knowledgeon the target genes for HSF1 and HSF2 and their cooperationduring stress responses is missing. Moreover, the cell cycle pro-gression creates profoundly different environments for tran-scription depending on whether the chromatin undergoesreplication or division or whether the cell resides in the gapphases. For transcription factors, the synthesis phase provides anopportunity to access the transiently unwound DNA, whereas inmitosis, most factors are excluded from the condensed chromatin(29–32). Importantly, throughout the cell cycle progression,epigenetic cues are required to maintain the cellular identity andfate (33).In this study, we investigated the genomewide transcriptional

response that is provoked in the acute phase of heat stress infreely cycling cells and in cells arrested in mitosis. We charac-terized the transactivator capacities and the genomewide targetloci for HSF1 and HSF2 and analyzed chromatin landmarks atthe HSF target sites. By comparing transcriptional responses incycling versus mitotic cells, we determined the ability of mitoticcells to respond to proteotoxic insults and the capacity of tran-scription factors to interact with chromatin that is condensed for

Significance

We determined the transcriptional program that is rapidlyprovoked to counteract heat-induced stress and uncovered thebroad range of molecular mechanisms that maintain cellularhomeostasis under hostile conditions. Because transcriptionalresponses are directed in the complex chromatin environmentthat undergoes dramatic changes during the cell cycle pro-gression, we identified the genomewide transcriptional re-sponse to stress also in cells where the chromatin is condensedfor mitotic division. Our results highlight the importance of thecell cycle phase in provoking cellular responses and identifymolecular mechanisms that direct transcription during theprogression of the cell cycle.

Author contributions: A.V. and L.S. designed research; A.V., J.V., M.A.H.B., and A.N.E.performed research; A.V. and C.S. performed computational data analysis; A.V., C.S., P.R.-M.,and L.S. analyzed data; and A.V. and L.S. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

Data deposition: The high-throughput sequencing data reported in this paper have beendeposited in the Gene Expression Omnibus (GEO) database, www.ncbi.nlm.nih.gov/geo(accession no. GSE43579).1To whom correspondence should be addressed. E-mail: lea.sistonen@abo.fi.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1305275110/-/DCSupplemental.

E3388–E3397 | PNAS | Published online August 19, 2013 www.pnas.org/cgi/doi/10.1073/pnas.1305275110

Dow

nloa

ded

by g

uest

on

May

4, 2

020

cell division. We discovered the broad range of molecularmechanisms that maintain cellular homeostasis in stressed cellsand provide unique mechanistic insights into the regulation ofgene expression during the cell cycle progression. Our resultsrevealed the cooperation of HSF1 and HSF2 in orchestratinggene expression in stressed cycling cells and identified theirprofoundly distinct capacities to coordinate transcription in cellswhere the chromatin is compacted for cell division.

ResultsGenomewide Identification of Target Sites for HSF1 and HSF2 inCycling and Mitotic Cells. ChIP coupled to massively parallel se-quencing (ChIP-seq) is a powerful method that enables genome-wide mapping of protein binding sites in a high-resolution andunbiased manner (34–36). We used ChIP-seq to characterize thebinding sites for HSF1 and HSF2 in cycling and mitotic cells thatwere either untreated or heat treated for 30 min at 42 °C. As themodel system, we chose human K562 erythroleukemia cells, whereHSF1 and HSF2 levels and regulatory mechanisms are wellcharacterized (17, 20, 37), and chromatin landmarks have beenidentified by the Encyclopedia of DNA Elements (ENCODE)consortium (38). The efficiency of cell cycle arrest was improvedby collecting the cells in S-phase before nocodazole treatment(39). Histograms of cells based on the DNA content are shown inFig. 1A, confirming the mitotic arrest (5% of cells in G1 and 85%in G2/M) compared with freely cycling cells (45% in G1 and 15%in G2/M). For sequencing, 10 PCR-verified ChIP-replicates werecollected per sample, and two controls, IgG and input, were in-cluded (Fig. S1).The ChIP-seq provided high-resolution maps of HSF1 and

HSF2 target loci in the human genome (Dataset S1; Gene Ex-pression Omnibus accession no. GSE43579). Under optimalgrowth conditions in cycling cells, 45 target sites were identifiedfor HSF1 and 148 for HSF2. On acute stress, both the number ofthe target sites (1242 for HSF1 and 899 for HSF2) and the foldenrichments of the targets were considerably higher, indicatinga rapid recruitment of HSF1 and HSF2 to their target loci inheat-stressed cycling cells (Fig. 1B; Dataset S1). In mitosis, theability of HSFs to bind chromatin was clearly different; HSF2occupied 50 loci under optimal conditions and 545 loci on acutestress. In contrast, HSF1 interacted only with the promoter ofHSPA1B/HSP70.2 in the absence of stress, and with 35 loci onheat stress (Fig. 1B; Dataset S1). Although effectively excludedfrom the dividing chromatin, HSF1 displayed prominent heat-induced occupancy on certain loci, including promoters ofHSPA1A/HSP70.1, HSPA1B/HSP70.2, HSPH1/HSP110, MRPS6

(mitochondrial ribosome protein 6), and DNAJB6 (Fig. 1C;Dataset S1). Enrichments of HSF1 and HSF2 on selected targetgenes are illustrated in Fig. 1 C–F.The HSPA1/HSP70 locus was strongly bound by HSF1 and

HSF2 in heat-treated cycling and mitotic cells (Fig. 1C), whereasDUSP1 (dual specific phosphatase 1) was occupied by HSF1 andHSF2 in cycling cells only, demonstrating the importance of thecell cycle phase in transcriptional responses (Fig. 1D). In Fig. 1 Eand F, the capacity of HSF1 and HSF2 to bind their individualtarget loci is indicated with promoters of GBA (glucosidase βacid) and MLL (myeloid/lymphoid or mixed-lineage leukemia).Intriguingly, HSF2 occupied the promoter of MLL in unstressedmitotic cells, although in cycling cells the binding was strictlyinduced by heat shock (Fig. 1F). Promoter-proximally pausedRNA polymerase II (RNPII) was originally identified at theHSP70 promoter (40, 41), and it is estimated to poise ∼30% ofhuman genes for rapid or synchronous activation (42). We in-vestigated the status of RNPII at selected HSF target promotersusing an existing ChIP-seq data on RNPII (wgEncodeEH000616;Snyder Laboratory, Yale University). ChIP cannot determinetranscriptional engagement, but recent global-run-on sequencing(GRO-seq) experiments revealed that the majority of promoter-associated polymerases are transcriptionally engaged and com-petent for elongation (43). Accordingly, we refer to RNPIIwhose enrichment on a promoter site exceeds at least five timesthe overall signal on the gene as paused RNPII (see below). InFig. 1 C–F, the distribution of RNPII is visualized.

HSF1 and HSF2 Recognize Similar Consensus DNA Sequences butDisplay Distinct Binding Profiles in the Human Genome. Binding ofHSF2 to target genes on stress has been considered to occur inan HSF1-dependent manner (16, 17). However, we identifiedtarget genes that are specific for either HSF1 or HSF2 (Fig. 1 Eand F; Fig. S2A; Dataset S1). In heat-treated cycling cells, HSF1and HSF2 shared a majority of their target sites, but under op-timal growth conditions and in mitosis, they displayed strikinglydifferent localization to the human genome (Fig. S2A). Despitetheir distinct binding sites, HSF1 and HSF2 recognized similarHSEs both in cycling and mitotic cells (Fig. S2B), demonstratingthat the DNA sequence alone is not sufficient to determine thebinding site for HSF1 vs. HSF2.Given that promoters and exons account for <2% of the hu-

man genome (Human Genome Project), HSF1 and HSF2 wereenriched on gene promoters and protein coding sequences (Fig.S3A). For example, 19% of HSF1 and 22% of HSF2 target lociwere found within promoters in heat-treated cycling cells (Fig.

Fig. 1. Binding sites for HSF1 and HSF2 in cycling and mitoticK562 cells. (A) Histograms of cycling and mitotic cells based onthe DNA content. (B) (Upper) Number of HSF1 (red) and HSF2(green) target loci in cycling and mitotic cells. (Lower) Foldenrichment over input of each identified HSF1 (red) and HSF2(green) target locus. The dashed gray line indicates fold en-richment five, which was set as cutoff criterion for HSF targetsites. (C–F) Enrichments of HSF1 (red) and HSF2 (green) on in-dicated target genes in cycling and mitotic cells. (C) HSPA1Aand HSPA1B promoters are bound by HSF1 and HSF2 in cyclingand mitotic cells. (D) DUSP1 promoter harbors prominent HSF1and HSF2 enrichments in stressed cycling cells only. (E) GBA isan HSF1-specific and (F) MLL is an HSF2-specific target pro-moter. Distribution of RNPII in nontreated cycling cells (blue) isrecovered from the ENCODE project (wgEncodeEH000616;Snyder Laboratory, Yale University). The scale for each HSF orinput sample is set to 0–100, except at the HSPA1 locus, where,due to high enrichments, the scale is set to 0–250 in cycling and0–125 in mitotic cells. C, control; HS, heat shock.

Vihervaara et al. PNAS | Published online August 19, 2013 | E3389

CELL

BIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

May

4, 2

020

S3A). On average, the enrichments of HSF1 and HSF2 werehigher on promoter regions than on coding sequences (Fig. S3B),but prominent HSF1 and HSF2 targets were found also withinintrons, exons, and intergenic regions as illustrated with PRKCAand PRKCB (protein kinase C α and β, respectively), KCNN1(calcium-activated channel N1), and an intergenic region up-stream of HSPB1/HSP27 (Fig. S3C).

HSF1 and HSF2 Bind Selected Chaperone Gene Promoters. HSF1 andHSF2 are best known for their stress-related regulatory functionson certain chaperone genes, such as HSPA1A/HSP70.1, HSPC/HSP90, HSPB1/HSP27, and DNAJB1/HSP40 (4, 17, 26, 27). The

ChIP-seq enabled unbiased analyses of HSF1 and HSF2 on everychaperone gene encoded in the human genome and revealedHSF occupancy on 70% of HSP, 90% of chaperonin, and 13% ofDNAJ genes (Table 1; Table S1). A majority of the HSF-targetedchaperone genes were bound by both HSF1 and HSF2 on pro-moters that contained paused RNPII (Table 1; Table S1).Exceptions were the small HSPs HSPB2/HSP27-2, HSPB5/CRYAB, and HSPB9, whose promoters lacked paused RNPII(Table 1). To address the impact of HSF1 and HSF2 on tran-scription, we depleted either HSF1 or HSF2, or both factorstogether, by shRNA-mediated degradation and analyzed a setof HSF-targeted chaperone genes for their mRNA expression

Table 1. HSF1 and HSF2 occupancy on HSPA, HSPB, HSPC, HSPH, and chaperonin genes

Family Gene Alias

HSF1 HSF2

Bindingsite

PausedRNPII*

Cycling Mitosis Cycling Mitosis

C HS C HS C HS C HS

HSPA HSPA1A HSP70-1 2.8 40.4 2.5 16.3 5.2 25.6 3.3 10.8 Promoter YesHSPA1B HSP70-2 4.3 45.1 5.3 20.5 5.4 26.8 5.4 15.5 Promoter YesHSPA1L HSP70-1L 2.8 40.4 2.5 16.3 5.2 25.6 3.3 10.8 Promoter YesHSPA2 HSPA3 — — — — — — — — — NoHSPA5 BIP, GRP78 — — — — — — — — — YesHSPA6 HSP70B’ — 15.2 — — — 7.3 — — Promoter YesHSPA7 HSP70 — 3.0 — — — — — — Promoter NoHSPA8 HSC70 4.3 29.5 2.8 5.3 20.2 26.8 8.3 12.3 Promoter YesHSPA9 GRP75 — 6.6 — — — 5.9 — — Promoter YesHSPA12A FLJ13874 — — — — — — — — — NoHSPA12B RP23-32L15.1 — — — — — — — — — NoHSPA13 Stch — — — — — — — — — YesHSPA14 HSP70-4 — — — — — — — — — Yes

HSPB HSPB1 HSP27-1 — 23.8 — 6.3 — 16.5 3.5 5.6 Promoter YesHSPB2 HSP27-2 — 14.1 — — — 10.4 — — Promoter NoHSPB3 HSPL27 — — — — — — — — — NoHSPB4 CRYAA — — — — — — — — — NoHSPB5 CRYAB — 14.1 — — — 10.4 — — Promoter NoHSPB6 HSP20 — — — — — — — — — NoHSPB7 cvHSP — — — — — — — — — NoHSPB8 HSP22 — — — — — — — — — NoHSPB9 FLJ27437 — 12.3 — 3.5 7.0 19.0 — 6.0 Promoter NoHSPB10 ODF1 — — — — — — — — — NAHSPB11 HSP16.2 — — — — — — 3.1 — Promoter Yes

HSPC HSPC1 HSP90AA1 — 13.2 — 6.0 — 10.5 3.2 5.3 Promoter YesHSPC2 HSP90AA2 — — — — — — — — — NAHSPC3 HSP90AB1 — 23.8 — 8.2 6.8 14.1 — 9.9 Promoter YesHSPC4 HSP90B1 — — — — — — — — — YesHSPC5 TRAP1, HSP90L — — — — — — — — — Yes

HSPD HSPD1 HSP60, GROEL 5.8 43.0 3.5 15.1 17.8 33.4 6.0 9.5 Promoter YesHSPE HSPE1 HSP10, GROES 5.8 43.0 3.5 15.1 17.8 33.4 6.0 9.5 Promoter YesHSPH HSPH1 HSP105/110 — 32.6 — 8.8 — 19.0 — 6.3 Promoter Yes

HSPH2 HSPA4 — 22.5 — 4.2 — 20.9 — 4.2 Promoter YesHSPH3 HSPA4L — 9.1 — 4.2 — 5.5 — 3.5 Promoter YesHSPH4 HYOU1 — — — — — — — — — Yes

TRiC CCT1 TCP1, CCTA 3.5 31.8 — 4.6 8.5 21.5 — 5.2 Promoter YesCCT2 CCTB — 8.7 — — — 6.3 — — Prom/Int YesCCT3 CCTG — 8.5 — — — 8.5 3.9 — Prom/Int YesCCT4 CCTD — 12.4 — 3.9 4.7 8.5 — — Promoter YesCCT5 CCTE — 14.0 — 2.7 — 15.3 — 4.9 Promoter YesCCT6A CCTZ — — — — — 5.7 — — Promoter YesCCT6B CCTZ2 — — — — — — — — — YesCCT7 CCTH — 12.5 — 3.5 — 11.4 — 3.9 Promoter YesCCT8 CCTQ — 5.0 — — — 7.1 — — Promoter Yes

Fold enrichments ≥2 are shown. The nomenclature of HSPs is according to ref. 85. C, control; HS, heat shock; NA, not analyzed; Prom/Int, promoter orintron depending on the transcript variant.*The RNPII density signal is recovered from ENCODE (wgEncodeEH000616; Snyder Laboratory, Yale University).

E3390 | www.pnas.org/cgi/doi/10.1073/pnas.1305275110 Vihervaara et al.

Dow

nloa

ded

by g

uest

on

May

4, 2

020

during heat shock (Fig. 2 A and B; Fig. S4A). In freely cyclingcells, HSPH1/HSP110, HSPH2, and DNAJB6 were identified asunique HSF-regulated genes whose heat-induced expression wasstrictly dependent on HSF1 (Fig. 2B; Fig. S4A). CCT1 (chaper-onin 1) has previously been shown to be regulated by HSFs (44),which we corroborated in K562 cells (Fig. 2B). The HSPH1/HSP110 mRNA expression rapidly increased to sevenfold,whereas the levels of DNAJB6 and CCT1 mRNA doubledduring the acute heat stress (Fig. 2B). DNAJB6 is the most po-tent chaperone in inhibiting protein aggregation, and HSPH1/HSP110 remodels the HSP70-HSP40 machinery to solubilizeand refold aggregated proteins in metazoan cells (45, 46). Theseresults indicate a considerably wider repertoire of HSF-inducedchaperones than earlier anticipated and highlight the importanceof managing aggregated proteins under proteotoxic conditions.

HSF1 Is Required for Heat-Induced Cochaperone Expression. Asshown in Fig. 2C and Dataset S1, HSF1 and HSF2 occupiedmany cochaperone genes, which have not previously been dem-onstrated to be under control of HSFs. Most cochaperonespossess intrinsic chaperone activity but are termed cochaperonesbecause they determine the activities of the HSP70 and/orHSP90 protein complexes (47, 48). As summarized in Fig. 2C,HSF1 and HSF2 showed a strong tendency to bind cochaperonegene promoters that contain the paused RNPII. We determinedthe mRNA levels of the HSP90 cochaperones AHSA1 (encodingAHA1), CDC37, and PTGES3 (encoding p23) in the presenceand absence of HSF1 or HSF2 (Fig. 2D). AHA1, which stim-ulates the ATPase cycle and is the most potent activator of the

HSP90 protein complex (49), showed a threefold heat-inducedexpression that required the presence of HSF1 (Fig. 2D). CDC37and p23 are involved in client protein recruitment and matura-tion, and they inhibit the HSP90 ATPase cycle (48, 50, 51). Wedid not detect any significant change in CDC37 mRNA levelsduring the 2-h heat shock but found a twofold HSF1-dependentincrease in p23 mRNA (Fig. 2D). Curiously, HSF1 and HSF2bound the promoters of AHSA1 and PTGES3, but the first intronof the noninducible CDC37 (Fig. 2C). We conclude that in ad-dition to controlling the set of chaperones induced by acute heatstress, HSFs also determine the expression patterns of cocha-perones, which are critical for the recognition and fate of theclients, as well as for the chaperoning efficacy.

Stress-Generated Demand for Protein Clearance Is Mitigated by HSF1-Dependent Ubiquitin Expression. On heat stress, damaged andaggregated proteins accumulate and generate an increased de-mand for protein clearance. The vast majority of proteins des-tined for degradation are marked by ubiquitin, which is encodedby four genes in humans (52). As revealed by ChIP-seq, HSF1and HSF2 selectively bound to the promoters of polyubiquitingenes UBB and UBC but were absent on the monoubiquitingenes RPS27A and UBA52 (Fig. 3A). Transcription of UBB orUBC enables a prompt increase in the ubiquitin levels becausetheir corresponding mRNAs encode a precursor for three andnine ubiquitin proteins, respectively (Fig. 3B). In comparison,transcription of RPS27A or UBA52 would generate one ribo-somal protein per each ubiquitin produced (Fig. 3B). In heat-treated cycling cells, the mRNA levels of UBB and UBC rapidly

Fig. 2. HSFs define the set of chaperones and cochaperonesinduced in response to acute heat stress. (A) HSF1 and HSF2protein levels during the 2-h heat stress in scrambled-trans-fected (Scr), HSF1-depleted (shHSF1), and HSF2-depleted(shHSF2) cycling cells. HSF2 levels decline on heat stress aspreviously reported (25). (B) mRNA levels of chaperones HSPH1,HSPH2, DNAJB6, and chaperonin CCT1 during heat stress in thepresence or absence of HSF1 or HSF2. (C) Cochaperone genesbound by HSF1 and HSF2 in cycling cells. Fold enrichments overinput as detected with ChIP-seq are indicated. (D) mRNA levelsof HSP90 cochaperones AHA1, CDC37, and p23 in nontreatedand heat-treated cycling cells with or without HSF1 or HSF2. InB and D, SDs of a minimum of three biological replicates areshown. The statistical analyses were conducted with two-tailedindependent Student t test and asterisks denote statisticalsignificance between indicated scrambled-transfected and HSF-depleted cells. *P < 0.1; **P < 0.05; ***P < 0.01.

Vihervaara et al. PNAS | Published online August 19, 2013 | E3391

CELL

BIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

May

4, 2

020

increased to two- and fourfold, respectively, whereas the levels ofRPS27A and UBA52 mRNA remained unchanged (Fig. 3 C andD). Depletion of HSF1 severely compromized the heat-inducedubiquitin expression in cycling cells (Fig. 3D), revealing thatHSF1 localization to the promoters of polyubiquitin genes isrequired for the cell to generate adequate levels of ubiquitinmRNA under proteotoxic stress.

HSF1- and HSF2-Driven Transcriptional Programming Extends toGenes Encoding Transcriptional and Translational Regulators, CellCycle Determinants, and Core Signaling Components. To investigatethe biological processes associated with HSF1 and HSF2 targetgenes in acute stress, we performed gene ontology analysis usingDatabase for Annotation, Visualization and Integrated Discovery(DAVID) (53). Because heat shock leads to a rapid transcriptionalresponse, we sought to understand whether HSFs are recruited togenomic loci that harbor RNPII before stress (wgEncodeE-H000529; Iyer Laboratory, University of Texas) and whether thepresence or absence of RNPII characterizes HSF target genesassociated with a specific function. Interestingly, 90% of HSF1- orHSF2-targeted loci on promoters were occupied by RNPII, and atthe gene bodies, approximately half of the HSF1 or HSF2 targetsites harbored RNPII (Fig. 4A).In heat-shocked cycling cells, HSF1 and HSF2 co-occupied

promoters of chaperones and cochaperones, transcriptional andtranslational regulators, and mediators of cell cycle progression(Fig. 4A; Dataset S2). On the gene bodies, HSF1 occupiedRNPII-containing loci of genes that mediate transcriptional re-pression, methylation, and inhibition of mRNA metabolism. As acomparison, HSF2 localized to the coding sequences of genesinvolved in activation of diverse cellular processes, specifically onthe sites that lacked RNPII (Fig. 4A; Dataset S2). Among thepromoter-targeted genes, we verified the mRNA levels of tran-scriptional regulators TAF7 (TATA-box associated factor 7) andMLL, translational components EEF1G (eukaryote elongationfactor 1γ) and MRPS6, as well as mitotic factors NUDC (nucleardistribution C homolog) and BANF1 (barrier to autointegrationfactor 1). Among the genes that harbored HSF on the codingsequence, mRNA levels of CTCF (CCCTC-binding factor) andPRKCA were investigated. In response to heat stress, an HSF1-dependent increase to 3-fold was detected for TAF7, 2-fold forEEF1G, and 1.5-fold for NUDC (Fig. 4B). Depletion of HSF2did not hamper the heat-inducible expression of any of the

studied genes, but it disrupted the expression kinetics of MLL(Fig. 4B). Moreover, absence of HSF2 caused more than a two-fold increase in mRNA levels of BANF1 and PRKCA underoptimal growth conditions (Fig. 4B).

Mitosis Inhibits Transcriptional Activation and Renders ChromatinInaccessible for HSF1. HSF1 bound to 35 target loci in heat-trea-ted mitotic cells, mainly on the promoters of chaperones andtranslational components (Fig. 5 A and B; Datasets S1 and S2). Itis noteworthy that HSF1 was unable to bind to 1,207 loci, whichit occupied in heat-stressed cycling cells (Fig. 5B). Among thegenes that lacked HSF1 in stressed mitotic cells were UBB andUBC (Fig. S4B), and subsequently, the ubiquitin gene expressionremained unchanged during heat treatment of mitotic cells (Fig.S4C). In contrast to the radically limited capacity of HSF1 tointeract with the mitotic chromatin, HSF2 occupied hundreds oftarget loci both in cycling and mitotic cells (Figs. 1B and 5B;Dataset S1). However, depending on the phase of the cell cycle,HSF2 localized to a distinct set of targets (Fig. 5 A and B;Datasets S1 and S2). We confirmed that depletion of HSF1 orHSF2 did not interfere with synchronization of cells to mitosis(Fig. 5 C and D) and investigated whether binding of HSF1 orHSF2 leads to induced transcription of their target genes inmitotic cells (Fig. 5E; Fig. S4D). Although HSF1 was a potenttransactivator in cycling cells (Figs. 2–4), no significant inductionof any of the studied HSF1 and HSF2 target genes, HSPH1/HSP110, HSPH2, NUDC, DNAJB6, or MRPS6, was detected inmitosis (Fig. 5E; Fig. S4D). These results indicate that besidesdisplaying an impaired ability to bind to the mitotic chromatin,the occupancy of HSF1 at promoters does not induce tran-scription in the mitotic environment, where the promoters likelylack RNPII and the condensed chromatin generates barriers fortranscriptional initiation and elongation (30). Moreover, mitoticcells have been shown to be highly susceptible to heat-inducedstress (54, 55), which is in agreement with increased cell deaththat we observed in heat-treated mitotic cells (Fig. S4E).The mRNA levels of the HSF2-specific target gene, MLL,

remained relatively constant in the mitotic cells exposed to acuteheat stress. However, depletion of HSF2 led to increased MLLexpression under nonstress conditions (Fig. 5E), suggesting thatHSF2 either inhibits transcription or modifies the chromatin en-vironment ofMLL. To elucidate the role of HSF2 in mitosis, whenthe overall transcription is silenced (30), we monitored the expres-

Fig. 3. HSF1 induces polyubiquitin gene expression in cyclingcells. (A) HSF1 and HSF2 bind to polyubiquitin (UBB and UBC)but not to monoubiquitin (RPS27A, UBA52) genes as shown byChIP-seq. (B) Schematic illustration of the four ubiquitin codinggenes in the human genome. (C and D) Heat stress leads toincreased expression of polyubiquitin but not monoubiquitingenes in cycling cells, and the heat-induced ubiquitin expres-sion is abolished in HSF1-depleted cells. The mRNA levels andstatistical significances were analyzed as in Fig. 2.

E3392 | www.pnas.org/cgi/doi/10.1073/pnas.1305275110 Vihervaara et al.

Dow

nloa

ded

by g

uest

on

May

4, 2

020

sion kinetics of MLL after the non–heat-shocked cells had beenreleased from the mitotic arrest (Fig. 6A). In scrambled-transfectedcells, the mRNA levels of MLL gradually doubled as the cellsprogressed from mitosis to G1 phase (Fig. 6A). These results arein agreement with a previous study where the MLL protein levelswere shown to increase after mitosis (56). However, the post-mitotic induction of MLL expression in HSF2-depleted cells wasonly minute and showed delayed kinetics (Fig. 6A). As a control,we used DUSP1, which was not bound by HSFs in the mitoticcells (Fig. 1D). DUSP1 is a heat-inducible gene (57), and noelevation in DUSP1 mRNA expression was detected after themitotic release, nor did the absence of HSF2 affect the DUSP1expression during cell cycle progression (Fig. 6A). In conclusion,the ability of HSF2 to bind to the MLL promoter in mitosis (Fig.1D) and to direct its postmitotic expression (Fig. 6A) indicatesthe involvement of HSF2 in restoring the transcription of MLLafter the mitotic silencing.The binding of HSF2 to mitotic chromatin (Figs. 1B and 5B;

Fig. S2A) argued that a great number of HSEs are accessible fortranscription factors during cell division. We compared the HSF

target sites in heat-treated mitotic cells to DNaseI hypersensitiveregions in mitosis (wgEncodeEH003472; Crawford Laboratory,Duke University) and found that open chromatin is a pre-requisite for HSF1 binding, as 97% of its target sites occurredwithin DNaseI sensitive regions (Fig. 6B). Instead, 42% of HSF2targets in mitosis were found within open chromatin (Fig. 6B).Although HSF2 was able to bind both to DNaseI hypersensitiveand insensitive regions, this did not explain why HSF1 was absentfrom 195 DNaseI hypersensitive loci that contained HSF2. Next,we addressed whether the exclusion of HSF1 from the mitoticchromatin is mediated by inhibiting the intrinsic DNA-bindingcapacity of HSF1 or by rendering the chromatin inaccessible forHSF1. As analyzed with oligonucleotide-mediated pull-down,both HSF1 and HSF2 bound to the exogenous HSE-containingoligonucleotides in cycling and mitotic heat-treated cells (Fig.6C). Curiously, HSF2 bound to DNA also in untreated cyclingcells, and its levels declined in mitosis (Fig. 6C). Besides bindingto the exogenous HSE, hyperphosphorylation of HSF1 wasdetected in mitosis (Fig. 6C), indicating HSF1 transactivatingcapacity (22). These results demonstrate that HSF1 was indeed

Fig. 4. HSFs control a broad range of cellular processes during stress. (A) Gene ontology analyses of HSF1 and HSF2 target genes in heat-treated cycling cellsin respect to genomic region and to presence or absence of RNPII. Promoter is defined as −1,200 to +300 bp from the TSS, and gene body is coding sequencefrom +301 bp to the end of the gene. HSF1-specific target gene groups are indicated in red, HSF2-specific target gene groups are in green, and their sharedtarget gene groups are in blue. (Lower) Fractions of HSF1 or HSF2 target sites that localize to RNPII-occupied sites. RNPII-containing sites in nontreated cyclingcells are recovered from the ENCODE project (wgEncodeEH000529; Iyer Laboratory, University of Texas). (B) mRNA levels of HSF target genes during heatshock. mRNA levels and statistical significances were analyzed as in Fig. 2.

Vihervaara et al. PNAS | Published online August 19, 2013 | E3393

CELL

BIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

May

4, 2

020

capable of binding to DNA in mitosis, but its access to chromatinwas dramatically reduced from 1,242 sites in heat-treated cyclingcells to only 35 sites in cells arrested to mitosis.An intriguing congruency has been reported during meiotic

division in mouse spermatocytes, where HSF1 was shown to bindto the exogenous HSE on heat stress but no binding to theHSP70 promoter could be detected (11, 58–60). Because HSF2-deficient mice display meiotic defects (61), we investigated byconfocal microscopy the subcellular localization of HSF1 andHSF2 in metaphase spermatocytes. HSF2 was abundant at thechromatin in meiotic divisions, whereas HSF1, albeit highly ex-pressed, was detected only outside the dividing chromatin (Fig.6D). These results suggest a common mechanism in cell divisionthat selectively allows HSF2 to bind mitotic and meiotic chro-matin from which HSF1 is efficiently excluded.

DiscussionCharacterization of the genomewide binding sites and trans-activating capacities for HSF1 and HSF2 discovered the extentof HSF-mediated transcriptional reprogramming under acuteheat stress. The results uncovered molecular mechanisms that

maintain homeostasis, highlighting the myriad of different cellularprocesses that are orchestrated under protein-damaging con-ditions. In stressed freely cycling cells, HSFs induced the ex-pression of polyubiquitin genes, which is a rapid means toproduce a large amount of ubiquitin (Fig. 3). Ubiquitin is a smallsignaling molecule with versatile functions in the cell, and it isrequired for the proteasomal degradation of damaged proteins(52). Polyubiquitin genes have been shown to be induced onexposure to heat (52, 62), and here we provide evidence for thedirect involvement of HSF1 and HSF2 in the regulation of pol-yubiquitin gene expression. HSFs also defined the expression ofchaperones (Table 1; Table S1; Fig. 2B) and cochaperones (Fig.2 C and D) that facilitate protein folding. Previously, a subset ofchaperone genes has been used as a model for induced tran-scriptional responses, but here we characterized the complete setof human chaperones that are transcriptionally regulated by HSF1and HSF2 (Table 1; Table S1). Importantly, we discovered thatHSF1 and HSF2 also determine the cochaperones that are in-duced during heat stress, demonstrating that HSFs orchestrate the

Fig. 5. HSF2 interacts with mitotic chromatin. (A) Gene ontology analyses ofHSF1 and HSF2 target loci in mitosis. The genomic regions were defined as inFig. 4. (B) The number of HSF1 (red) and HSF2 (green) target loci in cyclingand mitotic heat-treated cells. (C) HSF1 and HSF2 protein levels in scrambled-transfected (Scr), HSF1-depleted (shHSF1), and HSF2-depleted (shHSF2) mi-totic cells during heat stress. (D) Histograms of scrambled-transfected (Scr),HSF1-depleted (shHSF1), and HSF2-depleted (shHSF2) mitotic cells after a 1-hheat treatment, indicating that neither transfection nor lack of HSF1 or HSF2compromises the synchronization of cells to G2/M. (E) mRNA levels of HSPH1,HSPH2, NUDC, and MLL in mitotic cells with or without HSF1 or HSF2. mRNAlevels and statistical significances were analyzed as in Fig. 2.

Fig. 6. Chromatin compaction during mitosis and meiosis renders DNA in-accessible for HSF1. (A) mRNA levels of MLL and DUSP1 in the presence (Scr)or absence (shHSF2) of HSF2. The analyses were conducted in non–heatshock conditions at indicated times points after releasing the cells frommitotic arrest. mRNA levels and statistical significances were analyzed as inFig. 2. (Right) Progression of cells from mitotic arrest is illustrated with FACSprofiles. (B) Fraction of HSF1 or HSF2 target sites in mitosis that occur withinDNaseI hypersensitive regions. The DNaseI sensitive regions in G2/M are re-covered from the ENCODE project (wgEncodeEH003472; Crawford Labora-tory, Duke University). (C) Oligonucleotide-mediated pull-down in cyclingand mitotic cells. Bound fraction indicates binding of HSF1 or HSF2 to theHSE-containing oligonucleotide (+). The scrambled oligonucleotide sequence(−) lacks an HSE. WB indicates the total HSF1 and HSF2 protein levels in therespective samples. (D) Single sections of confocal microscope imagesshowing localization of HSF1 (red) and HSF2 (green) in mouse male germcells during meiotic division I. DNA (stained with Hoechst) is shown in blue.(Scale bar, 100 μm.)

E3394 | www.pnas.org/cgi/doi/10.1073/pnas.1305275110 Vihervaara et al.

Dow

nloa

ded

by g

uest

on

May

4, 2

020

whole machinery that maintains proteostasis. Beyond the proteinquality control, HSFs directed genes encoding transcriptional andtranslational regulators, mitotic determinants, and core compo-nents of various signaling pathways (Fig. 4; Dataset S2). For ex-ample, TAF7 has been suggested to function as a transcriptionalcheckpoint regulator by inhibiting RNPII phosphorylation at thepromoter (63, 64), whereas BANF1 and NUDC are key regulatorsin mitosis required for correct spindle assembly, cytokinesis, andnuclear envelope assembly (65–68). The stress response has dra-matic effects on cellular physiology, but the basic mechanisms ofthe adaptation processes such as the transcriptional and trans-lational silencing or the stalling of the cell cycle progression arenot understood. The characterized HSF1- and HSF2-driventranscriptional reprogramming will aid in elucidating the detailedmechanisms that rapidly adjust cellular processes during stress, aswell as clarify how the processes are reestablished once favorableconditions for proliferation are again restored.

Functions of HSF1 and HSF2 Are Tightly Interlinked, but Their Effectson Promoters Are Profoundly Distinct. Previous genomewide stud-ies in yeast (69), fly (70, 71), and humans (8, 72) identified theHSE for HSF1, showing a striking conservation of the HSF1-recognized DNA sequence over evolution. Our analyses cor-roborated the importance of inverted nGAAn pentamers forHSF1 binding both in cycling and mitotic cells (Fig. S2B). Byrevealing the HSF2-targeted HSEs in a genomewide scale, weexpected to find a major difference between the DNA sequencespreferred by HSF1 vs. HSF2. Despite their unique target sites inthe human genome (Figs. 1 B, E, and F and 5B; Fig. S2A), HSF1and HSF2 recognized a staggeringly similar HSE (Fig. S2B).Earlier footprinting studies have revealed that HSF2 is able tobind shorter HSEs than HSF1 (19, 37). The ability of HSF2 tobind less extensive HSEs could provide an advantage whenaccessing HSEs in the compacted chromatin regions. The over-lapping genomic coordinates and highly similar binding profilesof HSF1 and HSF2 at the shared target sites (Figs. 1 C and D and3A; Fig. S3C; Dataset S1) suggest their intimate interplay in tran-scriptional regulation. The resolution of our ChIP-seq analyses,however, cannot distinguish HSF1-HSF2 heterotrimers fromhomotrimers on adjacent pentamers of the same or nearby HSE.Cancer cells display a nononcogene addiction to HSF1 as a re-

sult of ameliorated proteotoxic stress (5, 73). Moreover, thebinding profile of HSF1 in nontransformed cells radically differsfrom that in cancer cells with high malignant potential (8). Be-cause K562 cells are an erythroleukemia cell line and express highbasal levels of HSPs, HSF1 could have cancer-specific target genesalso in these cells. However, the low fold enrichments and thelimited number of HSF1 target sites in nonstressed K562 cells(Fig. 1B) are in accordance with the HSF1 binding profiles innontransformed cells and in cells with low malignant potential (8).The detailed analyses of target gene expression revealed HSF1

as a potent transactivator that responds instantly to stress (Figs.1B and 2–4). Also HSF2 rapidly localized to the target sites (Fig.1B), but it did not enhance transcription during 2 h of heat stress(Figs. 2–4). HSF1 and HSF2 show similar recruitment kinetics tothe HSP70 promoter, but HSF2 is gradually degraded from theHSE after 30 min of heat stress (25), indicating the need fordelicate regulation of the HSF1-HSF2 composition at the pro-moter. Our preliminary results show an HSF2-mediated effecton transcription during prolonged stress when HSF2 levels havealready declined (Fig. S4F) (25) and suggest that HSF2 modifiesthe chromatin landscape for transcriptional responses. HSF2 isalso involved in many developmental processes when the epi-genetic features of the cells are determined (10, 12, 61), and ithas been shown to affect the chromatin compaction and integrity(10). These results argue for an HSF2-mediated establishment ofthe transcriptional environment and indicate profoundly distinctmechanisms for HSF1 and HSF2 during their dynamic interplay

in driving gene expression. The versatility of cellular processesdirected by HSF1 and HSF2 underscore the importance ofestablishing their specific activators and inhibitors, PTM sig-natures, and chromatin-associated factors that direct HSFs totheir individual target loci.

Is HSF2 an Epigenetic Regulator of Cell Fate?An unexpected findingof our studies was the ability of HSF2 to interact with the di-viding chromatin where HSF1 was hardly detected (Figs. 1B and6D). HSF2 has been reported to bind HSP promoters in mitosisand suggested to mediate bookmarking of stress-responsivegenes (74, 75). We identified both HSF1 and HSF2 on chaper-one genes in mitosis, including promoters of HSPA1A/HSP70.1,HSPA1B/HSP70.2, HSPH1/HSP110, and HSPE1/HSP10 (Fig.1C; Dataset S1). However, we did not detect HSF1-inducedtranscription in mitosis (Fig. 5E; Fig. S4D), indicating an im-paired ability of dividing cells to provoke transcriptional re-sponses. HSF1 has been shown to recruit chromatin modifiersand facilitate RNPII loading (76), which could explain why HSF1localized to 25 target gene promoters in heat-treated mitoticcells without inducing transcription (Fig. 5; Fig. S4D; DatasetS1). The unbiased analyses by ChIP-seq identified a broad rangeof mitotic target genes for HSF2 both under optimal growth andstress conditions (Fig. 5A; Datasets S1 and S2). One prominentunique HSF2 target gene in mitosis is MLL, a trithorax homologand a transcriptional coactivator that has been found to interactwith promoters in mitotic human cells, marking the genes forearly activation in G1 (77, 78). We discovered that HSF2 con-trols the cell cycle phase–dependent expression of MLL, which isprerequisite for an efficient induction of MLL after mitotic si-lencing (Fig. 6A). Moreover, recent computational analysesfound MLL target promoters to be enriched with HSEs (77),arguing for either cooperation or competition of MLL and HSFson the genome. Several studies have addressed how the memoryof cell fate is maintained during mitosis when most of the tran-scriptional regulators are erased from the chromatin (33). To thisend, the capacity of HSF2 to bind mitosis-specific target genesboth under optimal growth conditions and on stress (Figs. 1B and5B; Dataset S1) indicates that HSF2 directs gene expressionthroughout cell cycle progression.Genomewide analyses in high resolution enable broad views

on the interplay of transcriptional regulators and chromatin statein different cell types, cell cycle phases, and in response to variousstimuli. In this study, we characterized the HSF1- and HSF2-driven transcriptional response to acute stress in cycling and mi-totic cells. Our results revealed the molecular mechanisms thatmaintain cellular homeostasis in cycling cells and identifieda dramatically impaired capacity of mitotic cells to provoke tran-scriptional responses, even when challenged with proteotoxicity.We characterized the complete set of chaperones, cochaperones,and ubiquitin genes that is directed by HSFs under acute stressand highlighted the various processes that are transcriptionallyreprogrammed in proteotoxic conditions. We uncovered thestrikingly distinct mechanisms for HSF1 and HSF2 in orchestrat-ing transcription and discovered HSF2 as an epigenetic regulatorthat controls transcription throughout cell cycle progression. Fu-ture studies, expanding beyond acute heat shock, will establishwhether cells are able to epigenetically remember past environ-mental exposures, such as stress.

Materials and MethodsCell Culture, Synchronization of Cells to Mitosis, and Heat Treatments. HumanK562 erythroleukemia cells were maintained at 37 °C in a humidified 5% CO2

atmosphere and cultured in RPMI medium (Sigma) containing 10% (vol/vol)FCS, 2 mM L-glutamate, and streptomycin/penicillin. Cells were arrested tomitosis using thymidine and nocodazole (39). Briefly, after 16 h in 2 mMthymidine (Sigma), the cells were washed with PBS to allow cell cycle pro-gression for 8 h. The cells were collected in S-phase by a second thymidine

Vihervaara et al. PNAS | Published online August 19, 2013 | E3395

CELL

BIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

May

4, 2

020

treatment for 24 h. After cell cycle progression for 5 h, the cells werearrested in mitosis with nocodazole (100 ng/mL; Fluka) for 12 h. The noco-dazole was removed, and the cells were harvested, allowed to proceed thecell cycle at 37 °C, or heat treated. The heat shock treatments were con-ducted in a water bath at 42 °C. The DNA content of the cells was de-termined by propidium iodide (PI) staining (40 μg/mL; Sigma) andfluorescence-mediated counting by FACSCalibur (BD Biosciences). The FACShistograms were generated using Cell Quest Pro-6.0 (BD Biosciences) andFlowing Software 2.5 (Turku Centre for Biotechnology). The error bars instatistical analyses indicate SDs.

ChIP. HSF1- and HSF2-bound DNA fragments were isolated by ChIP usinga previously described protocol (11, 17) and the following ChIP-verifiedantibodies: HSF1 (Spa-901; Enzo) and HSF2 (17). IgG (sc-2027; Santa Cruz)was used as a negative antibody and AcH4 (06-866; Upstate) was used asa positive antibody. PCR primers are listed in Table S2.

High-Throughput Sequencing and Data Analysis. For sequencing, 10 ChIPreplicates per sample were collected. Sequencing libraries were generatedusing New England BioLabs NEBNext kits, and adapters and primers werefrom a Bioo Scientific AIR DNA Barcodes kit. Template amplification andcluster generation were performed using the cBot and Truseq SR Cluster kitfor cBot v, and 36 nucleotides were sequenced with Illumina Genome Ana-lyzer IIx using v5 TruSeq SBS sequencing kits. After quality trim and removal ofduplicates, the reads were mapped to the human genome (hg19) with Bowtie(79). The peaks were called with MACS 1.4 (80), and a minimum fold en-richment of five times over input was set as a cutoff criterion for target sites.Any site that exceeded the cutoff in the negative IgG sample was discarded.

Identification of HSF1- and HSF2-Targeted DNA Sequences and GenomicRegions and Functional Annotation of Target Genes. The consensus DNAsequences for HSF1 and HSF2 were identified by motif analysis of large DNAdatasets (MEME-ChIP) (81) using a 120-bp region centered on the summitpoint. HSF1 and HSF2 target sites were annotated to genomic regions usingexon and intron coordinates provided by RefSeq and by defining a corepromoter to span from −1,200 to +300 bp from the transcriptional start site(TSS). Fifty percent of peak length was centered on the summit point, andpeaks that fell on exon-intron boundaries are indicated as exons. Biologicalprocesses associated with HSF1 or HSF2 target genes in cycling and mitoticcells were analyzed with DAVID (53), which uses Fisher’s exact test for cal-culation of the P value for enriched gene ontology terms. The density signalsof HSF1, HSF2, and RNPII on the human genome were visualized with theIntegrative Genomics Viewer (82).

Depletion of HSF1 and HSF2 with RNAi by shRNA. HSF1 and HSF2 were de-pleted from the cells using shRNA constructs ligated into pSUPER vectors(Oligoengine) as previously described (17). The vector-encoded oligonu-cleotides were specific for HSF1 (GCT CAT TCA GTT CCT GAT C) or HSF2 (CAGGCG AGT ACA ACA GCA T). A scrambled sequence (GCG CGC TTT GTA GGATTC G) was used as a control. The shRNA constructs were transfected intocells by electroporation (970 μF, 220 mV), and the cells were incubated for72 h prior to harvesting. Synchronization of cells to mitosis was initiatedafter a 7-h recovery from the transfection, allowing for simultaneous samplepreparation of the cycling and mitotic cells.

Quantitative Real-Time RT-PCR. RNA was isolated using the RNeasy kit (Qia-gen), and 1 μg of total RNA was DNaseI treated (Promega) and reversetranscribed with Moloney murine leukemia virus reverse transcriptase RNaseH(–) (Promega). Real-time RT-PCR reactions were prepared and run as de-

scribed earlier (17) using ABI Prism 7900 (Applied Biosystems). The primerswere purchased from Oligomer (Helsinki) and the probes were purchasedfrom Roche Applied Science. Primer and probe sequences are listed in TableS3. Relative quantities of the target gene mRNAs were normalized toGAPDH, and the fold inductions were calculated against the respectivemRNA level in nontreated cells. All reactions were made in triplicate forsamples derived from at least three biological replicates. SDs were calculatedand are shown in the graphs. An independent two-tailed Student t test wasused to determine the P value when comparing mRNA levels betweenscrambled-transfected and shHSF1- or shHSF2-transfected cells.

Western Blotting. Cells were lysedwith buffer C [25% (vol/vol) glycerol, 20mMHepes, pH 7.4, 1.5 mM MgCl2, 0.42 M NaCl, 0.2 mM EDTA, 0.5 mM PMSF, and0.5 mM DTT], and the protein concentration in the soluble fraction wasmeasured using Bradford analysis. Twenty micrograms of total protein wasboiled in Laemmli sample buffer and subjected to SDS/PAGE followed by trans-fer to nitrocellulose membrane (Protran nitrocellulose; Schleicher & Schuell).Proteins were analyzed with primary antibodies against HSF1 (Spa-901; Enzo),HSF2 (3E2; Upstate), and β-TUBULIN (ab6046; Abcam). The secondary anti-bodies were HRP conjugated (GE Healthcare), and the blots were developedusing an enhanced chemiluminescence method (ECL kit; GE Healthcare).

Oligonucleotide-Mediated Pull-Down Assay. The oligonucleotide-mediatedpull-down assay was performed as described previously (83) using bio-tinylated oligonucleotides (Oligomer). The double-stranded HSE containedthe sequence 5′-biotin-TCG ACT AGA AGC TTC TAG AAG CTT CTA G-3′ andthe scrambled control 5′-biotin-AAC GAC GGT CGC TCC GCC TGG CT-3′.Buffer C extracts of 100–400 μg total protein were annealed to oligonucle-otide (0.5 μM) in binding buffer [20 mM Tris·HCl, pH 7.5, 100 mM NaCl, 2 mMEDTA, and 10% (vol/vol) glycerol] containing salmon sperm DNA (0.5 μg/μL).The samples were precleared, and the oligonucleotides were precipitatedwith a 50% (vol/vol) slurry of UltraLink streptavidin beads (Pierce). Boundfractions were washed three times with binding buffer containing 0.1%Triton X-100. DNA-bound proteins were eluted with Laemmli buffer anddetected by SDS/PAGE and Western blotting.

Immunofluorescence of HSF1 and HSF2 in Dividing Male Spermatocytes. Testesof 60- to 80-d-old C57BL/6N mice were isolated, fixed in 4% (wt/vol) para-formaldehyde, and sectioned to 4 μm. All mice were handled according tothe institutional animal care policies of the Åbo Akademi University (Turku,Finland). Confocal immunofluorescence analyses were performed as pre-viously described (11) using primary antibodies against HSF1 and HSF2 (20)and secondary antibodies conjugated to Alexa 488 or Alexa 568 (Invitrogen).The DNA was stained with Hoechst 33342 (H-1399; Molecular Probes).Images for all channels were sequentially captured from a single confocalsection using a Zeiss Meta510 confocal microscope. For analyzing the meioticdivisions, stage XII of the mouse seminiferous cycle was selected, and di-viding spermatocytes in meiotic division I were imaged. The channels weremerged using ImageJ (84).

ACKNOWLEDGMENTS. We thank Michael J. Guertin for insightful discus-sions and for sharing his expertise in bioinformatics. The FunctionalGenomics Unit (University of Helsinki) is acknowledged for performing thehigh-throughput sequencing, Jukka Lehtonen for expert assistance incomputational analyses, and Diana M. Toivola, Johanna K. Björk, and EvaHenriksson for constructive advice during manuscript preparation. This workwas financially supported by the Academy of Finland, the Sigrid JuséliusFoundation, the Finnish Cancer Organizations, Åbo Akademi UniversityFoundation, the Tor, Joe, and Pentti Borg Foundation, BioCenter Finland,and the Turku Doctoral Program of Biomedical Sciences (A.V. and A.N.E.).

1. Kourtis N, Tavernarakis N (2011) Cellular stress response pathways and ageing:

Intricate molecular relationships. EMBO J 30(13):2520–2531.2. Richter K, Haslbeck M, Buchner J (2010) The heat shock response: Life on the verge of

death. Mol Cell 40(2):253–266.3. Hartl FU, Bracher A, Hayer-Hartl M (2011) Molecular chaperones in protein folding

and proteostasis. Nature 475(7356):324–332.4. Åkerfelt M, Morimoto RI, Sistonen L (2010) Heat shock factors: Integrators of cell

stress, development and lifespan. Nat Rev Mol Cell Biol 11(8):545–555.5. Dai C, Whitesell L, Rogers AB, Lindquist S (2007) Heat shock factor 1 is a powerful

multifaceted modifier of carcinogenesis. Cell 130(6):1005–1018.6. Fujimoto M, et al. (2005) Active HSF1 significantly suppresses polyglutamine

aggregate formation in cellular and mouse models. J Biol Chem 280(41):34908–34916.7. McMillan DR, Xiao X, Shao L, Graves K, Benjamin IJ (1998) Targeted disruption of heat

shock transcription factor 1 abolishes thermotolerance and protection against heat-

inducible apoptosis. J Biol Chem 273(13):7523–7528.

8. Mendillo ML, et al. (2012) HSF1 drives a transcriptional program distinct from heat

shock to support highly malignant human cancers. Cell 150(3):549–562.9. Santagata S, et al. (2011) High levels of nuclear heat-shock factor 1 (HSF1) are

associated with poor prognosis in breast cancer. Proc Natl Acad Sci USA 108(45):

18378–18383.10. Åkerfelt M, et al. (2008) Promoter ChIP-chip analysis in mouse testis reveals Y

chromosome occupancy by HSF2. Proc Natl Acad Sci USA 105(32):11224–11229.11. Åkerfelt M, et al. (2010) Heat shock transcription factor 1 localizes to sex chromatin

during meiotic repression. J Biol Chem 285(45):34469–34476.12. Chang Y, et al. (2006) Role of heat-shock factor 2 in cerebral cortex formation and

as a regulator of p35 expression. Genes Dev 20(7):836–847.13. Fujimoto M, et al. (2004) HSF4 is required for normal cell growth and differentiation

during mouse lens development. EMBO J 23(21):4297–4306.14. Takaki E, et al. (2006) Maintenance of olfactory neurogenesis requires HSF1, a major

heat shock transcription factor in mice. J Biol Chem 281(8):4931–4937.

E3396 | www.pnas.org/cgi/doi/10.1073/pnas.1305275110 Vihervaara et al.

Dow

nloa

ded

by g

uest

on

May

4, 2

020

15. Loison F, et al. (2006) Up-regulation of the clusterin gene after proteotoxic stress:Implication of HSF1-HSF2 heterocomplexes. Biochem J 395(1):223–231.

16. Sandqvist A, et al. (2009) Heterotrimerization of heat-shock factors 1 and 2 providesa transcriptional switch in response to distinct stimuli. Mol Biol Cell 20(5):1340–1347.

17. Östling P, Björk JK, Roos-Mattjus P, Mezger V, Sistonen L (2007) Heat shock factor2 (HSF2) contributes to inducible expression of hsp genes through interplay withHSF1. J Biol Chem 282(10):7077–7086.

18. Shinkawa T, et al. (2011) Heat shock factor 2 is required for maintaining proteostasisagainst febrile-range thermal stress and polyglutamine aggregation. Mol Biol Cell22(19):3571–3583.

19. Kroeger PE, Sarge KD, Morimoto RI (1993) Mouse heat shock transcription factors 1and 2 prefer a trimeric binding site but interact differently with the HSP70 heat shockelement. Mol Cell Biol 13(6):3370–3383.

20. Sarge KD, Murphy SP, Morimoto RI (1993) Activation of heat shock gene transcriptionby heat shock factor 1 involves oligomerization, acquisition of DNA-binding activity,and nuclear localization and can occur in the absence of stress. Mol Cell Biol 13(3):1392–1407.

21. Yoshima T, Yura T, Yanagi H (1998) Heat shock factor 1 mediates hemin-inducedhsp70 gene transcription in K562 erythroleukemia cells. J Biol Chem 273(39):25466–25471.

22. Anckar J, Sistonen L (2011) Regulation of HSF1 function in the heat stress response:Implications in aging and disease. Annu Rev Biochem 80:1089–1115.

23. Björk JK, Sistonen L (2010) Regulation of the members of the mammalian heat shockfactor family. FEBS J 277(20):4126–4139.

24. Guertin MJ, Petesch SJ, Zobeck KL, Min IM, Lis JT (2010) Drosophila heat shock systemas a general model to investigate transcriptional regulation. Cold Spring Harb SympQuant Biol 75:1–9.

25. Ahlskog JK, et al. (2010) Anaphase-promoting complex/cyclosome participates in theacute response to protein-damaging stress. Mol Cell Biol 30(24):5608–5620.

26. Jedlicka P, Mortin MA, Wu C (1997) Multiple functions of Drosophila heat shocktranscription factor in vivo. EMBO J 16(9):2452–2462.

27. Wu C (1984) Activating protein factor binds in vitro to upstream control sequences inheat shock gene chromatin. Nature 311(5981):81–84.

28. Yao J, Munson KM, WebbWW, Lis JT (2006) Dynamics of heat shock factor associationwith native gene loci in living cells. Nature 442(7106):1050–1053.

29. Alabert C, Groth A (2012) Chromatin replication and epigenome maintenance. NatRev Mol Cell Biol 13(3):153–167.

30. Delcuve GP, He S, Davie JR (2008) Mitotic partitioning of transcription factors. J CellBiochem 105(1):1–8.

31. Fisher D, Méchali M (2003) Vertebrate HoxB gene expression requires DNAreplication. EMBO J 22(14):3737–3748.

32. Taylor JH (1960) Nucleic acid synthesis in relation to the cell division cycle. Ann N YAcad Sci 90:409–421.

33. Probst AV, Dunleavy E, Almouzni G (2009) Epigenetic inheritance during the cell cycle.Nat Rev Mol Cell Biol 10(3):192–206.

34. Johnson DS, Mortazavi A, Myers RM, Wold B (2007) Genome-wide mapping of in vivoprotein-DNA interactions. Science 316(5830):1497–1502.

35. Park PJ (2009) ChIP-seq: Advantages and challenges of a maturing technology. NatRev Genet 10(10):669–680.

36. Pepke S, Wold B, Mortazavi A (2009) Computation for ChIP-seq and RNA-seq studies.Nat Methods 6(11, Suppl):S22–S32.

37. Sistonen L, Sarge KD, Phillips B, Abravaya K, Morimoto RI (1992) Activation of heatshock factor 2 during hemin-induced differentiation of human erythroleukemia cells.Mol Cell Biol 12(9):4104–4111.

38. Consortium EP; ENCODE Project Consortium (2011) A user’s guide to the encyclopediaof DNA elements (ENCODE). PLoS Biol 9(4):e1001046.

39. Whitfield ML, et al. (2000) Stem-loop binding protein, the protein that binds the3′ end of histone mRNA, is cell cycle regulated by both translational andposttranslational mechanisms. Mol Cell Biol 20(12):4188–4198.

40. Rasmussen EB, Lis JT (1993) In vivo transcriptional pausing and cap formation on threeDrosophila heat shock genes. Proc Natl Acad Sci USA 90(17):7923–7927.

41. Rougvie AE, Lis JT (1988) The RNA polymerase II molecule at the 5′ end of theuninduced hsp70 gene of D. melanogaster is transcriptionally engaged. Cell 54(6):795–804.

42. Adelman K, Lis JT (2012) Promoter-proximal pausing of RNA polymerase II: Emergingroles in metazoans. Nat Rev Genet 13(10):720–731.

43. Core LJ, et al. (2012) Defining the status of RNA polymerase at promoters. Cell Rep2(4):1025–1035.

44. Kubota H, Matsumoto S, Yokota S, Yanagi H, Yura T (1999) Transcriptional activationof mouse cytosolic chaperonin CCT subunit genes by heat shock factors HSF1 andHSF2. FEBS Lett 461(1-2):125–129.

45. Hageman J, et al. (2010) A DNAJB chaperone subfamily with HDAC-dependentactivities suppresses toxic protein aggregation. Mol Cell 37(3):355–369.

46. Rampelt H, et al. (2012) Metazoan Hsp70 machines use Hsp110 to power proteindisaggregation. EMBO J 31(21):4221–4235.

47. Kampinga HH, Craig EA (2010) The HSP70 chaperone machinery: J proteins as driversof functional specificity. Nat Rev Mol Cell Biol 11(8):579–592.

48. Li J, Soroka J, Buchner J (2012) The Hsp90 chaperone machinery: Conformationaldynamics and regulation by co-chaperones. Biochim Biophys Acta 1823(3):624–635.

49. Retzlaff M, et al. (2010) Asymmetric activation of the hsp90 dimer by its cochaperoneaha1. Mol Cell 37(3):344–354.

50. Ali MM, et al. (2006) Crystal structure of an Hsp90-nucleotide-p23/Sba1 closedchaperone complex. Nature 440(7087):1013–1017.

51. Taipale M, et al. (2012) Quantitative analysis of HSP90-client interactions revealsprinciples of substrate recognition. Cell 150(5):987–1001.

52. Kimura Y, Tanaka K (2010) Regulatory mechanisms involved in the control ofubiquitin homeostasis. J Biochem 147(6):793–798.

53. Dennis G, Jr., et al. (2003) DAVID: Database for Annotation, Visualization, andIntegrated Discovery. Genome Biol 4(5):3.

54. Hut HM, Kampinga HH, Sibon OC (2005) Hsp70 protects mitotic cells against heat-induced centrosome damage and division abnormalities. Mol Biol Cell 16(8):3776–3785.

55. Martínez-Balbás MA, Dey A, Rabindran SK, Ozato K, Wu C (1995) Displacement ofsequence-specific transcription factors from mitotic chromatin. Cell 83(1):29–38.

56. Liu H, Cheng EH, Hsieh JJ (2007) Bimodal degradation of MLL by SCFSkp2 and APCCdc20assures cell cycle execution: A critical regulatory circuit lost in leukemogenic MLL fusions.Genes Dev 21(19):2385–2398.

57. Keyse SM, Emslie EA (1992) Oxidative stress and heat shock induce a human geneencoding a protein-tyrosine phosphatase. Nature 359(6396):644–647.

58. Sarge KD, Bray AE, Goodson ML (1995) Altered stress response in testis. Nature374(6518):126.

59. Vydra N, et al. (2006) Spermatocyte-specific expression of constitutively active heatshock factor 1 induces HSP70i-resistant apoptosis in male germ cells. Cell Death Differ13(2):212–222.

60. Widlak W, et al. (2007) Inducible 70 kDa heat shock protein does not protectspermatogenic cells from damage induced by cryptorchidism. Int J Androl 30(2):80–87.

61. Kallio M, et al. (2002) Brain abnormalities, defective meiotic chromosome synapsisand female subfertility in HSF2 null mice. EMBO J 21(11):2591–2601.

62. Lee HS, Simon JA, Lis JT (1988) Structure and expression of ubiquitin genes ofDrosophila melanogaster. Mol Cell Biol 8(11):4727–4735.

63. Gegonne A, et al. (2008) TFIID component TAF7 functionally interacts with both TFIIHand P-TEFb. Proc Natl Acad Sci USA 105(14):5367–5372.

64. Gegonne A, Weissman JD, Zhou M, Brady JN, Singer DS (2006) TAF7: A possibletranscription initiation check-point regulator. Proc Natl Acad Sci USA 103(3):602–607.

65. Aumais JP, et al. (2003) Role for NudC, a dynein-associated nuclear movementprotein, in mitosis and cytokinesis. J Cell Sci 116(Pt 10):1991–2003.

66. Bradley CM, Ronning DR, Ghirlando R, Craigie R, Dyda F (2005) Structural basis forDNA bridging by barrier-to-autointegration factor. Nat Struct Mol Biol 12(10):935–936.

67. Haraguchi T, et al. (2008) Live cell imaging and electron microscopy reveal dynamicprocesses of BAF-directed nuclear envelope assembly. J Cell Sci 121(Pt 15):2540–2554.

68. Nishino M, et al. (2006) NudC is required for Plk1 targeting to the kinetochore andchromosome congression. Curr Biol 16(14):1414–1421.

69. Hahn JS, Hu Z, Thiele DJ, Iyer VR (2004) Genome-wide analysis of the biology of stressresponses through heat shock transcription factor. Mol Cell Biol 24(12):5249–5256.

70. Gonsalves SE, Moses AM, Razak Z, Robert F, Westwood JT (2011) Whole-genomeanalysis reveals that active heat shock factor binding sites are mostly associated withnon-heat shock genes in Drosophila melanogaster. PLoS ONE 6(1):e15934.

71. Guertin MJ, Lis JT (2010) Chromatin landscape dictates HSF binding to target DNAelements. PLoS Genet 6(9):e1001114.

72. Trinklein ND, Murray JI, Hartman SJ, Botstein D, Myers RM (2004) The role of heatshock transcription factor 1 in the genome-wide regulation of the mammalian heatshock response. Mol Biol Cell 15(3):1254–1261.

73. Solimini NL, Luo J, Elledge SJ (2007) Non-oncogene addiction and the stressphenotype of cancer cells. Cell 130(6):986–988.

74. Wilkerson DC, Skaggs HS, Sarge KD (2007) HSF2 binds to the Hsp90, Hsp27, and c-Fospromoters constitutively and modulates their expression. Cell Stress Chap 12(3):283–290.

75. Xing H, et al. (2005) Mechanism of hsp70i gene bookmarking. Science 307(5708):421–423.

76. Fujimoto M, et al. (2012) RPA assists HSF1 access to nucleosomal DNA by recruitinghistone chaperone FACT. Mol Cell 48(2):182–194.

77. Blobel GA, et al. (2009) A reconfigured pattern of MLL occupancy within mitoticchromatin promotes rapid transcriptional reactivation following mitotic exit.Mol Cell36(6):970–983.

78. Tyagi S, Chabes AL, Wysocka J, Herr W (2007) E2F activation of S phase promoters viaassociation with HCF-1 and the MLL family of histone H3K4 methyltransferases. MolCell 27(1):107–119.

79. Langmead B, Trapnell C, Pop M, Salzberg SL (2009) Ultrafast and memory-efficientalignment of short DNA sequences to the human genome. Genome Biol 10(3):R25.

80. Feng J, Liu T, Qin B, Zhang Y, Liu XS (2012) Identifying ChIP-seq enrichment usingMACS. Nat Protoc 7(9):1728–1740.

81. Machanick P, Bailey TL (2011) MEME-ChIP: Motif analysis of large DNA datasets.Bioinformatics 27(12):1696–1697.

82. Robinson JT, et al. (2011) Integrative genomics viewer. Nat Biotechnol 29(1):24–26.83. Anckar J, et al. (2006) Inhibition of DNA binding by differential sumoylation of heat

shock factors. Mol Cell Biol 26(3):955–964.84. Schneider CA, Rasband WS, Eliceiri KW (2012) NIH Image to ImageJ: 25 years of image

analysis. Nat Methods 9(7):671–675.85. Kampinga HH, et al. (2009) Guidelines for the nomenclature of the human heat shock

proteins. Cell Stress Chap 14(1):105–111.

Vihervaara et al. PNAS | Published online August 19, 2013 | E3397

CELL

BIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

May

4, 2

020