Formation of nuclear stress granules involves HSF2 and coincides with the nucleolar localization of Hsp70

Cells have developed a range of processes to increase survival and adaptation in response to different forms of stress. One of these events is the induced expression of heat-shock genes coding for heat-shock proteins(Hsps). Hsps have an important role in cellular homeostasis and protection through their well-recognized properties as molecular chaperones(Bukau and Horwich, 1998). The expression of the heat-shock genes is primarily regulated at transcriptional level by specific DNA-binding proteins called heat-shock factors (HSFs), which bind to the heat-shock promoter element (HSE)(Pirkkala et al., 2001). Four members of this family, HSF1-HSF4, have been identified in vertebrates(Rabindran et al., 1991; Sarge et al., 1991; Schuetz et al., 1991; Nakai and Morimoto, 1993; Nakai et al., 1997), which has raised questions about specific or overlapping functions among the distinct HSFs.

HSF1 is the most studied heat-shock transcription factor and is the classical HSF, which responds to elevated temperatures and other forms of protein damaging stress (Pirkkala et al.,2001). In addition, HSF1 has been shown to be involved in extra-embryonic development and female fertility in mice(Xiao et al., 1999; Christians et al., 2000). Upon stress, HSF1 is rapidly converted from a monomer to a trimer, is inducibly phosphorylated and concentrated in the nucleus to activate heat-shock gene transcription (Baler et al.,1993; Rabindran et al.,1993; Sarge et al.,1993). In addition to HSF1, chicken HSF3 is activated by similar but more severe stressors (Nakai and Morimoto, 1993; Tanabe et al.,1997). In HSF3-deficient cells the formation of HSF1 trimers is hampered and Hsp expression is reduced upon heat shock(Tanabe et al., 1998). The interdependency between HSF1 and HSF3 in avian cells is so far the only example of cross-talk between different HSFs. However, no physical interaction has been reported between these two proteins.

A characteristic feature of cellular stress in human cells is the organization of HSF1 into specific subnuclear structures, termed stress granules. These irregularly shaped granules have been described to form under various stress conditions, including exposure to heat, cadmium, azetidine and proteasome inhibitors (Cotto et al.,1997; Jolly et al.,1997; Jolly et al.,1999; Holmberg et al.,2000). Stress-induced HSF1 granules have been found in all investigated primary and transformed human cells but not in rodent cells(Sarge et al., 1993; Mivechi et al., 1994; Cotto et al., 1997; Jolly et al., 1999). The appearance of stress granules correlates positively with the inducible phosphorylation and transcriptional activity of HSF1(Sarge et al., 1993; Cotto et al., 1997; Jolly et al., 1999; Holmberg et al., 2000). Furthermore, HSF1 dissociates from the stress granules and relocalizes diffusely in the cell during attenuation and recovery from stress; upon subsequent exposure to stress, the granules reappear at the same sites(Jolly et al., 1999). The stress granules have not been shown to represent other previously described nuclear structures (Cotto et al.,1997). In addition to HSF1, several RNA binding proteins such as heterogeneous nuclear ribonucleoprotein (hnRNP) HAP (hnRNP A1 interacting protein), hnRNP M, Src-activated during mitosis (Sam68) and certain SR(serinearginine) splicing factors have been identified in stress granules(Weighardt et al., 1999; Denegri et al., 2001). Recently, stress granules were revealed to associate with human chromosome 9q11-q12, corresponding to a large block of heterochromatin composed primarily of satellite III repeats adjacent to the centromere(Jolly et al., 2002). Furthermore, Denegri et al. (Denegri et al., 2002) have reported that, in addition to chromosome 9,chromosomes 12 and 15 also contain nucleation sites for stress granules. The functional significance of stress granules is still unknown.

The autoregulation of the heat-shock response is facilitated by a direct interaction between HSF1 and molecular chaperones, such as Hsp70 and Hsp90 with their cochaperones. This has been established in several biochemical,genetic and cell physiological studies(Abravaya et al., 1992; Baler et al., 1992; Kim et al., 1995; Ali et al., 1998; Shi et al., 1998; Zou et al., 1998; Bharadwaj et al., 1999; Morimoto, 2002). Under normal physiological conditions, HSF1 exists in a repressed state associated with molecular chaperones. Upon stress, it is rapidly released from the complex,trimerized and hyperphosphorylated, and acquires DNA-binding and transcriptional activity. When exposed to continuous moderate heat stress, the equilibrium is shifted back to the monomeric, dephosphorylated and Hsp-bound state. The downregulation of HSF1 activity during prolonged stress is referred to as attenuation of the heat-shock response, and leads to the suppression of HSF1 reactivation capacity upon subsequent stress. A similar repression of HSF1 activity is observed when cells recover from stressful conditions. This acquired adaptation of cells to stress stimuli is called thermotolerance(Morimoto, 2002).

Biochemical characterization of HSF2 has revealed that, unlike HSF1, which undergoes a monomer-to-trimer transition, HSF2 is mainly converted from a dimer to a trimer upon activation, and regulation of HSF2 appears not to include phosphorylation (Sarge et al.,1993; Sistonen et al.,1994). The regulation of HSF1 and HSF2 expression varies dramatically. In contrast to stably and constitutively expressed HSF1, the HSF2 expression levels are regulated both transcriptionally and by mRNA stabilization (Pirkkala et al.,1999). Rapid accumulation of HSF2 protein by downregulation of the ubiquitin proteolytic pathway has provided evidence that HSF2 is a labile protein (Mathew et al., 1998; Pirkkala et al., 2000). The regulation of HSF2 has been linked to certain development- and differentiation-related processes, such as gametogenesis and pre- and postimplantation development of mouse embryos(Mezger et al., 1994; Sarge et al., 1994; Rallu et al., 1997; Alastalo et al., 1998; Eriksson et al., 2000). Furthermore, the human HSF2 DNA-binding activity is induced during hemin-mediated erythroid differentiation of K562 cells(Sistonen et al., 1992). A recent study by Kallio et al. (Kallio et al., 2002) revealed that the disruption of mouse hsf2gene leads to brain abnormalities and defects in gametogenesis in both genders. No correlation between HSF2 expression and activity with Hsp expression has been obtained during differentiation-related processes,indicating existence of other HSF2 target genes(Rallu et al., 1997; Alastalo et al., 1998; Kallio et al., 2002).

A few reports have proposed that HSF2 is involved in the regulation of the stress response. According to Sheldon and Kingston(Sheldon and Kingston, 1993),HSF2 localizes in the nucleus and in granule-type structures in HeLa cells exposed to heat stress. Furthermore, Mathew et al.(Mathew et al., 2001) have reported that, in murine fibroblasts, HSF2 has the biochemical properties of a temperature-sensitive protein because, upon heat shock, HSF2 is localized to the perinuclear region, its solubility is decreased and DNA-binding activity is diminished. This intriguing difference between human and murine cells has also been observed with HSF1, because the stress-induced subnuclear compartmentalization of HSF1 has been detected only in human cells(Sarge et al., 1993; Denegri et al., 2002). Altogether, direct evidence of HSF2, human or murine, being a physiological transcriptional regulator of heat shock genes is missing.

Despite the characterization of several biochemical properties of HSF1 and HSF2, the actual stress sensors and the signaling pathways regulating these two factors have remained obscure. In this study, we provide evidence that HSF1 and HSF2 form a physically interacting complex and co-localize in the dynamically regulated nuclear stress granules, suggesting a novel role for HSF2. We also show data suggesting that the association of HSF1 and HSF2 with stress granules might be regulated by Hsp70, and that the association is severely impaired in thermotolerant cells. Taken together, our results suggest that the regulation of the subcellular distribution of Hsp70 contributes to the regulation of HSF1-mediated heat-shock responses.

Human K562 erythroleukemia cells and HeLa cervical carcinoma cells were maintained in RPMI-1640 medium and Dulbecco's modified Eagle's medium,respectively, supplemented with 10% fetal calf serum (FCS), 2 mM L-glutamine and antibiotics (penicillin and streptomycin) in a humidified 5%CO₂ atmosphere at 37°C. For experimental treatments, K562 cells were seeded at 5×10⁶ cells and HeLa cells at 3×10⁶ per 10-cm-diameter plate. Heat shocks were performed at 42°C in a water bath. Hemin (Aldrich) was added to a final concentration of 40 μM and cells were incubated at 37°C for the indicated time periods. Recovery periods were at 37°C.

Human HSF1 with C-terminal Myc tags was constructed by PCR and cloned into EcoRI and HindIII sites in the pcDNA3.1(-)MycHis A vector(Invitrogen) in frame with the MycHis tag(Holmberg et al., 2001). Mouse HSF1 with N-terminal Flag tags was a kind gift from R. I. Morimoto(Northwestern University, IL, USA). Mouse HSF2-α and HSF2-β with N-terminal Flag tags have been described previously(Pirkkala et al., 2000). Expression vectors encoding mouse HSF2-α and HSF2-β with C-terminal Myc tags were constructed by PCR and cloned into the EcoRI site in the pcDNA3.1(-) MycHis B vector (Invitrogen) in frame with the MycHis tag. The mouse HSF2-β deletion mutants (Fig. 4A) were constructed by PCR and cloned into the EcoRI and EcoRV sites in frame with the N-terminal Flag tag in pFLAG-CMV-2(Kodak). All PCR-amplified products were sequenced to exclude the possibility of second site mutagenesis. HSF1-Myc-His stably overexpressing cell line (4H9)was generated by electroporating 30 μg of HSF1-Myc-His to K562 cells as described below. After 2 days of recovery, neomycin-resistant cells were selected by growing the cells in medium containing 500 μg ml^-1G418 (Life Technologies) for 2 weeks. Neomycin-resistant single cell clones were picked out after serial dilutions and screened for HSF1-Myc-His expression by western blotting. Cells were routinely grown in the presence of 500 μg ml^-1 G418.

K562 and HeLa cells were transfected by electroporation (975 μF, 220 V)using a Bio-Rad Gene Pulser electroporator. For this procedure,5×10⁶ cells were washed, resuspended in 0.4 ml of Optimem(Gibco-BRL) and placed in a 0.4-cm-gap electroporation cuvette (BTX). Plasmid DNA (30 μg) was added and, after a brief incubation at room temperature,the cells were subjected to a single electric pulse. Thereafter, cells were cultured at 37°C for 40 hours prior to the indicated experimental treatments.

For immunofluorescence analysis, HeLa cells growing on coverslips were washed with PBS and simultaneously fixed and permeabilized for 15 minutes in 0.5% Tween 20 in PBS containing 3% paraformaldehyde, or the cells were fixed for 15 minutes in 4°C methanol-acetone (1:1). After three washes with PBS,cells were incubated for 1 hour with blocking solution (20% boiled normal goat serum in PBS). Rabbit anti-HSF1 (Holmberg et al., 2000), rat anti-HSF1 (Neomarkers), rabbit anti-HSF2(Sarge et al., 1993), mouse anti-Hsp70 (SPA-810; StressGen), mouse anti-Flag M2 (Sigma) or mouse anti-Myc(Sigma) antibody (1:500 dilutions) were added for 1 hour. After washes with PBS, the bound primary antibodies were detected using goat anti-mouse antibodies (1:400 dilution for 1 hour; Alexa 488, Molecular Probes),Cy3-conjugated donkey anti-rabbit antibodies (1:400 dilution for 1 hour;Jackson ImmunoResearch Laboratories), goat anti-rat antibodies (1:400 dilution for 1 hour; Alexa 568, Molecular Probes) or donkey anti-rabbit antibodies(1:400 dilution for 1 hour; Alexa 488, Molecular Probes). The DNA was stained with DAPI (4′,6-diamidino-2-phenylindol, Sigma) for 2 minutes before final washes and mounting with Vectashield (Vector Laboratories). The cells were analyzed and photographed using a Leica DMR fluorescence microscope equipped with a digital Hamamatsu ORCA CCD camera or Leica TCS40 confocal laser scanning microscope using the SCANware 4.2a program. Images were further processed using Adobe Photoshop and CorelDraw software.

For in vivo coimmunoprecipitation experiments, transiently transfected cells were lysed in 200 μl of lysis buffer [25 mM HEPES, 100 mM NaCl, 5 mM EDTA, 0.5% Triton X-100, 20 mM β-glycerophosphate, 20 mM para-nitrophenyl phosphate, 100 μM ortovanadate, 0.5 mM phenylmethylsulfonyl fluoride, 1 mM dithiotreitol, 1× complete mini protease inhibitor cocktail (Roche Diagnostics)] supplemented with 20 mM N-ethylmaleimide, followed by centrifugation for 25 minutes at 15,000 g at 4°C. After protein extraction, 200-500 μg total cell protein was preincubated with slurry of protein-G/Sepharose (Amersham Pharmacia Biotech) in TEG buffer (20 mM Tris-HCl pH 7.5, 1 mM EDTA, 10%glycerol) containing 150 mM NaCl and 0.1% Triton X-100 for 30 minutes at 4°C followed by a brief centrifugation. The precleared cellular lysate was incubated with anti-HSF1 (Neomarkers), anti-HSF2 (Neomarkers), anti-Flag or anti-Myc antibodies at room temperature for 30 minutes under rotation, after which 40 μl of a 50% slurry of protein-G/Sepharose was added to the reaction mixture and incubated for 12 hours at 4°C under rotation. After centrifugation, the Sepharose beads were washed with supplemented TEG buffer and the immunoprecipitated proteins were run on 8% SDS-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to nitrocellulose filter (Protran Nitrocellulose; Schleicher & Schuell) for immunoblotting. For detection of protein input and expression levels, 12 μg of protein was analyzed by SDS-PAGE as described above. HSF1 was detected by polyclonal antibodies specific to mouse and human HSF1 (Sarge et al., 1993; Holmberg et al.,2000), HSF2 by polyclonal antibodies specific to mouse HSF2(Sarge et al., 1993), the inducible form of Hsp70 by 4g4 (Affinity Bioreagents), Hsc70 by SPA-815(StressGen). Horseradish peroxidase (HRP)-conjugated secondary antibodies were purchased from Promega and Amersham. The blots were developed with an enhanced chemiluminescence method (ECL; Amersham Pharmacia Biotech).

Cells (5×10⁶ each of 4H9 and K562) were labeled with 0.25 mCi ml^-1 of TRAN35S Label (ICN) for 16 hours. Cells were lysed in 1 ml of modified RIPA buffer (50 mM TRIS pH 7.4, 150 mM NaCl, 1% NP-40, 0.1%deoxycholate, 1 mM EDTA, 100 μM orthovanadate, 2 mM NaF) and the immunoprecipitation with anti-c-Myc (Sigma) antibodies was performed as above. Immunoprecipitated proteins were detected from the gels by autoradiography. To identify the bands, 1×10⁹ 4H9 and K562 cells were lysed and immunoprecipitated as above. The gels were silver stained(O'Connell and Stults, 1997)and the bands were identified using standard matrix-assisted laser desorption ionization mass spectrometry.

Whole cell extracts were prepared from experimentally treated cells as previously described (Mosser et al.,1988) and incubated (12 μg protein) with a ³²P-labeled oligonucleotide representing the proximal HSE of the human hsp70 promoter. The protein-DNA complexes were analyzed on a native 4% polyacrylamide gel as described previously(Mosser et al., 1988).

The activation of HSF1 upon stress is associated with a series of biochemical changes, including oligomerization and phosphorylation, which correlate with the subnuclear organization of HSF1 in chromatin-associated stress granules (Pirkkala et al.,2001; Jolly et al.,2002). Upon exposure to heat shock, we found that another member of the HSF family, HSF2, also formed brightly staining nuclear granules similar to HSF1, as detected by immunofluorescence microscopy(Fig. 1A). Interestingly, the morphology and organization pattern of both HSF1 and HSF2 granules varied dramatically depending on duration of heat stress(Fig. 1B). Both endogenous HSF1 and HSF2 localized in small and barely detectable granules after only 5 minutes of heat shock (data not shown), and the intensity and size increased until 1 hour (Fig. 1B). At 2 hours of treatment, the granules began to mature into ring-like structures. These structures were most clearly detected at 2-6 hours of continuous heat shock, corresponding to the attenuation of the heat-shock response(Fig. 1B). At the same time as the staining experiment, samples were collected for gel mobility shift and western blotting analyses to compare the morphology of HSF1 and HSF2 granules to the DNA-binding activity and phosphorylation of HSF1. During 2-6 hours of heat treatment, the maturation of heat-shock granules into ring-like structures preceded the dissociation of HSF1 and HSF2 from the granules and the loss of HSF1 DNA-binding activity (Fig. 1D). In addition to the morphological analyses, the number of granule-positive cells during the course of heat shock was determined(Fig. 1C). Already at 30 minutes of heat shock, the number of granule-positive cells had increased prominently and, at 1-2 hours, almost 100% of cells contained granules. After 2 hours, the number of granule-positive cells began to decline(Fig. 1C). The dephosphorylation of HSF1 correlated temporally with the dissociation of HSF1 from the granules (Fig. 1B,D). A dramatic change in the number and morphology of both HSF1 and HSF2 granules were detected after a 1 hour recovery from heat stress and, after 3 hours,hardly any granules were found and both HSF1 and HSF2 were diffusely distributed (Fig. 1B,C). Interestingly, the number and morphology of HSF1 and HSF2 granules were almost identical, suggesting that they correspond to the same nuclear structure.

The heat-induced localization of HSF1 and HSF2 in granules did not occur randomly, but a co-localization of these two endogenous factors was detected after a 1-hour heat shock, as shown by double staining(Fig. 2A). To ensure that finding the HSF2 stress granules was not a result of cross-reactivity between the antibodies against HSF1 and HSF2, we constructed both N- and C-terminally Flag- and Myc-tagged HSF1 and HSF2 plasmids. Upon transient expression in HeLa cells, the subcellular localization of ectopic HSF1 and HSF2 was analyzed. Immunofluorescence analyses with monoclonal antibodies against Flag and Myc epitopes showed a similar nuclear pattern(Fig. 2B) to the experiments conducted with the antibodies against HSF1 and HSF2(Fig. 1A,B, Fig. 2A). The co-localization between HSF1 and HSF2 was further confirmed by confocal microscopy (data not shown). Double-staining experiments were also performed as in Fig. 1B to study the co-localization of HSF1 and HSF2 at different phases of granule development. In these studies, co-localization of HSF1 and HSF2 was observed from the compartmentalization to the dissociation of these factors from stress granules(data not shown). It is worth noticing that, although there was prominent co-localization of HSF1 and HSF2 in the same subnuclear structures, some heterogeneity occurred (i.e. stress granules were detected in which HSF1 but not HSF2 was present).

The dynamic localization of HSF1 with HSF2 in stress granules upon heat shock provided new evidence for HSF2 functioning as a heat-responsive factor,and for the stress-induced regulation of HSF1 and HSF2 being closely related. This finding prompted us to investigate whether a similar heat-induced subnuclear distribution of HSF1 and HSF2 could be caused by their physical interaction. Myc-tagged HSF1 was transiently or stably expressed in K562 cells and the endogenous HSF2 was coimmunoprecipitated with both transiently(Fig. 3A, left) and stably(Fig. 3A, right) overexpressed HSF1. Furthermore, monoclonal anti-HSF2 antibodies were used to immunoprecipitate the endogenous HSF2, and western blotting showed that the overexpressed HSF1 coimmunoprecipitated with HSF2(Fig. 3B).

To analyze the dynamics of the interaction during the course of heat-shock response and recovery periods, endogenous HSF1 from HeLa cells was immunoprecipitated with monoclonal anti-HSF1 antibody. The immunoprecipitated complexes were analyzed by western blotting, and dynamic interaction of HSF1 and HSF2-α and -β isoforms(Pirkkala et al., 2001) was observed at different time points (Fig. 3C). At control temperature and heat shock up to 30 minutes,HSF2-β was the major HSF2 isoform in the complex, whereas the amount of HSF2-α began to increase from 30 minutes to 1 hour and similar amounts of both isoforms were detected in the complex. After 1 hour of heat treatment,the levels of HSF2-β in the complex rapidly declined and, at 2-4 hours of continuous heat shock, HSF2-α was the major HSF2 isoform interacting with HSF1. Upon prolonged heat shock (>4 hours) and recovery, the amount of HSF2-β began to increase, restoring the ratio of the HSF2 isoforms observed at control temperature (Fig. 3C, Fig. 8B). The changes in the interaction stoichiometry between HSF1 and HSF2 isoforms, as shown in immunoprecipitation samples (Fig. 3C, top), were reflected also in western blotting of the lysates(Fig. 3C, bottom), indicating altered solubility of the HSF2 isoforms during different phases of the continuous heat shock.

After 3 hours of recovery from a 1-hour heat shock, when cells acquire thermotolerance, the interaction between HSF1 and HSF2 was downregulated(Fig. 3C, Fig. 8B). In contrast to the dynamic regulation of HSF2 isoforms detected both in the lysates and protein complex, the downregulation of HSF1-HSF2 interaction seen after recovery from stress cannot be explained by changes in solubility. This suggests that there are modifications in the HSF-associated protein complex during thermotolerance and further confirms the dynamic nature of HSF1-HSF2 interaction.

Different members of the HSF family share a similar structure, consisting of a leucine-rich oligomerization domain (HR-A/B) adjacent to the N-terminal DNA-binding domain, and another leucine-rich region near the C-terminus (HR-C)(Rabindran et al., 1991; Sarge et al., 1991; Schuetz et al., 1991). To examine whether the interaction of HSF1 and HSF2 depended on an intact oligomerization domain, we constructed a series of Flag-tagged HSF2 deletion mutants (Fig. 4A). The Myc-tagged HSF1 was transiently expressed with these HSF2 mutants in K562 cells, and HSF1 was immunoprecipitated with anti-Myc antibody, and the presence of HSF2 in the complex was analyzed. As shown in Fig. 4B, HSF2 mutants lacking HR-A/B (HSF2 203-517 and HSF2 Δ126-201) did not form complexes with HSF1.

Considering that HSF2 interacts with HSF1, we wanted to investigate whether the HSF1-HSF2 heterocomplex could bind in vitro to the HSE of human hsp70 promoter. Flag-tagged HSF1 and Myc-tagged HSF2 were transiently expressed in K562 cells and a strong heat-induced DNA-binding activity was detected in the lysates (Fig. 5A, left). The spontaneous DNA-binding activity in untreated samples was due to overproduction of HSFs. To analyze the DNA-binding complexes, we performed antibody perturbation assays with anti-Flag and anti-Myc antibodies to detect the presence of HSF1 or HSF2 molecules, respectively. As shown in Fig. 5A (right), the anti-Flag(HSF1) antibody supershifted the heat-induced DNA-binding complexes completely, whereas the anti-Myc (HSF2) antibody had no effect on the complex. Because hemin induces the DNA-binding activity of HSF2(Pirkkala et al., 1999), a hemin-treated sample expressing Myc-tagged HSF2 was included as a positive control for HSF2 DNA-binding activity and anti-Myc antibody. The existence of HSF1-HSF2 complex in the same lysates was determined by coimmunoprecipitation(Fig. 5B).

To exclude the possibility that the C-terminal Myc-epitope in the HSF2 fusion protein was being masked in the heat-induced DNA-binding complex and could therefore not be detected, we repeated the above mentioned experiments using C-terminally Myc-tagged HSF1 and N-terminally Flag-tagged HSF2. The results were identical to those in Fig. 5, with only HSF1 being detected in the HSE-binding complex (data not shown). These results were also confirmed by overexpressing HSF2 alone,and no HSF2 was detected in the heat-induced HSE-binding complex (data not shown).

Based on the obtained results, the HSF1-HSF2 complex formation could have significance in other steps of the stress-regulated signaling pathways unrelated directly to the HSE-binding and heat-shock gene expression. The appearance of stress-induced nuclear granules has been shown to coincide with HSF1 activation (Cotto et al.,1997; Jolly et al.,1999; Holmberg et al.,2000), so we investigated whether the HSF1-HSF2 interaction affects the subnuclear compartmentalization of HSF1. For this purpose, the Flag-tagged HSF2 deletion mutants (Fig. 3A) were transiently transfected into HeLa cells and double staining was performed with anti-Flag and anti-HSF1 antibodies. As shown in Fig. 6A, the full-length HSF2 co-localized with HSF1 during heat shock. Unexpectedly, the DNA-binding-domain deletion mutant (HSF2 108-517), which was able to interact physically with HSF1 (Fig. 4B), did not localize in the granules, and also the translocation of HSF1 into the granules was severely impaired (Figs. 6A,6D). The HR-A/B oligomerization domain deletion mutants (HSF2 203-517 and HSF2Δ126-201) could not interact with HSF1(Fig. 4B) but did not translocate into the granules; instead, these cells displayed normal localization of HSF1 in the granules (Figs. 6B,6D). The C-terminal deletion mutant (HSF2 1-391), which was capable of interacting with HSF1 (Fig. 4B), localized spontaneously to granules and, intriguingly, HSF1 showed clear co-localization with this HSF2 mutant at both control and elevated temperatures (Figs. 6C,6D). Taken together, the immunofluorescence analyses, combined with the coimmunoprecipitation results (Fig. 4B), indicate that HSF2 could affect the translocation of HSF1 into the nuclear stress granules, because either induction or prevention of HSF1 subnuclear compartmentalization was observed depending on which HSF2 mutant was used.

In addition to co-localization of HSF1 and HSF2 in the same dynamically regulated protein complex, we observed profound changes in subnuclear compartmentalization and HSF1-HSF2 interaction during attenuation and recovery of the heat-shock response. To address the question of how thermotolerance affects the localization of HSF1 and HSF2 in stress granules, we induced thermotolerance by exposing HeLa cells to a 1-hour heat shock, after which the cells were allowed to recover at normal temperature for 3 hours. During recovery, HSF1 and HSF2 dissociated from the granules and both proteins were partly translocated into the cytosol (Fig. 7A). After a second heat shock of 30 minutes or 1 hour, a decrease in granule-positive cells was detected, because almost 100% of primarily heat-shocked cells, but less than 15% of thermotolerant cells, exposed to a 1-hour heat shock contained HSF-positive granules(Fig. 7B). Because HSF1 and HSF2 behaved identically, only HSF1 was used as a marker for nuclear stress granules in the following experiments.

One of the mechanisms regulating thermotolerance and attenuation of the heat shock response is the inhibition of HSF1 activity by molecular chaperones such as Hsp70 and Hsp90 and their co-chaperones(Morimoto, 2002). However, the role of these chaperones in the regulation of nuclear stress granules has not been elucidated. Using immunoprecipitation of Myc-tagged HSF1 from stably transfected 4H9 cells after methionine labeling followed by identification of the interacting proteins by mass spectrometry, we found Hsc70 and Hsp70 forming a complex with HSF1 in a stoichiometric manner(Fig. 8A). During attenuation of the heat-shock response and in thermotolerant HeLa cells, dynamically regulated complex formation was detected between HSF1 and Hsp70(Fig. 8B). At the control temperature and upon heat shock of less than 30 minutes, HSF1 was found in a complex with Hsp70, whereas this complex disassembled after 30 minutes and was hardly detectable at 2 hours (Fig. 8B, data not shown). After 2 hours of continuous heat shock, the interaction began to increase. The interaction pattern with Hsp70 correlated well with the onset of attenuation seen previously in HSF1 DNA-binding activity and phosphorylation, as well as with the decline in granule-positive cells (Fig. 1). During recovery periods, complex formation between HSF1 and Hsp70 gradually increased(Fig. 8B).

Next, we investigated how the subcellular localization of Hsp70 corresponded to the interaction pattern seen in Fig. 8B and to the heat-shock granule formation. Hsp70 and its cochaperones such as Hsp40 are rapidly translocated from the cytosol to the nucleoli upon heat shock(Welch and Feramisco, 1984; Welch and Suhan, 1986; Hattori et al., 1993) but the significance of this rapid change in subcellular distribution is poorly understood. The localization of HSF1 and Hsp70 was investigated by double-staining immunofluorescence microscopy and, at normal temperature, HSF1 was found both in the cytosol and nucleus, whereas Hsp70 was mainly cytosolic (Fig. 8C). Upon heat shock of less than 1 hour, the increasing intensity and size of the stress granules coincided with the ongoing translocation of Hsp70 into the nucleoli (data not shown). At 1 hour, HSF1 was primarily detected in mature nuclear granules and Hsp70 concentrated in the nucleoli(Fig. 8C). Interestingly, at this time point, the interaction between HSF1 and Hsp70 was dramatically downregulated (Fig. 8A). During the attenuation of the heat-shock response (2-6 hours of continuous heat shock), both Hsp70 and HSF1 gradually dissociated from nucleoli and the stress granules, respectively, and HSF1-Hsp70 complex formation increased(Fig. 8A,C, data not shown). In thermotolerant cells, HSF1 and Hsp70 formed an interacting complex and were both mainly localized in the cytosol (HS+3R; Fig. 8A,C). Upon exposure of thermotolerant cells to a second heat shock, a high level of interaction persisted and HSF1-positive granules were detected only in cells in which Hsp70 was translocated into nucleoli (HS+3R+HS1h; Fig. 8C). The number of cells containing HSF1 in stress granules and Hsp70-positive nucleoli corresponded to the heat-shocked thermotolerant cells shown in Fig. 7B (data not shown). In these experiments, the nucleolar localization of Hsp70 coincided with the downregulation of HSF1-Hsp70 interaction and localization of HSF1 in nuclear stress granules. The reversible localization pattern of Hsp70 into and out of the nucleoli appeared to be closely related to the subcellular distribution of HSF1, and could thereby contribute to the regulation of HSF1 activity.

The physiological role of HSF1 in the regulation of the heat-shock response is well established but the possible interplay between HSF1 and other HSFs is largely unresolved. Although Sheldon and Kingston(Sheldon and Kingston, 1993)earlier reported that the intracellular localization of human HSF2 is regulated by heat stress, the function of HSF2 has ever since been solely linked to differentiation-related processes. In this study, we provide evidence of human HSF2 being an integral member of the cellular stress response pathway. Specifically, HSF2 is a novel component of the nuclear stress granules, and the localization of HSF2 in the granules coincides temporally with the localization and activation of HSF1. By analyzing both HSF1 and HSF2 separately and together, we detected a clear maturation pattern in the formation of stress granules. The mechanisms behind the transformation from small clusters to single globular units, followed by the maturation into ring-like structures, are still obscure. However, the morphological changes in the granules are temporally related to the alterations in the DNA-binding activity of HSF1. Both acquisition and attenuation of DNA-binding activity strictly correlate with the assembly and disassembly, respectively, of the stress granules.

In addition to the co-localization of HSF1 and HSF2 in nuclear stress granules, we provide the first in vivo evidence of complex formation between HSF1 and HSF2. Our data reveal a regulated interaction between HSF1 and HSF2 isoforms during the course of heat shock, possibly caused by regulated solubility of different HSF2 isoforms, as indicated by Mathew et al.(Mathew et al., 2001). We also show that the HSF1-HSF2 complex is efficiently disassembled in thermotolerant cells, suggesting modifications in the properties of the HSF1-interacting protein complex. Further experiments are needed to reveal whether the downregulation of HSF1-HSF2 interaction is involved in the suppression of HSF1 activity during thermotolerance, and how the HSF2 isoform composition reflects the overall heat shock response. By analyzing a set of HSF2 deletion mutants,we show that the physical interaction with HSF1 requires an intact HR-A/B oligomerization domain of HSF2. It is still an open question whether HSF1 and HSF2 form a heterodimer or heterotrimer, or whether the interaction occurs between two homo-oligomers. Considering our results and the structural similarity, it is likely that HSF1 and HSF2 form a hetero-oligomer.

Despite the prominent interaction between HSF1 and HSF2, we failed to detect HSF2 in the heat-shock-induced HSE-binding complex, which is in agreement with Mathew et al. (Mathew et al., 2001). Because these observations indicate that heat stress does not activate HSF2 HSE-binding activity, the HSF1-HSF2 complex formation together with the subnuclear co-localization represents a previously uncharacterized step in the regulation of the stress-signaling pathway in human cells. To address the question of whether HSF1-HSF2 interaction affects the localization of HSF1 in stress granules, we performed double-staining experiments using the HSF2 deletion mutants. Strikingly, expression of HSF2 mutant lacking the DNA-binding domain (HSF2 108-517) inhibited the recruitment of HSF1 into the stress granules, which could be due to the physical interaction between this mutant and HSF1. An attractive explanation is that the effect caused by HSF2 108-517 would lead to a defect in the interaction between the chromosomal nucleation sites and the HSF1-HSF2 108-517 heterocomplex. This is in good agreement with another study, in which the granule-forming-ability of HSF1 was shown to be strictly dependent on the DNA-binding and the oligomerization domains of HSF1(Jolly et al., 2002). By contrast, neither of the deletion mutants lacking the HR-A/B oligomerization domain (HSF2 203-517, HSF2 Δ126-201) was able to form a complex with HSF1 and thereby affect the compartmentalization of HSF1 in the stress granules. Therefore, the capability of HSF2 to oligomerize seems to be crucial for both interaction with HSF1 and localization in the stress granules. Unexpectedly, the C-terminal deletion mutant (HSF2 1-396), which both interacted and co-localized with HSF1, was spontaneously translocated into the stress granules with endogenous HSF1. This further emphasizes the interdependency between HSF1 and HSF2 in their stress granule localization.

A characteristic feature of the heat-shock response is acquired upon recovery from stressful conditions. The mechanisms regulating thermotolerance have been closely related to the inhibition of HSF1 by direct interaction between upregulated Hsp70 and its co-chaperones(Morimoto, 2002). Owing to the changes in HSF1 regulation upon acquired thermotolerance, we wanted to examine the localization of HSF1 and HSF2 in nuclear stress granules under these conditions. Our results demonstrate a suppression of HSF1 and HSF2 localization in stress granules in heat-shocked thermotolerant cells, which correlates well with the inhibition of HSF1 activity during thermotolerance and suggests a role for Hsps in the regulation of granule formation. We also show that the translocation of Hsp70 into and out of the nucleoli coincides with the localization of HSF1 to nuclear stress granules and interaction with HSF1. The dissociation of the HSF1-Hsp70 complex during heat shock corresponds to the translocation of Hsp70 into nucleoli. Simultaneously, the DNA-binding activity and hyperphosphorylation of HSF1, and the number of granule-positive cells are at maximum. During the attenuation of HSF1 activity, Hsp70 dissociates from the nucleoli, the complex formation between Hsp70 and HSF1 increases, and HSF1 relocates from the stress granules. Concomitant with the inhibition of HSF1 localization to granules in thermotolerant cells exposed to a second heat shock, fewer cells are observed that contain Hsp70 in their nucleoli. Taken together, our results show a strict correlation between the subnuclear compartmentalization of HSF1 and Hsp70 nucleolar localization, and,because the dissociation of HSF1 from stress granules during attenuation also coincides the translocation of Hsp70 out from the nucleoli, it is plausible that the nucleolar localization of Hsp70 contributes to the regulation of HSF1.

We thank K. M. Heiskanen, T. Katajamäki and H. Saarento for excellent technical assistance. We are grateful to L. Pirkkala for suggestions in the initial phase of this study and valuable comments on the manuscript. P. Roos-Mattjus is also acknowledged for discussions and critical comments. This work was supported by the Academy of Finland, the Sigrid Jusélius Foundation, the Wihuri Foundation, the Finnish Cancer Organizations, and the Finnish Cultural Foundation. T-PA, MH and VH were supported by the Turku Graduate School of Biomedical Sciences.