The Mammalian HSF4 Gene Generates Both an Activator and a Repressor of Heat Shock Genes by Alternative Splicing*

The expression of heat shock genes is controlled at the level of transcription by members of the heat shock transcription factor family in vertebrates. HSF4 is a mammalian factor characterized by its lack of a suppression domain that modulates formation of DNA-binding homotrimer. Here, we have determined the exon structure of the human HSF4 gene and identified a major new isoform, HSF4b, derived by alternative RNA splicing events, in addition to a previously reported HSF4a isoform. In mouse tissues HSF4b mRNA was more abundant than HSF4a as examined by reverse transcription-polymerase chain reaction, and its protein was detected in the brain and lung. Although both mouse HSF4a and HSF4b form trimers in the absence of stress, these two isoforms exhibit different transcriptional activity; HSF4a acts as an inhibitor of the constitutive expression of heat shock genes, and hHSF4b acts as a transcriptional activator. Furthermore HSF4b but not HSF4a complements the viability defect of yeast cells lacking HSF. Moreover, heat shock and other stresses stimulate transcription of target genes by HSF4b in both yeast and mammalian cells. These results suggest that differential splicing of HSF4 mRNA gives rise to both an inhibitor and activator of tissue-specific heat shock gene expression.

In response to heat shock, all organisms induce a set of proteins known as heat shock proteins (Hsps) 1 to protect themselves from cell death. Hsps are abundantly expressed in most organisms even under normal growth conditions and cooperatively act as molecular chaperones to facilitate many cellular processes including protein synthesis, folding and assembly, translocation, degradation, and the regulation of kinases and transcription factors (1)(2)(3)(4). The expression of Hsps is tightly controlled by changes in metabolic processes in cells such as cell cycle progression, differentiation, development, neuroendocrinological stress, infection, trauma, exercise, and the expression of oncoproteins and by stresses that cause protein denaturation such as heat shock, oxidative stress, and exposure to heavy metals and anti-cancer drugs (5).
The expression of eukaryotic Hsps is controlled mainly at the level of transcription by heat shock transcription factors (HSFs) that bind to the heat shock element (HSE) located in the promoter regions of all heat shock genes (6). In contrast to the single HSF gene in yeast and fruit fly, four HSF genes (HSF1, HSF2, HSF3, and HSF4) have been isolated in vertebrates, and biochemical features of HSF1, HSF2, and HSF3 have been established (7,8). HSF1 is the major factor that is activated by heat shock and by exposure to other environmental or physiological stresses (9,10). Mouse cells deficient for HSF1 are viable but are defective in the induction of heat shock genes (11). HSF2 is not activated by these stresses (12) but may function in developmental processes (13)(14)(15). In chicken cells HSF3 is activated by more extreme heat shock and other stresses compared with HSF1 (16,17). Disruption of the HSF3 gene markedly suppresses the induction of heat shock genes even in the presence of HSF1 in chicken B lymphocyte DT40 cells (18).
The heat shock factors HSF1-3 are localized primarily in the cytoplasm as inert forms in normally growing cells (9,10,16,17). In the absence of exogenous stress, HSF1 remains a monomer, whereas HSF2 and HSF3 are dimers in cultured cells; the precise nature of these complexes is not completely characterized (9,10,16,19). When cells are exposed to heat shock, each HSF relocalizes to the nucleus and is converted to a homotrimer that can bind to HSEs with high affinity (9, 10, 16, 19 -21). These changes in intracellular localization and oligomer formation are controlled mainly by two cis-regulatory domains HR-A/B and HR-C, which are composed of heptad repeats of hydrophobic amino acids (22)(23)(24)(25)(26). In unstressed cells, all of these heptad repeats appear to be involved in the stabilization of a monomer or dimer form, because mutation of specific amino acids in these heptad repeats causes constitutive trimer formation (22)(23)(24)(25)(26)(27). Upon heat shock, trimers are formed through intermolecular interactions across the HR-A/B domains (27). Heat and oxidative stress can directly cause HSF1 trimer formation in vitro; however, regulatory factors modulate the monomer-or dimer-to-trimer transition of HSF in vivo (28 -30). Hsp90 suppresses this transition (31)(32)(33), and probably to a lesser extent Hsp70 is involved in this process (34,35).
The acquisition of HSE binding activity by trimer formation is not sufficient to induce the expression of heat shock genes (36,37). In fact, the overexpression of HSF1 or HSF3 in cells causes a trimer form that can bind to HSEs (Ref. 10; see Ref. 38 for Drosophila HSF) but does not elevate the expression of heat shock genes without heat shock. 2 The transactivation domain of HSF1 is located near the carboxyl terminus and is repressed by a central cis-regulatory domain under nonstress conditions (39 -42). This regulatory domain is phosphorylated by MAPK/ ERK kinase under normal conditions, and mutation of the phosphorylation site abolishes this suppression of the activation domain in the absence of stress (43)(44)(45)(46). This suppression may be alleviated by heat shock and other stresses through phosphorylation or dephosphorylation (47)(48)(49). Hsp70 has also been demonstrated to suppress the activation domain by binding directly (50).
Although much information has been obtained regarding the mechanism of the activation of heat shock genes, few studies have examined the mechanisms controlling the cell-specific constitutive expression of heat shock genes or the mechanisms underlying the suppression of heat shock gene expression. We recently isolated a novel cDNA coding for human HSF4 (hHSF4) (51). hHSF4 does not have the HR-C domain that is necessary for the suppression of trimer formation, suggesting that hHSF4 exists as a trimer that can bind to HSEs without stress. In addition, hHSF4 does not have the potential to activate transcription, as shown by reporter gene analysis in transfection experiments. Actually, the overexpression of hHSF4 in HeLa cells suppressed the expression of heat shock genes in normally growing cells (51). These observations indicate that hHSF4 may play a role in modulating the constitutive expression of heat shock genes. In the present study, we have extended these observations by characterizing endogenously expressed mouse HSF4 and found the existence of a major splicing isoform of HSF4, HSF4b, that positively regulates heat shock gene expression.

EXPERIMENTAL PROCEDURES
Isolation of Genomic Clones of Human HSF4 -A FIXII library of human placenta DNA (Stratagene, La Jolla, CA, USA) was screened by hybridization using a 32 P-labeled insert of phHSF4 -7a (51). The filters were rinsed four times with 2ϫ SSC (1ϫ SSC is 0.15 M NaCl plus 0.015 M sodium citrate)-0.1% SDS at room temperature for 5 min per rinse, washed twice in 0.1ϫ SSC/0.1% SDS at 68°C for 30 min, dried, and exposed to x-ray film. Six positive clones were isolated, and the clone D has all exons corresponding to the open reading frame of phHSF4 -7a. After appropriate XbaI digestion fragments were identified by Southern blotting using a 5Ј-or 3Ј-terminal cDNA fragment, they were subcloned into pGEM7 vector (Promega, Madison, WI). Sequencing reactions were performed by using an AutoRead sequencing kit (Amersham Pharmacia Biotech) with synthetic oligonucleotides and Fluore-dATP labeling mix (Amersham Pharmacia Biotech). Sequences were analyzed by using ALF express DNA sequencer (Amersham Pharmacia Biotech) and assembled with GeneWorks (IntelliGenetics, Inc., Campbell, CA).
Isolation of cDNA Clones of Mouse HSF4 -A gt11 library of mouse brain cDNA was screened by hybridization using a 32 P-labeled insert of phHSF4 -7a as described above. Clones were subcloned into pGEM7 vector, and one clone, pmHSF4b-1, was sequenced in both directions. The insert of pmHSF4b-1 (1653 base pairs) lacked 28 base pairs from the putative translation start site. To isolate the 5Ј terminus of mouse HSF4 cDNA, we performed PCR for rapid amplification of cDNA ends using a Marathon cDNA amplification kit (CLONTECH, Palo Alto, CA). We used Marathon-Ready cDNA from mouse brain (CLONTECH) as a template. The amplified products were subcloned into pCR II vector (TA cloning kit, Invitrogen, NV Leek, Netherlands), and the longest clone was sequenced, yielding a 1686-base pair clone containing the mouse HSF4b cDNA. The sequences obtained were confirmed by the sequences of the genomic clones isolated from the FIXII library of mouse 129 Svj (Stratagene, La Jolla, CA).
Generation of Antiserum Specific for Mouse HSF4 -For creating antiserum to mouse HSF4, a pGEX2T-mHSF4CT plasmid was constructed that encoded the carboxyl-terminal region of mHSF4 (amino acids 395-492) fused to the GST protein. GST-mHSF4CT protein was electroeluted from the gel and was used as the rabbit immunogen to obtain a specific antiserum for hHSF4 (anti-mHSF4t) as described previously (51). An anti-mHSF4t serum recognizes a recombinant mouse HSF4 protein much stronger than antiserum raised against human HSF4 (anti-hHSF4b) (51).
Western Blot Analysis-To examine the tissue-specific expression of mHSF4, tissues were dissected from adult BDF1 mice and immediately frozen in liquid nitrogen. Tissues were homogenized in five volumes of buffer C (20 mM HEPES, pH 7.9, 0.42 M NaCl, 1.5 mM MgCl 2 , 0.2 mM EDTA, 0.5 mM phenylmethylsulfonyl fluoride, 1 g/ml of pepstatin A, 1 g/ml of leupeptin, and 0.5 mM of dithiothreitol) by using a Polytron 2 M. Tanabe and A. Nakai, unpublished results. homogenizer, and the protein concentration of each extract was determined by using the Bradford dye reagent (Bio-Rad). Western blot analysis was performed essentially as described previously (16) using an anti-mHSF4t serum (1:500 dilution), and polypeptides were detected by using the ECL system (Amersham Pharmacia Biotech). To confirm the specific bands for mouse HSF4 protein, nitrocellulose membranes containing 60 g of whole cell extracts were soaked for 1 h at room temperature in 1 ml of 2% dry milk/phosphate-buffered saline containing 2 l of anti-mHSF4t serum with 5 g of recombinant GST-mHSF4CT as a competitor or GST-cHSF1 (16) as a control. For the detection of hHSF4a and hHSF4b expressed in yeast cells, the 12CA5 mouse monoclonal antibody (Roche Molecular Biochemicals) was used to detect hemagglutinin HA-tagged hHSF4a and hHSF4b.
Gel Filtration and Subcellular Fractionation-Whole cell extracts were applied on a Superdex 200 HR column with a fast protein liquid chromatography system (Amersham Pharmacia Biotech) as described previously (16). Subcellular fractionation was performed as described (16), and equal volumes of cytoplasmic and nuclear fractions were subjected to Western blot analysis.
Northern Blot Analysis-Northern blot analysis was performed as described previously (51). To compare the levels of HSF4 mRNA in various mouse tissues, RNA blots containing 2 g each of poly(A) ϩ RNA from various mouse tissues (MTN blot purchased from CLONTECH) were hybridized with a 1.65-kilobase pair EcoRI fragment of pmHSF4 -1 and a human ␤-actin cDNA fragment (CLONTECH).
Reporter Analysis-The expression plasmids pGAL4-hHSF4b (amino acids 200 -493) and pGAL4-hHSF1 (amino acids 204 -529) were generated as described previously (51). COS7 cells maintained in 100-mm Petri dishes were transfected with 10 g of pCMV-GAL4, pGAL4-hHSF4a, pGAL4-hHSF4b, or pGAL4-hHSF, 5 g of a reporter plasmid (ptk-gal3-luc), and 0.5 g of an internal control vector, pCDM8-LacZ (a kind gift from Dr. J. Fujita, Kyoto University) by the calcium phosphate transfection method. At 6 h after transfection, cells were washed with phosphate-buffered saline and incubated further for 24 h in normal medium. A luciferase assay was performed exactly as described previously (16), and the transfection efficiencies were normalized by ␤-galactosidase activity. For the ␤-galactosidase assay, aliquots of transfected cells were suspended in 100 l of 0.25 M Tris-HCl, pH 7.5. After the cell suspension was frozen and thawed three times and centrifuged at 14,000 rpm for 5 min (MX-150, Tomy, Tokyo, Japan), 30 l of the supernatant was mixed with 66 l of O-nitrophenyl-␤-D-galactosidase (4 mg/ml in 0.1 M phosphate buffer, pH 7.0) and 204 l of Z buffer (60 mM Na 2 HPO 4 , 40 mM NaH 2 PO 4 , 10 mM KCl, 1 mM MgSO 4 , 50 mM ␤-mercaptoethanol, pH 7.0) and incubated at 37°C for 30 min. The reaction was stopped by adding 500 l of 1 M sodium carbonate, and the absorption at 420 nm was measured (Ultraspec 3000; Amersham Pharmacia Biotech).
Generation of Stable Lines Expressing hHSF4b-The human HSF4b expression vector pCMV-hHSF4b was generated by subcloning the EcoRI fragment of hHSF4b cDNA, containing the entire hHSF4 open reading frame, into pCMV/Blue vector (PharMingen, San Diego, CA). HeLa cells maintained in 100-mm Petri dishes were cotransfected with 30 g of pCMV-hHSF4b and 5 g of pH␤-Act-neo vector (16) for 4 h by the calcium phosphate transfection method. After two washes with phosphate-buffered saline, the cells were shocked with 1.5% glycerol for 2 min and incubated in normal medium for 24 h. Stable transformants were then selected in the presence of Geneticin disulfide (Wako) at a final concentration of 1.5 mg/ml.
Yeast Plasmids, Strains, Growth Conditions, RNA Analyses, and Acquired Thermotolerance Assays-Human cDNAs encoding HSF4a and HSF4b were subcloned into the yeast expression vectors p424GPD and p423GPD (56), respectively. p424PGPD-hHSF4aHA and p423GPD-hHSF4bHA were generated by the insertion of a triple influenza HA tag immediately upstream of the stop codon in each of the cDNAs. Other yeast plasmids, yeast strains, growth conditions, cell extract preparation, cross-linking, and acquired thermotolerance assays were conducted as described (57). RNase protection assay was performed as described (57,58).

RESULTS
Isoforms of Human HSF4 -In the course of the molecular cloning of human HSF4, we isolated some partial cDNAs whose sequences did not completely match that of the reported clone phHSF4 -7a (51). This prompted us to examine the HSF4 genomic locus and explore the possibility of alternative transcripts from the human HSF4 gene. We isolated hHSF4 genomic DNA from a human placental genomic library to re-veal its genomic structure. The hHSF4 gene corresponding to the open reading frame of the phHSF4 -7a clone was composed of 13 exons (Fig. 1A). We refer these as exon 1 to exon 13 in this paper. Exons coding for functional domains are shown at the top of Fig. 1A. The unique structural feature of hHSF4 is its lack of a carboxyl-terminal heptad repeat of hydrophobic amino acids (HR-C). We ruled out the possibility that the HR-C region might be encoded on the hHSF4 genome by sequencing the genomic fragment encoding hHSF4 (data not shown).
To assess the full functional repertoire of the hHSF4 gene, we systematically analyzed the hHSF4 mRNA isoforms. By using specific primers (primers a to f) (Fig. 1A), RT-PCR analysis of the transcripts in various human tissues was performed, and amplified DNA bands were visualized after hybridizing with the 32 P-labeled insert of phHSF4 -7a (Fig. 1B). Reverse transcription reactions using either oligo(dT) or random hexanucleotide primers gave the same results. A single band of 458 base pairs was amplified using primers a and b (Fig. 1B, panel I). In contrast, two bands were clearly amplified using primers c and d (Fig. 1B, panel II). We analyzed these fragments from skeletal muscle, heart, and testis by sequencing and found that the lower bands were composed of a fragment identical to that of phHSF4 -7a (4 clones of 38 clones sequenced) and other fragments. In all of the predicted transcripts with the fragments other than that of phHSF4 -7a, a termination codon was observed within the fragments (data not shown). The upper band was a single fragment of 705 base pairs. Although the nonlinearity of these PCR cycles did not allow quantitative comparison of the expression levels in various tissues, the expression level of the upper band was found to be relatively high in the brain, heart, and skeletal muscle compared with that of the lower band in each tissue (Fig. 1B, panel II). We discuss this 705-base pair fragment in detail below. RT-PCR analysis using primers e and f showed a single band of 379 base pairs in most tissues, except that a faint band of 309 base pairs was observed in the pancreas, brain, and testis (Fig. 1B, panel  III). DNA sequencing analysis revealed that this DNA fragment is derived from transcripts from which exon 12 was spliced out (Fig. 1A). The amount of mRNA corresponding to each hHSF4 isoform was not altered by heat shock in HeLa cells (data not shown). In summary, we have identified a major alternatively spliced product referred as hHSF4b, in addition to the previously reported hHSF4a (51).
Human HSF4b Is an Alternatively Spliced Product-Sequence analysis of the 705-base pair fragment detected by RT-PCR using primers c and d (Fig. 1, A and B) showed that the hHSF4b transcript lacked 14 base pairs in exon 8 compared with the hHSF4a transcript (Fig. 1C, exon 8a) and has 104 base pairs upstream of exon 9b (Fig. 1C, exon 9a). The sequences of the exon-intron boundaries in both the hHSF4a and hHSF4b matched well with the consensus sequences of mammalian splice site (Fig. 1D and Ref. 59). This strongly suggests that hHSF4a and hHSF4b mRNAs are generated by alternative splicing events. Consequently, the putative amino acid sequences of hHSF4b corresponding to exons 8 and 9 are completely different from those of hHSF4a, but the sequences of the carboxyl-terminal regions of both isoforms, corresponding to exons 10 -13, are identical (Fig. 1E).
Isolation of Mouse HSF4 cDNA Clones-To begin to understand the functional relevance of the HSF4a and HSF4b isoforms, the tissue distribution of their corresponding mRNAs was examined in mice. We performed RT-PCR analysis using primers designed to specifically detect the alternative transcripts in tissues from mice corresponding to human HSF4a and HSF4b (see "Experimental Procedures"). Unexpectedly, a 614-base pair fragment that corresponds to the b-form was a major band in all mouse tissues ( Fig. 2A, arrow). The faint lower bands amplified from C2C12 mRNA were sequenced and shown to possess termination codons (data not shown). However, among these sequences a mouse transcript corresponding to human HSF4a was still observed (two clones of ten clones sequenced; Fig. 2C). To exclude the possibility that the ampli-fication of only a b-form using mouse tissue RNA was due to the use of mouse specific primers, we simultaneously performed RT-PCR using human and mouse brain total RNA with another set of primers, the sequences of which were identical in human and mouse HSF4 cDNAs. Consistent with the data shown above, we observed two bands in reaction using human brain HeLa cells (lane 9). phHSF4 -7a cDNA was used as a template for a control PCR (lane 10). RT-PCR reactions were fractionated on a 1% agarose gel, and Southern blotting was performed using a 32 Plabeled insert of phHSF4 -7a as a probe. Primers used were a and b (I), c and d (II), and e and f (III) (see "Experimental Procedures"). The numbers of base pairs of these amplified bands are shown on the right. C, two forms of alternative 3Ј splice sites are shown. In both forms, the same two alternative 5Ј splice sites at the end of exons 7 and 8b are used. hHSF4a splicing occurs from alternative 3Ј splice sites at the starts of exons 8a and 9b. hHSF4b splicing occurs from alternative 3Ј splicing sites at the starts of exons 8b and 9a. The numbers of base pairs of each exon are 14, 125, 104, and 124 for exons 8a, 8b, 9a, and 9b, respectively. D, alignment of sequences of alternative 3Ј splice sites were shown. Arrows indicate boundaries between introns and exons. Y represents C and T, and N indicates any nucleotides. E, comparison of the structures of hHSF4a and hHSF4b. The amino acid sequences of the regions that are encoded by exons 8 and 9 are completely different between hHSF4a (amino acids 246 -290) and hHSF4b (amino acids 246 -320).
RNA and a single band of b-form (644 base pairs) in reaction using mouse brain RNA ( Fig. 2A, lanes 9 and 10). We concluded that mHSF4b is a dominant isoform in most mouse tissues compared with mHSF4a.
By screening a phage library of mouse brain cDNA, we next isolated cDNA clones corresponding to the mouse homologue of hHSF4b (Fig. 2B). The predicted amino acid sequence of mouse HSF4b is 86% identical to that of human HSF4b. Interestingly, the unique sequence for the b-form (amino acids 245-319 in mouse HSF4b; Fig. 2B, sequences between triangles) contains the site c region that all of the vertebrate HSFs have (22,51) and another conserved region between HSF4b and HSF1 (site g). A serine residue (amino acid 299 in hHSF4b and amino acid 303 in hHSF1) in this region was shown to be phosphorylated  2-7) to show the expression of a-and b-isoforms of mouse HSF4. An isolated mouse HSF4b cDNA clone, pmHSF4b-1, was used as a template for a control PCR (lane 8). A set of mouse specific primers was used for the PCR reaction (lanes 1-8). RT-PCR reactions were fractionated on 1% agarose gel, and Southern blotting was performed using a 32 P-labeled insert of phHSF4b-1 as a probe. The arrow indicates a fragment amplified from the mHSF4b transcript (614 base pairs (bp)), and the arrowhead represents bands amplified from alternative products. Internal control primers were used to co-amplify transcripts of mouse S16 ribosomal protein (53), and DNA bands were visualized by ethidium bromide staining. To exclude a possibility of a primer-specific amplification, a set of common primers for mouse and human HSF4 cDNA was used to perform RT-PCR using human and mouse brain total RNA (lanes 9 and 10). B, an alignment of putative amino acid sequences of mouse and human HSF4b. The numbers on the right indicate the amino acid positions of each protein. Identical amino acids are shaded. The borders of amino acids encoded by exons 8b and 9a are indicated by triangles. The boxed region I corresponds to the DNA binding domain; region II corresponds to the amino-terminal hydrophobic repeat (HR-A/B); and region IV corresponds to the downstream of the hydrophobic repeat (DHR). HSF4 lacks region III, which corresponds to the carboxyl-terminal hydrophobic repeat (HR-C) observed in all other vertebrate HSFs (51). The underlined regions indicated as sites a to g correspond to additional regions of identity among each of the four HSF proteins (22,51). Sites f and g are highly related regions in between HSF4b and HSF1 as revealed in this study, and a serine residue in site g is shown to be a putative phosphorylation site in HSF1 (43)(44)(45)(46). Sites a and c are found in all four HSFs. C, an alignment of putative amino acid sequences of mouse and human HSF4a. The putative amino acids encoded by exons 8a, 8b, and 9b in mouse and human HSF4a are aligned. by MAPK/ERK kinase or glycogen synthase kinase 3 in hHSF1 (43)(44)(45)(46); however, another site of phosphorylation at residue 307 in hHSF1 is not present in HSF4b. It is currently unknown whether HSF4, like HSF1, is regulated by phosphorylation within this region. In contrast to the unique sequences in the b-form, the sequences unique for the a-form contain no region related to sequences of other HSFs (Fig. 2C) (51).
Tissue-specific and Cell Type-specific Expression of Mouse HSF4 -To examine the tissue-specific expression of mHSF4, we performed Northern blot analysis (Fig. 3A) and Western blot analysis using antiserum raised against the carboxyl-terminal region of mHSF4 (Fig. 3B). We found a single band of 62 kDa specific for mHSF4 in brain and lung tissues (Fig. 3B) by comparing results obtained with immune versus preimmune serum (data not shown). The specificity of this band was confirmed by its disappearance when antiserum was preabsorbed with a recombinant GST-mHSF4CT protein but not by a recombinant GST-cHSF1 protein (Fig. 3C). In brain and lung, mHSF4 mRNA was also abundantly expressed (Fig. 3A). Although mHSF4 mRNA was also detected in liver and skeletal muscle, we did not detect mHSF4 protein in these tissues. We also detected expression of mHSF4 in mouse myoblast C2C12 cells and rat pheochromocytoma PC12 cells but not in all cell lines examined (Fig. 3D). The molecular size of rat HSF4 may FIG. 3. Expression of mHSF4 in mouse tissues and cell lines. A, Northern blot analysis of RNA from adult mouse tissues. RNA blots (MTN blot purchased from CLONTECH) were hybridized with a mHSF4b cDNA probe or a human ␤-actin probe. A 1.8-kilobase pair mRNA was detected in brain, lung, liver, and skeletal muscle. B, Western blot analysis in various tissues. Whole cell extracts (60 g) from various tissues of adult BDF1 mouse were subjected to 10% SDS-PAGE, followed by Western blot analysis using an anti-mHSF4t serum. The specific band for mHSF4 (62 kDa) is indicated at the left side by an arrow. An asterisk in the figure indicates a nonspecific band corresponding to that shown in C. C, competition experiment to confirm the specific band for mouse HSF4 protein. After blocking nitrocellulose membranes containing 60 g of protein from brain (B) and lung (L), these were soaked in 1 ml of 2% dry milk/phosphate-buffered saline containing 2 l of anti-mHSF4t serum without (lanes 1 and 2) or with 5 g of recombinant GST-mHSF4CT (lanes 5 and 6) as a competitor or 5 g of GST-cHSF1 (lanes 3 and 4) (37) as a control. The specific band for mHSF4 (62 kDa) was indicated by an arrow, and a nonspecific band (69 kDa) was indicated by an asterisk. D, protein levels of mHSF4, mHSF1, and mHSF2 in various rodent cell lines. Whole cell extracts (40 g) from non-neuronal (lanes 1-3) and neuronal (lanes 4 -9) cells were subjected to 10% SDS-PAGE, followed by Western blotting using an anti-mHSF4t serum, an anti-cHSF1␤ serum, or an anti-cHSF2␣ serum as the primary antibody. The arrow indicates the specific band for mouse HSF4 (62 kDa) in C2C12 cells or rat HSF4 (58 kDa) in PC12 cells. The asterisk indicates a nonspecific band corresponding to that shown in C. mHSF1 and mHSF2 proteins were ubiquitously expressed in all cells shown, whereas the expression of mHSF4 protein was restricted in C2C12 and PC12 cells.
be smaller than that of mouse HSF4. The level of expression was not changed when PC12 cells were differentiated by treatment with nerve growth factor, and mHSF4 protein was not observed in differentiated NCB20 neuroblastoma-brain hybrid cells (Fig. 3D). Compared with the ubiquitous expression of mHSF1 and mHSF2, the expression of mHSF4 is highly specific to some cells (Fig. 3D). We concluded that a single band of mHSF4 observed in this paper consists of the b-form for several reasons. First, RT-PCR analysis showed that mHSF4b is a major species in C2C12 cells and various tissues, whereas mHSF4a is a faint one ( Fig. 2A). Second, the apparent molecular sizes of hHSF4a (56 kDa) and hHSF4b (62 kDa) are clearly distinct (see Fig. 5D). The molecular size of mHSF4 (62 kDa) observed in mouse tissues and cells is identical to that predicted for hHSF4b. Third, hHSF4a immediately became an insoluble aggregate when HeLa cells expressing hHSF4a are incubated at 42°C, whereas hHSF4b is very stable when hHSF4b-expressing cells are heat-shocked (data not shown). Similar to hHSF4b, HSF4 in C2C12 cells was stable even when cells were heat-shocked at 43°C for 30 min (Fig. 4A).
Mouse HSF4b Is Multimeric under Nonstress Conditions-Our previous study revealed the lack of the HR-C in hHSF4a that is necessary for the suppression of trimer formation and showed that hHSF4a overexpressed in HeLa cells forms a trimer in the nucleus in the absence of stress (51). We examined the oligomeric state of endogenous mHSF4b in mouse tissues and cells. Gel filtration analysis showed that in C2C12 cells a substantial amount of mHSF4b was eluted in fractions 18 and 19 (Fig. 4A, row c), the same fractions in which activated mHSF1 (indicated at the bottom) and overexpressed hHSF4 (data not shown; see Ref. 51) were detected. Significant amounts of rat HSF4 in PC12 cells also eluted in these two fractions (row e). Furthermore, the elution profile of mHSF4b extracted from brain and lung are consistent with constitutive multimerization (rows a and b). We should note, however, that the elution profiles of mHSF4b did not show a sharp peak, rather its elution was observed from a fraction 17 to fraction 22 (Fig. 4A, rows a-c and e). We tested the possibility that the mHSF4 distribution would shift in response to stresses such as heat shock. The elution profile revealed that, like HSF1, mHSF4b showed a sharp peak at fractions 18 and 19 after C2C12 and PC12 cells were treated with heat shock (at 43°C for 30 min) or the proline analogue azetidine (at a concentration of 5 mM for 8 h) (Fig. 4A, rows d, f, and g). This suggests that mHSF4b may undergo conformational changes and or post-translational modifications in response to external stresses. Unfortunately, we could not observe a distinct supershifted band of mHSF4b by gel shift assay using an HSE probe in the presence of anti-mHSF4b antibody (data not shown). This may be due to the character of this antibody like that of anti-cHSF3 antibody (16), or the low abundance of mHSF4b relative to that of mHSF1 in tissue and cell extracts.
We then examined the subcellular localization of mHSF4b in C2C12 cells. Cytoplasmic and nuclear extracts were subjected to Western blotting by using antiserum against mHSF4 and cHSF1, which cross-reacts with mouse HSF1. The nucleus was a major pool of mHSF4b before and after heat shock, whereas the distribution of mHSF1 shifted from the cytoplasm to the nucleus upon heat shock. Moreover, a distinct mobility shift was observed for mHSF1 in response to heat shock, which was previously shown to be due to phosphorylation (Fig. 4B) (9, 10). Because the mobility of mHSF4b was also retarded by heat shock (Fig. 4, A, row d, and B), mHSF4b may be phosphorylated in response to heat shock.
Human HSF4b Elevates the Expression of Heat Shock Genes in the Absence of Stress-A previous study demonstrated that the human HSF4a isoform lacks the properties of a transcriptional activator (51). The overexpression of hHSF4a caused marked decreases in the basal expression of Hsp90 and Hsp27, and thus we postulated that hHSF4 acts as a repressor of heat shock gene expression under normal conditions (51). To reveal potential functional differences between hHSF4a and hHSF4b, we compared their ability to activate transcription of a GAL4 site-directed luciferase gene. The DNA-binding domain (DBD) of GAL4 was fused to the region downstream of HR-A/B (amino acids 200 -463 in hHSF4a, 200 -493 in hHSF4b, and 204 -529 in hHSF1) and co-transfected with a reporter gene (Fig. 5A). As was shown previously, GAL4-hHSF4a did not increase luciferase activity and rather suppressed the activity when expressed at very high levels (51) (Fig. 5B). In contrast, GAL4-hHSF4b elevated the luciferase activity by about 3-fold, although the activity was less than that produced by GAL4-hHSF1, in which transcriptional activity was repressed (39,40,60) (Fig. 5B). The activity of GAL4-hHSF4b was about one-tenth of that of the fusion protein containing the GAL4 DBD and the HSF1 activation domain (data not shown; see Ref. 51). Because a carboxyl-terminal region of hHSF4a (amino acids 296 -463), which was identical to that of hHSF4b (Fig. 1D), was shown to have the ability to activate transcription (51), these results  figure (data not shown). B, subcellular fractionation of mouse C2C12 cells. Cytoplasmic (C) and nuclear (N) extracts were prepared from control (cont) and heat-shocked (at 43°C for 30 min) (H.S.) cells. Equal volumes of extracts were subjected to SDS-PAGE, followed by Western blotting by using an anti-mHSF4t serum or an anti-cHSF1␤ serum. suggest that a unique region of hHSF4a corresponding to exons 8 and 9 suppresses the activity, whereas a unique region of the hHSF4b does not.
The results from the size fractionation experiments suggested that the activities of HSF4 isoforms might be regulated by stress stimuli like that of HSF1. We treated COS7 cells, transfected with the expression vectors for GAL4 DBD fusion proteins and reporter genes as described above, with heat shock, sodium arsenite, or the proline analogue azetidine. We found that GAL4-hHSF4a did not induce LUC activity by any treatment, and rather the activities were clearly decreased (Fig. 5C). In marked contrast, GAL4-hHSF4b induced the LUC activity 6 -16-fold when stressed, although the levels of its activity were much lower than that driven by the GAL4-hHSF1 fusion protein. This suggests that hHSF4b has stronger potential to activate transcription when cells are stressed, whereas, consistent with previous observations (51), hHSF4a does not activate gene transcription in this context.
To further explore the functional distinctions between hHSF4a and hHSF4b, we generated HeLa cell lines stably expressing hHSF4a and hHSF4b. The level of hHSF4b in each line was examined by Western blotting (Fig. 5D). Relatively high expression of hHSF4b was observed in the lines F4B-10, F4B-7, and F4B-18; however, the significance of protein doublets is unclear. In these cells, the Hsp70 and Hsp27 mRNA levels were higher (up to 2-fold) than that of parental HeLa cells in the absence of stress (Fig. 5E). The level of Hsp90 was only marginally increased in these lines. These results were completely different in HEF4d cells expressing hHSF4a, in which Hsp90 and Hsp27 mRNA levels were markedly repressed (Fig. 5E, HEF4d) (51). This suggests that hHSF4b could act as an activator of heat shock genes under normal growth conditions. The accumulation of Hsp70 mRNA in cells overexpressing hHSF4b after heat shock was similar to that in wild type HeLa cells (data not shown).
Human HSF4b Functionally Substitutes for the Saccharomyces cerevisiae HSF-To gain further insight into functional differences between the two splicing isoforms of hHSF4, we tested whether the isoforms could functionally substitute for the loss of the single essential HSF gene in the yeast S. cerevisiae. It was previously shown that human HSF2 and a constitutively active mutant of human HSF1 (hHSF1lz4m) can complement the viability defect of an S. cerevisiae hsf⌬ strain (57). Similar assays were carried out with hHSF4a and hHSF4b, using the S. cerevisiae strain PS145, which bears a chromosomal disruption of yHSF gene and an episomal plasmid-borne yHSF gene under the control of a galactose-inducible, glucoserepressible GAL1 promoter. As illustrated in Fig. 6A, the expression of hHSF4b allows the yhsf⌬ cells to grow on glucosecontaining medium, indicating that it can functionally substitute for yHSF to an extent similar to that of hHSF2 and hHSF1lz4m. Expression of hHSF4a, like wild type hHSF1, failed to support cell growth on glucose. We tested different expression levels of hHSF4a, using promoters of varying strength, and the same results were consistently obtained. A carboxyl-terminal HA tag added to hHSF4a and hHSF4b did not change the ability of hHSF4b to complement yhsf⌬, nor did it allow hHSF4a to complement yhsf⌬. Measurements of protein levels using either anti-hHSF4 antibody or anti-HA monoclonal antibody demonstrated the abundant expression of both hHSF4a and hHSF4b, suggesting that the lack of complemen-tation by hHSF4a is not due to a lack of expression (data not shown, also see Fig. 6B). We also tested whether co-expression of hHSF4a might have an adverse effect on the complementation by hHSF including hHSF2, hHSF1lz4m, and hHSF4b; however, no inhibition of growth was observed (data not shown). Therefore, we conclude that the hHSF4b protein, but not hHSF4a, can functionally substitute for the S. cerevisiae HSF to sustain viability of yeast cells. This functional difference between hHSF4a and hHSF4b correlates well with their trans-activation potential in mammalian cells, suggesting that complementation of the yhsf⌬ mutation by hHSF4b requires (51). Molecular mass standards (Amersham Pharmacia Biotech) are indicated: 97.4 kDa, phosphorylase b; 69 kDa, bovine serum albumin; 46 kDa, ovalbumin; 30 kDa, carbonic anhydrase. E, Northern blot analysis of HeLa cells, hHSF4b-overexpressing cell lines, a hHSF4a-overexpressing line HEF4d, and heat-shocked HeLa cells (45°C for 1 h, H.S.) using cDNAs for human Hsp90␣, human Hsp70, human Hsp27, and mouse ␤-actin as probes (39).
FIG. 6. Human HSF4b functionally substitutes for S. cerevisiae HSF. A, hHSF4b supports growth of yhsf⌬ cells. The S. cerevisiae hsf⌬ strain PS145, harboring a plasmid-borne GAL1-yHSF, was transformed with plasmids expressing yHSF, hHSF2, hHSF1, hHSF1lz4m, hHSF4a, or hHSF4b. Recepient cells were streaked onto synthetic complete medium containing either 2% glucose or 2% galactose as the carbon source, incubated at 30°C for 2 days and photographed. The individual transformants are indicated in the key shown above. B, both human HSF4a and HSF4b trimerize in yeast cells. EGS cross-linking was carried out using native whole cell extracts from PS145 cells expressing hHSF4a or hHSF4b containing a carboxyl-terminal HA tag (both behave identically to unmodified hHSF4a or hHSF4b) at 25°C. EGS-treated and Me 2 SO control extracts were electrophoretically fractionated on a 6% SDS-PAGE gel, followed by Western blotting by using HA monoclonal antibody. The positions of the molecular mass markers are indicated on the left. EGS concentrations used in the cross-linking were 0, 0.5, and 2.0 mM, and the ovals indicate the expected migration of HSF monomers, dimers and trimers. the ability to activate target gene transcription.
Previous work on hHSFs expressed in S. cerevisiae has shown a strong correlation between the ability of the HSF molecules to trimerize and their functional complementation of yhsf⌬ (57). Human HSF2 and all derivatives of hHSF1 that complement the yhsf⌬ strain are constitutively trimerized in yeast, whereas wild type hHSF1 and its nonfunctional derivatives exist primarily as monomers, with a small percentage of dimers, under either control temperature or heat shock conditions. To ascertain whether the lack of complementation by hHSF4a is due to its inability to trimerize in yeast, crosslinking by using ethylene glycol bis-(succinimidylsuccinate) (EGS) was performed to assess the oligomeric states of hHSF4a and hHSF4b expressed in yeast cells (Fig. 6B). The data clearly demonstrated that both hHSF4a and hHSF4b formed trimers under normal growth temperatures. Under these same conditions we cannot detect trimerized HSF1, whereas the HSF1lz4m mutant is essentially quantitatively trimerized (data not shown; see Ref. 57). Therefore, the inability of hHSF4a to complement the yhsf⌬ strain is not due to an inability to trimerize in yeast.
Human HSF4b Activates Target Gene Transcription with Human HSF1-like Specificity in Yeast-Mouse HSF1 and HSF2 have different DNA-binding site preferences in vitro (61).
It has been shown that both hHSF2 and hHSF1lz4m activate target gene transcription in yeast cells in response to heat shock (57). In addition, hHSF2 and hHSF1lz4m exhibit specificity toward different HSE-containing promoters in a manner consistent with their in vitro DNA binding preferences. Previous in vivo target gene activation experiments have shown that in response to heat shock hHSF1lz4m more strongly activates transcription of the yeast SSA3 gene, containing a consensus HSE with five consecutive pentameric nGAAn repeats, whereas hHSF2 selectively activates transcription of the yeast CUP1 gene, which has three nGAAn repeats with a gap between the second and third pentamers. Using RNase protection assays, we found that like hHSF1lz4m, hHSF4b also exhibits a selectivity toward the SSA3 promoter (Fig. 7A). hHSF4b activates SSA3-lacZ transcription in response to heat shock, to an extent similar to that by hHSF1lz4m, whereas, consistent with previous observations, SSA3-lacZ is not significantly activated by hHSF2. In contrast hHSF2 strongly activates CUP1-lacZ expression, whereas hHSF1lz4m and hHSF4b only weakly activate transcription from this reporter gene.
The activation of target gene transcription by hHSF4b in response to heat shock suggests that yeast cells expressing hHSF4b may exhibit acquired thermotolerance. We tested this hypothesis by assessing the acquired thermotolerance of FIG. 7. Human HSF4b activates target gene transcription and confers acquired thermotolerance in S. cerevisiae hsf⌬ cells. A, SSA3-lacZ and CUP1-lacZ mRNA levels in response to heat shock. Strain PS145 cells expressing yHSF, hHSF2, hHSF1lz4m, or hHSF4b as the sole source of HSF were independently transformed with SSA3-lacZ and CUP1-lacZ reporter plasmids. Transformants were grown at 25°C, subjected to control (C, 25°C) or heat shock (HS) treatment at 40°C for 15 min. Total RNA was isolated, and lacZ mRNA levels were analyzed by RNase protection assays. The S. cerevisiae actin mRNA was used as an internal control. ACT1, SSA3-lacZ, and CUP1-lacZ refer to the protected mRNA fragments using 32 P-labeled RNA probes specific for actin, SSA3-lacZ, and CUP1-lacZ mRNA, respectively. B, human HSF4b confers thermotolerance to yeast cells. PS145 cells expressing yHSF, hHSF2, hHSF1lz4m, or hHSF4b were grown to early log phase at 25°C, shifted to 37°C for 30 min, and subjected to 50°C heat treatment. Control cells were shifted directly from 25 to 50°C. Cell aliquots were withdrawn at the indicated times, diluted, and plated on YPD agar, and plates were incubated at 30°C. The percentage of cell survival was plotted against the zero time point samples. Data shown are the averages of three independent experiments.
hHSF4b-expressing cells in comparison with cells expressing yHSF, hHSF2, or hHSF1lz4m, which are known to be thermotolerant (57). A moderately high temperature (37°C) pretreatment of yeast cells can subsequently protect them from lethal temperature (50°C) treatment, presumably because of the induction of Hsp genes by the pretreatment. Expression of hHSF4b supported acquired thermotolerance to a level similar to that allowed by hHSF2 and hHSF1lz4m (Fig. 7B). These results demonstrate that hHSF4b is a functional molecule in yeast cells that can activate target gene transcription in response to heat shock and confer acquired thermotolerance. DISCUSSION In this paper, we found that a novel mouse HSF4b isoform was constitutively expressed specifically in some mouse tissues and cell lines. This is a first demonstration that three proteins of HSF family members (HSF1, HSF2, and HSF4) are coexpressed in mammalian cells. Among the vertebrate HSFs, HSF4 (both HSF4a and HSF4b) has unique structural features: it lacks the carboxyl-terminal heptad repeat of hydrophobic amino acids (HR-C) necessary for the inhibition of HSF trimer formation. Although a class of plant HSF that lacks the HR-C was thought to exist, biochemical features of these HSFs remain to be explored in detail (62). HSF is known to be activated by two steps. First, a monomeric or dimeric HSF is converted to a trimer that can bind to DNA (9,10,16,19). Second, the DNA-binding form of HSF undergoes additional changes to acquire transactivation activity (36,37,42,60). Consistent with the absence of an HR-C region, we found that mHSF4b forms a trimer in the absence of stress, indicating that mHSF4b bypasses the first step of the HSF activation (Fig. 4A). Thus, mHSF4b may bind to DNA constitutively. It is currently unknown whether HSF4b requires additional modifications for the acquisition of trans-activation function. Overexpression of human HSF4b showed that HSF4b can moderately elevate the expression of the heat shock genes (Fig. 5E). Human HSF4b can also complement the viability defect of S. cerevisiae in which the HSF gene is disrupted, whereas HSF1 cannot (Fig. 6A). These observations suggest that mHSF4b has the potential to activate transcription and can induce the expression of heat shock genes during normal growth conditions. In addition to this constitutive activity, it is interesting to note that the transcriptional activity of human HSF4b is increased in response to stress in both yeast and mammalian cells (Figs. 5C and 7A). The electrophoretic mobility of mHSF4b is also reduced in response to heat shock, and a stress-induced conformational change of mHSF4b is suggested by gel filtration analysis (Fig. 4A). Taken together, mHSF4b may regulate the constitutive expression of heat shock genes and could further induce the expression of these genes in response to stress in a manner analogous to yeast HSF.
The existence of isoforms of mammalian HSF1 and HSF2 was reported previously (13,63,64), and the ratio of the expression level of these isoforms diverges in various tissues. In both HSF1 and HSF2, exons coding for 22 and 18 amino acids, respectively, located downstream of HR-C (the DHR region in the present paper), are alternatively spliced out. Although the biological significance of these alternative forms is still unclear, an HSF2 isoform that was made from an alternatively spliced transcript was a less potent transcriptional activator (63) and acted as a repressor of hemin-induced heat shock gene expression (65). In contrast to these simple deletions of some amino acids, two alternative 5Ј splice sites (59) were observed in exons 8 and 9 in the hHSF4 transcript, and a frameshift occurred within these two exons in HSF4. Thus, the form of alternative RNA splicing in HSF4 transcripts is clearly different and more complex from those observed in HSF1 and HSF2. Alternative splicing events, leading to the generation of functionally distinct HSF isoforms, may be another level of regulation to control heat shock gene expression.
Because Hsps act as molecular chaperones to facilitate protein folding, assembly, synthesis, and translocation during normal growth conditions, substantial levels of Hsps are essential for cell growth. Conversely, overexpression of Hsps are detrimental (66,67). Thus, heat shock genes are expected to be strictly regulated depending on the types and conditions of cells. In budding yeast, a single HSF regulates the basal expression of heat shock genes as well as the stress-induced expression (68). In vertebrates, the heat shock regulation by HSF1 and chicken HSF3 has been well studied, whereas less attention has been paid to the roles of HSF in the expression of heat shock genes during normal cell growth. Actually, overexpression of vertebrate HSF1 or chicken HSF3 in cells results in their trimerization (Ref. 10; see Ref. 38 for Drosophila HSF) but does not affect the constitutive expression of heat shock genes at all. 2 In contrast, the characterization of HSF4 suggests that it may regulate the expression of heat shock genes during normal growth conditions. Although the potential of HSF4b to activate heat shock genes is weak compared with that of an activated HSF1, a severalfold increase in the levels of heat shock proteins may have great effects on processing many proteins in the absence of stress.
Consistent with heat shock activation of gene expression and acquired thermotolerance in yeast cells expressing hHSF1lz4m and hHSF2, we observed both acquired thermotolerance and heat-inducible gene expression in cells expressing hHSF4b but not hHSF4a. This is in marked contrast to the findings obtained in vertebrate cells, as is suggested by the fact that activation domains of several transcription factors act differently in budding yeast and mammalian cells (69). In mouse cells deficient for mHSF1, heat shock induction of heat shock genes was not detected (11), and heat shock induction was also hardly detected in chicken cells deficient for cHSF3 even in the presence of cHSF1 and cHSF2 (18). In these cells, thermotolerance was not acquired. Although heat shock response caused by hHSF2 or hHSF4b may be unique in yeast cells, the present findings provide some important information. First, the transcriptional activities of hHSF2 and hHSF4b are induced by external stimuli. As shown in Fig. 5C, the activity of hHSF4b was moderately induced by heat shock, sodium arsenite, and azetidine. This suggests that hHSF4b, which has the ability to activate transcription under normal growth conditions, could become a more potent activator in response to some stresses in vertebrate cells. Second, consistent with a putative role in negative regulation in human cells, hHSF4a cannot complement the yeast hsf1⌬ viability defect, even though it is constitutively trimerized. Third, SSA3-lacZ was strongly activated by hHSF1 and hHSF4b but not highly activated by hHSF2. Conversely, hHSF2 activated CUP1-lacZ expression strongly, whereas hHSF1 and hHSF4b activated expression from this promoter only slightly. This similarity between hHSF4b and hHSF1 is consistent with the higher sequence identity of the DNA-binding domains between hHSF4 and hHSF1 (76%) compared with that between hHSF2 and hHSF1 (69%). Thus, experiments conducted in yeast support the notion that HSF4b may activate gene transcription in mammalian cells.
We and others have examined the expression of members of HSFs in chicken 3 and mouse (13,14) tissues. In contrast to the ubiquitous expression of HSF1, HSF2, and chicken HSF3, the expression of mouse HSF4 is observed to be restricted to the brain and lung tissues. This may suggest that HSF4 may be a mediator for physiological stress in these tissues rather than for external stress. Examination of its function in the whole body could reveal unexpected physiological roles for HSF isoforms.