Regulation of Stress-Inducible Phosphoprotein 1 Nuclear Retention by Protein Inhibitor of Activated STAT PIAS1

Stress-inducible phosphoprotein 1 (STI1), a cochaperone for Hsp90, has been shown to regulate multiple pathways in astrocytes, but its contributions to cellular stress responses are not fully understood. We show that in response to irradiation-mediated DNA damage stress STI1 accumulates in the nucleus of astrocytes. Also, STI1 haploinsufficiency decreases astrocyte survival after irradiation. Using yeast two-hybrid screenings we identified several nuclear proteins as STI1 interactors. Overexpression of one of these interactors, PIAS1, seems to be specifically involved in STI1 nuclear retention and in directing STI1 and Hsp90 to specific sub-nuclear regions. PIAS1 and STI1 co-immunoprecipitate and PIAS1 can function as an E3 SUMO ligase for STI. Using mass spectrometry we identified five SUMOylation sites in STI1. A STI1 mutant lacking these five sites is not SUMOylated, but still accumulates in the nucleus in response to increased expression of PIAS1, suggesting the possibility that a direct interaction with PIAS1 could be responsible for STI1 nuclear retention. To test this possibility, we mapped the interaction sites between PIAS1 and STI1 using yeast-two hybrid assays and surface plasmon resonance and found that a large domain in the N-terminal region of STI1 interacts with high affinity with amino acids 450–480 of PIAS1. Knockdown of PIAS1 in astrocytes impairs the accumulation of nuclear STI1 in response to irradiation. Moreover, a PIAS1 mutant lacking the STI1 binding site is unable to increase STI1 nuclear retention. Interestingly, in human glioblastoma multiforme PIAS1 expression is increased and we found a significant correlation between increased PIAS1 expression and STI1 nuclear localization. These experiments provide evidence that direct interaction between STI1 and PIAS1 is involved in the accumulation of nuclear STI1. This retention mechanism could facilitate nuclear chaperone activity.

Elimination of STI1 in yeast sensitizes cells to Hsp90 inhibitors, but it is not by itself lethal (19). STI1 can also be eliminated in C. elegans, although it results in decreased life span (20). In contrast, STI1 mutant mice do not survive E10.5 and present several morphological defects, owing to decreased levels of several Hsp90-client proteins (21). Mouse embryonic fibroblasts obtained from STI1-deficient embryos also fail to thrive and present increased levels of the DNA damage marker ␥-H2AX, suggestive of increased cellular stress (21). Hence, in mammals STI1 seems to play additional roles in cellular survival that are not yet fully understood.
STI1 is abundantly expressed in the cytoplasm of cells, but can also be found in the Golgi (22), in vesicles and in multivesicular bodies (11). Moreover, this cochaperone has been shown to shuttle between the cytoplasm and the nucleus in cell lines (23). Cellular stress, arrest in G1/S phase of the cell cycle and phosphorylation are factors that seem to regulate STI1 nuclear localization (23,24). Presumably nuclear STI1 can regulate chaperone activity, but whether it can interact with nuclear proteins is unknown.
Previous experiments using cell lines have shown that knockdown of STI1 increases susceptibility of cells to irradiation (25). Whether changes in STI1 levels in primary differentiated cells, such as astrocytes, may affect their response to irradiation stress is unknown. This is of interest, as astrocytes, which can give rise to distinct tumor cells, are highly radioresistant (26). Indeed, astrocytes have a noncanonical DNA damage response (DDR) to irradiation (26). Here we show that STI1 undergoes nuclear translocation in astrocytes after ␥-radiation-induced DNA damage. Moreover, astrocytes haploinsufficient for STI1 are more susceptible to cell death induced by irradiation. To understand potential mechanisms involved with STI1 nuclear retention, we have performed yeast-two hybrid screenings to identify STI1 nuclear partners. We identified protein inhibitor of activated STAT (PIAS1) as a direct interactor of STI1 and provide evidence that it acts as a small ubiquitin-like modifier (SUMO) E3 ligase for STI1. We show this interaction is involved with STI1 nuclear retention after irradiation. Interestingly, tissue microarray analysis demonstrated that higher PIAS1 levels are found in glioblastoma multiforme (GBM) when compared with non-neoplastic tissue. Furthermore, we uncovered a positive relationship between increased PIAS1 expression in GBMs and augmented STI1 nuclear localization. Our results reveal a novel mechanism by which increased expression of PIAS1, as observed in GBM, can increase the retention of nuclear STI1, a critical regulator of the chaperone machinery.

EXPERIMENTAL PROCEDURES
Plasmids-YFP-Ubc9 and FLAG-Pc2 were kindly provided by Dr. David Wotton (Professor of Biochemistry and Molecular Genetics of University of Virginia, Charlottesville, US). The plasmids Myc-PIAS4 and Myc-PIAS1 were kindly provided by Dr. Lienhard Schmitz (Institute of Biochemistry, Medical Faculty Friedrichstrasse, Justus-Liebig University, Giessen, Germany). GFP-STI1 and YFP-STI1 were generated by cloning STI1 in the SalI and KpnI restriction sites, and XhoI and KpnI restriction sites of pEGFP-C1 and pEYFP-N1, respectively (Clontech, Mountain View, CA). PML-YFP was kindly provided by Marc Tini (Department of Physiology and Pharmacology, Schulich School of Medicine and Dentistry, University of Western Ontario). HA-STI1 was constructed by cloning STI1 in pCMVHA (Clontech) using the restriction enzymes SalI and SfiI (BioLabs). Mutants of STI1 were generated in the pCMVHA construct. A mouse N-terminal (bp 1-870) portion of STI1, or smaller fragments within this region, were amplified and ligated in frame with the yeast Gal4 DNA-binding domain (Gal4 DBD ) in the vector pGBKT7 (BD Biosciences, pGBKT7-N-STI1). Full-length and truncated PIAS1 sequences were cloned into pACT2 (Clontech) in frame with the yeast Gal4 transcriptional activation domain (Gal4 AD ). Site-directed mutagensis of PIAS1 was performed by overlap extension PCR.
Mouse Line Generation-Genetically modified mice were generated using standard homologous recombination techniques, using C57BL/6j ES cells. Mice were generated by Ozgene (Australia). Chimeric mice were bred to C57BL/6j mice and germline transmission of the mutant STI1 allele was identified by Southern blot (not shown). F1 mice were crossed to constitutively expressing Cre mice to remove loxP flanked regions. Cre recombined mice were then crossed to C57BL/6j mice and progeny bearing the recombined STI1 allele, but lacking the Cre transgene, were identified by Southern analysis (not shown). These mice were then used to expand the colony. STI1 Ϫ/ϩ mice were then intercrossed to generate STI1 Ϫ/Ϫ mice. These mice presented early embryonic lethality (E10.5) and a range of developmental defects (21). We have attempted to generate mice with a conditional STI1 floxed allele, however we found that after removal of the Neo-cassete this particular floxed allele was null and also caused embryonic lethality (results not shown). Hence, astrocytes could be isolated only from STI1 Ϫ/ϩ mice. Haploinsufficient astrocytes have no difference in their cell cycle and proliferative properties compared with wild-type astrocytes (21).
Primary Culture of Astrocytes-Animals were maintained and handled in the mouse vivarium at University of Western Ontario Animal Care and Veterinarian Service (ACVS). Procedures were conducted in accordance with the animal care guidelines of the CCAC. This study was approved by the AUS committee at the University of Western Ontario (Protocol # 2008 -127). Nineteen day time-pregnant mice (C57BL6/j) were euthanized and embryos were quickly removed and decapitated to isolate cortical astrocytes (10). Cortical hemispheres were dissected and separated from the striatum. Cortices were dissociated in 5 ml Dulbecco's modified Eagle's medium supplemented with 10% (v/v) fetal bovine serum and 1% (v/v) penicillin/streptomycin by pipetting up and down several times with a serological pipette and plated with additional 5 ml of medium in a 100 mm Petri dish. Cultures were maintained in incubator at 37°C, 5% CO 2 . Astrocytes were used after 10 -11 days in vitro.
Cell Culture, Transfections and Co-Immunoprecipitation-CF-10 cells were cultured and transfected as previously described (13). HEK293 cells were grown and transfected as described (14). Fourtheight hours after transfection, cells were suspended in 300 l of lysis buffer containing 50 mM Tris-HCl,150 mM NaCl pH 7.5, with 0.5% or 1% Triton X-100, 20 mM N-ethylmaleimide (NEM) and Protease Inhibitor Mixture (Sigma). A total of 800 -900 g of protein extract were incubated with 50 l of anti-HA agarose-conjugated beads (Sigma); subsequent steps followed the manufacturer's protocols. Cell lines were usually transfected with 80% efficiency. Astrocytes were transfected with 30 -40% efficiency.
Fluorescence Imaging-Confocal microscopy was performed using a Bio-Rad MRC 1024 laser scanning confocal system running the Lasersharp 3.0 software coupled to a Zeiss microscope (Axiovert 100) with a 100ϫ/1.4 or 63ϫ/1.3 oil-immersion lens (Zeiss); a LSM 510 Meta laser-scanning confocal system coupled to a Zeiss microscope or a Leica SP5 laser-scanning confocal microscope using a 63ϫ/1.2 water-immersion or a 63x/1.4 oil-immersion lens (Leica). Image analysis and processing were performed with Lasersharp (Bio-Rad), Confocal Assistant, Adobe Photoshop, Metamorph, Leica Application Suite Advanced Fluorescent Lite, and ImageJ (version 1.24) software.
Irradiation and Nuclear STI1 and Foci Counting-Cells were plated, transfected with constructs of interest and after 48 h the dishes were placed in a Faxitron RX-650 to receive a dose rate of 2 Gy/min. Cells were lysed or fixed for different periods of time after irradiation. Ten random fields of each dish were imaged and then analyzed by number of cells, number of cells with nuclear STI1 (defined as homogenous labeling in the cells, compared with nonirradiated cells that present no staining in the nucleus for STI1) and total number of foci. The results were then normalized by the total number of cells, which was Ͼ200 for each experiment.
Live/Dead Assay-STI1 Ϫ/ϩ and STI1 ϩ/ϩ astrocytes were plated in glass-bottom culture dishes (MatTek, Ashland, MA) irradiated and used 48 h later. Subsequent steps were performed according to the Live/Dead Assay Kit (Invitrogen). Ten random fields were imaged per dish and live or dead cells were then counted with the cell counting tool from Image J software.
LDH Release Assay-STI1 Ϫ/ϩ and STI1 ϩ/ϩ astrocytes were counted and 2 ϫ 10 5 cells were plated on 35-mm Petri dishes. Dishes were irradiated for 10 min at 2 Gy and 48 h later 50% of their medium were collected and cleared from debris by centrifuging. Subsequent steps were done following the manufacturer's protocols (LDH assay Sigma kit #MAK066 -1KT).
Yeast Two-Hybrid Screening-Yeast strain AH109 was transformed with pGBKT7 (Empty vector) or pGBKT7-N-STI1 and transformants were selected on S.D. (synthetic dropout) plates, in which the media lacks Trp. These selected yeast cells were transformed with a mouse brain cDNA library (title 4.2 ϫ 10 7 cfu/ml) fused with Gal4ad in the vector pACT2 (Leu) (Clontech). 1.0 ϫ 10 6 independent clones were screened and yeast cells were selected for growth on Ade/His/Leu/Trp -deficient medium and tested for lacZ expression according to instructions provided by the manufacturer (Clontech). The same STI1 construct was used for another screening using the yeast mating protocol with a BD Matchmaker TM pretransformed human brain cDNA library (title 1.3 ϫ 10 8 cfu/ml) fused with Gal4ad in the vector pACT2. Procedures were carried out according to manufacturer's protocols (Clontech). The Matting efficiency was 9.4% and 3.9 ϫ 10 7 clones were screened. Attempts to express a C-terminal STI1 construct is yeast did not work, so we restricted our analysis to the N terminus.
Yeast two-hybrid assays for mapping the interaction domains of PIAS1 and STI1 were carried out using diploid yeast cotransformants produced by mating yeast strains Y187 and Y2HGold (Clontech) transformed with pACT2-and pGBKT7-based plasmids, respectively.
Surface Plasmon Resonance-Surface plasmon resonance was studied using Biacore X system (GE Healthcare, Pittsburgh, PA) equipped with a CM5 chip. Recombinant STI1 and PIAS1 peptides were produced using pE-SUMOstar Amp Kit (LifeSensors, Malvern, PA) and purified to Ͼ95% purity estimated by SDS-PAGE. STI1 was covalently bound to the chip using standard amine-coupling NHS/ EDC procedure (30) to the level of ϳ8000 response units (RU). Before injections the chip was equilibrated in the running buffer (25 mM HEPES, 150 mM NaCl, pH 8.0). Different concentrations of PIAS1 peptides in the running buffer were injected at 5 l/min rate for 6 min. After that 2-min "off" reactions were recorded followed by washing with running buffer. Between injections the chip was additionally washed at 100 l/min with 1-min injections of 10 mM HCl. The background signal was obtained by injecting the same peptides through a control flow cell with no bound STI1. Binding curves were analyzed with Biacore software and GraphPad Prism 5 (GraphPad Software, San Diego, CA). "On" curves were fitted with a one-site interaction model. Off curves were fitted with an exponential decay model.
Identification of SUMOylation Sites by Mass Spectrometry-For in vitro STI1 SUMOylation, 10 g of His tagged STI1 (mouse), 3 g of SUMO-activating enzyme 1 (Aos1/Uba2) (human recombinant), 25 g of untagged Ubiquitin Conjugating Enzyme UBC9 (SUMO E2) (human), 10 g of His6-tagged SUMO3 mutant protein (human) were added to SUMOylation Buffer containing 0.6 M Mg-ATP. This mixture was incubated at 37°C under agitation for 1.5 h. 8 M urea was then added to the final concentration of 4 M. Proteins were reduced with 5 mM tris-(2-carboxyethyl)-phosphine (TCEP, Pierce) for 20 min at 37°C and then alkylated with 50 mM chloroacetamide (Sigma-Aldrich) for 20 min at 37°C. Excess chloroacetamide was neutralized by addition of 50 mM dithiothreitol. The solution was diluted to 1 M urea with 50 mM ammonium bicarbonate and digested overnight with modified trypsin (1:50, enzyme:substrate ratio) at 37°C with high agitation. The digest was acidified with trifluoroacetic acid (TFA), desalted using an Oasis HLB cartridge (Waters, Milford, MA) and dried in a speed vacuum concentrator.
LC-MS/MS analyses were performed on a nano-LC 2D pump (Eksigent) coupled to a LTQ-Orbitrap Velos mass spectrometer via a nanoelectrospray ion source (Thermo Fisher Scientific). Peptides were loaded on an Optiguard SCX trap column (5 m particle, 300Å, 0.5 ID ϫ 23 mm, Optimize Technologies) and eluted on a 360 m ID ϫ 4 mm, C 18 trap column before separation on a custom-made 150 m ID ϫ 10 cm nano-LC column (Jupiter C 18 , 3 m, 300 Å, Phenomenex). Tryptic digests were loaded on the SCX trap and sequentially eluted using salt plugs of 0, 250, 500, 750 mM, 1, and 2 M ammonium acetate, pH 3.5. Peptides were separated on the analytical column using a linear gradient of 5-40% acetonitrile (0.2% formic acid) in 53 min with a flow rate of 600 nL/min. The mass spectrometer was operated in data dependent mode to automatically switch between survey MS and MS/MS acquisitions. The conventional MS spectra (survey scan) were acquired in the Orbitrap at a resolution of 60,000 for m/z 400 after accumulation of 10 6 ions in the linear ion trap. Mass calibration used a lock mass from ambient air [protonated (Si(CH 3 ) 2 O)) 6 ; m/z 445.120029], and provided mass accuracy within 5 ppm for precursor and fragment ion mass measurements. MS/MS spectra were acquired in HCD activation mode using an isolation window of 2 Da. Precursor ions were accumulated to a target value of 30,000 with a maximum injection time of 100 ms and fragment ions were transferred to the Orbitrap analyzer operating at a resolution of 15,000 at m/z 400. The dynamic exclusion of previously acquired precursor ions was enabled (repeat count 1, repeat duration: 30 s; exclusion duration 45 s).
MS data were acquired using the Xcalibur software (version 2.1 build1139). Peak lists were generated using Mascot distiller (version 2.3.2.0, Matrix science) and MS/MS spectra were searched against the IPI human database containing 75429 forward sequences (version 3.54, released Jan 2009) using Mascot (version 2.3.2, Matrix Science) with a mass tolerance of 10 ppm for precursor ions and 0.05 Da for fragments. The number of allowed missed cleavage sites for trypsin was set to 3 and oxidation (M), deamidation (NQ), carbamidomethylation (C) and SUMOylation (K) (GGTQN: SUMO3) were selected as variable modifications. A software application was developed to search mascot generic files (mgf) for specific SUMO3 fragment ions (e.g. m/z 132.0768, 226.0822, 243.1088, 344.1565, 401.1779; and neutral losses of SUMO3 remnant) to produce a mgf file containing only MS/MS spectra of potential SUMOylated peptide candidates. SUMO fragment ions were removed from the corresponding mgf files and searched again using Mascot as indicated above. Manual inspection of all MS/MS spectra for modified peptides was performed to validate assignments.
Mutations-About 50 -100 ng of pCMV5-HA STI1 were used to generate point mutations or deletion mutants. Primers were designed using Stratagene Software tool for QuikChange® Multi Site-Directed Mutagenesis Kit. PCR mix, cycle parameters and transformation in competent cells were done following the kit manufacturer's protocols. Sequence of primers is available on request. All mutants were confirmed by sequencing.
Quantification of Fluorescence-Intracellular distribution of STI1 was evaluated using the MetaMorph software. Total fluorescence in the nucleus was measured and normalized by the total of fluorescence of the cytoplasm. The area of each cell was considered during analyzes. Regions of interest were determined and total fluorescence was detected automatically and independently by the software. Results were expressed as the mean of total fluorescence in the nucleus/ cytosol per cell. Colocalization analysis was evaluated using Image J software. Each image first had its channels split; cells nuclei were the chosen region of interest (ROI) and backgrounds were subtracted before colocalization measurement. The software detects the pixels that colocalize between the channels and calculates the Manders M coefficient for each image, which identifies the percentage of pixels on one channel overlapping with the other.
siRNA Transfection-A total of 0.8 g of PIAS1 siRNA (NM_ 016166) (Qiagen) and fluorescent nonrelevant siRNA controls (Invitrogen) were used to transfect astrocytes with the Effectene transfection kit (QUIAGEN). For these experiments cells were incubated with the reagents for 24 h and used 72 h after transfection.
Patients and Tissue Microarrays-Tissue microarrays (TMA) were prepared from formalin-fixed, paraffin-embedded GBMs (from Department of Pathology of AC Camargo Cancer Center) and 14 nonneoplasic samples (from surgically remediable patients with intractable mesial temporal lobe epilepsy from the Department of Pathology, Federal University of Parana, Brazil). Samples were prepared after approval from AC Camargo Cancer Center Research and Ethics Committee (Process 1613/11 and 1692/12). From May 1980 to December 2004, 91 patients with GBMs underwent surgical treatment at the Neurosurgery Department of AC Camargo Cancer Center. Analysis of PIAS1 and STI1 expression was possible in 87 cases (60% males; mean age 53.5 Ϯ 18.3 y). None of these patients were treated before surgery.
Statistical Analysis-Mean values of at least three independent datasets are shown in the figures; the error bars represent S.E. Student's t test was used to compare two groups. ANOVA followed by appropriate post-hoc tests was used for multiple comparisons. PIAS1 expression and STI1 nuclear localization in samples from patients with gliblastomas were correlated using Pearson's correlation coefficient. For all tests, results were considered statistically significant when p value was Ͻ 0.05.

STI1 Accumulates in the Nucleus of Astrocytes in Response
to ␥-Radiation-Previous experiments have shown that increased cellular stress by heat shock, or treatment with hydroxiurea, which arrests cell cycle, causes nuclear retention of STI1 in cell lines (23,24). To test the possibility that STI1 participates in the response to genotoxic stress in astrocytes, we used ionizing radiation (IR, 2Gy/10min). To monitor DNA damage in response to IR we measured phosphorylated histone H2AX (␥-H2AX S139), a well-established marker of double-strand breaks (DSBs) and accumulation of nuclear foci (31), as this is the most suitable marker for DNA damage in astrocytes (26). We found that IR led to fast recruitment of STI1-YFP (Fig. 1A) or endogenous STI1 to the nucleus (Fig.  1B), detected as increased proportion of astrocytes presenting nuclear STI1 labeling (Fig. 1C). This increased nuclear localization of STI1 persisted up to 24 h after IR treatment. Concomitant increase of ␥-H2AX foci was also observed (Fig.  1). Therefore, in support of previous reports using non-neuronal cells and other types of cellular stress (23,24), nuclear localization of STI1 is increased in irradiated astrocytes.
STI1 Levels Regulate Astrocyte Cell Death After Irradiation-To test if STI1 could have a role in the responses of astrocytes to IR, we used astrocytes obtained from STI1 heterozygous knockout mice (21), which present 50% reduction in STI1 levels ( Fig. 2A). By using Live-Dead staining we observed that wild-type astrocytes showed a small, but significant increase in cell death in response to IR (2Gy/10 min, Fig. 2B and 2C). Interestingly, STI1-deficient astrocytes presented increased cell death in response to irradiation when compared with control astrocytes (Fig. 2B and 2C). We also measured cell death by determining lactate dehydrogenase (LDH) leakage from dying cells. In this experiment we could not detect a significant increase in LDH leakage from wildtype astrocytes in response to irradiation (Fig. 2D), likely because of lower sensitivity of this method. In contrast, astrocytes deficient for STI1 showed increased LDH leakage in response to ␥-radiation (Fig. 2 D).
Increased nuclear localization of STI1 in response to irradiation suggested a potential role for this protein in genotoxic stress response in astrocytes. To identify potential nuclear targets of STI1, two independent yeast-two hybrid screenings were performed, one using a mouse and the other using a human brain cDNA library. As expected, a number of positive clones encoded for heat shock proteins of the Hsp70 family (Table I), indicating that we were able to identify known interactors of STI1 in the distinct screenings. Interestingly, in both screenings we identified several genes coding for nuclear proteins. These include UBC9, the unique E2 SUMO conjugation enzyme (32); and all members of the PIAS family (PIAS1, PIAS2, PIAS3 and PIAS4), which have SUMO E3 ligase activity and also function as transcription and repression factors (33,34). Pc2, a polycomb protein that has E3 SUMO ligase activity and is involved with DNA repair (35), was also isolated in the screening (Table I). Although other cDNAs were also identified, we focused further work on genes related to the SUMOylation pathway.
To investigate the functional significance of these potential interactions we examined the colocalization of these proteins with STI1-YFP fusion constructs (36) in CF-10 cells because of their neuronal-precursor like features (13). STI1-YFP is localized mainly in the cytoplasm of CF-10 cells (Fig. 3A), as previously described in other cell lines (23). In contrast, UBC9, Pc2 as well as PIAS1 and PIAS4 (Fig. 3A), used as representatives of the PIAS family, show a predominant nuclear localization. Interestingly, cotransfection of PIAS1 with STI1-YFP led to increased nuclear localization of STI1-YFP in cells (Fig.  3B) and colocalization between STI1 and PIAS1. In contrast, transfection of PIAS4, Pc2 and UBC9 did not affect the localization of STI1 (Figs. 3C, 3D, and 3E respectively). Quantification of fluorescence intensity showed that the nuclear/cytoplasmic (N/C) STI1 ratio was significantly increased with overexpression of PIAS1, but not PIAS4 (Fig. 3F).
Localization of STI1, PIAS1 and UBC9 in astrocytes showed essentially the same pattern observed in CF-10 cells (compare Fig. 3 to Fig. 4). Moreover, overexpression of PIAS1 resulted in increased localization of STI1-GFP in the nucleus and colocalization of PIAS1 with STI1-GFP is astrocytes (Fig.  4B). Nuclear retention of endogenous STI1 was also increased by overexpression of PIAS1, but not by overexpres- PIAS1 has been shown to be present in PML bodies, which are sites of protein-protein interaction in the nucleus (37)(38)(39)(40).
To determine if STI1 could also be present in PML bodies in PIAS1 overexpressing astrocytes we transfected these cells with PIAS1 and PML-YFP and monitored the localization of endogenous STI1. The results show that part of the endogenous STI1 accumulates in PML bodies only in PIAS1 transfected cells (Figs. 5A and 5B, arrowheads). Quantification of FIG. 2. STI1 levels influence astrocyte death after irradiation. A, Western blot representing STI1 expression in primary culture of astrocytes obtained from STI1 ϩ/ϩ and STI1 Ϫ/ϩ mice. B, Primary culture of astrocytes obtained from STI1 ϩ/ϩ and STI1 Ϫ/ϩ mice were submitted to 2Gy of gamma-radiation for 10 min (20Gy of IR). Cells were stained 48 h later by using live/dead kit (green live cells/red dead cells) or had their cultured media collected for LDH release assay after 48 h. Representative images of astrocytes stained using the live/dead kit in the indicated conditions. C, Quantification of cell death in four independent experiments. D, LDH release in four independent experiments for the indicated conditions. Data were analyzed and compared by one-way ANOVA (p ϭ 0.0002) with Newman-Keuls post hoc test ***, p Ͻ 0.0001 **, p Ͻ 0.001 and *, p Ͻ 0.01. images from multiple cells showed colocalization of STI1 and PML bodies in the presence of PIAS1 (0.28 Ϯ 0.03, Mean Ϯ S.E.) when compared with cells in which PIAS 1 was not overexpressed (0.09 Ϯ 0.02, Mean Ϯ S.E.; p Ͻ 0.001 when the two colocalization indexes are compared). Moreover, because STI1 can biochemically and functionally interact with Hsp90 we examined the localization of this chaperone in astrocytes. STI1-GFP and Hsp90 show extensive colocalization in astrocytes mostly in the cytoplasm (Fig. 5C). Remarkably, increased expression of PIAS1 caused both STI1 and Hsp90 to be in the nucleus (Fig. 5D). Furthermore, colocalization analysis in several images from distinct dishes suggest that STI1, Hsp90 and PIAS1 colocalize in the nucleus of PIAS1 overexpressing astrocytes (Manders colocalization coefficient of M1 ϭ 0.92 Ϯ 0. 10  PIAS1 is a SUMO E3 ligase and it is well known that SUMOylation can increase nuclear localization and retention of proteins (41)(42)(43). We envisioned two potential mechanisms by which augmented expression of PIAS1 could increase the nuclear localization/retention of STI1: (1) PIAS1 could SUMOylate STI1, which would lead to its increased nuclear retention; (2) PIAS1 and STI1 could interact directly and this interaction would serve to retain nuclear translocated STI1 into the nucleus. To discriminate between these two possibilities we first determined if STI1 could be SUMOylated. We performed in vitro SUMOylation assays using recombinant STI1, UBC9 and SUMO1, SUMO2, and SUMO3. ATP-dependent SUMOylation of STI1 in vitro was observed as an upward shift in the apparent molecular mass of the protein, with several SUMOylated species of STI1 being detected (Fig.  6A, arrowheads, small protein fragments lower than STI1 are likely proteolytic products). This experiment showed that UBC9 can poly-SUMOylate STI1 in vitro in the absence of any E3-ligase and that SUMO1, 2, and 3 can all be used to SUMOylate STI1 (Fig. 6B). To further investigate STI1 SUMOylation in cells, and test the possibility that PIAS1 could function as E3-ligase for STI1, we conducted assays in HEK 293 cells overexpressing SUMO and PIAS1. Extracts from cells cotransfected with SUMO1, PIAS1 and STI1-HA were immunoprecipitated using an antibody against HA and immunoprecipitated material (Fig. 6C) as well as cell lysates (Fig.  6D) were analyzed by Western blotting. Immunoprecipitates from cells transfected with STI1 and SUMO1 showed SUMOylated STI1 species (Fig. 6C), which were more pronounced when PIAS1 was cotransfected with SUMO1 and STI1. Similar results were obtained when cells were transfected with STI1 and SUMO3 (not shown). STI1 SUMOylated bands were also detected with SUMO antibodies (arrowheads Fig. 6C), suggesting that indeed PIAS1 can act as an E3ligase for STI1. PIAS1 and presumably SUMOylated PIAS1 were also coimmunoprecipitated with STI1 in this condition (Fig. 6C, anti-Myc). The different species of SUMOylated STI1 could be easily and specifically identified in cell extracts using the STI1 antibody (Fig. 6D, anti-STI1).
STI1 SUMOylation Sites-Our data indicated that STI1 can be poly-SUMOylated, therefore we used bioinformatic analysis to identify potential STI1 SUMOylation sites. Our analysis indicated the presence of 62 potential SUMOylation consensus site (using SUMOsp 2.0 software). To determine the precise location of modified lysine residues on STI1, we performed an in vitro SUMOylation assay using a His-tagged form of SUMO3 that facilitates the identification of SUMOylation sites (44). STI1 was incubated in the presence of UBC9, SAE1/2, Mg/ATP, and this SUMO3 form, and proteins were subsequently digested with trypsin and analyzed by LC-MS/MS to identify modified tryptic peptides bearing the remnant side chain from the mutant SUMO3. Confirmation of modified residues was obtained from the high resolution HCD MS/MS  Fig. S1A-S1E). These analyses indi-cated that STI1 is SUMOylated in vitro at lysine residues 123, 210, 312, 395, and 486. All these sites are found in regions predicted to be solvent exposed. We then generated a series of lysine (K) to arginine (R) STI1 mutants, including point mutations of each of the SUMOylation sites above, as well as combinations of these mutations. We compared SUMOylation of mutants with SUMOylation of STI1-HA in HEK 293 cell lysates. For these experiments, HEK 293 cell were transfected with SUMO1 and PIAS1, as STI1 is similarly SUMOylated by both SUMO1 and SUMO3 (Fig. 6A). To detect only transfected mutated STI1 (mutant or wild-type), we used an anti-HA antibody for the immunoblots. The anti-HA antibody was less effective than the STI1 antibody to detect minor SUMOylated species of STI1, and only the three stronger SUMOylated STI1 bands were detected (supplemental Figs. S1F and S1G). We detected three main protein bands; whereas the first and second bands ran close together around 120 kDa, the third band showed higher molecular mass (200 kDa). The K123R mutant showed a selective decrease in the first and third bands, whereas the second band seemed preserved (supplemental Fig. S1F). In contrast, the intensity of the three SUMOylated STI1 bands was weaker in all the mutants in which K210 was changed (supplemental Figs. S1F and S1G), suggesting that K210 may provide hierarchal control of STI1 SUMOylation. Individual mutants K312R, K395R, and K486R did not show large effects on SUMOylated STI1 species (supplemental Fig. S1G). A mutant containing changes in these five amino acids (K123-210-312-395-486R) presented impaired SUMOylation (supplemental Fig. S1F, arrow, note that the HA antibody does recognize a nonspecific band that migrates close to the SUMOylated STI1 bands).
Following the identification of STI1 SUMOylation sites we assessed whether SUMOylation was responsible for the increased nuclear retention of STI1 on PIAS1 overexpression. For this experiment, we used the STI1 construct with all SUMOylation sites mutated (K123-210-312-395-486R). HAtagged wild-type STI1 and the mutant STI1 presented similar cytoplasmic localization in astrocytes (Fig. 7A). PIAS1 overexpression increased the nuclear retention of both wild-type and mutant STI1 (Figs. 7B and 7C). Hence, preventing STI1 SUMOylation did not affect its nuclear accumulation, sug-gesting the possibility that a direct interaction between STI1 and PIAS1, rather than SUMOylation, is responsible for the increased STI1 nuclear localization.
STI1 and PIAS 1 Interact Directly-Yeast-two hybrid screening (Table I) suggested that STI1 and PIAS1 interact directly. Additionally, cotransformation of yeast with a construct containing the N-terminal region of STI1 and PIAS1 cDNA allowed cell growth in restrictive media, whereas no growth was observed in yeast transformed with the empty vector pGBKT7 and PIAS1-pACT2 (not shown), further supporting a direct interaction between these two proteins. To map the interaction sites we generated STI1 and PIAS1 deletion constructs. Multiple regions of STI1 containing the TPR domains were sufficient for interaction with PIAS1 (Fig. 8A). In contrast, a specific region in PIAS1 (a.a. 450 -480) containing the SUMO-interacting motif (SIM), was required for the inter-FIG. 6. STI1 is SUMOylated and PIAS1 is a SUMO E3 ligase for STI1. Recombinant STI1 was subject to an in vitro SUMOylation assay by using recombinant SUMO1, SUMO2, SUMO3, and the SUMO conjugating enzymes SAE1/2 and UBC9. A, The higher molecular weight bands detected for STI1 (arrowheads) in the presence of ATP and SUMO1, 2, and 3, indicate that all SUMO isoforms are able to SUMOylate STI1 in an ATP dependent process. B, UBC9 expression (arrowhead). C, HEK293 cells were cotransfected with HA-STI1, SUMO1 (or SUMO3not shown) and PIAS1. STI1 was pulled down using HA agarose beads and blots were processed. IP of STI1 in cells expressing PIAS1 and SUMO1. Arrowheads indicate SUMOylated STI1. D, Lysates of cells used in C. The data are representative of four experiments. action with STI1 in the yeast-two hybrid assay (Fig. 8B). Substitution of alanine for each of the hydrophobic residues (V457, V459, I460, L462) in the SIM domain of full-length PIAS1 abolished interaction with STI1, whereas mutation of other residues flanking the SIM within aa 450 -480 in PIAS1 did not affect the interaction (Fig. 8C). These results indicate that in the yeast two-hybrid assay, the SIM in PIAS1 is critical for interaction with STI1. To further confirm the direct interaction of STI1 with PIAS1 we obtained recombinant proteins in bacteria and used surface plasmon resonance (SPR). Full length PIAS1 interacted with full length STI1 immobilized on the SPR chip in a concentration-dependent way (Fig. 8D, . A C-terminal peptide (amino acids 450 -651) containing the SIM domain of PIAS1 interacted with full length STI1 with similar affinity (Fig. 8E, K D ϭ 2.3 Ϯ 0.3 M). In contrast, a peptide containing only the RING motif of PIAS1 did not interact with full length STI1 (Fig. 8F). These results suggest that the two proteins can interact directly and that the C-terminal region of PIAS1 containing the SIM motif is sufficient for this interaction. Our observation that PIAS1 coimmunoprecipitates with STI1 in HEK293 cells (Fig. 6C) gives further support for this interaction. To test if STI1 could form a complex with PIAS1 in mammalian cells in the absence of transfected SUMO, we transfected HEK 293 cells only with STI1-HA and PIAS1-Myc. Pull-down of STI1 with HA-beads co-immunoprecipitated PIAS1 (Fig. 8G). PIAS1 was not coimmunoprecipitated from HEK 293 cells with anti-HA antibod-ies when cells were not transfected with STI1-HA, demonstrating specificity.
PIAS1 is Required for Nuclear Retention of STI1 After Genotoxic Stress in Astrocytes-To further determine if the nuclear accumulation of STI1 was indeed because of the interaction with PIAS1, we generated a PIAS1 mutant lacking amino acids 450 -480 (PIAS1-Myc⌬450 -480). Localization of the PIAS1 mutant was indistinguishable from that of wild-type PIAS1 (Fig. 9A). Importantly, whereas overexpression of PIAS1 induces nuclear accumulation of STI1, overexpression of the mutant does not (Figs. 9B and 9C).
To test whether PIAS1 plays a role in the nuclear retention of STI1 in response to IR, we used siRNA to knockdown PIAS1 in astrocytes (Fig. 10). siRNA transfection decreased PIAS1 expression by 75% and abolished immunostaining of PIAS1 in the nucleus of astrocytes (Figs. 10B and 10C, respectively), whereas a control scrambled siRNA had no effect on PIAS1 expression when compared with nontransfected astrocytes (Figs. 10A-10C). Irradiated astrocytes transfected with PIAS1 siRNA presented STI1 mainly in the cytoplasm (Fig. 10D). A correlation analysis indicated a relationship between PIAS1 levels and STI1 nuclear localization (Fig. 10E). These results indicate that retention of nuclear STI1 after irradiation depends on its interaction with PIAS1 in the nucleus.
GBMs Express Higher Levels of PIAS1 and Nuclear STI1 than Non-neoplasm Samples-To evaluate the potential significance of increased PIAS1 expression and STI1 nuclear localization in astrocytes, we evaluated these proteins in GBM, the most aggressive and common CNS tumor in adults. GBMs present altered DDR (45) and are fairly resistant to irradiation (46). IHC analysis showed that PIAS1 is expressed only in the nucleus of GBM samples (Fig. 11A), whereas STI1 is highly expressed in the cytoplasm of non-neoplasm and tumor cells (Fig. 11C). Remarkably, GBMs presented higher PIAS1 expression (Fig. 11B) and an increment of STI1 nuclear localization (Fig. 11D), when compared with normal brain tissue. Furthermore, there was a significant positive correlation among GBMs with higher levels of PIAS1 expression and increased STI1 nuclear localization (r ϭ 0.364, p Ͻ 0.0005). These findings are consistent with our observations in astrocytes suggesting that PIAS1 can modulate nuclear STI1 localization. DISCUSSION We show that decreased levels of STI1 increase the sensitivity of astrocytes to ␥-radiation-induced cell death, suggesting a role for STI1 in response to this specialized form of stress. Our data show that STI1 is retained in the nucleus in response to ␥-irradiation and we identified several nuclear proteins that can interact with STI1. Functional overexpression assays identified PIAS1, a critical nuclear protein that regulates the DDR and gene transcription (47,48), as a direct interactor for STI1. Although PIAS1 has E3-ligase activity toward STI1, we demonstrated that STI1 nuclear accumulation is independent of its SUMOylation, but it seems to depend on the interaction between STI1 and PIAS1. This interaction requires multiple domains on STI1 and the SIM domain on PIAS1. Our biochemical experiments however show that the two proteins can interact independently of SUMOylation. Importantly, localization of STI1 in the nucleus in response to ␥-irradiation depends on PIAS1, suggesting that PIAS1 is a novel interacting partner for STI1 that promotes its nuclear retention during genotoxic stress.
It seems that a number of proteins involved in signaling for canonical DDR response, such as ATM and p53, play less important roles in astrocytes when compared with cancer cells (26). Furthermore, it has been shown that astrocytes do not activate apoptosis in response to irradiation (26). Although we did not examine apoptosis, our results are compatible with these previous observations, and show that wild-type astrocytes are indeed highly resistant to IR. Using the Live-Dead assay we were able to detect a small, but significant increase in astrocyte cell death after IR. Importantly, astrocytes isolated from STI1 heterozygous knockout mice showed increased cell death after ␥-irradiation when compared with wild-type astrocytes, suggesting the possibility that STI1 is involved in modulating cell survival in response to DNA damage. These results are in agreement with previous observations showing that knockdown of STI1 specifically sensitizes YAC-1 cells (mouse lymphoma cells) to irradiation mediated cell death (25).
Nuclear accumulation of STI1 in response to IR in astrocytes was fast (within minutes), persisted for at least 24 h and was dependent on PIAS1 expression. Interestingly, although it is known that STI1 accumulates in the nucleus of cell lines in response to other types of cellular stress (23), its nuclear functions are poorly understood. STI1 has both nuclear localization signals (e.g. a.a. 222-239) and nuclear export signals (23). Inhibition of nuclear export of proteins with leptomycin B increases the localization of STI1 in the nucleus, suggesting that the protein normally shuttles between nuclear and cytoplasmic locations (23).
Nuclear STI1 did not appear to be initially recruited to DNA breaks, as it did not colocalize with ␥-H2AX foci (Fig. 1), a marker of DNA damage. Thus, the pattern of STI1 nuclear localization is not completely identical to that of PIAS1, which is recruited to sites of DNA damage (48). However, PIAS1 has also been found to colocalize with PML in PML nuclear bodies (NB) (49); thus the STI1-PIAS1 interaction may serve to direct STI1 to specific locations in the nucleus (such as PML NBs) or to specific client proteins. In agreement with the potential for STI1 to act as a cochaperone when recruited by PIAS1, Hsp90 was also recruited to the same sites. STI1 was previously reported to function in DNA damage/ repair pathways as a cochaperone for CHK1 (50) and the DNA repair protein Ssl2, the yeast homolog of XPB (51), a DNA helicase that can function in DNA repair pathways. Interestingly, a recent report pointed out that CHK1 activation in Fig. 8. STI1 interacts with PIAS1. A-C, Yeast cotransformed with plasmids directing expression of the indicated fusion proteins, or vector with no insert (-), were patched onto solid medium lacking histidine and supplemented with the histidine analog 3-amino-1,2,4-triazole (3-AT), and interaction of the fusion proteins was assessed based on growth of yeast after incubation at 28°C for 5 days. Positions of truncation points or point mutations relative to functional domains in PIAS1 and STI1 are indicated. D, Kinetics of PIAS1 peptides binding to STI1 immobilized on a CM5 chip were followed by surface plasmon resonance (SPR) signal using Biacore X system (GE, USA). Graph showing binding curves for the full-length PIAS1 (a. a. 1-651). Darkening shades of gray represent increasing (0.31 to 5 M) concentrations of full-length PIAS1. "On" and "off" kinetics were analyzed as described in Methods to give K D ϭ 2.1 Ϯ 0.   10. STI1 retention in the nucleus after genotoxic stress is dependent on PIAS1 expression. A, Expression of PIAS1 in control astrocytes, astrocytes transfected with PIAS1 or control siRNA. B, Western-Blot quantification of PIAS1 expression in siRNA-transfected astrocytes. C, PIAS1 expression determined by immunolocalization in astrocytes transfected with the indicated siRNA. Green staining represents anti-␤ tubulin used to determine cell morphology. Note that transfection of control siRNA has no effect on PIAS1 expression. D, Cellular distribution of STI1 and expression of PIAS1 in control or irradiated astrocytes, the cells presented a tranfection efficiency of 30%. E, Irradiated cell transfected with siRNA for PIAS1 were quantified for nuclear STI1 and PIAS1 expression and a correlation between the levels of PIAS1 expression and nuclear STI1 was determined. Nuclear localization of STI1 was correlated to decreased PIAS1 expression. R 2 ϭ 0.38, Pearson r ϭ 0.6164 and p ϭ 0.0064. Data are representative of three independent experiments for each condition in which Ͼ50 cells were imaged and analyzed. Scale bars, 25 m. response to DNA damage is dependent on the integrity of PML NBs (52). Although the failure to activate CHK1 in response to DNA damage may be as a result of impaired ssDNA formation in PML depleted cells (52), it is tempting to speculate that PML NBs localization could facilitate STI1 interaction with CHK1. PML NBs are populated by various factors involved in DNA repair, cell cycle checkpoint and apoptosis and have emerged as important regulators of the DDR (38,53). Several reports have described the relocalization of a number of proteins to PML NBs in response to DNA damage, including p53 (54), IKK family member IKK (55) and Mre11 (56). We have recently shown that loss of STI1 causes a decrease in the levels of p53 (21). Although functioning in distinct pathways, these proteins all have in common a protective function against the threat of DNA damage. PML bodies serve also as sites for accumulation of SUMOylated proteins. We show that STI1 is SUMOylated and defined five SUMOylation sites on STI1. The precise role of STI1 SUMOylation on STI1 cochaperone activity will need to be studied in the future. However, STI1 SUMOylation is not required for its nuclear retention. It is likely that on recruitment of STI1 to the nucleus, the interaction of STI1 with PIAS1 will allow for SUMOylation of the former, which may help STI1 to regulate its nuclear interaction network.
Consistent with the possibility that STI1 may serve as a cochaperone in the nucleus, we found that Hsp90 colocalized with nuclear STI1. Interestingly, elimination of embryonic STI1 is lethal likely because of loss of Hsp90 client proteins such as Grk2, STAT3 and p53 (21), without changes in the levels of Hsp90. Nuclear Hsp90 has recently emerged as a regulator for the DDR (57). Phosphorylation of Hsp90 by IR led to its accumulation in ␥-H2AX foci and knockdown of Hsp90 led to less efficient DNA repair as well as a decrease in ␥-H2AX foci with both events occurring concomitantly 3 h after irradiation (57). Moreover, BRCA1, a critical protein involved in DSBs has recently been shown to be a Hsp90 client (58).
There are different subtypes of human brain tumors, however GBMs are the most aggressive and common tumors in the adult CNS. These tumors are highly proliferative, present diffuse infiltration, propensity for necrosis and robust neovascularization. Despite their histological similarities GBMs have recently been suggested to represent heterogeneous diseases (59). Therapeutic options to approach these tumors increased in the last years, nevertheless the prognosis is still poor (60). We show here that GBMs express higher levels of PIAS1. In agreement with results in mouse astrocytes, we detected a positive correlation between increased levels of PIAS1 and increased nuclear STI1 in GBMs, suggesting the possibility of pathological relevance for the interaction we described. It is known that mTOR signaling is one of the key pathways altered in GBMs and resistance to mTOR inhibitors has been associated to PML overexpression (61). Nuclear PMLs can be regulated by arsenic trioxide, a drug used to treat acute promyelocytic leukemia (62) and also proposed to be used in GBMs (61). This agent is known to induce PML-NB SUMOylation followed by polyubiquitination and proteasomal degradation (62) or to modulate sequestration of PMLs in cytoplasmic complexes of PML and nucleoporins (63). Our data showed that STI1 colocalizes with PML-NB when PIAS1 is overexpressed. It is unknown if the complex STI1/PIAS1 can modulate PML-NB SUMOylation or have any chaperone activity on PML-NB, but this possibility deserves further investigation.
The precise role for nuclear STI1 in irradiation and particularly its consequences in GBM treatment with radiotherapy will need to be further explored, but few possibilities can be envisioned. Some of the subnuclear structures, such as PML NB, that have crucial functions in the DDR present high levels of accumulated proteins (40). Having functional Hsp90-STI1 chaperone machinery in the nucleus may allow higher concentration of client proteins in these structures without promiscuous interactions or formation of higher order aggregates. Indeed, STI1 has been shown in yeast to cure prions via Hsp104 (64), a mechanism that is not well understood, but may depend at least in part on protein disaggregation. Moreover, in the absence of embryonic STI1 cells present increased ␥-H2AX sites and commit to apoptosis (21). In agreement with the notion that STI1 may indirect affect DNA repair, p53, a Hsp90 client, is decreased by 50% in embryos lacking STI1 (21). Interestingly, interference with the interaction between STI1 and Hsp90 induces cytotoxicity in GBM cell lines, by reducing levels of p53 and other Hsp90 client proteins (65). STI1 may also serve as a scaffold for bridging Hsp90 to other proteins, as recently suggested for regulation of Piwi and its piRNAs in canalization mechanisms in Drosophila (66). Our results provide a novel mechanism for interplay among STI1, Hsp90, PIAS1, and DNA damage in astrocytes. Further identification of interactions of STI1 and nuclear proteins may shed new light on multiple functions of STI1 as a cochaperone protein and scaffolding molecule.