Evidence for involvement of HMGB1 protein in human DNA mismatch repair.

Defects in human DNA mismatch repair predispose to cancer, but many components of the pathway have not been identified. We report here the identification and characterization of a novel component required for mismatch repair in human cells. A 30-kDa protein was purified to homogeneity by virtue of its ability to complement a depleted HeLa extract in repair of mismatched heteroduplexes. The complementing activity was identified as HMGB1 (the high mobility group box 1 protein), a non-histone chromatin protein that facilitates protein-protein interactions and recognizes DNA damage. Evidence is also presented that HMGB1 physically interacts with MutSalpha and is required at a step prior to the excision of mispaired nucleotide in mismatch repair.

applied onto an Amersham Biosciences Superdex 200 (S200) gel filtration column (24 ml) equilibrated with buffer A containing 0.15 M KCl and 10% glycerol. The column was developed with the same buffer, and nearly homogenous complementing activity (ϳ85 g) eluted with volumes of 15-16.5 ml.
Mismatch Substrates and Repair Assays-DNA heteroduplexes used in this study contained a strand break either 5Ј (5Ј substrate) or 3Ј (3Ј substrate) to the mismatch (Fig. 1). The 5Ј substrates were derived from f1MR phage series by hybridizing Sau96I-linearized double-stranded DNA of an f1MR phage with ssDNA of another phage as described previously (42). A G-T mismatched heteroduplex containing a 36-nt gap 5Ј to the mismatch was also constructed in a way similar to the construction of the nicked substrate, except that the gap was created by digesting f1MR3 double-stranded DNA with Sau96I and DrdI (Fig. 1, SspI fragment) before annealing with f1MR1 ssDNA. The 3Ј G-T substrate was derived from the M13mp18-UKY phage series by hybridizing PstI-linearized double-stranded DNA from M13mp18-UKY1 with ssDNA from M13mp18-UKY2 as described (43). In all cases, the mismatch was replaced in overlapping recognition sites for two restriction endonucleases ( Fig. 1), so that the correction of the mismatch can be monitored by these restriction enzymes.
Unless otherwise specified, MMR complementation assays were performed in reactions containing a receptor (30 g of FII and 150 ng of RPA), 2-4-l individual column fractions, 100 ng (24 fmol) heteroduplex DNA, 10 mM Tris-HCl, pH 7.6, 5 mM MgCl 2 , 1.5 mM ATP, and 0.1 mM dNTPs in the presence or absence of 0.5 l of [␣-32 P]dCTP (3000 Ci/mmol) as described (13). After incubation at 37°C for 15 min, the DNA samples were recovered by phenol extraction and ethanol precipitation and digested with restriction enzymes to score repair. The reaction products were analyzed by 1% agarose gel electrophoresis and visualized by UV illumination for reactions without [␣ 32 P]dCTP or by autoradiography for reactions with [␣-32 P]dCTP. Repair rate (relative repair) was calculated by dividing the amount of repair products in individual complementation reactions with the amount of repair products in the reaction containing 50 g of HeLa nuclear extracts, where ϳ40% of heteroduplexes (100 ng) are repaired.
Immunoprecipitation and Western Blot-All of the immunoprecipitation steps were performed at 0 -4°C as described (32) with minor modifications. Nuclear extract (400 g) was incubated with 5 l of protein A-positive Staphylococcus aureus for 30 min and recovered by centrifugation (16,000 ϫ g for 2 min). The clarified nuclear extract was added to a reaction containing 20 mM Tris-HCl, pH 7.6, 50 g/ml bovine serum albumin, 5 mM MgCl 2 , 1.5 mM ATP, 1 mM dithiothreitol, 1 mM glutathione, 0.1 M KCl, and 1 g of double-stranded f1MR3 DNA (homoduplex) or G-T heteroduplex DNA, followed by an overnight incubation with 3 g of antibodies against HMGB1 (BD Transduction Laboratories), MLH1 (PharMingen) or MSH2 (Oncogene Sciences). After incubation with 15 l of prewashed protein A-Sepharose (Roche Applied Science) for 1 h, the precipitates were recovered by centrifugation (350 g, 2 min), washed three times with fresh Nonidet P-40 lysis buffer (50 mM Tris-HCl, pH 8.0, 0.5% Nonidet P-40, 0.15 M NaCl, 5 mM EDTA, 1 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, and 1 mM NaVO 4 ), and resuspended in 1.5ϫ SDS loading buffer. The samples were resolved in 12.5% SDS-polyacrylamide gels and transferred onto nitrocellulose membranes (Fisher). The membranes were blotted with antibody against MSH2, MLH1, or HMGB1. Bound antibodies were detected by chemiluminescence using a secondary antibody conjugated with horseradish peroxidase (Amersham Biosciences).
Co-immunoprecipitation-Western blot (IP-Western) was also performed using purified MutS␣, MutL␣, and the FII-RPA complementing activity. The procedure for these experiments was identical to that of the IP-Western analysis described above using crude nuclear extracts. MutS␣ and MutL␣ were purified from HeLa nuclear extracts as described (15,19).
Expression and Purification of Recombinant RPA and HMGB1-The RPA expression vector was kindly provided by Dr. Marc Wold (University of Iowa), and the RPA protein was purified as described (29,44). Expression and purification of human recombinant HMGB1 were performed as described (45)(46)(47). The HMGB1 expression vector, a gift from Dr. Carol Prives (Columbia University), was transformed into bacterial strain BL21 (DE3) pLysS for expression. One liter of culture was induced with 0.75 mM isopropyl-␤-D-thiogalactopyranoside when the optical density at 600 nm reached 0.6. The cells were induced for 6 h at 28°C and harvested for protein isolation. The recombinant protein was purified by chromatography using a 5-ml HiTrap chelating nickel column and a Mono Q column. Purified HMGB1 was analyzed by SDS-PAGE and detected by silver staining. The recombinant protein was a single major polypeptide with a molecular size of ϳ30 kDa by silver staining (see below).
Peptide Sequencing-Peptide sequencing was performed at the Harvard Microchemistry Facility using microcapillary reverse-phase HPLC nano-electrospray tandem mass spectrometry on a Finnigan LCQ DECA quadrupole ion trap mass spectrometer. The FII-RPA complementing activity was electrophoresed through 12% polyacrylamide in the presence of SDS, and the protein was excised from the gel after staining with Coomassie Blue. The protein was digested by trypsin in situ, tryptic peptides were separated by microcapillary HPLC, and selected fractions were sequenced using mass spectrometry as described (48).

Purification of a Novel Protein by Complementation of an MMR-deficient HeLa
Fraction-A complementation assay for heteroduplex repair was developed to identify novel proteins for MMR in vitro. MMR-deficient HeLa cell fractions were isolated by fractionating HeLa nuclear extract using ammonium sulfate precipitation and ssDNA cellulose column chromatography. This procedure yielded fractions SS1, SS2, and FII, which have been previously characterized (29). These fractions contain many known MMR components (e.g. SS1 and FII contain MSH2, MLH1, and proliferating cell nuclear antigen), but each fraction alone and each combination of two fractions are deficient in MMR in vitro. SS2 can be substituted by purified recombinant RPA, a human ssDNA-binding protein that protects template DNA and facilitates mismatch excision and DNA resynthesis in human MMR (29). To identify novel MMR components, column fractions were assayed for their ability to stimulate MMR in the presence of FII and RPA (referred to as FII-RPA). An FII-RPA complementing activity was identified FIG. 1. DNA substrates. The heteroduplexes were constructed to contain a single mismatch and a strand break either 125 nt 5Ј (5Ј substrate) or 172 nt 3Ј (3Ј substrate) to the mismatch. In each case, the mismatch is located within overlapping recognition sites (indicated by a line either above or below the sequence) of two restriction enzymes so that the substrate is resistant to hydrolysis by both enzymes. The strand-specific (or nicked directed) MMR in human cells restores the sensitivity of the substrate to one of the restriction enzymes, which can be used for scoring repair. The solid bars indicate the location where oligonucleotide probes hybridize in Southern analysis (see Fig. 4 for details).

Involvement of HMGB1 in Mismatch Repair
in the eluate from a phosphocellulose column. The activity was purified to near homogeneity using Amersham Biosciences Hi-Trap Desalting (Sephadex G-25), Mono Q, and Superdex 200 (S200) column chromatography ( Table I). The peak of activity co-eluted from the S200 column with a single ϳ30-kDa polypeptide ( Fig. 2A). Fig. 2B shows restoration of MMR to FII-RPA by the 30-kDa protein from the S200 column in repair of a G-T mismatched heteroduplex with a 5Ј strand break (5Ј G-T; Fig. 1). As expected, little repair was detected when FII-RPA was incubated with the G-T heteroduplex. However, repair activity increased significantly (ϳ10-fold) when 100 ng or more of the S200 fraction were added to FII-RPA. Similar results were observed with other heteroduplex DNA substrates including base-base mismatches and small insertion/deletion mispairs with a strand break either 5Ј or 3Ј to the heterology (Table II). These results suggest that the 30-kDa protein in the S200 fraction is required for bi-directional repair of base-base and insertion/deletion mismatches during MMR in vitro.
The FII-RPA Complementing Activity Is HMGB1-The molecular size of the FII-RPA complementing factor does not match previously characterized MMR proteins. Therefore, tryptic peptide sequencing by mass spectrometry was carried out to determine the identity of the 30-kDa protein in the active S200 fraction. Twenty-seven unique tryptic peptides were se-quenced as described (48), and the peptide sequences were used to search the NCBI nonredundant protein sequence data base using BLAST. This search revealed a perfect match for each of the individual peptides analyzed to the internal regions of the high mobility box group 1 protein (HMGB1) ( Table III). Many of the peptides overlapped because of incomplete digestion in arginine and/or lysine-rich protein regions. This result indicates that the FII-RPA complementing activity is probably the HMGB1 protein.
Confirmation of the identity of the FII-RPA complementing activity as HMGB1 was based on several independent lines of evidence. First, the electrophoretic mobility of the FII-RPA represents the unrepaired product. The repair rate (relative repair) for individual complementation reactions (an assay based on 32 P incorporation) was determined using the repair rate in HeLa extracts as a standard. HL, HeLa nuclear extract; S-200, the FII-RPA-complementing activity from the S200 column.

Involvement of HMGB1 in Mismatch Repair
complementing activity on an SDS-polyacrylamide gel was similar to that reported for HMGB1 (49). Second, an HMGB1 antibody specifically recognized the peak fraction from the S200 column (Fig. 3A, lane 3). Third, the HMGB1-depleted HeLa nuclear extract contained little MMR activity, but the addition of the FII-RPA complementing factor restored MMR to the depleted extract (Fig. 3B), Fourth, the FII-RPA complementing activity was heat-stable, a characteristic of HMGB1 (49). As shown in Fig. 3C, although purified hMutS␣ was capable of restoring MMR to extracts (lane 8) of the hMSH2deficient NALM-6 cell line (lane 7) (50), the protein heated at 95°C for 5 min failed to rescue MMR in the NALM-6 extract (lane 9). However, both heated (Fig. 3C, lanes 5 and 6) and unheated (Fig. 3C, lanes 3 and 4) peak fraction from the S200 column restored MMR to FII-RPA at the same level, indicating that heating at 95°C had no effect on the MMR restoration ability of the S200 fraction. Finally, strong ratification of the identity of the FII-RPA complementing activity as HMGB1 came from testing recombinant human HMGB1 protein (rH-MGB1) purified from bacteria. As shown in Fig. 3D, His-tagged rHMGB1 was capable of complementing the FII-RPA fraction in repair of a G-T substrate. Taken together, we conclude that the FII-RPA complementing factor is HMGB1.
HMGB1 Participates in Repair Initiation/Excision-The MMR reaction can be divided into initiation, excision, and resynthesis, and the structure of the DNA substrate changes at each of these stages. Individual MMR components play a specific role in one specific stage of the repair reaction. To explore the specific role of HMGB1 in MMR, the structure of repair intermediates of a G-T heteroduplex containing a 36-nt gap 5Ј to the mismatch was analyzed in reactions containing FII-RPA in the presence or absence of HMGB1 using Southern blot analysis (43,51). The gapped heteroduplex of this kind has been recently shown to be a better substrate for the in vitro MMR (52). After incubation of the substrate with the protein mixtures, DNA samples were digested with restriction enzymes SspI and HindIII, and the products were subjected to Southern blot analysis (Fig. 4A). As expected, three major bands, 744, 544, and 414 nt, were detected in a reaction containing an MMR competent HeLa nuclear extract (Fig. 4A, lane  2). The 414-nt band was derived from a homoduplex product that is sensitive to HindIII as a result of the repair of the gapped heteroduplex by the HeLa extract. The 744-nt DNA fragment is an unrepaired ligated DNA substrate, and the 544-nt fragment is an unrepaired unligated DNA substrate; both of these DNA fragments are resistant to HindIII. Similar analysis was carried out in the presence of FII-RPA but lacking HMGB1. A very small amount of 414-nt product was detected in this reaction (Fig. 4A, lane 3). However, ϳ50% of the DNA substrate was either unchanged (i.e. 544-nt species) or converted to products slightly faster migrating than 544 nt or migrating between 544 and 414 nt (Fig. 4A, lane 3). The latter species are likely to be unrepaired unligated reaction intermediates that have been partially excised. However, these putative reaction intermediates decreased, and the 414-nt repair product increased when purified HMGB1 (the S200 fraction) was added to the reaction (Fig. 4A, lanes 4 -6). Similar analysis was used to examine the role of HMGB1 in 3Ј substrate repair. As shown in Fig. 4B, FII-RPA could only convert a limited amount (7%) of heteroduplex to homoduplex (i.e. molecules sensitive to NsiI (lane 2)), but the amount of the NsiI-sensitive species in reactions supplemented with HMGB1 increased to  2 and 3); purified His-Tag rHMGB1 (lanes 4 and 5). Lanes 2 and 5 were silver-stained, and lanes 3 and 4 were analyzed by Western blot. B, MMR deficiency in HMGB1-depleted HeLa nuclear extract. HeLa nuclear extract (0.4 mg) was incubated with anti-HMGB1 antibody (9.0 g) on ice for 3 h, and the mixture was then incubated with protein A-agarose on ice for 1 h, followed by centrifugation to remove immunoprecipitates. The supernatant (the HMGB1-depleted extract, 50 g) was used for MMR assay in reactions containing 100 ng of the 5Ј G-T substrate and nonlabeled dNTPs in the presence or absence of the FII-RPA complementing activity (S-200) as described (13). DNA was digested with HindIII and Bsp106 and visualized by UV illumination after electrophoresis on an agarose gel. C, the FII-RPA complementing activity is heat-resistant. MMR assays were performed as described in the legend to Fig. 2 except that S200 fraction was either heated or unheated at 95°C for 5 min as indicated. For a control, heated or unheated MutS␣ was assayed for its ability to restore MMR to the MSH2-deficient NALM6 extract. D, restoration of MMR to FII-RPA by rHMGB1. Repair assays were carried out as described in the legend to Fig. 2 but using rHMGB1. HL, HeLa nuclear extract; S-200, the FII-RPA-complementing activity from S200 column. more than 24% (lanes 3 and 4). Given the fact that FII-RPA is capable of conducting MMR-associated DNA resynthesis (29) and that there was no detectable small (Ͻ496 nt) excision intermediates in the reaction containing only FII-RPA (Fig. 4B,  lane 2), the deficiency in the reaction is not at the step of DNA synthesis but at or prior to the step of excision. Taken together, these observations strongly suggest that HMGB1 is involved in bi-directional MMR at or prior to the mismatch excision. It is noteworthy that background repair/excision was observed in the reaction containing FII-RPA only (Fig. 4, A, lane 3, and B,  lane 2), and this could be due to the presence of a limiting amount of HMGB1 in FII. A quantitative Western blot analysis indeed revealed that FII contains ϳ20% of the total HMGB1 in HeLa nuclear extracts (data not shown).
HMGB1 Physically Interacts with MutS␣-The results presented above suggest that HMGB1 plays a role in initiation/ excision of MMR. Human MutS and MutL homologs are also required for this process, raising the possibility that HMGB1 and MutS and/or MutL interact with one another. This idea was tested using IP-Western analysis with HeLa nuclear extracts. Antibody against HMGB1 was incubated with HeLa nuclear extracts in the presence or absence of homoduplex or heteroduplex DNA, and the immunoprecipitates were analyzed by Western blot using antibodies against MSH2 and MLH1. As shown in Fig. 5, MSH2 and MLH1 co-precipitate with the HMGB1 antibody, especially in the presence of heteroduplex (lane 2). HMGB1 was also co-precipitated by anti-MLH1 antibody (data not shown). Unfortunately, the available MSH2 antibody was not suitable for immunoprecipitation. To determine whether HMGB1 interacts with MSH2 or MLH1, or both, IP-Western was carried out using the HMGB1 antibody in the presence of purified proteins HMGB1 and MutS␣ or HMGB1 and MutL␣. The HMGB1 antibody precipitated MSH2 regardless of the presence of DNA substrates (Fig. 5, lanes 4 -6) but did not pull down MLH1 (Fig. 5, lanes 7-9). A gel shift analysis revealed that HMGB1 does not bind homoduplex or a G-T heteroduplex DNA under the immunoprecipitation conditions (data not shown), ruling out the possibility that the co-precipitation of HMGB1 and MutS␣ is due to their independent abilities to interact with a common DNA substrate. These observations suggest that HMGB1 and MutS␣ specifically interact with each other during in vitro MMR. This interaction appears to be stimulated by heteroduplexes, as judged by the fact that a strong interaction between these two proteins was observed only in the presence of heteroduplex DNA in crude nuclear extracts, although the specificity of the interaction was lost with homogeneous proteins in the presence or absence of heteroduplex DNA (Fig. 5). DISCUSSION This study demonstrates that HMGB1 is required to reconstitute MMR in vitro. In this system, HMGB1 is essential for bi-directional repair of both base-base and insertion-deletion mispairs. As a well characterized member of the HMG box protein superfamily, HMGB1 has been shown to play important cellular roles in transcription, cellular signaling, inflammatory response, and tumor metastasis (38,39,53). However, previous studies did not anticipate a role for HMGB1 in MMR. On the other hand, although the searching in human cells for homology to E. coli MMR proteins has largely been successful and has identified human homologs of MutS, MutL, exonuclease I, RPA, and polymerase ␦, E. coli MMR does not appear to require a functional equivalent of human HMGB1, indicating that human MMR has higher complexity than E. coli MMR and involves nontraditional MMR components. This observation is consistent with the fact that a substantial fraction of hereditary non-polyposis colorectal cancer kindreds do not appear to be mutants of known human MMR genes (10).
Our data indicate that HMGB1 is involved in the in vitro MMR at or prior to the step of excision, as judged by the fact that the addition of HMGB1 to FII-RPA converted species between the 544-nt marker and the 414-nt marker to the repair products (414-nt species) (Fig. 4). However, how HMGB1 facilitates the excision is not clear. It is well known that HMGB1 can bend DNA (40,54,55), can recognize some unusual DNA structures including cisplatin-modified DNA (56), and promotes protein-protein interactions (38,39). One or all of these properties could be required for MMR. It is also known that the binding of a mismatch by MutS and its homologs is associated with bending of the DNA substrate at the mismatch site (57,58). Therefore, it is highly likely that the role of HMGB1 in MMR is to enhance heteroduplex bending by MutS homologs so   FIG. 4. HMGB1 promotes mismatch excision. A, 5Ј nick-direct repair. MMR assays were performed as described under "Experimental Procedures" using a G-T heteroduplex containing a 36-nt gap 125 bp 5Ј to the mismatch. DNA samples were recovered and digested with restriction endonucleases HindIII (to score for the conversion of heteroduplex to homoduplex) and SspI (see Fig. 1). The resulting products were electrophoresed in a denaturing 5% polyacrylamide gel and transferred to a nylon membrane. The membrane was blotted with a 32 Plabeled oligonucleotides 5Ј-CAAAGCAACCATAGTACGC-3Ј to map the excision tracts of 5Ј termini of the SspI restriction fragment in the nicked strand (see Fig. 1). B, 3Ј nick-direct repair. The repair assays were performed as in A but using a G-T heteroduplex containing a strand break 173 nt 3Ј to the mismatch. DNA products were digested with NsiI (to score for repair) and SspI-DraIII (see Fig. 1) and subjected to Southern blot analysis using a 32 P-labeled oligonucleotide probe, 5Ј-AATTTAACGCGAATTTTAAC-3Ј. Corresponding indirectly labeled repair products or intermediates and their sizes are shown to the left of the gel, and a schematic diagram showing the fragments of interest and restriction sites is shown to the right. Black bars indicate where the probes hybridize. HL, HeLa nuclear extract.
that the DNA substrate is more accessible to attack by helicase(s) and/or nuclease(s). This could explain the requirement for HMGB1 in the repair of both 5Ј and 3Ј heteroduplexes (Table II). Although direct evidence for HMGB1-induced enhancement of DNA bending is lacking, this possibility is strongly supported by our co-IP experiments, where HMGB1 and MutS␣ physically interact each other, and the interaction is enhanced in the presence of a heteroduplex DNA (Fig. 5).
In addition to co-IP experiments, the interaction between HMGB1 and MutS␣ is also supported by a possible interaction domain for HMGB1 in MutS␣. MSH6, a subunit of MutS␣ heterodimer, seems to contain a putative HMGB1 interaction domain, based on a recent study by Dintilhac and Bernues (59), who showed that HMGB1 interacts with other proteins through many short peptide motifs that have no apparent homology or similarity. For example, the motifs PXXPXP and WXXW (where X can be any amino acid) can interact with box A and box B of HMGB1, respectively. Peptide TLTTPIL is another peptide that is recognized by HMGB1 box A (59). A similar sequence, TLTTPAL, is present in E. coli MutS (amino acids 189 -195), and the equivalent residues in human MSH6 are SKTLRTL (amino acids 631-637). If TXXXL functions as a consensus motif for the binding of HMGB1, MutS␣ interacts with HMGB1 via the MSH6 subunit. Interaction of MutS␣ with other proteins, e.g. proliferating cell nuclear antigen (33)(34)(35), through the MSH6 subunit has been documented.
It is also possible that HMGB1 acts as a molecular chaperone to recruit MMR proteins for early steps of MMR because the removal of a mismatch requires a coordinating interaction among MutS homologs, MutL homologs, proliferating cell nuclear antigen, DNA helicase(s), and exonuclease(s). Interestingly, although we did not detect a direct interaction between MLH1 and HMGB1, human PMS1, a heterodimeric partner of MLH1 in MutL␤, contains an HMG box domain (60). However, the function of the HMG box domain of PMS1 in MMR remains unclear. Additional experiments are needed to clarify the precise role of HMGB1 in human MMR.