Oxidative Stress Induces Nuclear-to-Cytosol Shift of hMSH3, a Potential Mechanism for EMAST in Colorectal Cancer Cells

Background Elevated microsatellite alterations at selected tetranucleotide repeats (EMAST) is a genetic signature observed in 60% of sporadic colorectal cancers (CRCs). Unlike microsatellite unstable CRCs where hypermethylation of the DNA mismatch repair (MMR) gene hMLH1’s promoter is causal, the precise cause of EMAST is not clearly defined but points towards hMSH3 deficiency. Aim To examine if hMSH3 deficiency causes EMAST, and to explore mechanisms for its deficiency. Methods We measured −4 bp framshifts at D8S321 and D20S82 loci within EGFP-containing constructs to determine EMAST formation in MMR-proficient, hMLH1−/−, hMSH6−/−, and hMSH3−/− CRC cells. We observed the subcellular location of hMSH3 with oxidative stress. Results D8S321 mutations occurred 31-and 40-fold higher and D20S82 mutations occurred 82-and 49-fold higher in hMLH1−/− and hMSH3−/− cells, respectively, than in hMSH6−/− or MMR-proficient cells. hMSH3 knockdown in MMR-proficient cells caused higher D8S321 mutation rates (18.14 and 11.14×10−4 mutations/cell/generation in two independent clones) than scrambled controls (0 and 0.26×10−4 mutations/cell/generation; p<0.01). DNA sequencing confirmed the expected frameshift mutations with evidence for ongoing mutations of the constructs. Because EMAST-positive tumors are associated with inflammation, we subjected MMR-proficient cells to oxidative stress via H2O2 to examine its effect on hMSH3. A reversible nuclear-to-cytosol shift of hMSH3 was observed upon H2O2 treatment. Conclusion EMAST is dependent upon the MMR background, with hMSH3−/− more prone to frameshift mutations than hMSH6−/−, opposite to frameshift mutations observed for mononucleotide repeats. hMSH3−/− mimics complete MMR failure (hMLH1−/−) in inducing EMAST. Given the observed heterogeneous expression of hMSH3 in CRCs with EMAST, hMSH3-deficiency appears to be the event that commences EMAST. Oxidative stress, which causes a shift of hMSH3’s subcellular location, may contribute to an hMSH3 loss-of-function phenotype by sequestering it to the cytosol.


Introduction
DNA mismatch repair (MMR) dysfunction drives the process of microsatellite instability (MSI), observed in nearly all Lynch Syndrome colorectal cancers (CRCs) and in 15-20% of sporadic CRCs [1]. MSI is observed when both alleles of a major DNA MMR gene is inactivated by mutation in Lynch tumors or by hypermethylation in sporadic tumors, causing MMR protein production or function to fail, and allowing frameshifts to occur at DNA microsatellite sequences throughout the genome [1][2][3]. Classic MSI is determined by the occurrence of frameshifts in mono/dinucleotide markers, and has a defined panel for which to compare studies [4]. This panel, or variations of these mono/ dinucleotide markers, readily detects hMLH1and hMSH2deficiency (as dysfunction of these two proteins completely inactivates DNA MMR), and occasionally hMSH6-deficiency [1,4]. The MSI panel is not as specific to detect hMSH3-deficiency due to the nature of repair that this protein contributes to MMR, based on yeast studies [5,6]. DNA MMR is an enzyme system whose main cellular function is to repair nucleotide mispairs and insertion/deletion loops (IDLs) during DNA replication. The specificity of mismatch recognition is directed by hMutSa and hMutSb; hMutSa (hMSH2-hMSH6 heterodimer) binds to single mispairs and small IDLs (#2 nucleotides), whereas hMutSb (hMSH2-hMSH3 heterodimer) binds larger IDLs [1,[5][6][7]. We and others have observed DNA lesion binding specificities with purified hMutSa and hMutSb as well as made specific observations on binding using human CRC cells of differing DNA MMRdeficient backgrounds harboring constructs containing specifiedlength IDLs or nucleotide mispairs [8][9][10][11][12].
In order to test if hMSH3-deficiency is a principal cause of EMAST, we generated tetranucleotide repeat-containing constructs that were fused out-of-frame with EGFP, such that when frameshifted by 24 bp (i.e. deletion of one tetranucleotide unit), EGFP would be expressed. This allowed observation of the generation of EMAST in real-time, and afforded us the ability to calculate mutation frequencies and mutation rates in multiple MMR-deficient backgrounds. In this report, our data clearly shows that hMSH3-deficiency is a cause of EMAST in human CRC cells and we further provide a possible mechanism that may contribute to the acquired hMSH3-deficiency in human CRCs.

Establishment of a Measuring System for EMAST Tetranucleotide Frameshifts
Because some studies have indicated that EMAST is associated with heterogeneous expression of hMSH3 in colorectal tumors [13][14][15][16], our approach was to generate a system to measure EMAST in a quantitative fashion as a means to directly examine if hMSH3-deficiency is the cause of EMAST. Human tetranucleotide D8S321 and D20S82 loci sequences (containing 12 and/or 16 repeats of AAAG, respectively) have shown the highest frequency for a 24 bp frameshift in colorectal tumors [14,16]. These two loci were therefore selected for study using CRC cells with different MMR genetic backgrounds to ascertain subsequent mutation. We have previously established a similar experimental system where deletion of 1 bp of (A) n would lead to the expression of the reporter gene EGFP when a construct was placed +1 bp out-of-frame [10][11][12].
To generate the experimental constructs, we inserted the D8S321 (D8) and D20S82 (D20) loci as 60-mers containing the tetranucleotide repeats and surrounding nucleotides immediately after the EGFP start codon [ Table S1]. Experimental constructs [D8-OF (D8-out-of-frame) and D20-OF] were made +1 bp out-offrame such that deletion of one AAAG repeat will shift the EGFP reading frame (+1 bp to 23 bp) into the correct frame to cause EGFP expression. Mutation resistant (MR) plasmids (MR-D8 and MR-D20), serving as negative controls, were constructed by replacing the middle AA nucleotides within the AAAG repeats with C/G nucleotides in every third AAAG, interrupting the series of tetranucleotide repeats to prevent frameshift mutations. The MR-D8 and MR-D20 plasmids were placed out-of frame (OF; D8/D20-MR-OF) and/or in-frame (IF; D8/D20-MR-IF) to be used as negative and positive controls for EGFP expression (also see Materials and Methods). These constructs were transfected into human CRC cell lines with different MMR genetic backgrounds and stable cell lines were established via hygromycin B selection. Single cell clones were subsequently established. DNA sequencing was employed to correctly confirm the transfected EMAST constructs ( Table S1). As expected, cells carrying MR-IF (positive control) were EGFP positive, whereas cells containing MR-OF (negative control) were EGFP negative.
To test if hMSH3-deficiency could cause EMAST, nonfluorescent cells carrying the different stably transfected EMAST constructs were sorted by flow cytometry and exponentially grown for four to six weeks. Each week, cells were analyzed in duplicate by flow cytometry for EGFP expression, presumably resulting from deletion of one AAAG repeat from the transfected D8S321 and/or D20S82 loci. EGFP positive cells began emerging as soon as one week after EGFP negative cells were sorted and plated (data not shown). To ensure that our experimental system allowed us to correctly assess the occurrence of EMAST, the EGFP negative and positive cells from hMLH1 2/2 and/or hMSH3 2/2 cells harboring the experimental constructs (D8-OF and D20-OF) were sorted to isolate genomic DNA for sequencing. As expected, essentially all clones (greater than 96%) from EGFP negative cells retained wild type (WT) sequences for each of D8S321 and D20S82 loci in both hMLH1 2/2 and hMSH3 2/2 cell lines, whereas clones from EGFP positive cells acquired deletion of one AAAG repeat ( Figure 1A and Figure S1A and S1B). We also observed expansion frameshift mutations at each locus by DNA sequencing (included in ''others'' in Figure 1A). The most frequent one (ranging from 12.5% to as high as 30% of total EGFP-positive cells) was insertion of two AAAG repeats [(AAAG) 12 to (AAAG) 14 ] in D8S321 that changed the reading frame from +1 bp to +9 bp (listed as ''others'' in Figure 1A and Figure S1A). In addition, more than one AAAG repeat deletion (e.g. 4 or 10 AAAG repeat deletion) was rarely observed (,8% of total, listed in ''others'' in Figure 1A). Importantly, deletion/ insertion mutations were always detected within the microsatellite sequences while no mutation occurred in the flanking sequences surrounding AAAG repeat ( Figure S1). These observations indicate that we have established a reliable experimental system to monitor the occurrence of EMAST in CRC cells.

Frameshift Mutations at Tetranucleotide Repeats of D8S321 and D20S82 Loci are Dependent on the MMR Background
To examine the effect of MMR background on frameshift mutations at the tetranucleotide D8S321 and D20S82 loci, we calculated mutation frequencies of the markers by comparing each experimental group (OF) to its negative control (MR-OF) using the following formula: ''(EGFP positive cells/total live cells from OF)/ (EGFP positive cells/total live cells from MR-OF)''. hMLH1 2/2 cells (completely devoid of DNA MMR activity) and hMSH3 2/2 cells showed 7-12 fold higher mutation frequencies compared to the controls for both loci ( Figure 1B). In particular, the mutation frequencies observed in hMSH3 2/2 were comparable to that in hMLH1 2/2 for both loci ( Figure 1B). The 24 bp mutation rates for both loci in each MMR background were also calculated ( Table 1). With complete MMR-deficiency (hMLH1 2/2 ), mutation rates for both loci were 27 and 56610 24 mutation/cell/ generation. Similarly, hMSH3 2/2 cells generated 33 and 34610 24 mutations/cell/generation. Thus, the mutation rates with hMSH3-deficiency are very comparable to those observed for complete MMR-deficiency. This is in sharp contrast to hMSH6-deficiency (,1610 24 mutation/cell/generation), essentially identical to MMR-proficient cells. These findings strongly suggest the hMSH3-deficiency is the driver for EMAST formation.

Direct Knockdown of hMSH3 Expression in MMRproficient Cells Causes EMAST
To confirm a role of hMSH3 in EMAST, we knocked down hMSH3 expression in MMR-proficient cells harboring the D8S321-EGFP construct by transfecting plasmids carrying hMSH3-specific shRNA and/or scrambled control, then measured D8S321 tetranucleotide frameshift mutations as above. Prior to shRNA transfection, hMSH3 expression levels in chosen MMRproficient clones were comparable to each other ( Figure S2). After generation of shRNA stable cell lines (single cell clones), hMSH3 expression was significantly reduced in hMSH3 shRNA transfected cells ( Figure 2; varying between 26-34% of the protein levels of the scramble controls).   Using the shRNA stably-transfected cells containing the D8S321 locus, we examined the cell lines for occurrence of EMAST by flow cytometry. An EGFP-positive population appeared in cells where hMSH3 expression was knocked down. hMSH3 KD cells showed 5-fold (clone 1) and 6-fold (clone 2) increase in mutation frequencies ( Figure 3A) compared to scrambled shRNA control cells. Reduction in hMSH3 expression greatly increased mutation rates of the D8S321 construct to 12.7 and 18.4610 24 mutation/ cell/generation, compared to 1610 24 mutation/cell/generation in scramble controls ( Table 2). EGFP-negative and -positive cells were separated out for DNA sequencing to examine frameshift mutations. In both clones, greater than 90% of the EGFP negative cells retained WT sequence [(AAAG) 12 ], whereas approximately 90% of the EGFP-positive cells revealed a deletion of one repeat of AAAG ( Figure 3B). Expansion of the microsatellite repeats was also detected (included as ''others'' in Figure 3B). Overall, our observations indicate that direct KD of hMSH3 causing hMSH3deficiency alone is able to generate EMAST in previous MMRproficient CRC cells. Our studies above further provide evidence that reduction of hMSH3 expression alone can induce EMAST in human CRC cells. We began to explore how hMSH3 might become deficient in EMAST-positive tumors with the observed reduction from full to heterogeneous expression. In CRC, EMAST is associated with inflammatory cells [13,14,16] that may facilitate oxidative stress, which has been shown to impair MMR function [23]. Using H 2 O 2 as a surrogate, we investigated the effect(s) of oxidative stress on hMSH3 expression. MMRproficient cells were treated with increasing concentrations of H 2 O 2 (50 mM -500 mM) for up to 24 hours. We did not detect any reductions in total hMSH3 protein levels in any conditions we tried (data not shown). However, upon treating cells with 50 mM of H 2 O 2 for 4 hours, hMSH3 was found homogenously in both the nucleus and the cytosol in a significant percentage of cells (40-60%; second panel column in Figure 4A). Strikingly, hMSH3 was observed to completely vacate nuclei in some cells [10-20% of total cells (third panel column in Figure 4A], in sharp contrast to the untreated cells where hMSH3 predominately remained localized in the nucleus (left panels in Figure 4A). The shift of hMSH3 localization could be detected as soon as 2 hours into treatment, and the effect peaked between 4 and 8 hours after H 2 O 2 treatment started (data not shown). Unlike hMSH3, hMLH1, hMSH2, and hMSH6 did not alter their cellular localization and remained in nuclei ( Figure 4B and Figure  S3A, S3B, and S3C). Cell fractionation to separate nuclear and cytosolic fractions matched our immunocytochemistry observations (Lanes 1-3 and 5-7 in Figure 4B).

Oxidative Stress Causes Subcellular Compartmental Shift of hMSH3 from the Nucleus to the Cytosol
To examine if the effect of H 2 O 2 on hMSH3 was reversible, we replaced H 2 O 2 containing media with fresh media after 4 hours of treatment. hMSH3 was found to be nuclear by 24 hours after removal of H 2 O 2 (right panels in Figure 4A and lanes 4 and 8 in Figure 4B). This result may indicate that the effect of H 2 O 2 on hMSH3 was reversible, where hMSH3 is allowed to re-enter into the nucleus once oxidative stress abated. Alternatively, the hMSH3 that was in the cytosol might have been damaged and degraded and thus was incapable of re-entering into nuclei, and might indicate that the observed nuclear hMSH3 was newlysynthesized. To determine which scenario could be correct, we treated cells with 20 mg/ml of cyclohexamide upon removal of H 2 O 2 to stop protein synthesis for 24 hours. hMSH3 was found to be in the nucleus (data not shown), demonstrating that the hMSH3 that was shifted to the cytosol was indeed able to re-enter into the nucleus.

Discussion
DNA MMR studies from bacteria and yeast indicate that hMutSb, a heterodimer of hMSH2 and hMSH3, binds and directs repair of insertion-deletion loops 62 nucleotides [1,5,6]. Thus, hMSH3-deficiency would be expected to allow large insertiondeletion loop frameshift mutations, which has been strongly suggested in previous studies [13][14][15][16]. Determining the consequence of hMSH3-deficiency was not previously of prime importance, as there has never been described a germline mutation for hMSH3 to cause Lynch Syndrome, and there was no correlation with sporadic CRC until the description of EMAST in CRCs [13][14][15][16]. The linkage of EMAST with heterogeneous loss of hMSH3 expression in sporadic CRCs suggested that this DNA MMR protein could drive EMAST. In our study, we set out to test this by creating a human cell model to measure EMAST formation in a quantitative fashion. Our study shows that (a) EMAST formation is dependent on the MMR background, with hMSH3-deficiency matching complete MMR-deficiency in causing EMAST, (b) hMSH3-deficiency mutation rates for EMAST formation are markedly higher compared to that of hMSH6deficiency (essentially null for EMAST formation), (c) knockdown of hMSH3 alone initiates formation of EMAST, and (d) oxidative stress may be one of the factors causing deficiency of hMSH3, as this MMR protein shifts its subcellular location with H 2 O 2 . Altogether, our study offers strong support to the concept that hMSH3-deficiency drives EMAST formation and provides a possible mechanism that may contribute to hMSH3-deficiency in human CRCs.
Tetranucleotide frameshifts that define EMAST in CRC tumors can be by insertion or deletion of tetranucleotide repeats. In our experiments, we designed constructs to detect a 24 bp deletion; however upon DNA sequencing some insertions (expansion) of the microsatellites were also detected. Indeed, the majority of frameshift mutations detected in our experiments were due to deletion of one AAAG unit ( Figure 1A and 3B). We selected the D8S321 and D20S82 loci for our experiments because deletion of one AAAG unit had been shown to be the most prevailing mutation found among loci utilized to detect EMAST [14,16]. Because our system was designed to only detect one type of mutation, our rates calculated are likely an underestimation of the mutability of these loci with hMSH3-deficiency. Individual sequence context and flanking sequences may also contribute to the different outcome of the mutation (e.g. insertion vs. deletion) [11,12,24]. At the D20S82 locus, only about one-third of EGFPpositive hMSH3 2/2 cells demonstrated one AAAG unit deletion and this could be due to this particular clone containing multiple copies of the construct. This is contrasted by more than 70% of the EGFP positive cells carrying the same construct in hMLH1 2/2 showing deletion of one AAAG unit, indicating the construct behaved as designed. For hMSH3 2/2 cells, mutation on one of the copies would lead to EGFP expression, while the rest remained WT, confounding sequencing analysis. This, however, would not affect the mutation frequency and mutation rate as the calculation was based on the number of the EGFP-positive cells. It is not fully clear for EMAST-positive tumors how hMSH3 is downregulated to initiate EMAST. There is clear evidence that EMAST-positive tumors are strongly associated with inflammatory cells, particularly within the epithelial components of the tumor and in the stroma surrounding the tumor (so called ''tumor nests'') [16]. Thus, one hypothesis is that inflammation triggers an acquired reduction of hMSH3, which subsequently drives EMAST. Interestingly, H 2 O 2 treatment (simulating oxidative stress) leads to a shift of hMSH3 into the cytosol away from its functional site in the nucleus without diminishing its overall expression levels. With short-term H 2 O 2 treatment, we could not demonstrate detectable levels of EMAST formation in our model (data not shown). A longer and consistent presence of oxidative stress may be required to commence EMAST, considering EMAST appears to be associated with chronic inflammation and with advanced histology of tumors [14]. Alternatively, other components of inflammation and/or the tumor milieu may work synergistically with oxidative stress to cause EMAST. EMAST appears to be a poor prognostic factor for patients with CRC [16,18]. This is in contrast to classic MSI for which patients with CRC have better survival at the same staged tumor of patients with MSS tumors [1]. It remains to be determined if this is intrinsic to the type of MMR defect (hMLH1 vs. hMSH3) or some other factors, such as the degree or type of inflammation in the tumor.
In summary, we show through the creation of a novel cell model that hMSH3-deficiency drives EMAST. Our data strongly complements the observation of heterogeneous loss of hMSH3 expression in EMAST-positive tumors, and suggests that acquired hMSH3-deficiency, driven by oxidative stress, may be the most common DNA MMR defect among human CRCs.

D20S82-EGFP Plasmids
The two human loci, D8S321 and D20S82, have been reported to have the highest mutation frequency in EMAST colonic tumors [13][14][15][16], and were selected to construct plasmids. D8S321 and Two independent hMSH3 and scramble KD clones as well as the parental cells were included for this study. EMAST mutation frequencies were calculated as described in Figure 1B  D20S82 sequences contain 12 and/or 16 repeats of AAAG, respectively. Sense and anti-sense oligos containing D8S321 or D20S82, surrounding sequences, as well as PmeI and AscI cloning sites (Table S1) were synthesized (Integrated DNA Technologies, Inc., San Diego, CA) and subjected to T4-polynucleotide kinase reactions to phosphorylate the 59 ends (Invitrogen). The sense and anti-sense oligos were allowed to anneal and cloned into pIREShyg2-EGFP plasmids via PmeI-AscI sites immediately after the start codon of the EGFP gene as described previously [10][11][12]24]. Experimental plasmids were constructed +1 bp out of frame, such that one deletion of a tetranucleotide repeat within D8S321 or D20S82 (24 bp frameshift mutation) would shift the reading frame into the correct frame to allow EGFP expression. Mutation resistant plasmids [MR-out-of-frame (OF)] were constructed by using sense and anti-sense oligos in which two nucleotides within every third repeat of AAAG were changed to prevent frameshift mutations. The mutation resistant in frame (MR-D8-IF) plasmid, a positive control for EGFP expression, was constructed by deleting 1 bp from D8S321 sequence in the MR- . Parental cells (Parental) and cells that were serum-starved without the subsequent H 2 O 2 treatment (2FBS) were used as controls. Cellular proteins were separated into nuclear and cytosolic fractions using a nuclear fractionation kit. Equal volume of proteins from each group was used for SDS-PAGE and subsequent Western Blotting to check levels of hMSH3. Anti-histone H3 was used for nuclear fraction loading equality, and tubulin was used for cytosolic fraction loading equality. doi:10.1371/journal.pone.0050616.g004 D8-OF plasmids using Quikchange II site-directed mutagenesis kit (Stratagene, La Jolla, CA).
hMSH3 shRNA Construction hMSH3 shRNA construct and its paired empty vector were kindly provided by Dr. Ajay Goel, Baylor University Medical Center, Dallas, TX. The GFP sequence of this construct was removed by digesting the construct with ApaI and PacI, filling-in using Klenow fragment to form blunt ends, and then performing blunt-end ligation. To make a scrambled control, the GFP sequence from the empty vector was first removed as described above and the scrambled sequence, removed from a commercially available construct (GeneCopoeia, Rockville, MD), was inserted onto the vector via EcoRI and XhoI sites. The existence and the accuracy of the scramble and/or hMSH3 shRNA sequences on the plasmids were confirmed by DNA sequencing.

Transfection and Generation of Stable Cell Lines
Cells were transfected with the different pIREShyg2-D8S321-EGFP and pIREShyg2-D20S82-EGFP plasmids (described in Table S1) by using Nucleofector kits (Amaxa, Cologne, Germany). Stable cell lines were established by hygromycin B (Invitrogen) selection. Each cell line was subjected to single cell sorting using FACS ARIA (Becton Dickinson, San Jose, CA) to establish individual clones. All single cell clones were examined by DNA sequencing to ensure the existence and accuracy of the EMAST construct they carried.
For hMSH3 knockdown (KD) cells, MMR-proficient cells (SW480) carrying D8-OF and/or D8-MR-OF constructs were transfected with plasmids harboring scramble or hMSH3-specific shRNA using Fugene 6 (Roche, Mannheim, Germany), followed by puromycin (Invitrogen) selection to establish stable lines. Two independent clones from each group were used for the studies.

Mutation Analysis by Flow Cytometry
Four hundred thousand non-fluorescent stable-transfected cells were sorted onto each well on 6-well plates by FACS ARIA (Becton Dickinson) and cultured for 4 to 6 weeks. Cells were maintained to keep them in exponential growth. To quantitate EGFP-positive populations, cells from each 100-mm dish were trypsinized, collected, and resuspended in a total volume of 2 ml of cold PBS containing 1 mg/ml of propidium iodide (PI; Sigma) right before analysis. Each group was analyzed in duplicate on a FACSCalibur (Becton Dickinson) for each time point.
For the hMSH3 KD experiments, non-fluorescent MMRproficient cells containing D8-MR-OF or D8-OF were sorted one week before scramble or hMSH3-specific shRNA transfection. Puromycin-selected scramble or hMSH3 KD cells were analyzed by flow cytometry to quantitate EGFP-positive cells as described above.

Polymerase Chain Reactions and DNA Sequencing
Genomic DNA from sorted EGFP-negative and/or -positive cells was PCR-amplified using a cloned Pfu DNA polymerase (Stratagene). The PCR products were subjected to TA cloning (Promega, Madison, WI) to obtain individual clones. Multiple clones from EGFP negative and/or positive cells were sequenced to examine the frameshift mutations. The primers used for sequencing are as following: Forward primer: 59-GAGCTCG-GATCTGTACAGGC-39; Reverse primer: 59TGCCGTCGTCCTTGAAGAAGA-39.

Immunocytochemistry
Twenty-five thousand cells were seeded onto each well on 8-well chamber slides (Invitrogen) and incubated overnight. After serum starvation for 24 hr, cells were treated with 50 mM H 2 O 2 for 4 hours. Cells were fixed with cold acetone on ice for 5 min, airdried, and stored at 4uC until staining. During the staining process, slides were kept in a humidified chamber at room temperature. Slides were re-hydrated with 5% FBS/PBS (blocking/washing buffer) for 20 min before incubating with antibodies against hMLH1 (1:100), hMSH2 (1:100), hMSH3 (1:100), or hMSH6 (1:500) for 90 min and then Alexa 488 conjugated anti-mouse antibody (1:2000) for 1 hr in dark. Slides were mounted with Prolong Gold mounting media with DAPI (Invitrogen). Pictures were taking using an Olympus DP72 Fluorescent microscope. Figure S1 Chromatographs for DNA sequencing of tetranucleotide microsatellites. (A) The EMAST construct carrying D8S321 marker contained 12 copies of AAAG, which underwent deletion of one copy of the AAAG unit or insertion of two copies of AAAG repeat as shown. (B) The D20S82 construct harboring 16 copies of AAAG underwent deletion of one repeat of AAAG. Please note that the flanking sequences remained identical to the wild-type when EMAST markers underwent deletion/ insertion. (TIF) Figure S2 Expression levels of MMR proteins in clones prior to shRNA transfection. Note that clones contained equivalent amounts of hMSH3 protein before attempted knockdown of hMSH3. (TIFF) Figure S3 Subcellular localization of hMLH1, hMSH2, and hMSH6 upon H 2 O 2 treatment. After 24 hours serumstarvation and then 4 hours treatment with H 2 O 2 , cells were fixed and stained with antibodies against (A) hMLH1, (B) hMSH2, and/ or (C) hMSH6, and then Alexa 488 conjugated anti-mouse antibody. No subcellular shift was observed for these MMR proteins.