DNA Mismatch Repair Complex MutSβ Promotes GAA·TTC Repeat Expansion in Human Cells*

Background: Friedreich ataxia (FRDA) is a progressive, debilitating and lethal disease caused by GAA·TTC repeat expansion. Results: Expression of mismatch repair complex MutSβ, particularly the MSH3 subunit, is necessary for GAA·TTC repeat expansion in model cells and FRDA patient fibroblasts. Conclusion: MutSβ promotes GAA·TTC expansion in FRDA. Significance: MSH3 may be a potential therapeutic target for slowing GAA·TTC expansion. While DNA repair has been implicated in CAG·CTG repeat expansion, its role in the GAA·TTC expansion of Friedreich ataxia (FRDA) is less clear. We have developed a human cellular model that recapitulates the DNA repeat expansion found in FRDA patient tissues. In this model, GAA·TTC repeats expand incrementally and continuously. We have previously shown that the expansion rate is linked to transcription within the repeats. Our working hypothesis is that structures formed within the GAA·TTC repeat during transcription attract DNA repair enzymes that then facilitate the expansion process. MutSβ, a heterodimer of MSH2 and MSH3, is known to have a role in CAG·CTG repeat expansion. We now show that shRNA knockdown of either MSH2 or MSH3 slowed GAA·TTC expansion in our system. We further characterized the role of MutSβ in GAA·TTC expansion using a functional assay in primary FRDA patient-derived fibroblasts. These fibroblasts have no known propensity for instability in their native state. Ectopic expression of MSH2 and MSH3 induced GAA·TTC repeat expansion in the native FXN gene. MSH2 is central to mismatch repair and its absence or reduction causes a predisposition to cancer. Thus, despite its essential role in GAA·TTC expansion, MSH2 is not an attractive therapeutic target. The absence or reduction of MSH3 is not strongly associated with cancer predisposition. Accordingly, MSH3 has been suggested as a therapeutic target for CAG·CTG repeat expansion disorders. Our results suggest that MSH3 may also serve as a therapeutic target to slow the expansion of GAA·TTC repeats in the future.

Friedreich ataxia (FRDA) 3 is a progressive, neurodegenerative disease caused by expansion of a GAA⅐TTC trinucleotide repeat (TNR) in the first intron of the frataxin (FXN) gene (1). TNR expansion is the causal agent in a growing number of diseases (2,3). The largest known class of TNR expansion disorders are dominant diseases, such as Huntington disease, that are caused by a CAG⅐CTG repeat expansion within a coding region. However, recessive DNA repeat expansion disorders like FRDA are a rapidly growing subclass. In the FXN gene, the normal range of GAA⅐TTC repeats is from 6 to 36 triplets, while those in FRDA patients typically range from 600 to 900 (4 -6). The GAA⅐TTC expansion in FXN does not alter the protein sequence. Although the precise mechanism remains unclear, expansion ultimately leads to frataxin insufficiency in a repeat length-dependent fashion (1,(7)(8)(9)(10)(11). Regardless of how the repeats repress FXN transcription, understanding the mechanism through which repeat expansion occurs may reveal a means to slow FRDA progression.
While a contribution by the mismatch repair (MMR) pathway has been established in mouse models of several CAG⅐CTG repeat expansion diseases (12)(13)(14)(15)(16), our understanding of a role for DNA repair in GAA⅐TTC expansion is limited. MMR involvement in CAG⅐CTG repeat expansion has been explained in part, by the ability of the individual strands of the repeat to form CNG hairpin structures with mismatched central bases (17)(18)(19)(20). DNA mismatches in human cells are recognized by the MutS protein heterodimers MutS␣ and MutS␤ (21). MutS␣, a heterodimer of MSH2 (MutS homologue 2) and MSH6, is the dominant MutS complex that recognizes base-base mismatches and short insertion/deletion loops (22). MutS␤, a complex of MSH2 and MSH3, is less abundant than MutS␣ in most cell types. For the most part, although it may better recognize and bind to larger insertion/deletion loops, MutS␤ appears to be functionally redundant to MutS␣ (23,24). Unlike the CNG motif, GAA⅐TTC sequences do not form hairpins with mismatches. While MutS␤ was found to contribute to GAA⅐TTC mediated chromosomal breaks and rearrangements in a yeast model (25), MMR has not traditionally been considered an active agent in GAA⅐TTC expansion. Recently however, it was determined that knockdown of MSH2 in FRDA patient-derived induced pluripotent stem cells (iPSCs) caused a slower rate of GAA⅐TTC expansion (26). Furthermore, it has also been reported that the MMR system plays a role in intergenerational GAA⅐TTC repeat dynamics in an FRDA mouse model (27). Thus it appears that regardless of the ability of a particular trinucleotide sequence to form hairpin structures, the contribution of DNA mismatch repair may be a common denominator in TNR expansion disorders.
Tissue-specific somatic instability leads to GAA⅐TTC allele size mosaicism in FRDA patients (6); in general, a contraction bias occurs during aging (28). Significantly, a somatic expansion bias is evident in small pool PCR studies of FRDA disease relevant tissues such as the dorsal root ganglia (DRG) (29). While there are several excellent models available to study TNR contraction (30,31), the lack of higher eukaryotic cellular or animal models that exhibit robust TNR expansion has limited our understanding of this phenomenon. To fill this gap, our laboratory has developed a human cellular model of GAA⅐TTC expansion that recapitulates the propensity for expansion seen in affected tissues of FRDA patients (32). Our model has several key advantages over using authentic FRDA patient cells or their derivatives to study GAA⅐TTC repeat expansion. For example expansion is rapid; a repeat starting at 176 triplets can double in size in two months through incremental and continuous expansion (32). Furthermore, expansion does not have lethal consequences because the repeats are not linked to an essential gene. Thus we are able to avoid the well-known problems associated with selection against frataxin knockdown cells in culture (33). A deficit of our model is that it is not the endogenous FXN locus. Therefore, experiments in this study were performed using both these cells and primary FRDA patient fibroblasts. Each experimental system revealed the same essential role for MutS␤ in GAA⅐TTC expansion, further substantiating the validity of our cellular model.
Frataxin is an essential protein (1) and is transcribed to some degree in all cells; therefore, we surmise that the tissue-specific expansion seen in FRDA patients likely reflects the expression of one or more DNA repair pathways that contribute to expansion in these tissues. Our working hypothesis is that during transcription within the GAA⅐TTC repeat, a structure forms attracting DNA repair enzymes that then facilitate the expansion process.
In this work we study the contribution of DNA repair to the rate of GAA⅐TTC expansion. We used shRNA to reduce expression of mismatch repair proteins in cells and monitored the expansion rate. Knockdown of MSH2 and MSH3 expression in our cellular model slowed GAA⅐TTC expansion; knockdown of MSH6 had little effect on expansion rate. Furthermore, ectopic expression of MSH3 triggered repeat expansion in primary fibroblasts from three different FRDA patients. These cells have shown no previous propensity for repeat instability in their native form.

EXPERIMENTAL PROCEDURES
Cell Lines-The production of HEK 293 cells with single copy constructs bearing defined GAA⅐TTC repeats integrated into the same genomic location via Flp recombinase has been described previously (32). The capped in vitro ligation strategy used to create the uninterrupted (GAA⅐TTC) n repeat inserts has also been previously described (34). Cell lines were maintained in Dulbecco's modified Eagle's medium (DMEM) high glucose (Invitrogen) with 5% fetal bovine serum (Sigma) at 37°C in an atmosphere containing 5% CO 2 . Primary FRDA fibroblasts GM03665, GM0816, and GM04078 (Coriell Cell Repository) were maintained in the same conditions, with the exception that 10% FBS was used.
PCR Analysis of GAA⅐TTC Repeats-Genomic DNA was isolated from HEK293 cells bearing repeat inserts as described in Ditch et al. (32). DNA was extracted using the DNAzol Reagent (Invitrogen) following the manufacturers protocol. Typically, 50 l reactions were performed with 100 ng of template, 200 nM primers, 250 M each dNTP (Stratagene) and 2.5 units of Paq5000 DNA polymerase (Stratagene). Primer pairs for model GAA⅐TTC expansions in HEK293 cells were TAN2767 (GAG-GACGCTGTCTGAAGTCC) and MGR3537 (TGAGCAACT-GACTGAAATGCCTCAA) annealed at 64°C for 32 cycles. Primer pairs for FRDA primary fibroblasts were GAA517F (GGCTTGAACTTCCCACACGTGTT) and GAA629R (AGG-ACCATCATGGCCACACTT) annealed at 62°C for 34 cycles. Amplified products were resolved by electrophoresis on 1% agarose gels with the 1 kb Plus DNA Ladder (Invitrogen) as a marker. DNA was visualized by staining with ethidium bromide and images were acquired with a Kodak Gel Logic 440 Imaging System. Images were analyzed with Carestream Molecular Imaging Software (5.0.2.26 for Mac OS). Bar graphs were created using KaleidaGraph 4.1 for Mac OS. Student's t test for unpaired data with unequal variance was used for statistical analyses of GAA⅐TTC expansion.
Expression Plasmids-The plasmid pIRES2EGFP (Clontech) that expressed only Enhanced Green Fluorescent Protein (EGFP) from an Internal Ribosomal Entry Site (IRES) was used as a control. Human MSH2 cDNA was cut from plasmid pOTB7 (Open Biosystems) with restriction enzymes BamH1 and XhoI and cloned between the compatible BglII and Sal1 sites in pIRES2EGFP to make pMSH2IRES2EGFP (expresses both MSH2 and EGFP). The IRES and EGFP were cut out from pIRES2EGFP with restriction enzymes NheI and BsrGI and were cloned into the polylinker region of pNL-EGFP/CMV/ WPREdU3 (36) that had been linearized with NheI and BsrGI to make PNL-IRES2EGFP. PNL-MSH2-IRES2EGFP, which expresses both MSH2 and EGFP, was made in the same way. The human MSH3 cDNA was excised from pFAST-BacI-MSH3 (generous gift of Minna Nystrom) from restriction site SpeI to XmaI and inserted into PNL-IRES2EGFP linearized with NheI and XmaI to make PNL-MSH3-IRES2EGFP, which expresses both MSH3 and EGFP.
Cell Transfections-Transient transfections of plasmids were carried out using Lipofectamine 2000™ (Invitrogen) as per the manufacturer's suggestions. Transfection with the empty pLKO.1 or pIRES2EGFP vector served as a control. Following transfection, to select for successful transfection and integration, the cells receiving pLKO.1-derived vectors were passaged in media containing 2 g/ml puromycin. In specific cases, cells containing expression vectors co-expressing EGFP were sorted for expression via FACS.
Lentiviral Production and Cell Transductions-Lentiviral particles were produced from vectors derived from the pLKO.1 vector (Open Biosystems) or the pNL-EGFP/CMV/WPREdU3 vector (36). Viral particles were made via a three-plasmid expression system as described in Marino et al. (37) and Reiser et al. (36) except that the HEK293T cells were transfected using Lipofectamine 2000 (Invitrogen). Transductions were performed with HEK293 cells containing (GAA⅐TTC) 176 tandem reporter constructs in the presence of 8 g/ml of Polybrene (Sigma). Knockdowns were selected for with 1 g/ml of puromycin.
Protein Isolation and Western Blot Analysis-Cell extracts were lysed in 4x Laemmli Buffer (20% glycerol, 4% SDS, 100 mM Tris (pH 6.8), fresh 1 mM DTT). Samples were boiled at 100°C for 10 min. 10 g of each cell lysate was separated on pre-cast 4 -20% Tris-glycine gels (Bio-Rad Mini-PROTEAN TGX System) at 20 mA constant current. MagicMark XP (Invitrogen) was used as a size ladder and SeeBlue Plus2 Prestained Standard (Invitrogen) was used to establish transfer efficiency. Protein was transferred from the gel to an Immobilon-P membrane (Millipore) for 40 min at 20 V using a Bio-Rad semi-dry transfer apparatus. Membranes were blocked for 45 min using 10% evaporated Carnation Milk and 90% phosphate-buffered saline with 0.1% Tween (PBST). The membranes were rotated with primary antibodies MSH2 (Calbiochem), MSH3 (BD Transduction Labs), MSH6 (BD Transduction Labs), and ␤-actin (Sigma) diluted in 10% evaporated Carnation Milk and 90% PBST for 90 min at room temperature or overnight at 4°C. Simultaneous Western blotting of multiple MMR proteins has also been done by others (38,39). After several washes with PBST, the membranes were incubated for 45 min with horseradish peroxidase-conjugated goat anti-mouse (or goat antirabbit) secondary antibody (Pierce) in 10% evaporated Carnation Milk and 90% PBST. After several PBST washes, the ECL Advance chemiluminescence kit (Amersham Biosciences) was used to produce a signal captured by a Kodak Gel Logic 440 Imaging System. Image analysis was done with Carestream Molecular Imaging Software version 5.0.2.26 for Mac OS. Statistical analysis was done using the Student's t test for unpaired data with unequal variance.

Mismatch Repair Has a Major Role in GAA⅐TTC Repeat
Expansion-To determine the contribution of DNA repair to the expansion of GAA⅐TTC repeats in our human cellular model we used shRNA to knockdown components of Global Genome Repair (GGNER), Transcription Coupled Repair (TCNER) and Mismatch Repair (MMR). Knockdown of either MSH2 or MSH3 slowed GAA⅐TTC repeat expansion substantially, whereas knockdown of MSH6 had little effect on GAA⅐TTC repeat expansion (Fig. 1A). Protein expression levels were evaluated by Western analysis of biological triplicates. The average residual expression of the MSH2, MSH3, and MSH6 knockdowns were determined to be 35, 24, and 38% of normal expression. Representative expression of each MSH subunit knockdown is shown in Fig. 1B. Knockdown of the DNA repair protein XPA, which is central to both the GGNER and TCNER pathways, did not reduce GAA⅐TTC repeat expansion in our system (Fig. 1A) when reduced to 61% of its endogenous protein expression level (Fig. 1B). That MSH2 knockdown can slow GAA⅐TTC expansion in FRDA patient-derived induced pluripotent stem cells (iPSCs) has recently been reported (26), lending validity to our model. Based on these findings, we used lentiviral shRNA delivery to knockdown MSH2 and MSH3. Lentiviral transduction was more efficient at producing stable, long-term knockdowns necessary for our time-course studies than the plasmid-based shRNA delivery system. Importantly, the results of both plasmid-and lentiviralmediated knockdown experiments were in agreement. In addition to technical replicates, all experiments were performed using multiple, independent isolates of the target clonal cells to ensure biological reproducibility. For example, plasmid-medi- Although the DNA profiles of the clones have different appearances ( Fig. 2A), the repeats gained in the absence or presence of the MSH2 knockdown are tightly clustered (Fig. 2B). In the plasmid-mediated system, a modest reduction of MSH2 protein expression, averaging 65% of its normal level, (Fig. 2C) resulted in statistically significant changes in the repeat expansion rate of three different clonal isolates (Fig. 2B) In all cases, irrespective of the method used, GAA⅐TTC repeats in MSH2-or MSH3-knockdown cells expanded more slowly than those transduced with the empty vector control (see Fig. 3, A and B). At the protein level, a 60% decrease of MSH2 expression caused a 75% reduction in MSH3 and an 89% reduction of MSH6 (Fig. 3C, lane 3). Thus, because all MSH subunits were reduced, changes in expansion rate of MSH2knockdown cells could not be solely attributed to reduction of MSH2. These results are consistent with observations that stability of individual MSH proteins is dependent upon heterodimerization (40,41) and that, as in cancer cell lines, reduced levels of MSH2 decrease the levels of both MSH3 and MSH6 (24,42).
The initial shRNA sequences used to knockdown MSH3 (MSH3-1 and MSH3-2) in early experiments (such as those shown in Fig. 1) were moderately effective, either in combination or individually. To achieve a more robust knockdown we used the RNA folding program Mfold (43) and the siRNA design program Sfold (35) to model MSH3 mRNA structures and target accessible loops. MSH3 mRNA (NM_002439.3) is over 4.5 kilobases long and Mfold analysis revealed many predicted folded structures. We picked three loop regions predicted to exist in non-overlapping subsets of structures. We targeted these three accessible loops with shRNA sequences MSH3-3, MSH3-4, and MSH3-5. When these constructs were tested individually, none were particularly effective (not shown), however, when these three shRNA were co-expressed in the cell, the knockdown of MSH3 was robust (Fig. 3, lanes   176 inserts, the rate of repeat expansion decreased in a manner similar to that seen with the MSH2 knockdown (Fig. 3A). In contrast to what we found for MSH2 however, MSH3-targeted shRNA did not substantially affect the protein levels of MSH2 or MSH6 (Fig. 3C, compare lanes MSH2sh with MSH3sh), which were expressed at 84 and 73% of normal, respectively. These data suggest that a complex containing MSH3 could be the real basis of expansion. While MSH2 knockdown cells had a greater decrease in the rate of expansion than MSH3 knockdown cells (Fig. 3, A and B), MSH2-knockdown cells were indirectly diminished for the expression of MSH3 and MSH6 (Fig. 3C). Due to the effects observed with both the MSH2-and MSH3 knockdowns and because reduction of MSH6 to 38% of its endogenous levels did not significantly affect the rate of expansion (Fig. 1, A and B), we conclude that the major active expansion agent in our system is MutS␤ (complex of MSH2 and MSH3) rather than Mut-S␣ (MSH2 and MSH6).
The specificities of both the MSH2-and MSH3-targeted knockdowns were confirmed via rescue of the expansion phenotype. Rescue was accomplished with transduction of MSH2 or MSH3 cDNA-bearing lentiviral constructs that co-expressed EGFP (Fig. 4). Supplementation of MSH2-knockdown cells with lentiviral-mediated MSH2 cDNA expression rescued the rapid expansion phenotype. This led to an increase in the number of repeats gained after 3 weeks (Fig. 4, A and G, lane  ϩM2). Quantified repeat gains are shown in Fig. 4, G and H. Furthermore, lentiviral delivery of ectopic MSH3 cDNA fully restored rapid expansion to the MSH3 knockdown cells (Fig. 4,  C and H, lanes ϩM3). Interestingly, lentiviral-mediated expression of MSH3 ( Fig. 4A and quantified in 4G, lane ϩM3) did not rescue the expansion phenotype of MSH2-knockdown cells, and vice versa, ectopic MSH2 expression did not restore rapid expansion to MSH3-knockdown cells.
Western blot analyses of rescue experiments demonstrate both the importance of the MutS␤ complex to the expansion process and the specificity of our knockdowns (Fig. 4, B and D). Rescue of MSH2 knockdowns by ectopic MSH2 expression not only restored the rapid expansion phenotype, but also restored the expression of MSH3 and MSH6 (Fig. 4B, lane ϩM2). Therefore, the reduction in protein levels of MSH3 and MSH6 seen with the MSH2 shRNA knockdowns were likely due to the lack of MSH2 protein to form stable dimers with MSH3 or MSH6, not off-target activity of the shRNA. The competitive nature of MSH2 binding to its subunits is reflected in the reduction of endogenous MSH6 protein in the presence of ectopic expression of MSH3 (Fig. 4D, lane ϩM3). The high accumulation of both MutS␤ subunits in MSH3-supplemented cells correlates with the highest expansion rate in these sets of experiments (Fig. 4, C and H, lane ϩM3). In the presence of an empty vector control, ectopic expression of MSH2 and MSH3 yielded an increase of protein expression of 2-fold and 8.2-fold, respectively, (mock knockdown Fig. 4, E and F). This demonstrates that the effects of MutS␤ subunits can be saturated and that the

rate of expansion is not increased by an overabundance of MutS subunits.
Ectopic Expression of MSH3 Is Sufficient to Induce GAA⅐TTC Repeat Expansion in FRDA Patient Fibroblasts-Primary fibroblasts from FRDA patients (GM04078, GM03816, and GM03665; NIGMS Coriell Cell Repository) homozygous for expansions in both FXN alleles do not ordinarily undergo additional expansion when carried in tissue culture. However, the repeats in some of these cells (GM04078 and GM03816) can expand when the cells are converted to induced pluripotent stem cells (26). We reasoned that GAA⅐TTC repeats do not expand in primary fibroblasts either because: 1) the FXN gene is not transcribed as highly as it is in affected FRDA tissues such as neurons, or 2) because a factor needed for expansion was missing. We tested the second part of this hypothesis by ectopically expressing MutS␤ subunits in primary FRDA fibroblasts either individually, or in combination. We transduced FRDA fibroblasts with lentivirus expressing EGFP alone, MSH2 plus EGFP, MSH3 plus EGFP or MSH2, MSH3 and EGFP. We cultured the cells for 6 weeks, prepared DNA, and monitored the GAA⅐TTC repeat lengths of both alleles of the endogenous FXN loci. Expression of MSH3 was sufficient to cause the GAA⅐TTC repeat to expand. Representative examples shown in Fig. 5 indicate that MSH3 might be a limiting factor for GAA⅐TTC expansion in FRDA patient fibroblasts.
We have previously found that replication rate does not alter expansion rate in our HEK293-based GAA⅐TTC expansion model (32). However, a somatic expansion bias exists within post-mitotic neurons of the spinal cord in FRDA patients (29) and mouse models (44,45). To investigate whether a lack of replication contributed to this bias, we performed parallel experiments with both dividing and non-dividing primary FRDA patient fibroblasts. So that contact inhibition would block replication, cells in the non-dividing group were seeded at a high density and were not split for the duration of the 6-week time course of the experiment. A representative profile of the repeats from non-dividing cells is shown in Fig. 5A. Cells in the second, dividing experimental group were passaged normally and the profile of the repeats is shown in Fig. 5B. The expansion pattern of GAA⅐TTC repeats at either FXN allele, in both dividing and non-dividing populations of primary fibroblasts, were the same (Fig. 5, A and B). Thus, although changes in repeat stability correlated with MSH3 expression, they were not associated with the rates of cell division.
Analysis of protein expression of primary patient fibroblasts illustrated that MMR proteins are in low abundance; at best, only endogenous MSH2 was detectable (Fig. 5C, lane GFP). Consequently, quantification of absolute levels of endogenous protein expression relative to overexpression of MutS subunits was not possible. Ectopic expression of MSH2 did not visibly induce expression of MSH3 (Fig. 5C, lane ϩM2), nor did it produce expansion of the endogenous FXN GAA⅐TTC repeats to as great a degree as ectopic expression of MSH3 alone (lane ϩM2 in Fig. 5, A and B). Ectopic expression of MSH3 induced a statistically significant increase in the rate of repeat expansion. Western analysis revealed the presence of detectable amounts of both MSH2 and MSH3 following ectopic expression of MSH3 (Fig. 5C, lane ϩM3).
The Effect of MutS␤ Expression on the Rate of Expansion Can Be Saturated-When both MSH2 and MSH3 were ectopically expressed together, the two subunits accumulated to a greater degree than when either subunit was expressed independently (Fig. 5C, ϩM2ϩM3). However, co-expression of MSH2 and MSH3 did not increase the rate of expansion to a greater degree than ectopic expression of MSH3 alone (compare lanes ϩM2ϩM3 with ϩM3 in Fig. 5, A and B). Quantitative analysis of the GM04078 line revealed a statistically significant increase in the rate of expansion in both the large and small allele under both conditions. However, changes in the rate of expansion for GM04078 cells co-expressing MSH2 and MSH3 were not as robust as in cells with ectopic expression of MSH3 alone (1.37 triplet/week versus 0.82 for large alleles, 1.34 triplet/week versus 0.9 for small alleles). In FRDA fibroblast line GM03665, the change in the rate of expansion in cells with co-expression of MSH2 and MSH3 did not reach statistical significance. The variability of expansion rates seen in fibroblasts expressing both subunits simultaneously may be partly attributable to deleterious effects of overexpressing MSH2. Primary cells transduced with virus expressing MSH2, either alone or in combination with MSH3, did not divide as rapidly as controls or those cells transduced with virus expressing MSH3 alone. Furthermore, cells expressing MSH2 in fibroblast cultures that were passaged during the 6-week time course were slow to reach confluency. Effects on checkpoint activation and enhanced apoptosis have been previously described in connection with MSH2 over expression (46 -48). These reasons might contribute to the results seen with co-expression of MSH2 and MSH3.

DISCUSSION
This study establishes that the mismatch repair complex MutS␤ is a major contributor to GAA⅐TTC repeat expansion in human cells. MSH2 and MSH3 form a hetero-multimer called MutS␤ (23,49). While MutS␤ has been shown to be essential for CAG⅐CTG expansion in mice (12)(13)(14)(15)(16), there is little understanding of its role in GAA⅐TTC expansion. Our results demonstrate that knocking down either MSH2 or MSH3 using shRNA in our FRDA expansion model reduces the rate of GAA⅐TTC expansion.
MSH2 is central to DNA mismatch repair and its knockdown causes a marked reduction of both MSH3 and MSH6 protein expression levels. Thus, while MSH2 knockdown is associated with the greatest reduction of GAA⅐TTC expansion rate, the change in rate cannot be separated from the reduction of its binding partners, MSH3 and MSH6. In contrast, MSH3 knockdown reduces the rate of GAA⅐TTC expansion and does not cause obvious changes in the expression of MSH2.
In contrast to MutS␤, neither MSH6, a component of MutS␣, nor XPA, a protein central to both GGNER and TCNER, appeared to contribute to expansion in our system. It might be necessary to completely deplete or severely reduce the amount of XPA to observe a phenotypic effect. Our XPA knockdown was not robust, therefore we cannot rule out a role for nucleotide excision repair. One might also argue that partial knockdown of MSH6 (as in Fig. 1) was not sufficient to reach a physiological tipping point. However, it should be noted that even a partial knockdown of MSH2 or MSH3 clearly and significantly reduced GAA⅐TTC expansion rates. Furthermore, ectopic expression of MSH3 in control cells (no knockdown) caused a reduction in MSH6 (Fig. 4F, lane ϩM3) with no change the rate of repeat expansion (Fig. 4E, lane ϩM3). Similarly, primary patient fibroblasts do not express detectable levels of MSH6 either endogenously or with ectopic expression of MSH2 (Fig. 5C, compare GFP control to ϩM2). It has been suggested that effects of MSH6 on CAG⅐CTG repeat dynamics in mice are likely to be mediated through competition with MSH3 for MSH2 rather than direct enzymatic processes (13,15). While our data suggest that similar dynamics may operate in the context of GAA⅐TTC expansion in cultured human cells, we cannot formally exclude a role for MSH6.
To extend our observations, we studied the effects of ectopic expression of MutS␤ subunits at the endogenous FXN gene in primary FRDA patient fibroblasts. FRDA patient fibroblasts have no known propensity for instability in their native state. Previously, these same primary cells showed instability at the FXN locus after being transformed into iPS cells (26). In our case, FRDA fibroblasts were maintained in their native state and transduced with lentiviral delivery of MSH2, MSH3, or both. Western blot analysis indicated that ectopic expression of MSH3 increased the amount of endogenous MSH2 in the fibroblasts (Fig. 5C, compare lane GFP to ϩM3). This can likely be attributed to formation of stable MutS␤ protein heterodimers.
Compared with the rapid expansion model developed in our laboratory, the expansion rates in FRDA patient fibroblasts were much slower. For example, ectopic MSH3 expression caused a gain of about one repeat a week in both dividing and non-dividing fibroblasts. The slower expansion in fibroblasts may reflect: 1) the need for additional factors that affect expansion rate or 2) the low level of FXN transcription in fibroblasts. Nonetheless a gain of one repeat a week is roughly 50 repeats a year or 500 repeats in a decade. The reproducibility of the slow rate of expansion in three separate FRDA patient cell lines suggests that these fibroblasts may represent a remarkable paradigm for the somatic expansion that could underlie FRDA disease onset and progression. Furthermore, induced expansion by ectopic expression suggests that environmental insults capable of inducing the DNA repair complex MutS␤ may stimulate GAA⅐TTC repeat expansion in otherwise non-expanding tissues.
While we had expected to see synergy with co-expression of MSH2 and MSH3 in primary FRDA fibroblasts, it actually resulted in a lower rate of expansion than ectopic expression of MSH3 alone. We attribute this outcome to the deleterious effects of MSH2 over expression. Previous reports indicate that MSH2 over expression may lead to a block in replication or enhanced apoptosis (46 -48). These findings are consistent with the poor growth we noted in our MSH2-transduced fibroblasts. Our hypothesis is that effects on cell growth combined with the potential toxicity of MSH2 over expression may have preferentially reduced or eliminated the population of cells expressing MSH2 at high levels. This may have diluted the potential impact of MSH2 and MSH3 co-expression on the rate of expansion in these cells.
That higher expression of both subunits did not lead to more rapid expansion might also suggest that ectopic expression of MSH3 alone was able to combine with endogenous MSH2 to provide all the MutS␤ necessary for expansion in the fibroblasts. In such a circumstance, some other factor(s) became limiting to the expansion. In previous work, we, and others, have found that transcription is one factor involved in repeat expansion (32,50). We suggest that the low level of FXN expression in fibroblasts may have also contributed to the slower rate of expansion. Post-mitotic neurons in the DRG are characteristic sites of GAA⅐TTC repeat expansion (29) and neurodegeneration (51) in FRDA patients. These neurons are also particularly high in FXN gene expression (1), in contrast to the low level of transcription in native FRDA fibroblasts.
Interestingly, MSH3 is expressed to a higher degree in neuronal cells when compared with other somatic tissues; this may contribute to the neuronal bias for TNR expansion (16,52). HEK293 cells, upon which our system is built, have been shown to be positive for a large number of proteins expressed only, or preferentially, in neuronal cells. For example, rather than expressing filament proteins typical of kidney such as desmin or high molecular weight keratins, HEK293 cells express neurofilaments (53). The preferential transformation of neuronal cells by adenovirus 5 and cells of neuronal lineage in embryonic kidney are thought to be the reason HEK293 is most likely of neuronal origin (53,54). Thus, our ability to study the role of DNA repair on TNR expansion with an accelerated time course is aided by the neuronal nature of HEK293 cells.
The mechanisms that drive expansion in mammalian cells are still unclear; however, transcription and DNA repair might provide a unifying theme in TNR expansion. The instability of disease-associated repeat sequences is generally attributed to the ability of these sequences to adopt non-B DNA structures: GAA⅐TTC repeats have been shown to adopt triplex and triplex-associated structures (7-10, 55, 56) and hairpin structures are formed by CAG⅐CTG (57)(58)(59) or CGG⅐CCG (60) trinucleotide repeats.
In Fig. 6, we present a model for small, incremental increases in GAA⅐TTC that are initiated by transcription and mediated by MutS␤. The model may be applicable to all TNR expansions. Fig. 6A shows part of an expanded GAA⅐TTC repeat at rest, each GAA is base-paired in register to its corresponding TTC partner. In vivo, these pairs are only likely to be separated during replication or transcription. Because most somatic expansion occurs in non-dividing cells, in this particular case, transcription is likely the pertinent strand-separating process. In fact, we have shown that transcription is linked to GAA⅐TTC expansion (32), and others have shown that transcription increases CAG⅐CTG expansion (50) or contraction (31,50).
We have evidence that transcription into a GAA⅐TTC repeat can lead to polymerase stalling and RNA⅐DNA hybrid formation both in vitro (10,56) and in bacteria (56). Fig. 6B illustrates our theory that structures formed during transcription initiate the expansion process. Further evidence that this structure might attract DNA mismatch repair enzymes was illustrated by Ku et al. (26) in showing enriched binding of MSH2 and MSH3 (but not MSH6) at the promoter distal end of the repeat in an iPS derivative of FRDA fibroblasts. Finally, our finding that GAA⅐TTC expansion preferentially occurs in the promoter distal end of a repeat in human cells (32) also supports this model. In the CNG repeat model, other structures may form that might include but are not limited to RNA⅐DNA hybrids (61,62). Regardless of the structure formed during transcription, the next step in our hypothesis is that a misalignment occurs during structure resolution. This misalignment might loop out one or a few triplets (as shown in Fig. 6C). Although most early models have assumed a replication-based mechanism and relatively large jumps, such slipped strand structures have long been suspected to play a role in repeat expansion (63,64). Recently, these earlier models have evolved to include small loop-outs of CAG⅐CTG repeat units that are targeted by MutS␤ (39), in close agreement with our data and model for GAA⅐TTC expansion (see below).
Our data indicate that somatic expansion of GAA⅐TTC repeats occurs in very small increments such as the gain of one repeat per week in MSH3-supplemented FRDA patient fibroblasts (Fig. 5); these small gains point to a specific role of very small loops in GAA⅐TTC repeat expansion. Furthermore, functional tests and structural modeling have indicated that small insertions (Ͻ5) occupy a unique niche in the spectrum of mismatches that MutS␤ works with (65). Insertions of this size might be more difficult for the MutS␤ protein to accommodate (65). It is possible that discrimination can occur with respect to binding short loops of differing flexibility. A recent report on the crystal structures of MutS␤ bound to small DNA loops showed that it bound three phosphates in the loop (66). This finding suggests that a three-base loop is the minimal substrate for MutS␤ and on a broader level might also explain the seeming preferential instability of trinucleotide repeats. It is possible that MutS␤ repairs small loops less effectively or asymmetrically, and this leads to a gain of DNA (Fig. 6D) thus mediating the expansion process.
We suggest that errant DNA mismatch repair may be a common denominator among the array of diseases caused by DNA repeat expansion. Our results in combination with findings from Ku et al. (26) and Ezzatizadeh et al. (27) strongly implicate MMR as a critical component of GAA⅐TTC expansion. Although formal proof that somatic expansion contributes to FRDA progression has yet to be established, from the perspective of identifying therapeutic targets, it may eventually prove to be a way to address disease progression. For instance, histone deacetylase (HDAC) inhibitors have been an area of intense focus as potential FRDA therapeutics (67)(68)(69). Recently, it has been reported that histone deacetylase complexes promote instability in threshold size (CAG⅐CTG) 20 repeats in budding yeast and human astrocytes. HDAC inhibitors targeting those complexes strongly reduced the propensity of the (CAG⅐ FIGURE 6. Transcription leads to small loop targets for MutS␤ in model of GAA⅐TTC repeat expansion. A, part of a GAA⅐TTC repeat is depicted with the purine (or R) strand in red, and the pyrimidine (or Y) strand in yellow. The numbers associated with the bases indicate that the bases are registered in alignment. B, during transcription the two strands may be separated by formation of a variety of structures, one example is depicted here (56). C, resolution of a structure can lead to an out-of-register re-annealing within the repeat. The slipped annealing leads to a small GAA bubble at triplet 86. This small loop becomes a substrate for MutS␤. A compensatory bubble in the TTC strand elsewhere in the molecule may provide an alternative target for MutS␤ (not shown). D, shows TNR expansion has occurred with the addition of a single trinucleotide (*) after repair initiated by MutS␤. CTG) 20 repeats to expand (70). If this effect translates to the longer GAA⅐TTC repeats in FRDA, then the HDAC inhibitors currently being studied as potential FRDA therapeutics may serve double duty by: 1) increasing transcription through the repeat and 2) suppressing further expansion.
MutS␤ is key to GAA⅐TTC expansion; if somatic expansion is confirmed to contribute to FRDA progression, then MutS␤ should be targeted therapeutically. Notably however, the translational potential of MSH2, an essential component of MutS␤, is limited because MSH2 inactivation is strongly associated with hereditary nonpolyposis colorectal cancer (HNPCC) (71,72). MSH3 however, has limited association with cancer pathology. Multiple studies have shown that MSH3 both contributes and is rate-limiting to CAG⅐CTG repeat expansion (12)(13)(14)(15)(16). MSH3 has been considered a potential therapeutic target in the context of CAG⅐CTG repeat expansions (3,16,52). Accordingly, because we have demonstrated that MSH3 expression is required for GAA⅐TTC expansion in primary FRDA patient cells, it is possible to extend this consideration to FRDA.