MSH2-deficient Human Cells Exhibit a Defect in the Accurate Termination of Homology-directed Repair of DNA Double-Strand Breaks¹

MMR³ is a highly conserved repair system that corrects mismatches arising during DNA replication and safeguards genomic integrity (1, 2, 3, 4, 5). Defective MMR is strongly associated with HNPCC, an autosomal dominant inherited disease characterized by early onset colon tumors, as well as cancers of the endometrium, stomach, upper urinary tract, small intestine, and ovary (6, 7). MMR consists of at least seven proteins, including four bacterial MutL homologues, hMLH1, hMLH3, hPMS1, hPMS2, and three MutS homologues, hMSH2, hMSH3, and hMSH6. Defects in hMSH2 and hMLH1 appear to be the cause of most HNPCCs (8).

MMR proteins function as heterodimeric complexes. hMSH2 can form a heterodimer with either hMSH3 or hMSH6. hMSH2 is a uniformly essential component of all of the heterodimers, whereas hMSH3 and hMSH6 modify substrates specificity. hMSH2-hMSH6 heterodimer recognizes single bp mismatches and small IDLs, whereas hMSH2-hMSH3 heterodimer recognizes an overlapping set of small IDLs as well as larger IDLs, containing up to 12 nucleotides (9, 10). The relatively low number of hMSH6 and hMSH3 mutations in HNPCC kindreds has been attributed to their imparted redundancy in substrate specificity. hMLH1 can form a heterodimer with hPMS2, hPMS1, or hMLH3 (11). The three types of hMutL complexes are presumably functionally redundant. hMutS plays a central role in initiating mispair recognition and binding, and hMutL acts as a molecular matchmaker between hMutS and downstream proteins to complete the repair process (1, 2, 3, 4, 5). Therefore, both hMSH2 and hMLH1 are critically important to MMR activity, as reflected by their predominant alteration in HNPCC patients, as well as in most sporadic tumors with microsatellite instability.

The majority of cells deficient in MMR exhibit a mutator phenotype characterized by a 100-1000-fold increment in the rate of spontaneous mutation at miscrosatellites and coding sequences (12, 13, 14, 15). However, studies on mice with MMR gene knockout indicate that mutations of MMR genes other than hMSH2 and hMLH1 may induce a mutator phenotype but not necessarily an increase in colon tumors, implying that a mutator phenotype alone may not be sufficient for intestinal tumor formation (16). In addition to editing replication, MMR proteins have also been implicated in the editing of recombination between divergent sequences, in transcription-coupled repair of DNA damage, and are also thought to function as lesion sensors for certain types of DNA damage that kill by triggering apoptosis (1, 3, 4, 5, 17). Moreover, MMR proteins inhibit gene amplifications (18, 19) and have been associated with the fidelity of DNA DSB repair, although the underlying mechanisms remain unknown (19, 20, 21, 22, 23).

DNA DSBs are common lesions that occur in all of the cells, and DSB repair is a fundamental mechanism of genome protection. There are two major DSB repair pathways in mammalian cells: homologous recombination and NHEJ (24). Defects in either pathway have been linked with genome instability and cancer (25, 26, 27). In mammalian cells, DSB repair by homologous recombination occurs almost exclusively by GC: a nonreciprocal transfer of genetic information from an intact homologous duplex to a broken duplex (28, 29, 30). In this pathway (31), DSBs are first processed to yield 3′-ended single-strand tails (Fig. 1,a) that invade a homologous donor duplex (Fig. 1,b). DNA synthesis is then primed from the 3′ end of the invading strand, which results in the copying of donor information (Fig. 1,c). DSB repair may be completed through displacement of the newly synthesized strand from the donor template and its annealing with the 3′ noninvading single-stranded end to elicit GC only (Refs. 32, 33, 34, 35, 36; Fig. 1,d). DSB repair by GC occurs with fidelity and reflects the capacity of cells to precisely restore the damaged DNA with no loss or gain of nucleotide sequences (Fig. 1,d). NHEJ differs from GC in that there is no requirement for extensive homology between DSB ends, and is potentially error-prone, as nucleotide insertions and deletions are tolerated at the rejoining sites. GC and NHEJ compete with one another (37), but very little is known about how eukaryotic cells elect one or the other (38). Under certain circumstances, eukaryotic cells can couple GC and NHEJ to seal DSBs and ensure their genome stability (Fig. 1 e; Refs. 30, 32, 33, 34, 39, 40, 41, 42, 43, 44, 45). In these cases, GC intermediates are interrupted at various stages and shunted to NHEJ, being characterized by deletions, insertions, or duplications of DNA sequences at NHEJ junctions (30, 32, 33, 34, 39, 40, 41, 42, 43, 44, 45, 46). Such rearranged GC products can be as frequent as PGC products (32, 33, 34, 39, 40, 41).

Here, we report that the proportions of precise and rearranged GC products vary according to the MMR status of the cells. In MMR-proficient or MLH1-deficient cells, >50% of GC products were of the precise type as has been reported with several distinct MMR proficient backgrounds (30, 32, 33, 34, 39, 40, 41, 42, 43, 44, 45), whereas in two MSH2-deficient cell lines, this proportion decreased to 8%, whereas that of rearranged products increased by 2-fold. These results suggest a defect in the accurate termination of GC in MSH2-deficient cells and seem to predict a novel way by which MSH2-deficiency could promote mutation: deletion or insertion mutations associated with DSB repair.

The tumor cell lines used in these experiments originated from the American Type Culture Collection (Rockville, MD). They were: the MMR-proficient human colorectal carcinoma cell line SW480 (47), the hMLH1-deficient colorectal carcinoma cell line HCT116 (47), the hMSH2-deficient colorectal carcinoma cell line LoVo (48), and the hMSH2-deficient uterine tumor cell line SK-UT-1 (49) grown in Leibovitz’s L15, McCoy’s, Ham’s F12, and MEM media, respectively. All of the media were supplemented with 10% fetal bovine serum (20% for LoVo cells), and gentamicin (50 μg/ml). Except for SW480 cells, which were cultured without CO₂, all of the other cells were cultured at 37°C in a humidified 5% CO₂ incubator.

The recombination reporter plasmid pST100, shown in Fig. 2, was linearized with I-Sce1 restriction enzyme (Fig. 2) and introduced with the plasmid pGKpuro into cells by electroporation as described previously (44). The electroporated cells were divided in two separate cultures. Forty-eight h later, puromycin (Sigma, St. Louis, MO) containing medium was added to one third of the transfected cells, and G418 (Geneticin; Life Technologies, Inc., Grand Island, NY) containing medium was added to the remaining two thirds. The media were changed two to three times per week, and the surviving colonies were counted, picked, and expanded in their corresponding medium for additional analysis.

Genomic DNA from individual G418^R colonies (recombinants) was extracted and digested with the restriction enzyme Nae1 that cut once in pST100 at the neo3′ cassette. Colonies with single integration events were determined by Southern blot hybridization, using the Hyg gene as a probe. More than 80% of G418^R colonies analyzed contained single pST100 recombinant molecules. For analysis of the recombinants (Fig. 2), the following primer pairs were used in PCR: (a) Neo952 (5′-ccacgacgggcgttccttgcgcag-3′) and neo1300 (5′-gtcacgacgagatcctcgccgtc-3′) amplify the 1.4 kb fragment, or 1.4 and 2.4 kb fragments, between the realigned neo cassettes; (b) Neo800 (5′-gaatagcctctccacccaag-3′) and hyg4419 (5′-gctgtgtagaagtactcgccg-3′) amplify the 2.1 kb fragment between the transferred neo5′ cassette and the Hyg gene; (c) Neo1300 and hyg4419 amplify the 3.6 kb fragment between the neo3′ cassette and the Hyg gene; (d) Neo800 and pUC469 (5′-tgaccatgattacgccaagct-3′) amplify the 1.7 kb fragment between the neo5′ cassette and pUC sequences; and (e) Neo1300 and pUC469 amplify the 3.2 kb fragment between the inverted neo3′ cassette and pUC sequences (Fig. 2). PCR amplifications were conducted with the Expand High Fidelity PCR System kit (Roche Diagnostic, Indianapolis, IN) in a thermal cycler for 30 cycles. The PCR products were analyzed by restriction enzymes, Southern blot hybridization, and DNA sequencing as described previously (44).

Complete hMLH1 and hMSH2 cDNAs used to correct the MMR deficiency of HCT116, LoVo, and SK-UT-1 cells, respectively, were generous gifts from Drs. Nick Nicolaides (Magainin Institute of Molecular Medicine, Plymouth Meeting, PA) and Richard Kolodner (Ludwig Institute for Cancer Research, La Jolla, CA), respectively. These cDNAs were subcloned in pCEP-4 under control of the CMV promoter. The plasmid pCEP-4 (Invitrogen, Carlsbad, CA) replicates autonomously in human cells and contains the Hyg gene. After transfection with pCEP-cDNA constructs, 20 HYG^R clones were picked for each cell line, and their protein extracts were subjected to Western blot analysis, using antibodies against hMSH2 (Ab-3) and hMLH1 (Ab-2) at 2 μg/ml, and affinity-purified rabbit antiserum against these antibodies at a dilution of 1:1000 (Oncogene, Boston, MA). SW480 cells express hMSH2 and hMLH1, and nontransfected LoVo and SK-UT-1 cells express only hMLH1, whereas nontransfected HCT116 cells express hMSH2 but not hMLH1 (data not shown). However, except for 3 clones expressing MLH1 at a barely detectable level, none of the 60 HYG^R clones analyzed was expressing hMSH2 or hMLH1. Because the Hyg gene is upstream of the CMV promoter and both are transcribed in the same orientation, we thought that the lack of cDNA expression might be because of transcriptional interference: epigenetic silencing of the CMV promoter by transcriptional activity of the upstream Hyg gene (50). Therefore, we inverted the orientation of the Hyg gene relative to the CMV promoter. We analyzed 20 new HYG^R clones from each cell line, but again none of these was expressing the transfected cDNA. While conducting these studies, it was reported that the expression of MSH2 and MLH1 cDNAs is toxic to LoVo and HCT116 cells (51). This would explain why, in our hands, only HYG^R clones that did not express hMSH2 or hMLH1 grew in culture.

The intron-based inverted repeat recombination assay system used in this study was described previously (44). Briefly, recombination between two inverted introns, L1 (L1Md) sequences, will realign two flanking inverted neomycin-resistance (Neo) gene truncated cassettes (neo5′ and neo3′; Fig. 2). This produces a spliceable Neo gene that confers resistance to G418 drug. The use of intron allows the detection of both precise and rearranged products (Ref. 44; Figs. 1 and 2).

A DSB was introduced in one copy of the intron L1Md at the I-Sce1 restriction site in the reporter plasmid pST100 (Fig. 2). The linear plasmid pST100 was introduced with circular pGKpuro into the following human cancer cell lines: MMR-proficient SW480, hMSH2-deficient LoVo and SK-UT-1, and hMLH1-deficient HCT116. These cell lines have been widely used to study the effects of MMR on mutation rates, DNA damage tolerance, microsatellite instability, and genome rearrangements (1, 3, 4, 5, 18, 19). The plasmid pGKpuro contains the puromycin gene that confers puromycin (PURO) drug resistance to cells; its inclusion was a convenient way for assessing transfection efficiencies and for normalizing recombination frequency data obtained in different experiments. It allowed the comparison of normalized recombination frequencies (44, 52, 53), expressed here as the number of stable G418^R colonies divided by the number of PURO^R colonies. Recombinant molecules of pST100, which do not replicate autonomously, must integrate randomly by nonhomologous recombination into the genome of transfected cells to generate stable G418^R colonies. Analysis of the recombinants was done by PCR on genomic DNA of stable G418^R colonies that had integrated single copy recombinant molecules as determined by Southern blot hybridization, using the Hyg gene as a probe (see “Materials and Methods”).

MMR-proficient cells recombined transfected DNA more efficiently than MMR-deficient cells (Table 1). The frequency of homologous recombination (homologous:nonhomologous ratio) with MMR-proficient SW480 cells was 8-fold higher than with either MLH1-deficient HCT116 or MSH2-deficient SK-UT-1 cells, and 2-fold higher than with MSH2-deficient LoVo cells.

DSB repair by GC will transfer, presumably through DNA synthesis (Fig. 1), the neo5′ cassette from the intact L1Md copy to the broken L1Md copy (Fig. 2). GC can occur within one pST100 molecule (Fig. 2) or between two pST100 molecules (data not shown). Such events can be characterized by PCR using different sets of primer pairs on genomic DNA of G418^R colonies. A GC event would generate PCR fragments of 1.4 kb between the two realigned neo3′ and neo5′ cassettes, 2.1 kb between the transferred neo5′ cassette and the Hyg gene, 3.6 kb between the neo3′ cassette and the Hyg gene, and 2.4 kb between the two neo5′ cassettes when the original neo5′ cassette is maintained after integration of the recombinant molecule (Fig. 2).

PCR analysis of G418^R colonies with MMR-proficient SW480, MLH1-deficient HCT116, MSH2-deficient LoVo, or SK-UT-1 cells revealed that DSB repair products were almost exclusively GC events (Table 2). Analysis of G418^R colonies in previous studies on the MMR-proficient mouse LTK⁻ cell line (44) gave similar results (Table 2). Representatives of these recombinants can be seen in Fig. 3.

With MMR-proficient SW480 and LTK⁻ cells, and MLH1-deficient HCT116 cells, >50% of GC products were precise, whereas in MSH2-deficient LoVo and SK-UT-1 cells, only 8% were of the precise type (Table 2). Representatives of the precise type are shown in Fig. 3,A (a and b), and their corresponding full-length PCR fragments (1.4 and 2.4 kb, 2.1 kb, 1.7 kb, 3.6 kb, and 1.4 kb) appear in Fig. 3,B (Lanes 2–6, respectively). Restriction enzyme analysis of 1.4-kb PCR fragments from 40 distinct recombinants disclosed similar patterns, and sequencing of 3 revealed no small deletions, insertions, or substitutions of nucleotides compared with the parental L1Md sequence (data not presented). The remaining GC products were rearranged, because they generate PCR fragments that are smaller or bigger than those generated with precise products (Fig. 3,A, panels c–e; Fig. 3,B, Lanes 7–14). Restriction enzyme analysis of these fragments showed deletions and insertions of variable sizes, and DNA sequencing of 5 reduced-size PCR fragments smaller than 1.4 kb (Fig. 3,B, Lanes 10–12) revealed NHEJ flanked on both sides by pST100 L1 sequences (Fig. 3,A, panels d and e), whereas sequencing of three bigger size PCR fragments, between the transferred neo5′ cassette and the Hyg gene, evinced NHEJ with insertions and duplications of L1 sequences (Fig. 3,A, panel c; Fig. 3,B, Lanes 7, 13, and 14). The NHEJ junctions between the transferred neo5′ cassette and the Hyg gene incurred both deletions and insertions (Fig. 3,A, panel c), whereas those between the two realigned neo3′ and neo5′ cassettes all presented deletions (Fig. 3 A, panels d and e). Insertions that would lengthen the intron were not detected; they probably interfere with intron splicing (44).

The decreased frequency of homologous recombination (homologous:nonhomologous ratios) observed with MMR-deficient cells relative to wild-type cells could be attributable either to reduced homologous recombination efficiency or to increased nonhomologous integration efficiency. Although our results cannot discriminate between these two possibilities because of unsuccessful complementation of MSH2- or MLH1-deficient cells (see “Materials and Methods”), recent studies have indicated the heightened efficiency of nonhomologous integration of linearized plasmid or retroviral DNAs in hMSH2- and hMLH1-deficient cells (19). However, our frequency data show no correlation between the efficiency of nonhomologous integration and the proportions of precise and rearranged GC products (Tables 1 and 2). The ratio of homologous:nonhomologous recombination obtained with MMR-proficient cells was 8-fold higher than with MLH1-deficient HCT116 cells, and yet both cell lines gave similar proportions of precise and rearranged products. The ratio observed with MSH2-deficient SK-UT-1 cells was similar to that obtained with HCT116 cells, yet both cell lines presented different proportions of precise and rearranged products. Similarly, the ratio with LoVo cells was 4-fold higher than with SK-UT-1 cells, but these two cell lines gave similar proportions of precise and rearranged products.

The NHEJ joints in the rearranged GC products analyzed were flanked on both sides by pST100 sequences, indicating that such joints occurred before nonhomologous integration of the recombinants in the chromosomes of host cells. Such rearranged GC products occur in all of the organisms studied (30, 33, 34, 39, 40, 41, 42, 43, 44, 45, 54) and are best explained with the one-sided invasion model proposed for DSB repair (Fig. 1), in which only one DSB end invades the homologous template and primes DNA synthesis, whereas the other end terminates DSB repair either by annealing or by NHEJ (32, 33, 34, 39). In this model, the length and nature of the sequences acquired by the recipient broken molecule are only dependent on the extent of DNA polymerization. In this way, if DNA synthesis extends beyond the homologous region shared by the two participant duplexes and continues into flanking nonhomologous sequences, NHEJ of the released, newly synthesized strand and the noninvading end would generate a recipient molecule with insertions/duplications (30, 32, 33, 34, 39, 40, 41, 42, 43, 44, 45, 46). However, if DNA synthesis does not extend across the shared homologous region, NHEJ would generate a recipient molecule with deletions. Unlike insertions/duplications, deletion events would be detected only under nonselective conditions (32, 33, 34, 39, 54).

In MMR-proficient cells, the proportion of GC associated with NHEJ events can reach but never exceed that of PGC events, even when no constraints are imposed on such events (32, 34, 39, 40, 41, 42, 44). However, in 2 MSH2-deficient cells, the proportion of PGC products decreased by 7-fold, and that of rearranged products increased by 2-fold, compared with MMR-proficient or MLH1-deficient cells (Table 2). These results suggest a defect in the accurate termination of GC in MSH2-deficient cells. Although the exact molecular mechanism underlying such a defect remains to be elucidated, two possibilities can be entertained. First, because NHEJ became highly efficient in MMR-deficient cells (19), this process would out-compete homologous recombination (strand annealing) in terminating DSB repair (Fig. 1). However, the fact that NHEJ increased in both MSH2- and MLH1-deficient cells (19), and that inaccurate termination of DSB repair decreased only in MSH2-deficient cells argues against this possibility. Second, increased NHEJ events may be a consequence of the formation of unstable GC intermediates; e.g., the annealed region between the newly synthesized strand and the noninvading end could be very short (Fig. 1,d) and/or contain protruding 3′ single-strand tails (Fig. 1 f). The latter would be expected to form when the released, newly synthesized strand is longer than the noninvading end. In yeast, removal of such protruding 3′ nonhomologous tails from GC intermediates involves the participation of the nucleotide excision repair endonuclease RAD1-RAD10 and the heterodimer MSH2-MSH3, but not MSH6, MLH1, or PMS1 proteins (55). It has been postulated that the MSH2-MSH3 complex stabilizes GC intermediates by binding to unpaired 3′ nonhomologous single strands at the ends of the annealed region, and this allows RAD1/RAD10 to locate and cleave such unpaired strands (55). Mutations in XPF/ERCC1, the mammalian homologue of RAD1/RAD10, also lead to an increased proportion of homologous recombination associated with NHEJ when nonhomologous termini are present at DSBs (56). Taking these studies into consideration, it is easy to imagine that unstable GC intermediates would unwind and be shunted to NHEJ, which could explain the shift from precise to rearranged GC products.

The results reported show a defect in the accurate termination of DSB repair by GC, and an increase in the coupling of GC and NHEJ in MSH2-deficient cells. This seems to predict an additional way by which MSH2 deficiency could contribute to the accumulation of genomic rearrangements and, thus, to cancer predisposition.

We thank Drs. Nick Nicolaides (Magainin Institute of Molecular Medicine, Plymouth Meeting, PA) and Richard Kolodner (Ludwig Institute for Cancer Research, San Diego, CA) for providing the cDNA of hMLH1 and hMSH2, respectively. We also thank Shona Teijeiro, Fatima Zouanat, Nathalie Marçal, Zeinab Daher, and Caroline Lejeune for their technical assistance, and Ovid Da Silva for editing this text.