Apoptosis Induced by Overexpression of hMSH2 or hMLH1¹

HNPCC³ is an autosomal dominant inherited disease that is characterized by early-onset colon tumors as well as cancers of the endometrium, stomach, upper urinary tract, small intestine, and ovary (1, 2). Mutations in one of two mismatch repair genes (hMSH2 and hMLH1) appear to be the cause of the great majority of HNPCC patients (3). The remaining active allele of the mismatch repair gene is typically mutated as an early event in the development of tumors in these individuals (4), leading to the loss of function of this repair pathway and the induction of microsatellite instability.

hMSH2 is a member of a family of proteins that are homologous to the Escherichia coli MutS protein, which is responsible for the initial recognition of mismatched nucleotides generated during DNA replication (5, 6). Heterodimers formed between hMSH2 and two other MutS homologues (hMSH3 and hMSH6) are required for the repair of these DNA replication errors (789, 10, 11). Although hMSH2 is a uniformly essential component of all of the heterodimers, it appears that the hMSH3 and hMSH6 modify substrate specificity. Interestingly, the hMSH2-hMSH6 heterodimer recognizes single bp mismatches and small nucleotide IDLs, whereas hMSH2-hMSH3 heterodimers recognize an overlapping set of small IDLs as well as larger IDLs, containing up to 12 nucleotides (9, 12). The relatively low number of hMSH6 mutations in HNPCC kindreds and the lack of hMSH3 mutations in HNPCC kindreds have been attributed to the redundancy in substrate specificity imparted by hMSH3 and hMSH6. Three human homologues of the E. coli mismatch repair gene MutL have been identified (hMLH1, hPMS1, and hPMS2; Refs. 13, 14, 15). hMLH1 and hPMS2 proteins form a heterodimer that is essential for mismatch repair (16). Cell lines that are deficient in either of these components are deficient in the repair of all these types of replication errors (17, 18).

Most cells deficient in mismatch repair develop a mutator phenotype characterized by a 100-1000-fold increase in the rate of spontaneous mutation at microsatellites and coding sequences (4, 19, 20, 21). Because mutations at tumor suppressor genes in tumors from HNPCC patients fall in hot spots similar to those found in mismatch repair-deficient cells (22, 23, 24, 25), the mutator phenotype appears to play a crucial role in the progression of this disease. However, studies of mice with knockouts of the mismatch repair genes indicate that mutations of mismatch repair genes other than Msh2 or Mlh1 (in particular, Pms2) may induce a mutator phenotype but not necessarily an increase in colon tumors (26). Thus, a mutator phenotype alone may not be sufficient for intestinal tumor formation (26).

In the process of investigating the role of mismatch repair gene mutations on the mutator phenotype of various tumor cell lines, we found that the overexpression of some mismatch repair proteins from cDNA constructs induced apoptosis. Here, we used microinjection of these expression constructs to show that overexpression of hMSH2 or hMLH1 but not the other known mismatch repair proteins induces apoptosis in repair-proficient or -deficient cells. We further demonstrate that primary fibroblasts derived from Msh2-deficient mice lose the ability to induce apoptosis after treatment with the DNA-damaging agent MNNG. Considering that mutations of hMSH2 and hMLH1 are found in the great majority of HNPCC patients, we propose that an altered apoptotic pathway associated with defects in the hMSH2 or hMLH1 genes may be an additional factor in the development of this disease.

The tumor cell lines used in these experiments originated from the American Type Culture Collection (Manassas, VA). These were the hPMS2-deficient human endometrium adenocarcinoma cell line HEC-1-A (27), the hMSH2-deficient uterine tumor cell line SK-UT-1 (28), and the hMLH1-deficient colorectal carcinoma cell line HCT116 (17). The human embryonic kidney epithelial cell line 293 was obtained from Dr. Don Ayer, Huntsman Cancer Institute, University of Utah (Salt Lake City, UT). Wild-type (Msh2 +/+) and knockout (Msh2 −/−) mouse primary cultures were obtained from 13.5 days postcoitum whole mouse embryos by mechanical disaggregation, followed by sieving the cells through a 70-μm cell strainer and plating out. Cells were grown in DMEM (Life Technologies, Inc., Rockville, MD) supplemented with 10% fetal bovine serum (HyClone) at 37°C in a 5% CO₂ atmosphere.

Monoclonal antibodies against hMLH1 (Oncogene Research Products) and hPMS2 (PharMingen) were used at dilutions of 1:1000 for Western blotting and 1:50 for indirect immunofluorescence. The affinity-purified rabbit antiserum against hMSH2 (Oncogene Research Products) and the rabbit antiserum against hMSH6 were used at a dilution of 1:1000 for Western blotting and 1:20 for immunofluorescence.

The complete cDNA for hMSH2, hMSH3, hMSH6, hMLH1, and hPMS2 were subcloned from laboratory isolates (9) into the pcDNA3 vector (Invitrogen), placing the expression of these cDNAs under the control of the CMV promoter. The expression construct for GFP (pGFP), which also places the expression of this gene under the control of the CMV promoter, was obtained from Life Technologies, Inc. (pGreen Lantern-1).

Cells were plated onto uncoated etched grid glass coverslips (Bellco, Vineland, NJ) at least 1 day prior to microinjection. The expression constructs together with pGFP (at a concentration of 0.5 μg/μl each in 0.5 × PBS) were injected into the nuclei of cells using an Eppendorf 5242 microinjector attached to a 5171 motorized micromanipulator with a glass microcapillary. For each experiment, at least 100 cells were injected. The pcDNA3 vector and pGFP were injected on the same coverslip for negative control to minimize variations caused by differences in culture conditions, immunostaining, or TUNEL assays.

At 24 h postmicroinjection, cells were washed with PBS, fixed with 1% formaldehyde (methanol free) in PBS for 15 min at 4°C, and permeabilized with 70% cold ethanol for a minimum of 1 h at −20°C. The cells were then washed with PBS and incubated with indicated antibody for 30 min at 37°C. Following washing with PBS, cells were incubated for 30 min at 37°C with the corresponding secondary antibody conjugated to Texas Red (Southern Biotechnology Associates, Inc., Birmingham, AL) or rhodamine (Boehringer Mannheim, Indianapolis, IN). Nuclei were counterstained with Hoechst 33258 (Aldrich Chemical Co., Milwaukee, WI) in antifade.

Cells were plated into the 35-mm plates with or without etched grid glass coverslips. The following day, cells were preincubated with 20 μm O⁶-benzylguanine for 1 h at 37°C and exposed to different concentrations (10, 50, and 100 μm) of MNNG (Aldrich) in serum-free medium containing 20 μm O⁶-benzylguanine for 45 min. MNNG-treated cells were then grown in complete DMEM for 24 h at 37°C and used for TUNEL assay and cell viability assay. Colonies formed after 2 weeks were stained and counted.

Cells were fixed 24 h after microinjection with 1% formaldehyde (methanol free; Polysciences, Inc., Warrington, PA) in PBS for 15 min at 4°C and permeabilized with 70% cold ethanol for a minimum of 1 h at −20°C. TUNEL assay (Boehringer Mannheim) was carried out following manufacturer’s instructions with 2.5 units of terminal deoxynucleotidyl transferase in a 100-μl reaction. After washing with PBS, cells were stained with avidin-Texas Red (Boehringer Mannheim), and nuclei were counterstained with Hoechst 33258 in PBS (10 μg/ml), followed by examination by fluorescence microscopy (Zeiss; Axiovert 100).

To correct the mismatch repair defect in two human tumor cell lines, we transfected cDNA expression constructs for hMSH2 and hPMS2 into tumor cell lines with mutations of these repair genes (SK-UT-1 and HEC-1-A, respectively; Refs. 27 and 28). Transient expression of the two proteins was followed by Western blotting and indirect immunofluorescence. Expression of hMSH2 was detected in SK-UT-1 cells 8 h after transfection. However, cells overexpressing hMSH2 developed severe nuclear abnormalities, characteristic of cells undergoing apoptosis. After 72 h, no cells expressing hMSH2 were detectable. In contrast, HEC-1-A cells expressing hPMS2 persisted, and colonies stably expressing this protein were isolated (29).⁴ Although these data suggested that overexpression of hMSH2 but not hPMS2 was toxic to cells, it was difficult to score satisfactory numbers of cells expressing these proteins and to follow their fate by transient transfection. To obviate these technical problems, we microinjected cDNA expression constructs of the mismatch repair genes hMSH2, hMSH3, hMSH6, hMLH1, and hPMS2 (phMSH2, phMSH3, phMSH6, phMLH1, and phPMS2) into both repair-deficient and -proficient cell lines. To follow the fate of the microinjected cells over several days, we coinjected the mismatch repair expression constructs with a GFP expression construct (pGFP). Similar transient and high-copy expression studies have been used successfully in Saccharomyces cerevisiae as a mechanism to unmask genetic interactions (30, 31) and in human cells to identify novel forms of genetic instability (32).

When phMSH2 and pGFP expression vectors were coinjected into SK-UT-1 cells, apoptotic bodies appeared by 24 h (Fig. 1). Moreover, at 96 h, 50–70% of the cells expressing GFP (including the apoptotic cells found at earlier times) were no longer detectable (Table 1). Coinjection of the empty pcDNA3 vector with pGFP revealed a significantly (P < 0.001) lower level of cell death. Coinjection of phMLH1 and pGFP into the hMLH1-deficient line HCT116 revealed a similar percentage of cells undergoing cell death (Table 1; Fig. 1). Microinjection of the empty expression vector and pGFP again revealed a significantly (P < 0.001) lower level of death. Microinjection of the expression constructs into the mismatch repair-proficient human embryonic kidney epithelial cell line 293 gave nearly identical results (Fig. 2). Up to 67% of 293 cells injected with phMSH2 or phMLH1 were lost from the fields, and apoptotic bodies were detected in up to 19% of cells expressing GFP over the course of the experiment. Only 7.5% of 293 cells injected with the empty expression vector were lost over the same time period.

When HEC-1-A cells were microinjected with phPMS2 and pGFP, the fraction of cells observed with an apoptotic morphology was lower (Table 1), and this amount of cell death was not significantly different (P > 0.5) from that found in HEC-1-A cells injected with the empty vector and GFP. There appeared to be a higher background of apoptotic cells in the injected HEC-1-A cells, but this was not increased by injection of phPMS2. The total fraction of cells that were lost after injection of phPMS2 was 2.8–3-fold lower than the level obtained after injection of phMSH2 or phMLH1. The diminished response of HEC-1-A cells does not appear to be a consequence of some difference in the genetic background of this tumor cell line as 293 cells injected with phPMS2 together with pGFP also had a substantially lower level of apoptosis and death (Fig. 2). In fact, there was no significant difference (P > 0.98) between the fraction of GFP-expressing 293 cells surviving after injection with phPMS2 and the fraction surviving injection with the empty expression vector. Similarly, injection of phMSH3 and phMSH6 did not induce death in 293 cells (Fig. 2). Ninety-two percent of 293 cells injected with phMSH3 and 85% of cells injected with phMSH6 survived. No apoptotic cells were observed in cells injected with phMSH3, and only 1% of GFP-expressing 293 cells injected with phMSH6 were apoptotic.

To determine whether mutations of the mismatch repair genes eliminated the effect on cell death, expression constructs of hMSH2 containing nonsense mutations were injected into SK-UT-1 cells together with pGFP (Table 1). One of the mutant constructs containing a nonsense mutation (E114*) at amino acid 114 of hMSH2 substantially reduced the level of cell death and apoptosis (Table 1). In contrast, a second nonsense mutant allele (Q601*) at amino acid 601 induced cell death in about the same fraction of injected cells as the wild-type allele. The difference in response was not due to differences in the level of the two mutant proteins because both mutant proteins could be detected at approximately equal levels by indirect immunofluorescence using antibodies directed against hexahistidine tags present in these constructs. These results imply that there is a region of the hMSH2 protein responsible for inducing the apoptotic response that appears to be localized between amino acids 114 and 601.

To be certain that cells expressing GFP also expressed the appropriate mismatch repair protein, we fixed the mismatch repair-deficient cell lines HCT116, HEC-1-A, and DLD-1, injected with phMLH1, phPMS2, or phMSH6, respectively, and analyzed them for expression of the mismatch repair protein by indirect immunofluorescence. This revealed that ∼50–60% of cells expressing GFP also expressed the mismatch repair protein (Fig. 3). This was a particularly important observation for cells injected with phPMS2 or phMSH6 because the reduced fraction of dead cells could simply have been due to the absence of expression. Because 50–70% of cells injected with phMSH2 and phMLH1 died in previous experiments (Table 1, Fig. 2), this observation suggested that the surviving GFP-expressing cells may have lost expression of the respective mismatch repair protein. To test this possibility, the GFP-expressing cells remaining in microinjected SK-UT-1 or HCT116 cells after 96 h were analyzed for the presence of the appropriate mismatch repair protein by indirect immunofluorescence. We found that hMSH2 and hMLH1 were no longer detectable in the surviving GFP-expressing cells (data not shown).

The altered nuclear morphology seen in cells injected with phMSH2 or phMLH1 appeared to be the result of apoptosis. To obtain further support for this conclusion, 293 cells coinjected with phMSH2, phMSH3, phMSH6, phMLH1, or phPMS2 and with pGFP were fixed, and free 3′ DNA ends that accumulate during apoptosis were labeled using terminal deoxynucleotidyl transferase (TUNEL assay). This revealed that a high proportion of the cells injected with phMSH2 and phMLH1 expression constructs stained strongly as a result of the accumulation of these ends, whereas those injected with phMSH3, phMSH6, or phPMS2 did not (Fig. 4).

The observation that overexpression of specific mismatch repair proteins induced apoptosis raised the possibility that these repair proteins participate in a pathway leading to apoptosis. A prediction of this hypothesis is that mutations of these mismatch repair genes should directly affect the ability of cells to induce death in response to some types of DNA damage, particularly damage recognized by the repair protein. To test this hypothesis, we measured the frequency of apoptotic cells in repair-proficient and -deficient primary mouse embryo fibroblasts treated with the DNA-alkylating agent MNNG or ionizing radiation. Primary mouse embryo fibroblasts were used in these studies because they would have accumulated minimal changes that may influence the apoptotic response. Mismatch repair-deficient tumor cell lines (19, 33, 34, 35) and mouse embryo fibroblast lines developed from mice containing knockouts of the mismatch repair genes (36) are resistant to the cytotoxic effects of MNNG. To determine whether an Msh2 deficiency reduced the frequency of apoptotic cells, primary fibroblast cultures derived from Msh2 +/+ and −/− mice were exposed to increasing levels of MNNG, and the fractions of cells labeled in the TUNEL assay were measured 24 h posttreatment (Fig. 5). All of the concentrations of MNNG used were highly toxic to the Msh2 +/+ cells, whereas only the highest concentration reduced the survival of the Msh2 −/− cells (∼20-fold). The level of apoptotic cells was substantially lower in the cell line derived from Msh2 −/− mice compared to cell lines derived from the Msh2 +/+ mice (Fig. 5 A). Furthermore, the frequency of apoptotic cells in Msh2 −/− cultures did not increase after treatment with higher levels of MNNG. In contrast, the frequency in Msh2 +/+ cells undergoing apoptosis showed a clear dose response. Importantly, the maximum frequency of apoptotic cells in the MSH2 +/+ cultures (7.5%) was similar to that measured in etoposide treated rat primary fibroblast cells (37). It is also interesting to note that the level of apoptosis is higher in untreated Msh2 +/+ cells than in the Msh2 −/− cells. These results support the idea that Msh2 plays a role in the induction of apoptotic cell death in response to endogenous DNA damage.

The effect of ionizing radiation on the fraction of Msh2 +/+ and −/− primary mouse embryo fibroblasts undergoing apoptosis was also determined. Similar to MNNG-treated cells, the Msh2 +/+ fibroblasts displayed a higher frequency of apoptotic cells than did Msh2 −/− cells (Fig. 5 B). However, the frequency of apoptotic cells in the Msh2 −/− culture was found to increase, although the response was significantly lower than that observed in the Msh2 +/+ cells. Similar responses in the level of apoptosis in Msh2 −/− and +/+ mouse embryonic stem cells has recently been reported by DeWeese et al. (38).

Although mismatch repair protein levels similar to those obtained after microinjection are unlikely to occur in vivo, we propose that this overexpression unmasks an interaction of hMSH2 and hMLH1 with components of a pathway that leads to or influences the induction of apoptosis that also occurs under physiological conditions. Furthermore, this interaction may be important for the induction of apoptosis in response to certain forms of DNA damage that may be recognized by the repair proteins (e.g., damage induced by DNA alkylating agents; Refs. 19 and 33, 34, 35). The loss of the induction of apoptosis in MNNG treated Msh2-deficient mouse embryo cells argues that such an interaction is not an artifact of overexpression and provides evidence that Msh2 is involved in the apoptotic response in vivo. The observations, reported here and by DeWeese et al. (38), that Msh2−/− cells exposed to ionizing radiation have a reduced frequency of cells undergoing apoptosis suggests that this effect may not be limited to DNA-alkylating agents and may extend to a wide range of DNA-damaging agents. Thus, cells deficient in hMSH2 or hMLH1 (responsible for the great majority of HNPCC patients) may become defective in the ability to induce apoptosis in response to many types of DNA base damage. This would have direct consequences for cellular turnover and expansion in environments in which damage accumulates.

One possible explanation for the induction of apoptosis in cells overexpressing hMSH2 or hMLH1 is that the increased levels of repair proteins may sequester or alter the interaction of a protein or proteins that are essential for cell cycle progression or the induction of apoptosis. A potential candidate for such an effect is PCNA because some mismatch repair proteins have been shown to interact with this protein in yeast two-hybrid screens (39). Overexpression of these mismatch repair proteins in injected cells might sequester PCNA from its role in DNA synthesis. As a result, DNA synthesis may be arrested, and apoptosis may be induced. Interestingly, induction of p21 also inhibits S-phase progression through sequestration of PCNA after DNA damage (40). Yet there is little indication that overexpression of p21 leads to apoptosis (41, 42, 43). On the basis of this model, it would appear unlikely that sequestration of PCNA by the overexpressed mismatch repair proteins would account for the rapid induction of apoptosis observed here. Moreover, the differential response following microinjection of constructs containing nonsense alleles of hMSH2 has enabled us to identify a candidate region responsible for the induction of cell death (amino acids 116–601). Fine mapping of the domain responsible for the induction of cell death should greatly facilitate our ability to identify protein partners participating in this response. Moreover, we have found that a missense mutation of hMSH2 (K675A) that is deficient for mismatch repair still induces apoptosis when overexpressed.⁵ This observation combined with the data of the nonsense mutant E601*, which also induces apoptosis but is no longer competent for mismatch repair, suggests that the domains responsible for these two functions are distinct. Nevertheless, because the vast majority of mutations in HNPCC are nonsense mutations (∼90%; Ref. 3) that encode truncated peptides that appear rapidly turned over, the apoptotic response as well as repair proficiency is likely to be lost in most patients.

It is important to note that overexpression of other types of repair proteins has not been reported to induce cell death. Although the nucleotide excision repair proteins XPB and XPD are involved in the p53-mediated apoptotic pathway as well as being essential components of the transcription initiation machinery, microinjection of expression constructs of these proteins does not induce death (44). Furthermore, cell lines stably expressing these repair proteins have been obtained (45).

The differential effects of mismatch repair protein overexpression are particularly striking, given that mutations of hMSH2 and hMLH1 are most common among HNPCC families. It has been proposed that the loss of repair proteins such as hMSH2 and hMLH1 may have a more indirect effect on tumor development than mutations in tumor suppressor genes such as APC or p53. However, the potential role of these two repair proteins in inducing cell death in response to DNA damage (introduced by exogenous agents or even by endogenous metabolites) may mark a distinction between these two repair genes and others genes within and without the mismatch repair family. Thus, the roles of these proteins in the control of spontaneous mutation and the induction of apoptosis in response to environmental agents may be crucial to the control of colon cancer. In this respect, the induction of apoptosis following overexpression hMSH2 and hMLH1 appear to emphasize their similarity to tumor suppressor genes, such as APC (46) or p53 (Refs. 44, 47, and 48) rather than other DNA repair genes.

The mechanism by which hMSH2 and hMLH1 may control apoptosis is unknown. An intriguing possibility is that hMSH2 and hMLH1 may accomplish this through their participation in a signal transduction pathway that uses the hMSH2-dependent ADP→ATP switch function (49, 50). We consider the possibility that hMSH2-dependent lesion recognition, in concert with hMLH1 (with the likely participation of their respective heterodimeric partners), functions as the cellular signal for damage assessment and the subsequent decision to induce apoptosis.

We thank Eric Phillips for help with the apoptosis assay and Anil Ganesh for help with irradiating mouse embryo fibroblasts, S. Guerrette and S. Acharya for helpful discussions, and H.Alder and the Kimmel Nucleic Acids Facility for sequence analysis.