Sporadic microsatellite mutations are frequently observed in lung, bladder, and head and neck tumors with intact DNA mismatch repair. AAAG tetranucleotide repeats appear to be especially prone to the accumulation of these mutations. We hypothesized that occurrences of microsatellite mutations in these cancers may be linked to DNA damage caused by exposure to carcinogens in tobacco smoke. To test this hypothesis, we developed a model system based on reactivation of green fluorescent protein (GFP) in which a plasmid vector carries a microsatellite repeat that places the GFP sequence out of frame for protein translation. In this reporter system, DNA slippage mutations can restore the GFP reading frame and become detectable by flow cytometry as GFP-positive cells. Pools of stably transfected RKO cells were treated at four dose levels each of γ-irradiation, benzo(a)pyrene diol epoxide, N-methyl-N-nitro-N-nitrosoguanidine (MNNG), t-butyl hydrogen peroxide, and UV irradiation and assayed for GFP-positive cells 48 h later. We studied the microsatellite repeats AAAG, ATAG, CAGT, and CA, as well as a control sequence lacking any repetitive elements. A log-linear regression approach was used to discriminate between the effects of repeat unit and dose for each agent. A statistically significant increase in GFP-positive cells was found with increasing dose with all agents, although repeat unit-specific response patterns were only observed with MNNG, t-butyl hydrogen peroxide, and UV irradiation. With MNNG, significant differences in response were observed between dinucleotide and tetranucleotide repeat units. The effects of UV irradiation were consistent with the predicted number of pyrimidine dimers/repeat unit, with higher GFP activation in repeats that had large numbers of adjacent pyrimidines. We found no evidence to indicate that the AAAG repeat responded to any of the DNA-damaging agents with higher levels of GFP activation than other repeat units. These results provide evidence that DNA damage can induce slippage mutations and increase mutation rates in repeated sequences and that there are sequence-specific responses to different types of DNA damage. Our results are compatible with the hypothesis that sporadic microsatellite mutations in human cancer may reflect DNA damage caused by carcinogen exposure.
Microsatellites are short sequences of tandem repeats dispersed throughout the mammalian genome. Repeat units range from 1 to 6 bp in length, and the entire sequence of a typical repeat tract is less than 100-bp long. Microsatellite instability is characterized by the insertion or deletion of one or more repeat units. Widespread microsatellite instability has been reported in colorectal tumors with defective DNA mismatch repair (1) . Somatic microsatellite mutations have also been described in a variety of tumor types that have intact DNA mismatch repair (2) . The pattern of microsatellite mutations in tumors that are not defective in DNA mismatch repair is quite different from that observed in tumors with defective DNA mismatch repair (3) . In tumors with intact DNA mismatch repair, these mutations are usually found as a single aberrant sequence without the typical “ladder” effect seen in tumors with defective DNA mismatch repair. Presumably, these mutations arose at some point in the life of the original cell(s) that eventually expanded to become the tumor. Evidence for such a mechanism comes from the observation that mutations in the primary tumor are retained in identical form in distant metastases (4) . Thus, the molecular profile of microsatellite mutations in a DNA mismatch repair-proficient tumor may reflect the genetic insults that became fixed in the cell(s) of origin for a particular tumor.
Microsatellite Mutations in Lung, Bladder, and Head and Neck Cancer.
Sporadic microsatellite mutations have been observed in tumors with no known mutations in mismatch repair genes, such as those of the lung, bladder, and head and neck (5) . This observation suggests that mechanisms other than mismatch repair gene deficiency may contribute to microsatellite mutations in these tumors, especially in selected microsatellites with the AAAG or ATAG repeat units (5, 6, 7) . In human lung tumors, sporadic microsatellite mutations might reflect the high mutational load experienced by the epithelial lining of the airways in smokers. It was reported previously that sporadic mutations in lung cancer are more frequently found in tetranucleotide repeats such as (AAAG)n than in dinucleotide repeats such as (CA)n (5 , 7) . In one study, the occurrence of (AAAG)n mutations was associated with mutations in the p53 tumor suppressor gene (6) . In addition, several AAAG tetranucleotide repeats show hypermutability in germ line and/or somatic cells, whereas similar findings were reported for the repeat sequence ATAG (8, 9, 10, 11) . These observations suggest that certain repeat units are prone to accumulate slippage events and that this may be related to the sequence of the repeated DNA.
AAAG and Non-B DNA Secondary Structures.
Some tetranucleotide repeats can form non-B DNA secondary structures under conditions of negative supercoiling of the DNA. For example, sequences such as (AAAG)n, but not (CA)n, are mirror repeats that can form triplex DNA in vitro (12) . In triplex DNA, triple-stranded DNA is formed by folding back of a single strand into the major groove of repeat DNA. This third strand forms Hoogstein base pairing with the double-stranded part of the repeat, leaving the remaining fourth, single strand exposed outside of the triplex DNA. One can speculate that these rare secondary DNA structures might change the sensitivity of the repeated DNA to carcinogens. In the case of triplex DNA, the single exposed DNA strand may be more accessible to mutagenic action (13) . Alternatively, non-B DNA structures may be less efficient substrates for DNA repair enzymes.
Mutations in Microsatellites and DNA Mismatch Repair.
Spontaneous mutation rates vary widely between types of microsatellites and between individual loci with identical repeat units (for review, see Ref. 14 ). Such mutations have been reported not only in the germ line of various species but also as somatic mutations in cells with intact mismatch repair systems. The average germ-line mutation rate for human dinucleotide repeats was estimated at 6 × 10−4/germ-line transmission, whereas estimates for tetranucleotide repeats were at least 2-fold higher (1–3 × 10−3; for review, see Ref. 14 ). At least one AAAG repeat (D21S1245) had a very high mutation rate of up to 2 × 10−2/germ-line transmission (8) . This instability was also found in somatic cells, both in lymphocytes directly after immortalization and during subsequent culturing of immortalized lymphocytes. In contrast, the stability of a (CA)17 repeat was about 10-fold less than that of a (AAAG)17 repeat in both DNA mismatch repair-proficient and -deficient eukaryotic cells (15) . The same (CA)17 repeat in normal human fibroblasts had a mutation rate of 1 × 10−5/cell generation (16) . These results clearly indicate that multiple factors play a role in observed microsatellite mutation rates. These factors include the length and specific sequence of the repeat, its chromosomal location and associated flanking sequences, and the genetic background in which the mutations are measured.
Microsatellite Model Systems.
Microsatellite slippage mutations can be the result of mistakes in DNA repair processes that follow either spontaneous or induced DNA damage. Oxidative DNA damage can lead to microsatellite slippage mutations in several model systems (17 , 18) . In human cell lines, a (CA)13 repeat showed an increase in microsatellite mutations of up to 27-fold after oxidative DNA damage (18) . Because cigarette smoke can cause many types of DNA damage, the observed tetranucleotide microsatellite instability in lung cancer may be linked to this exposure. Other evidence for a link between microsatellite mutations and DNA damage comes from analyses of second primary lung and breast tumors in non-Hodgkin’s lymphoma patients treated with DNA-damaging chemotherapeutic agents (19) .
GFP 2 -based Mutation Reporter System.
We developed a GFP reporter assay to assess the detection of microsatellite slippage mutations induced by several DNA-damaging agents and applied it using a panel of different microsatellite repeats. Plasmid constructs were made to represent the tetranucleotide repeat units AAAG, ATAG, and CAGT and the dinucleotide repeat unit CA. The repeat units AAAG and ATAG were included because of their reported instability and potential to form non-B DNA, whereas the repeat unit CAGT was included because it lacks this capacity and because it has been extensively studied in yeast (12 , 20) . The CA repeat unit has been commonly used in studies of microsatellite instability and was included in our study for comparison (2 , 16) .
Different agents were chosen in this study to represent the major classes of exposures known to damage DNA. Several reflect carcinogenic insults related to tobacco smoke exposure. One important class of carcinogens from tobacco smoke are the polycyclic aromatic hydrocarbons, of which benzo(a)pyrene is one of the best-documented representatives (21) . This compound requires metabolic activation to its 7,8-diol-9,10-epoxide (BPDE), which we used in this study. BPDE binds predominantly to guanine bp in the DNA, causing a bulky adduct that is repaired by nucleotide excision repair. Another class of tobacco-related carcinogens are N-nitroso compounds. We chose to study a prototype of this group, MNNG, a monofunctional alkylating agent, which transfers a methyl group to DNA (22) . MNNG-induced DNA alterations are mostly repaired through base excision repair. Cigarette smoke contains several components that cause oxidative damage in exposed tissues (21) . Hydroxyl radicals and other reactive oxygen species cause base damage or loss and DNA single-strand breaks (23) . In this study, t-BHP was chosen to test the effects of oxidative damage. Other exposures were γ-irradiation and UV irradiation. The effects of γ-irradiation are either directly on DNA or indirectly via radicals formed through ionization, to cause single- and double-strand breaks or base and nucleotide damage (23) . UV irradiation is one of the most extensively studied agents of DNA damage. It causes pyrimidine dimers by covalent linkage of adjacent pyrimidine bp (23) . The main DNA repair mechanism to correct these lesions is nucleotide excision repair.
The main goals of this study were to assess the effects of DNA damage on mutation frequency in microsatellite sequences and to determine any sequence-specific responses to DNA damage that may explain sporadic microsatellite mutations observed in carcinogen-related human cancer. Using the GFP reporter assay, we demonstrate that DNA damage can differentially increase the number of these mutations, depending on the agent and on the microsatellite repeat unit.
MATERIALS AND METHODS
GFP Reporter Plasmids.
We developed a microsatellite mutation reporter system based on the detection of slippage events that restores the reading frame of the GFP coding sequence. The main features of the reporter system are outlined in Fig. 1 ⇓ . The plasmid encodes a fusion protein with the first 80 amino acids of hygromycin fused to EGFP with the normal ATG start codon for protein translation removed. For each of the microsatellite repeat units, 16 repetitions were inserted between the hygromycin and EGFP sequences. The inserted microsatellites shifted the protein translation-reading frame of the downstream EGFP so that the plasmids do not encode an active EGFP. Slippage events within the microsatellite repeats can cause restoration of the reading frame, leading to expression of EGFP. Because there are three possible reading frames, GFP activation can result from one of two possible frameshifts in the original sequence. For the tetranucleotide repeat units, deletion of one repeat unit or insertion of two repeat units will lead to restoration of the EGFP reading frame (see Table 1 ⇓ ) and be detected by flow cytometric measurements of green fluorescing cells. Thus, a deletion slippage event resulting in the deletion of 4 bp (1 repeat unit) or an insertion slippage event resulting in the addition of 8 bp (2 repeat units) would lead to expression of the EGFP protein. Larger deletion events (4 repeat units) or insertion events (5 repeat units) may theoretically also lead to restoration of EGFP expression. The CA repeat was constructed so that, similarly to the tetranucleotides, deletion of 4 bp (2 repeat units) or insertion of 2 bp (1 repeat unit) would lead to EGFP expression. In addition, small insertions or deletions in nonrepeating DNA sequences could also potentially lead to GPF activation. These events are expected to occur equally in the repeat plasmids and in the no-repeat control plasmid.
The plasmids used in this study are based on the backbone of pDsRed-N1 with coding sequences from pHygEGFP replacing the DsRed sequence (both plasmids were from Clontech, Palo Alto, CA). To achieve the desired configuration, the KpnI/NotI fragment from pHygEGFP was cloned into the pDsRed1-N1 plasmid digested with KpnI and NotI, replacing the DsRed sequence with HygEGFP. The microsatellite repeat units with flanking sequences were inserted between the EcoRI and BseRI recognition sites in the hygromycin/EGFP fusion sequence. Specifics of these repeat elements are shown in Table 1 ⇓ . The resulting plasmids contain the ATG start codon from HygEGFP and the first 242 nucleotides of hygromycin up to the EcoRI recognition site. This sequence is directly followed by the repeat units and the first 9 codons of EGFP up to the BseRI sequence (Fig. 1) ⇓ . A positive control plasmid was generated with the hygromycin ATG in-frame with the EGFP sequence, and this plasmid was shown to encode a functional EGFP and conferred resistance to G418 as a result of the neomycin gene present in the pDsRed1 backbone plasmid (data not shown). Similarly, a no-repeat control plasmid was generated that did not contain any repeat sequences but had EGFP out of frame with the hygromycin start codon in the identical sequence background. Similarly to the plasmids harboring repeat units, this plasmid contained the inserted oligonucleotide listed in Table 1 ⇓ plus 243 nonrepetitive nucleotides upstream of the BseRI site in EGFP corresponding to the 5′-end of hygromycin coding sequence.
Stable Transfection and Generation of Cell Line Pools.
The human colorectal carcinoma cancer cell line RKO was obtained from American Type Culture Collection (Manassas, VA) and grown in DMEM/Ham’s F-12 supplemented with 10% fetal bovine serum and penicillin/streptomycin. Cells were seeded in 6-well plates 1–2 days before each experiment at 10–20% density to ensure exponential growth at the time of treatment. RKO was chosen because it is defective in DNA mismatch repair through inactivation of MLH1. Thus, any spontaneous or induced microsatellite mutations are expected to become fixed in the genome and lead to GFP expression in the reporter system. RKO is not mutated in p53, has normal p53 expression levels, and has normal p53 tumor suppressor function (24) .
Transfections were performed with FuGENE reagent (Roche Biochemicals, Indianapolis, IN), using conditions recommended by the manufacturer. RKO cells were transfected by one of the following plasmids linearized by BsaI digestion: pMGFP-AAAG, pMGFP-CA, pMGFP-CAGT, pMGFP-GATA, pMGFP-P, and pMGFP-N (Table 1) ⇓ . One day after transfection, selection was started at the required dose for RKO of 2 mg/ml neomycin analogue G418 (Geneticin; Invitrogen Life Technologies, Inc., Carlsbad, CA). After 1–2 weeks of selection, pools of at least 100 individual transfectant clones were obtained from each plasmid construct to minimize the effects of clonal variation. These pools typically harbored 1–2% GFP-positive cells, presumably because of transfection-associated slippage events in the repeat sequences. These GFP-fluorescing cells were removed from the pools of transfectants using a FACS. This selection reduced the background levels of GFP-positive cells to below 5 × 10−5. This background level varied between the different cell pools, but it was stable for several months of continuous cell culture during which experimentation was performed. Because of the absence of DNA slippage events, background levels of GFP-positive cells were below 5 × 10−5 in the no-repeat control, and these cells were not subjected to the FACS enrichment before experimentation.
We studied five different DNA-damaging agents. Cell pools were treated at four different dose-levels: (a) zero dose (untreated); (b) low dose at a sublethal level; (c) high dose at the highest level that allowed reliable measurement of GFP at 48 h; and (d) an intermediate dose arbitrarily chosen between low and high. For γ-irradiation, cells were irradiated in culture medium using a cesium source at a dose rate of 5 Gy/min for 1 min or at a dose rate of 10 Gy/min for 4 and 12 min, creating non-zero doses of 5, 40, and 120 Gy, respectively. For BPDE treatment, growth medium was removed, and the cells were rinsed with PBS, followed by a 30-min incubation with BPDE (Midwest Research Institute, Kansas City, MO) using tetrahydrofurane as a solvent. The volume of tetrahydrofurane, an anhydrous solvent with no known genotoxicity (25) , was limited to 10 μl in a 2-ml volume. After incubation, the BPDE-containing PBS was removed, and growth medium was added back. Dose levels were 0.5, 1, and 2 μm BPDE. For MNNG treatment, MNNG (Sigma-Aldrich, St. Louis, MO) was added to the growth medium for a 1-h incubation. After incubation, medium was removed, cells were rinsed twice with PBS, and fresh growth medium was added. Cells were incubated with 5, 25, and 50 μm MNNG using DMSO as a solvent. For t-BHP exposure, t-BHP was added to the growth medium covering the cells, followed by incubation for 1 h. After this incubation, the medium was removed, and fresh growth medium was added. Dose levels were at final concentrations of 1, 2, and 3 mm t-BHP (Sigma-Aldrich). t-BHP is a relatively stable chemical oxidant that permeates cell membranes readily. For UV treatment, cells were irradiated in culture medium at room temperature using a Stratalinker (Stratagene, La Jolla, CA). Dose levels of UV were 15, 60, and 100 J/m2.
Cell Analysis by Flow Cytometry.
The experimental conditions of GFP detection by flow cytometry have been described in detail previously (26) . After any treatment, cells were allowed to recover for 48 h, after which a sample of 500,000 cells was analyzed using a LSR flow cytometer (Becton Dickinson, San Jose, CA). Green and orange fluorescence were simultaneously measured by exciting the cells using a 488 nm argon laser. Green emission was detected at 530 nm, whereas orange fluorescence was measured at 575 nm as an internal standard. For each sample of cells, green fluorescence signals were plotted against the orange fluorescence signals. In such plots, GFP-negative cells fall along the orange/green diagonal, whereas GFP-positive cells appear shifted from this diagonal (26 , 27) . The number of GFP-activating microsatellite mutations per 106 cells was calculated by dividing the number of green-fluorescing cells observed by the total number of cells counted and multiplying the result by 106. Because a typical analysis of green fluorescence was performed on 500,000 cells, the lowest level at which microsatellite slippage events could be detected was thus 2 in 106 events (a minimum of 1 event/analysis).
In statistical terms, our experiment had a factorial design involving three factors: (a) DNA-damaging agent (five levels); (b) dose (four levels); and (c) repeat unit (five levels). Each of the 100 treatment combinations of a particular agent, dose, and repeat unit was replicated three times in independent experiments. The response analyzed was the calculated number of GFP-positive cells per million cells. We analyzed these data using generalized linear model techniques (28) , employing a Poisson error distribution and the corresponding logarithmic link function. A Poisson error distribution is justified because GFP-positive cells are rare among all cells counted. We fit the resulting log-linear models using the GENMOD procedure of SAS/STAT statistical software (SAS Institute, Cary, NC). These models estimate the logarithm of the expected GFP-positive cell count in each treatment combination so that the measures of effect are differences in logarithms of expected counts. Consequently, the graphical presentation uses a logarithmic transformation of the numbers of GFP-positive cells per 106 cells.
Because dose levels were not quantitatively comparable among DNA-damaging agents, we analyzed data from each agent separately. However, preliminary analysis revealed extra-Poisson variation that was relatively constant across agents. We accommodated this variation by estimating a scale parameter based on all 100 treatment combinations using a previously described method (29) , and we used that same scale parameter in the analysis of each agent. We approached the analysis for each agent in the same way, guided by common ideas for studying main effects and interactions in two-factor experiments (30) . Our primary interest focused on nonparallel dose-response patterns across repeat units. This pattern is described in statistical terms by “interaction” effects between repeat units and dose levels. The finding of statistical interaction between dose and repeat unit suggests that patterns of change across dose levels in the frequency of GFP positivity differ among repeat units. Interaction effects are more important than repeat unit main effects in this regard because each cell line pool, after transfection and cell sorting, has its own background level of GFP-positive cells that is unique to each cell pool. Repeat unit main effects are based on averages across all four dose levels for each repeat unit and include nonslippage contributions to GFP-positive cell counts. Interaction effects or dose main effects, on the other hand, are based on comparing changes between dose levels and, consequently, are free of any repeat unit-specific background levels. Whenever relevant, estimates of interaction, dose, or repeat unit effects were corrected for the observed response in the appropriate no-repeat control. Because we chose dose levels with the expectation that they would cause DNA damage and thereby increase the frequency of mutation, detailed study of dose main effects held less interest than study of interaction effects.
With four dose levels and five repeat units for each agent, the overall test for interaction has (4 − 1) × (5 − 1) = 12 degrees of freedom. Whenever this test indicated the presence of interaction (P < 0.10), we examined the source of that interaction more closely by partitioning this overall test into 12 single degree of freedom tests based on contrasts among the treatment combinations. These contrasts describe certain expected relationships among the mean response levels at different treatment combinations. We constructed the single degree of freedom contrasts by specifying comparisons among dose levels and among repeat units that we found meaningful and building them into interaction contrasts. Three contrasts were made to compare the four dose levels. To examine whether the dose of an agent changed response, we compared the response at zero dose with the average response of three non-zero dose levels (“zero dose versus non-zero dose”). To examine whether an increase in dose affected the response, we compared the response at the high dose with that at the low dose (“low versus high dose”). A third contrast, constructed to be orthogonal to the previous two, compared the response at the intermediate dose with that average response at the low and high doses. This contrast examines whether response at the intermediate dose is midway between those at the low and high doses. If these three doses were equally spaced, this contrast would formally test for a nonlinear dose-response relationship.
We created four contrasts to compare the five distinct repeat units. First, we compared the response in the no-repeat cells with the average response in cells from the other four repeat units (“repeat units versus no repeat”). Second, we compared the response of cells with the dinucleotide repeat to the average response of cells from the three tetranucleotide repeats (“dinucleotide versus tetranucleotide”). Third, we compared the response of the AAAG repeat unit with that of the CAGT repeat unit in an effort to examine a possible effect of the capacity to form triplex DNA on the response (“AAAG versus CAGT”). The fourth repeat unit contrast, constructed to be orthogonal to the previous three, compared the response of the ATAG repeat unit with that average response of the AAAG and CAGT repeat units (“AAAG + CAGT versus ATAG”). The latter contrast tests whether mutation frequency for ATAG is midway between mutation frequencies for AAAG and CAGT, an ordering that might reflect increasing potential to form triplex DNA. The 12 single degree of freedom interaction contrasts that we considered arise from combining a dose contrast with a repeat unit contrast. For example, one interaction contrast asks whether the change in response from zero dose to the average of the non-zero doses (D1) differs in magnitude between the dinucleotide repeat and the average of the tetranucleotide repeats (R2). Another interaction contrast asks whether the change in response between low and high dose (D2) differs between AAAG repeats and CAGT repeats (R3).
With 100 treatment combinations, the problem posed by multiple comparisons can become severe. One advantage of the approach using contrasts that we have described is that it restricts an unmanageable number of potentially highly correlated comparisons to a circumscribed set of nearly independent comparisons that have been specified a priori. In addition, the strategy of requiring statistical significance from a multiple degree of freedom test before examining individual contrasts further reduces the opportunity for false positives. Consequently, we report Ps that were not further adjusted for multiple comparisons.
Flow cytometry results obtained with the DNA-damaging agents γ-irradiation, BPDE, MNNG, t-BHP, and UV irradiation showed that there was broad variation in GFP activation between the repeat units AAAG, ATAG, CAGT, and CA and the no-repeat control (Fig. 2) ⇓ . Except for γ-irradiation, exposure to each agent led to larger relative increases in the number of GFP-positive cells in cells harboring repeat units than in the no-repeat control cells. The observed number of GFP-positive cells averaged over three separate experiments was between 3.3 and 740 per million cells analyzed (Fig. 2) ⇓ . The largest increase over zero-dose cells was for UV irradiation in ATAG cells, in which we measured 735 GFP-positive cells per million at the highest UV dose. Under the assumptions that all GFP-positive cells had frameshift mutations that activated GFP and that approximately half of all slippage events led to GFP positivity, we calculated a mutation frequency of 1.5 × 10−3 for ATAG under UV exposure. For MNNG, the highest observed frequency of GFP-positive cells was 5.4 × 10−4, which would translate into a mutation frequency of 1.1 × 10−3 under the assumption that insertion and deletion mutations occur with equal probability. Similarly, the highest estimated mutation frequencies for γ-irradiation, BPDE, and t-BHP were 2.1 × 10−4, 4.6 × 10−4, and 3.5 × 10−4, respectively.
Detailed statistical analysis of the flow cytometry results used log-linear models to further understand how the effects of dose, repeat unit, and their interactions contributed to the observed variation in GFP-positive cells (Table 2) ⇓ . We investigated whether the number of GFP-positive cells, averaged for dose levels, was different among the five separate cell line pools. We found highly significant differences between the cell line pools with all five DNA-damaging agents (Table 2 ⇓ , main effect of repeat unit). However, this result can be largely attributed to the variation introduced through cell sorting that was necessary to remove cells with transfection-associated GFP-activating events. This sorting resulted in different baseline levels of GFP-positive cells in the different cell line pools. Thus, the levels of GFP-positive cells reflect the combined effects of transformation-associated activation events, the efficiency of removing these GFP-positive cells by FACS, and the changes induced by the exposures. For this reason, any main effect result for repeat unit will incorporate differences between cell pools that are not related to DNA slippage mutations in response to DNA damage. Consequently, we will not further consider these differences between repeat units.
For each DNA-damaging agent, we examined dose-related changes in the number of GFP cells averaged across all cell lines. We observed highly significant differences in the average response with increasing dose for each agent (Table 2 ⇓ , main effect of treatment dose). For every agent, the number of GFP-positive cells was significantly higher after treatment compared with the zero-dose level cells (Table 2 ⇓ , D1). In addition, statistically significant (P < 0.05) increases from low to high dose were observed for every agent, except t-BHP (Table 2 ⇓ , D2, P = 0.99). Thus, increasing dose from 1 to 3 mm t-BHP did not lead to higher numbers of GFP-positive cells on average across repeat units. The response at the intermediate dose was about midway between the responses at the low and high doses for every agent, except UV irradiation (Table 2 ⇓ , D3). For UV irradiation, the increase in response from low to intermediate dose was significantly larger than the increase from intermediate to high dose (P < 0.0001). In these comparisons, the baseline differences observed between cell lines with different repeat units do not affect the validity of the findings because they are subtracted away by examining dose-to-dose changes. Overall, the results obtained by averaging across repeat units indicate that, at the dose levels used, the DNA-damaging agents were effective in causing DNA frameshifts leading to increased numbers of GFP-positive cells.
One of the goals of this study was to investigate whether dose-response profiles for any DNA-damaging agent differed among the repeat units and when compared with the no-repeat control. These questions translate into statistical questions about the repeat unit by dose interaction effects for each agent. The overall interaction test for differences among repeat units in the shape of dose-response curves was significant at the 0.10 level for MNNG, t-BHP, and UV irradiation, but not for BPDE and γ-irradiation (Table 2 ⇓ , repeat by dose interaction effects). For BPDE and γ-irradiation, the shape of the dose-response curve appeared similar for all repeat units, including the no-repeat control, within the level of resolution provided by the tests for interaction with our data. We will present detailed dose-response patterns for MNNG, t-BHP, and UV-irradiation in separate subsections.
Effects of MNNG Exposure.
The overall test for differential effects of MNNG treatment with the different repeat units was highly significant (Table 2 ⇓ , repeat by dose interaction effects, P < 0.0001). Two comparisons contributed to this observation. First, the increase in response from low to high dose of MNNG was larger on average for the cell lines with repeat units than for the no-repeat control cells (Table 2 ⇓ , R1 × D2 interaction, P = 0.008). This effect is illustrated by the lack of a dose response to MNNG treatment for the no-repeat control cells compared with the strongly increasing response in most cell lines harboring repeat units (Fig. 2C) ⇓ . This result indicates that activation of GFP by MNNG depended on repeated DNA sequences upstream of the coding sequence of GFP and suggests that such activation was caused by DNA slippage events in this repetitive DNA. Second, the increase in GFP-positive cells from the low to high dose was significantly larger in the cell lines harboring tetranucleotide repeat units than in the cell line with the CA dinucleotide repeat unit (Table 2 ⇓ , R2 × D2 interaction, P < 0.0001). This effect can be seen by comparing the small increase in GFP-positive cells between low- and high-dose MNNG in the CA repeat cells with the much larger increases in each of the cell lines with tetranucleotide repeat units (Fig. 2C) ⇓ . No differences in the pattern of response to increasing MNNG dose were found among the tetranucleotide repeat cell lines (see Fig. 2C ⇓ and Table 2 ⇓ ). In summary, the results obtained with MNNG indicate that mutation frequencies were increased more in response to increasing doses of MNNG in cells with repeats than in no-repeat control cells and that the increase in response was greater in cells with tetranucleotide repeats than in cells with CA repeats.
Effects of t-BHP Exposure.
The overall test of interaction was significant for t-BHP treatment (Table 2 ⇓ , repeat by dose interaction effects, P = 0.052), but less so than the corresponding tests for MNNG or UV irradiation. Only the AAAG repeat unit cells demonstrated a stepwise increase with dose in the number of GFP-positive cells, whereas the other repeat cells had only a single increase from zero to low dose, with no further increase in GFP-positive cells with higher dose levels (Fig. 2D) ⇓ . The no-repeat control cells had essentially the same response at all dose levels. Two comparisons contributed to the repeat by dose interaction for t-BHP. First, the low- to high-dose increase in GFP-positive cells was larger in cell lines harboring the AAAG repeat unit than in those harboring the CAGT repeat unit (Table 2 ⇓ , R3 × D2 interaction, P = 0.0008). Second, the low- to high-dose increase in GFP-positive cells averaged across the AAAG and CAGT cell lines was larger than that in the ATAG cell line (Table 2 ⇓ , R4 × D2 interaction, P = 0.02). Both these findings can be traced to the lack of change in response from low- to high-dose t-BHP observed with the ATAG and CAGT cell lines compared with the large corresponding increase observed with the AAAG cell line. This result suggests that the effect of t-BHP on the AAAG repeat unit continued to increase up to the 3 mm dose, whereas the effect on the other repeat units was saturated at the 1 mm dose.
Effects of UV Irradiation.
The dose-response pattern for UV in the cell lines harboring repeat units was a generally small or absent increase in response from zero to low dose and a large increase from low to intermediate dose, with little further increase from intermediate to high dose (Fig. 2E) ⇓ . The no-repeat control cells did not show any significant increases with increasing UV dose. The overall test for differential effects of UV doses on the different cell lines was highly significant (Table 2 ⇓ , repeat by dose interaction effects, P < 0.0001). Four separate comparisons contributed to this interaction, of which three were based on observed differences in the low- to high-dose increase in GFP-positive cells. First, the average increase in response from low- to high-dose UV in cell lines harboring repeats was significantly larger than that in the no-repeat control cells (Table 2 ⇓ , R1 × D2 interaction, P = 0.003). This result indicates that the activation of the GFP reporter caused by UV irradiation was specific for the presence of a repeated element upstream of the reporter gene, which suggests that the activating DNA frameshifts were caused by DNA slippage events. Second, we observed a greater increase in GFP-positive cells from low to high dose of UV irradiation with the tetranucleotide repeat cell lines on average than with the CA repeat line (Table 2 ⇓ , R2 × D2 interaction, P = 0.009). This finding suggests that the CA repeat cell line was less responsive to UV treatment than the tetranucleotide repeat cell lines. Third, the AAAG repeat cell line had a larger increase in GFP-positive cells from low to high dose of UV than did the CAGT repeat cell line (Table 2 ⇓ , R3 × D2 interaction, P = 0.0012). The fourth significant comparison also represented a difference in dose-response pattern between the AAAG and CAGT repeat cell lines. The amount by which the response at the intermediate dose differs from the average of the low- and high-dose responses was larger for the AAAG repeat cell line than for the CAGT cell line (Table 2 ⇓ , R3 × D3 interaction, P = 0.04).
Effects of the Presence of Repeated Sequences.
One of the questions for this study was to investigate whether the presence of repeated DNA sequences was related to an altered dose-response relationship under conditions of DNA damage and repair. In our reporter system, such differences were expected to show up as increasing numbers of GFP-positive cells in response to increasing doses of DNA-damaging agents in cell lines harboring repeat units and possibly in the no-repeat control cell line. The clearest and most direct evidence for such differential effects comes from the repeat unit by dose interaction comparisons for each agent, especially those that compare repeat unit effects (R1) with any of the dose effects (D1, D2, and D3). Only for MNNG and UV irradiation did our data convincingly demonstrate such repeat unit versus no-repeat unit differences in the response to treatment (Table 2) ⇓ . In the case of t-BHP, where some interactions were statistically significant, the observed differences did not involve a comparison between repeat unit cells and the no-repeat control cells. One possibility is that these interaction effects do exist but that they are below the resolution of our study. Indirect evidence for a difference in dose-response relationship between the repeat cell lines and the no-repeat control may be obtained by examining dose effects for each cell line separately. A significant dose-response pattern in repeat cell lines together with a nonsignificant pattern in the no-repeat control cell line might be regarded as less than conclusive support that differences in dose-response patterns exist. We carried out separate 3 degree of freedom tests for a dose effect for every cell line under each exposure (25 tests). For the no-repeat control cell line, none of the five agents exhibited statistically significant dose effects (P = 0.39, 0.99, 0.78, 0.73, and 0.79 for γ-irradiation, BPDE, MNNG, t-BHP, and UV irradiation, respectively). In contrast, for each of the four repeat-harboring cell lines, every one of the 20 separate tests was statistically significant (highest P obtained was 0.008). Thus, although we have strong direct evidence for differences in dose-response patterns between repeat unit cell lines and no-repeat control cell lines only for MNNG and UV irradiation, we cannot rule out the existence of such differences for γ-irradiation, BPDE, and t-BHP.
To investigate the possible link between sporadic microsatellite mutations observed in human tumors and DNA damage caused by carcinogens, we constructed a GFP reporter assay. Our assay measures levels of microsatellite mutations using flow cytometric determination of GFP-expressing cells as the end point. The assay consists of a plasmid vector carrying different microsatellite sequences (AAAG16, ATAG16, CAGT16, or CA16) that place a GFP gene out of frame for protein translation. Microsatellite mutations can restore the reading frame and become detectable as they confer green fluorescence to cells, which can be measured by flow cytometry. This approach has several advantages over detection methods based on activation of antibiotic resistance genes (18 , 31) . The activation of GFP can be measured directly and does not depend on survival of a single cell to yield a scorable colony. This feature also reduces the time necessary to take the measurement. Because of the fast assessment of GFP activation, the spontaneous mutation rates during that timeframe are negligible. On the other hand, some cells do not tolerate high levels of GFP well, which may reduce the number of positive cells. Also, assay by flow cytometry requires a large number of cells, a requirement that can be time-consuming and still limits the detection level to about 1 GFP-positive cell among 105 GFP-negative cells. Because we used pooled clones of transfected cells, the absolute number of sequences that could potentially be activated is not known exactly. This indeterminacy makes an absolute assessment of the mutation frequency problematic. However, the maximum observed mutation frequencies after treatment were about 100-fold over background levels, indicating the presence of sufficient numbers of cells with the potential to reactivate GFP. Also, the observed mutation frequencies of up to 1.5 × 10−3 correspond well with those found using other detection systems with dinucleotide microsatellite repeats (18 , 32) .
We observed that each agent caused a significant increase in GFP-positive cells when the average of all dose levels was compared with zero-dose cells across all transfected cell pools. This observation may be an indication that the agents caused DNA slippage events leading to frameshift mutations or, alternatively, that epigenetic effects or other frameshift mutations were responsible for the increase in GFP-positive cells. Dose effects in the no-repeat control cells did not approach statistical significance; nevertheless, some apparent patterns were potentially interesting. Treatment of the no-repeat controls with γ-irradiation appeared to result in a larger increase in GFP-positive cells than treatment with other DNA-damaging agents, a pattern that might indicate a larger number of deletion or insertion events or other (epigenetic) treatment effects of γ-irradiation. In the no-repeat control cells, response sometimes appeared to drop off, particularly at the highest dose, as, for instance, with t-BHP. This pattern could perhaps reflect selective removal of GFP-positive cells at the highest doses as a result of cell cycle arrest or cell death. However, these qualitative patterns need to be interpreted with extreme caution because the changes in response were below the resolution of the assay, and none of them were statistically distinguishable from random variation.
Only MNNG, t-BHP, and UV irradiation, but not γ-irradiation or BPDE exposure, resulted in statistically significant interactions between dose and repeat unit with respect to microsatellite slippage events. These observations are consistent with the hypothesis that sporadic microsatellite mutations found in human tumors may be associated with carcinogen exposures and may be the direct result of the action of carcinogens. However, several limitations to the use of GFP in pools of stable transfectants make it difficult to estimate the exact frequencies of these mutations. These frequencies may depend on several factors such as the length and type of repeat unit in the microsatellite, the number of slippage events that are expected to result in activation of the reporter, the copy number and chromosomal site(s) of integration of the transfected plasmid, and the effect of the specific lesion on DNA slippage events. In this study, we have kept the same number of repeat units among reporter plasmids and constructed the upstream sequences to be of comparable size to allow a better comparison between the different constructs. Also, by using pools of transfected cells in these experiments, we expected to average out any effects of clonal variation that could be due to the reporter plasmid copy number and/or the chromosomal integration site.
Averaged across all cell lines, γ-irradiation led to an increase in GFP activation from zero-dose to higher-dose cells, suggesting the creation of frameshift mutations. However, we did not find any evidence for sequence-specific changes in dose response to γ-irradiation. This observation may be related to the variety of DNA damage caused by γ-irradiation, such as single- and double-strand breaks and base and nucleotide damage.
BPDE causes mostly GC to TA transversions, often at CpG sites, but it can also induce other types of mutations, such as frameshift alterations (33) . In one study, the frameshift mutation frequency for BPDE has been estimated at 3 × 10−5 for a nonrepeated sequence, a frequency that was at least 1000-fold above background (34) . In our study, BPDE exposure resulted in an increase in GFP-positive cells in all transfectants, which would be consistent with the occurrence of frameshift mutations leading to GFP activation. However, the observed increases were similar across repeat units, indicating that the effects of BPDE were independent of the DNA sequence upstream of GFP.
For MNNG, the strongest component of interaction was that the increase in response from low to high dose differed between the average of the tetranucleotide repeat units and the CA repeat unit. The absolute increase in response from low to high dose in the CA repeat cells was small. This finding was unexpected because the CA repeat sequence is rich in guanine bases that should provide a target for MNNG activity. Interestingly, resistance to MNNG has been reported in the absence of a functional DNA mismatch repair system (35 , 36) . The RKO cell line is deficient in recognizing methylated guanines and thereby avoids DNA repair and associated slippage events that could lead to reactivation of GFP. Consequently, the situation in which mismatch repair-proficient cells are exposed to alkylating agents and experience repair-associated DNA slippage events is beyond the scope of the present study. Our data did not demonstrate any differential dose-response pattern between the AAAG and CAGT repeats, which suggests that triplex DNA structures are not driving mutations with MNNG.
Increasing the dose of t-BHP clearly led to increased levels of GFP-positive cells, but most of this increase was attributable to the comparison of the zero dose with the average of the non-zero doses rather than to differences among the non-zero doses. This result may reflect a limited capacity to cause frameshift mutations by oxidative DNA damage or a rapidly saturating effect of t-BHP. For the ATAG, CAGT, and CA repeat units, the highest response was already observed with the low dose of t-BHP. Interestingly, the AAAG repeat responded differently to t-BHP than did the CAGT repeat, in that the AAAG repeat cells exhibited a significantly larger increase from low- to high-dose t-BHP. This difference was the only indication in the study for possible effects of triplex DNA-forming capacity. Previous reports indicate that oxidative damage to DNA can lead to increased mutation frequency in E. coli both in vitro and in vivo (17 , 37) . Similar observations were made with two human lung cancer cell lines harboring a CA repeat upstream of the hygromycin resistance gene. There, oxidative DNA damage caused an increase of up to 27-fold in mutation frequency after selection for resistant colonies (18) . We observed a comparable 22-fold increase in GFP reactivation in the CA repeat cell pools over the no-repeat control, but that increase was not statistically significant. These and other results suggest that oxidative DNA damage is an important factor in smoking-related carcinogenesis and that it might contribute to the observed mutations in microsatellite sequences in lung tumors (21) . However, our results do not provide any indication that mutations in AAAG repeats occur more frequently in response to oxidative damage than mutations in the other microsatellite repeats. On the contrary, there was some evidence that the AAAG repeat cells were less sensitive to the effects of BPDE than the other repeat unit cell lines.
The cellular effects of UV irradiation are due in part to direct DNA damage in the form of pyrimidine dimers and in part to oxidative stress resulting from the absorption of photons. In our system, UV irradiation resulted in large increases in GFP reactivation in the reporter system, with significant differences between the cell pools harboring repeat units and the no-repeat control and among the different repeat units. The average response of the tetranucleotide repeats was much higher than that exhibited by the CA dinucleotide repeat, whereas a significant difference was found between the CAGT and AAAG repeats. UV irradiation causes pyrimidine dimers, which are generally repaired by nucleotide excision repair. The rank order of mutation frequency for the four repeats is consistent with the available positions for pyrimidine dimer formation. Four pyrimidine dimers can be formed in either DNA strand per AAAG repeat unit, 2 dimers can be formed in either DNA strand per ATAG repeat unit, 1 dimer can be formed in either DNA strand per CAGT repeat unit, and none can be formed in either DNA strand in the CA dinucleotide repeat. In the case of the CA repeat, there is a pyrimidine sequence directly adjacent to the repeat unit, which may explain some of the observed effects of UV irradiation (see Table 1 ⇓ ). We did find significant differences in dose-response profile between the AAAG and CAGT repeat cells, but within our experimental setting, we cannot distinguish between effects of the number of potential pyrimidine dimers and triplex DNA-forming capacity. Thus, our results provide evidence that the AAAG repeat cells responded differently to treatment with t-BHP and UV irradiation. For UV irradiation, the response of the different repeat units corresponded with the potential to form pyrimidine dimers.
A limitation of the present study is that RKO is mismatch repair deficient and has intact p53 function. In the absence of mismatch repair, the pattern of differential responses may be more reflective of the direct effects of damage than of those associated with repair of premutagenic lesions. The presence or absence of p53 function may also influence preferential repair of mismatched DNA. It was reported that p53 may have differential binding specificity to mismatches in DNA and can bind triple-stranded DNA (38) . In one study, an association between sporadic mutations in AAAG microsatellite repeats and p53 mutations was observed (6) . Thus, although we were able to demonstrate a differential microsatellite response to environmental agents depending on specific repeat units, we may not have captured all influences leading to microsatellite mutations observed in human tumors. An additional factor that could affect preferential mutation of certain microsatellites might be the chromosomal location of some of these tetranucleotide repeats. A recent report found that some unstable minisatellite repeats are located near recombinational hot spots in the genome and concluded that location near such a hot spot is the most likely cause of the observed high mutation rates (39) . Future research will be necessary to determine the specific relationships among p53 status, DNA mismatch repair status, and chromosomal location and the observed differential responses of microsatellite repeats to DNA-damaging agents described in this study.
In conclusion, we observed sequence-specific dose-response patterns with MNNG, t-BHP, and UV irradiation, but not for γ-irradiation or BPDE. Nevertheless, even for the latter two agents, GFP reactivation increased with dose among cell lines harboring repeat units. To our knowledge, this report is the first to indicate that microsatellite sequences have a higher mutation frequency than nonrepetitive sequences after DNA damage in human cells. Our results suggest that some of the observed “sporadic” microsatellite mutations in lung, head and neck, and bladder tumors may be caused by exposure to carcinogens.
We thank Dr. Carl Bortner for assistance with flow cytometry and cell sorting.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
↵1 To whom requests for reprints should be addressed, at National Institute of Environmental Health Sciences, Laboratory of Molecular Carcinogenesis, Maildrop C3-01, P. O. Box 12233, Research Triangle Park, NC 27709. Phone: (919) 541-2694; Fax: (919) 541-2511; E-mail:
↵2 The abbreviations used are: GFP, green fluorescent protein; BPDE, benzo(a)pyrene diol epoxide; t-BHP, t-butyl hydrogen peroxide; EGFP, enhanced GFP; MNNG, N-methyl-N-nitro-N-nitrosoguanidine; FACS, fluorescence-activated cell sorter.
- Received April 18, 2002.
- Accepted August 26, 2002.
- ©2002 American Association for Cancer Research.