We compared the binding of human MutSα (hMutSα) to synthetic DNA substrates containing U opposite either G (U•G) or A (U•A), using a gel mobility shift assay. hMutSα bound well to substrates containing U•G mismatches (apparent k_D of 70 nM; Figure 1A), but poorly to substrates containing U•A pairs, (apparent k_D > 135 nM). Competition analysis showed that binding of hMutSα to a labeled duplex substrate containing a single U•G mismatch was unaffected by the presence of a 50-fold molar excess of unlabeled duplex containing a U•A pair, but diminished more than 5-fold by a comparable amount of duplex DNA containing a U•G mismatch (Figure 1B). These properties are consistent with previous binding studies showing that hMutSα preferentially recognizes U•G or UU/GG heteroduplex oligonucleotides relative to homoduplexes 151617; and evidence that duplexes containing U•G but not U•A pairs activate MutSα ATPase activity 18.

Substrate preference of purified recombinant human UNG (hUNG) was analyzed by comparing deglycosylation activity on duplex substrates containing a U•G mispair or U•A pair. Deglycosylation creates an abasic (AP) site, which is alkali-labile, allowing ready quantitation of UNG activity on end-labeled substrates by treatment with alkali. hUNG had comparable activity on substrates containing U•A or U•G (Figure 1C). This is consistent with the documented ability of UNG to recognize unpaired uracil 19; and with the ability of repair complexes associated with UNG to initiate short-patch base excision repair on DNA containing U opposite either G or A 20. Thus, in the repair of genomic uracil, base excision repair responds to U•A or U•G while mismatch repair may specifically target U•G heteroduplexes.

Specific recognition by hMutSα of DNA duplexes containing U•G mispairs (Figure 1) suggested that, in human cells, genomic uracil is corrected by MutSα-directed mismatch repair pathway. To test this possibility, we developed circular ds DNA substrates containing a single U at a defined position, which could be used to assay repair by either pathway. These substrates were modeled on those used to define the mechanism of MutSα-driven repair at single-base mismatches and chemically modified bases, which contain a mismatched or damaged base at a defined position and a single-strand DNA discontinuity (nick or short gap) to direct repair to the strand containing the damage 21222324252627. A synthetic oligonucleotide carrying a U within the single HindIII restriction site in the lacZα gene was annealed to single-stranded circular M13mp18 DNA (Figure 2A, left); and then extended with the high-fidelity Phusion polymerase (New England Biolabs), which lacks both 5'-3' exonuclease and strand displacement activity. This created a duplex circular molecule, containing a single U•G mismatch that interrupts the HindIII restriction site, and a strand discontinuity (nick or short gap) 38 nt from the mismatch (Figure 2A, center). This substrate, referred to henceforth as M13-U•G, cannot be cleaved by HindIII, which requires the correct palindromic sequence (AAGCTT) on both DNA strands. The strand discontinuity will direct mismatch repair to the strand containing the single U, and repair of U•G to C•G will create a site cleavable by HindIII (Figure 2A, right). Substrates contain a single Bme 1580 I site remote from the mismatch. Simultaneous digestion of the unrepaired substrate with Bme 1580 I and HindIII will create a single linear fragment 7.3 kb in length; while digestion of repaired DNA will generate two fragments, of 4.2 and 3.1 kb. The ratio of these two fragments to total DNA provides a measure of efficiency of repair.

We confirmed that the M13-U•G ds substrates could report on repair by the UNG-dependent base excision pathway by assaying reconstitution of the HindIII site after sequential incubation with purified hUNG, hAPE1, and DNA polymerase β (pol β), members of the UNG-directed base excision repair pathway. HindIII was unable to cleave M13-U•G following incubation with UNG, or UNG and APE1; however, subsequent incubation with pol β resulted in 68% cleavage by HindIII (Figure 2B). Therefore, the M13-U•G circular substrates sensitively monitor reconstitution of the HindIII site by in vitro repair.

To distinguish the contributions of MutSα and UNG to repair of U•G mismatches, we first asked how inhibition of UNG affected repair of M13-U•G circular substrates in human cell extracts. The small protein, uracil glycosylase inhibitor (Ugi), encoded by bacteriophage PBS2, is specific for UNG homologs and forms a 1:1 complex with the DNA binding domain of UNG, blocking uracil excision in vitro and in vivo 28293031. We prepared nuclear extracts from three human cell lines: HeLa, derived from an adenocarcinoma 2232; Ramos, derived from an actively hypermutating B cell lymphoma 33; and LoVo, derived from an MSH2-deficient colorectal carcinoma 34. All three extracts showed high levels of uracil DNA glycosylase activity; and Ugi completely inhibited deglycosylation in all three extracts (Figure 3A). There are other uracil DNA glycosylases in human cells, including MBD4, TDG, and SMUG1, which are divergent from UNG 35 and therefore should be insensitive to Ugi. The fact that uracil DNA glycosylase activity could not be observed in the presence of Ugi confirms that UNG is the predominant activity for deglycosylation of uracil in these extracts; and shows that treatment of human nuclear extracts with Ugi can be used to distinguish UNG-directed base excision repair from UNG-independent repair.

We then assayed repair of M13-U•G circular substrates in nuclear extracts prepared from HeLa and the constitutively hypermutating B cell line, Ramos, in the presence or absence of Ugi. HeLa nuclear extracts supported repair of 52% of the molecules containing U•G mismatches, and this diminished to 35% in the presence of Ugi (Figure 3B). Ramos supported repair of 64% of the molecules containing U•G mismatches, and this diminished to 32% in the presence of Ugi (Figure 3B). Repair values in the absence of Ugi are consistent with published values for correction of single base mismatches by human mismatch repair in vitro 2223242532. U•G mismatch repair in Ugi-treated extracts was reproducible. Repair efficiency of U•G in three independent experiments averaged 33% repair (S.E. 1.76%) in Ugi-treated HeLa extracts; and 32% (S.E. 1.45%) in Ugi-treated Ramos extracts. Thus, Ugi-sensitive, UNG-directed repair accounts for 30–50% of repair of U•G mispairs in Ramos or Hela extracts, while the remainder (and major fraction) of repair is Ugi-insensitive and UNG-independent. Because Ugi blocked detectable uracil glycosylase activity in HeLa and Ramos nuclear extracts (Figure 3A), Ugi-insensitive repair must be carried out by activities independent of base excision repair.

To ask if MutSα can direct repair of U•G mispairs, we assayed repair of the M13-U•G circular substrates in Ugi-treated extracts derived from the MSH2-deficient cell line LoVo. Mismatch repair is defective in LoVo extracts but can be restored in vitro by addition of purified hMutSα 2324323637. Nuclear extracts of LoVo, treated with Ugi to inhibit UNG (e.g. Figure 3A), supported repair of only 8% of the M13-U•G substrates (Figure 4A). Addition of purified hMutSα to Ugi-treated LoVo nuclear extracts increased repair 3-fold, to 24% (Figure 4A). Thus, MutSα promotes efficient repair of U•G mispairs.

We then asked if MutSα-dependent repair of U•G mispairs requires pol δ, ATP, and PCNA, all essential for canonical mismatch repair in vitro 383940. Pol δ activity and mismatch repair are both blocked by aphidicolin 3841. In a control experiment, we established that aphidicolin does not block the mutagenic polymerase pol η, which is strongly implicated in MutSα-directed hypermutation 4243, by showing that aphidicolin did not affect extension by pol η of a 5'-end-labeled 29-mer oligonucleotide annealed to a 50-mer template to produce a predicted 44-mer duplex extension product (Figure 4B). However, aphidicolin almost completely prevented repair of M13-U•G mispair substrates in Ugi-treated HeLa and Ramos nuclear extracts, diminishing repair to 5%, from 35% (HeLa) and 32% in Ramos (Figure 4C). This shows that pol δ participates in MutSα-directed repair of U•G mispairs. The low level of repair evident in Ugi-treated nuclear extracts in the presence of aphidicolin may be due to participation of aphidicolin-resistant polymerases.

PCNA binding protein (PBP) inhibits excision and DNA resynthesis during mismatch repair by blocking PCNA function 3840. Repair of M13-U•G mispair substrates was essentially abolished when PBP was added to Ugi-treated nuclear extracts (Figure 4C). Thus, UNG-independent repair of U•G mispairs depends upon PCNA.

Omission of ATP caused the percentage of molecules repaired to decrease from 35% to 10% in Ugi-treated HeLa nuclear extracts; and from 32% to 9% in Ugi-treated Ramos nuclear extracts (Figure 4C). Low levels of ATP in the extracts may account for repair in the absence of exogenous ATP.

The dependence of repair of M13-U•G substrates on MutSα, PCNA, pol δ, and ATP, establishes that the canonical mismatch repair pathway can correct U•G mismatches in duplex DNA molecules.

Comparable levels of U•G repair were observed in nuclear extracts of HeLa and the hypermutating B cell line, Ramos (Figures 3 and 4), suggesting that hypermutation does not reflect compromised function of faithful repair pathways. To further characterize repair fidelity, we used an assay that measures inactivation of β-galactosidase due to mutagenesis that occurs during synthesis across a gap in the lacZα gene in M13mp18 phage, quantitating the fraction of clear or pale blue plaques produced upon transformation of E. coli followed by plating on a helper lawn on X-gal plates 44. This assay provides convenient comparison of mutation frequencies under different conditions, but underestimates total mutation frequency because not all mutations affect β-galactosidase activity. As a positive control, we used the M13-Gap51 substrate, in which a 51 nt gap, extending from the site of the mismatch at the HindIII site (position 6281) to the EcoRI site (position 6230) near the 5' end of lacZα (Figure 5A), recapitulates the short excision tract created during mismatch repair of the M13-U•G substrate. We compared mutagenesis following incubation of M13-Gap51 substrates with either the faithful polymerase, Taq, or the mutagenic polymerase, pol η, to fill the gap. The proportion of clear or pale blue plaques was 2.5 × 10^-3 following by gap filling by Taq, and 1.3 × 10^-2 following gap filling by pol η (Figure 5B). These numbers are significantly different (P = 0.001, two-tailed Fisher's exact test), confirming that this assay provides a sensitive measure of mutagenesis.

We used the same assay to determine the frequency of β-galactosidase inactivation in the M13-U•G substrates. The background frequency of β-galactosidase inactivation evident upon direct transformation of E. coli with unrepaired M13-U•G substrates was 1.3 × 10^-3 (Figure 5B). When M13-U•G substrates were incubated in Ugi-treated Ramos extracts to allow repair to occur prior to transformation, the proportion of clear or pale blue plaques was 1.8 × 10^-3, not significantly different from unrepaired M13-U•G substrates (P = 0.77, two-tailed Fisher's exact test; Figure 5B). Parallel analysis showed that the fraction of repaired substrates, as assayed by HindIII-sensitivity, was 34%, comparable to the efficiency of repair in Ugi-treated Ramos nuclear extracts (not shown). Thus, the extracts were active for repair, but mutagenesis appeared not to accompany repair.

The plaque color-based mutagenesis assay scores only mutations that inactivate β-galactosidase. To ensure that this assay did not greatly underestimate the mutation frequency, we sequenced randomly selected blue plaques arising from M13-U•G incubated in the repair extract prior to transformation of E. coli. A majority of the excision tracts initiated by in vitro mismatch repair of nick-circular substrates terminate between 90 and 170 bases from the mismatch 45. We therefore analyzed a 200 nt region bordered by the strand discontinuity and encompassing DNA upstream of the mismatch site (Figure 2A) for mutations induced in the course of nick-directed uracil repair. Within this region, we did not find significant mutagenesis of UG-M13 substrates incubated with Ugi-treated HeLa extract (57 plaques, no mutations, 11,400 nt sequenced) or from Ramos nuclear extract (93 plaques, 1 mutation, 18,600 nt sequenced). Thus, in contrast to pol η-directed gap filling, repair of U•G by the human nuclear extracts was not mutagenic.

Human UNG2 (hUNG) was PCR amplified from cDNA from the Ramos cell line using primers 5'- CACCATGATCGGCCAGAAGACGCTCTACTCCTTTTTCTC and 5'- TCACAGCTCCTTCCAGTCAATGGGCTTCTTGCC; and cloned downstream of the T7 promoter in pET100 (Invitrogen, Carlsbad, CA), providing an N-terminal His₆ tag. Following sequence confirmation, the hUNG plasmid was used to transform E. coli BL21 DE3 pLysS (Invitrogen), in which T7 RNA polymerase is expressed under control of the lactose repressor. Mid-log cells were cultured at 37°C for 2 hr with 1 mM IPTG to induce T7 polymerase expression; collected by centrifugation; frozen; resuspended in 50 ml of lysis buffer containing 20 mM Tris pH 7.7, 300 mM NaCl, 0.5 mM DTT, and a protease inhibitor mini-tablet (Roche, Indianapolis, IN); sonicated twice for 20 seconds, and cell debris removed by centrifugation. hUNG was purified by Ni⁺⁺ chromatography (Sigma, St. Louis, MO) according to manufacturer's instructions, followed by S-Sepharose gravity flow chromatography (GE, Piscataway, NJ). hAPE1 was purified as described previously 52. Pol β was purchased from Trevigen (Gaithersburg, MD). MutSα was overexpressed by co-infecting Sf9 insect cells with baculovirus vectors expressing MSH2 and MSH6 and purified as previously described 15. HeLa, Ramos, and LoVo cells were supplied by the National Cell Culture Center (Minneapolis, MN) as pellets stored on wet ice, and nuclear extracts were prepared as previously described 37. Protein concentrations were determined using BSA as a standard.

Synthetic duplexes carrying U•G mismatches were created by annealing EDL56, 5'-GCCGAATTTCTAGAAT

U

U

M13-U•G double-stranded (ds) circular substrates for mismatch repair assays were designed to carry a single U•G mismatch within a HindIII restriction site, and a 5' strand discontinuity (nick or short gap) on the strand targeted for repair 222558. M13mp18 circular single-stranded (ss) (+) strand DNA (NEB) was annealed to PAGE purified oligonucleotide EDL330, 5'-CCCAGTCACGACGTTGTAAAACGACGGCCAGTGCCAAG

U

Uracil repair and binding assays results were confirmed by repetition and representative results are presented. Images were captured and analyzed by phosphorimager (GE) or captured and quantified by an AlphaInnotech HD2 system and presented as reversed ethidium bromide stained images as indicated for each assay.

MutSα binding assays were performed in 30 μl reactions containing 25 mM Tris pH 7.7, 100 mM KCl, 1 mM EDTA, 1 mM DTT, 7.5% glycerol, with indicated concentrations of purified MutSα, 10 fmol radiolabeled heteroduplex, and indicated concentrations of homoduplex oligonucleotide competitor DNA (annealed EDL365/EDL366). Following incubation on ice for 5 min, reactions were brought to 0.7% Ficoll, and then resolved by 6% native PAGE in 0.5 × TBE. For competition experiments, indicated concentrations of unlabeled U•G (EDL56/EDL364) or U•A (EDL56/EDL365) duplexes were added prior to addition of protein.

UNG activity was assayed by quantitating production of the alkaline-labile abasic site generated upon deglycosylation of uracil. Purified recombinant hUNG (40 to 2.5 fmols) and 50 fmols of U•G (EDL56/EDL364) or U•A (EDL56/EDL365) synthetic duplexes were incubated in 25 μl reactions containing 10 mM Tris, 1 mM EDTA, 37°C for 15 min; treated with 0.25% SDS, 10 mM EDTA and 100 μg/ml Proteinase K (Promega, Madison, WI) at 37°C for 10 min; and then with 0.1 N NaOH at 37°C for 5 min. Following addition of an equal volume of formamide, cleavage products were resolved by 15% denaturing PAGE. Images were captured and quantified by phosphorimager. UNG activity present in human nuclear extracts was determined by a similar procedure, using 50 μg of nuclear extract and 50 fmols of 5'-end-labeled EDL58 oligonucleotide in 30 μl reactions containing 20 mM Tris pH 7.7, 100 mM KCl, 5 mM MgCl₂, 1 mM glutathione, 1.5 mM ATP, 0.1 mM each dNTP, and 50 μg/ml BSA for 20 min at 37°C. Ugi (NEB, 2 U in 1 μl) was added to inhibit UNG.

Mismatch repair assays followed a published protocol 22. M13-U•G ds circular substrates (13.6 fmol) were incubated with 50 μg of nuclear extract in 30 μl reactions containing 20 mM Tris pH 7.7, 100 mM KCl, 5 mM MgCl₂, 1 mM glutathione, 1.5 mM ATP, 0.1 mM each dNTP, and 50 μg/ml BSA for 20 min at 37°C. In some cases, recombinant human MutSα (hMutSα) was added (1.6 pmol), or pol δ was inhibited with aphidicolin (183 μM; Sigma); or PCNA was inhibited by addition of PCNA binding protein (PBP, American Peptide Company Sunnyvale, CA), as described 24. Reactions were terminated by addition of 15 μl (one-half volume) 1% SDS, 25 mM EDTA pH 8, 0.1 mg/ml Proteinase K; and following phenol extraction and ethanol precipitation, DNA was resuspended in 1× NEB Buffer 2 supplemented with 100 μg/ml RNAse A, 100 μg/ml BSA. DNA was then cleaved by simultaneous digestion with Bme 1580 I (1 U), which linearizes M13mp18 at position 2088; and HindIII (1 U), which cleaves only those molecules in which the U•G mismatch has been converted to a C•G pair in the course of DNA resynthesis, creating a HindIII site (AAGCTT) at position 6281. Digestion products were resolved by 1% agarose gel electrophoresis, gels stained with ethidium bromide, and images captured and quantified using an AlphaInnotech HD2 imaging system. Representative images are presented as reversed ethidium bromide stained images.

The activity pol η was measured by extension of 5'-³²P-labeled oligonucleotide EDL145, annealed to EDL119, with 160 fmols purified recombinant pol η (Enzymax, Lexington, KY), in buffer conditions identical to those used for mismatch repair assays (20 mM Tris pH 7.7, 100 mM KCl, 5 mM MgCl₂ 1 mM glutathione, 1.5 mM ATP, 0.1 mM each dNTP, and 50 μg/ml BSA), in the presence or absence of 180 μM aphidicolin. The 44-mer extension product was distinguished from the EDL145 29-mer primer by 15% denaturing PAGE.

The assay for fidelity of U•G heteroduplex repair was based on an assay developed for quantitating DNA polymerase fidelity on gapped M13mp2 molecules 44. Control M13-Gap51 substrates, containing a 51 nt gap within the lacZα gene of M13mp18, were generated by annealing a 10-fold molar excess of M13mp18 ssDNA circles to a slightly shorter complementary strand, produced by digestion of duplex M13 with EcoR1 and HindIII, and removal of the 51 bp fragment using a PCR purification column (Qiagen). Unpaired linear single strands and any circular single-stranded M13mp18 were removed by BND cellulose chromatography and duplex linear DNA was removed by digestion with Exonuclease V (USB). The gap in M13-Gap51 was filled with either Taq polymerase (1 U, NEB) or 160 fmols pol η (12.5 ng), under manufacturer's standard conditions. Repair was carried out in 30 μl reactions in the same conditions as the mismatch repair assay (above). Recovered DNA was resuspended in 30 μl TE, and repair at the HindIII site was quantitated by digestion of a 5 μl aliquot with Bme 1580 I and HindIII followed by gel electrophoresis, as described above. Repair fidelity was analyzed by a plaque assay for β-galactosidase activity, in which E. coli MC1061 was transformed with 1.4 ng of DNA by electroporation, transformants plated with the helper strain CSH50, on minimal glucose media containing X-gal (Sigma) and unmutated (blue), or mutated (pale blue or clear) plaques quantitated. E. coli strains were provided by Dr. Lawrence A. Loeb, University of Washington, Seattle, WA.