Conformational trapping of Mismatch RecognitionComplex MSH2/MSH3 on repair-resistant DNA loopsWalter H. Langa,1,2, Julie E. Coatsb,1, Jerzy Majkaa, Greg L. Huraa, Yuyen Linb, Ivan Rasnikb,3, and Cynthia T. McMurraya,c,d,3

aLawrence Berkeley National Laboratory, Life Sciences Division, 1 Cyclotron Road, Berkeley, CA 94720; cDepartment of Molecular Pharmacology andExperimental Therapeutics; dDepartment of Biochemistry and Molecular Biology, Mayo Foundation, 200 First Street, Rochester, MN 55905; andbDepartment of Physics, Emory University, 400 Dowman Drive, MSC N214, Atlanta, GA 30322

Edited* by Peter H. von Hippel, Institute of Molecular Biology, Eugene, OR, and approved August 3, 2011 (received for review April 6, 2011)

Insertion and deletion of small heteroduplex loops are commonmutations in DNA, but why some loops are prone to mutation andothers are efficiently repaired is unknown. Here we report that themismatch recognition complex,MSH2/MSH3, discriminates betweena repair-competent and a repair-resistant loop by sensing the con-formational dynamics of their junctions. MSH2/MSH3 binds, bends,and dissociates from repair-competent loops to signal downstreamrepair. Repair-resistant Cytosine-Adenine-Guanine (CAG) loops adopta unique DNA junction that traps nucleotide-bound MSH2/MSH3,and inhibits its dissociation from the DNA. We envision that junctiondynamics is an active participant and a conformational regulatorof repair signaling, and governs whether a loop is removed byMSH2/MSH3 or escapes to become a precursor for mutation.

DNA repair ∣ mismatch repair ∣ smFRET ∣ trinucelotide expansion

Insertion or deletion of small extrahelical loops is one of themost common mutations in human cancers (1–3), but the me-

chanism by which they occur is unknown. Small loops, bulges,or kinked DNA occur frequently in DNA, and provide signalsfor p53 recognition (4–7), recombination (8, 9), and/or most oftenremoval by the mismatch repair system (10–14). Two hetero-dimeric mismatch recognition complexes, MSH2/MSH6 andMSH2/MSH3, operate in mammals with distinct, but overlappingspecificities (12–14). The crystal structure (15–18), Atomic ForceMicroscopy (AFM) (19), and single molecule fluorescence reso-nant energy transfer (smFRET) (19, 20) confirm that MSH2/MSH6 and Escherichia coli (MutS) preferentially bind single basemismatches or two base pair bulges. MSH2/MSH3 can recognizesome base-base mismatches (21), but has a higher apparent affi-nity and specificity for small DNA loops composed of 2–13 bases(12–14, 22–24). Thus, defects in repair mediated byMSH2/MSH3are poised to be a major source of insertion-deletion mutations.

The mechanism by which MSH2/MSH3 discriminates betweenrepair-competent and repair-resistant loops (24–26), howeverremains enigmatic. A small ðCAÞ4 loop of DNA can be faithfullyrepaired by MSH2/MSH3 both in vitro (24, 26) and in vivo (25,27, 28). In contrast, hydrogen bonded CAG hairpin loops are notexcised, and confer genomic instability through insertion andamplification of CAG repetitive tracts (29–31). Although ðCAÞ4loops and CAG hairpins both harbor three-way junctions, MSH2/MSH3 interacts with them distinctly (24, 26). Why one templateis repaired better than the other is not known, but the conse-quence is remarkable: Inefficient repair of CAG loops results inmutations that underlie more than 20 hereditary neurodegenera-tive or neuromuscular diseases (30–33).

Here, we address the underlying basis for discriminating repair-competent and repair-resistant DNA loops by MSH2/MSH3. Wefind that MSH2/MSH3 binds with similar affinity to a repair-competent ðCAÞ4 loop and to repair-resistant CAG hairpins. How-ever, the three-way hairpin junction adopts a conformational statethat traps nucleotide-bound MSH2/MSH3, and inhibits its disso-ciation from the hairpin. The biochemical and smFRET resultsimply that repair-resistant CAG hairpins provide a unique butnonproductive binding site for nucleotide-bound MSH2/MSH3,

which fails to effectively couple DNA binding with downstreamrepair signaling. We envision that conformational regulation ofsmall loop repair occurs at the level of the junction dynamics.

ResultsConformational Integrity of MSH2/MSH3 and the DNA “Junction” Tem-plates.We characterized the DNA-binding affinity and nucleotidebinding properties of MSH2/MSH3 bound to looped templates,either a ðCAÞ4 loop or CAG hairpin loops of either 7 or 13(ðCAGÞ7 and ðCAGÞ13) triplet repeats (Fig. 1A). Both loop andhairpin templates were constructed from two single-stranded oli-gonucleotides (Fig. 1A). Neither the ðCAÞ4 loop nor the CAGstem had complementary sequences within the duplex portionof the template. Thus, the junction templates folded into stableextrahelical loops, which have been previously characterized insolution (26, 31). Folding of the CAG loops creates A/A mis-matches every third base pair in the stem, for a total of threemismatches in the ðCAGÞ7 template or a total of six mismatchesin the ðCAGÞ13 template. DNA templates were analyzed by gelelectrophoresis to (i) ensure the absence of any traces of single-stranded DNA and (ii) that the DNA loops were intact, as judgedby an increase in the loop size and gel mobility (Fig. 1B). Unlessspecifically noted, all DNA templates were synthesized contain-ing a duplex base of 18 bases (Fig. 1A).

The purified human MSH2/MSH3 protein (hereafter referredto as MSH2/MSH3) was also of high quality (Fig. 1C). The full-length MSH2/MSH3 was expressed and copurified as a hetero-dimer, and each subunit, when resolved by PAGE, migrated as asingle band according to the expected molecular mass (Fig. 1C).To further test the conformational integrity of the protein, wevisualized full-length MSH2/MSH3 using small angle X-ray scat-tering (SAXS) (34, 35)(Fig. 1D). Interestingly, the high-resolu-tion SAXS structure revealed that the N-terminal portion of theMSH3 subunit formed an unfolded “panhandle” structure, whichextended beyond the heterodimeric interface of MSH2/MSH3.The handle undergoes a visible broadening and conformationalchange upon binding the ðCAÞ4 loop. But otherwise, the DNA-bound heterodimeric portion of humanMSH2/MSH3 was similarin conformation to that of the human MSH2/MSH6 bound totemplate containing a G-T mispaired base (16). Using these well

Author contributions: W.H.L., J.M., I.R., and C.T.M. designed research; W.H.L., J.E.C., J.M.,G.L.H., and Y.L. performed research; W.H.L., J.E.C., J.M., G.L.H., I.R., and C.T.M. analyzeddata; and W.H.L., J.M., I.R., and C.T.M. wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor.

Freely available online through the PNAS open access option.1W.H.L. and J.E.C. contributed equally to this work.2Present address: Department of Surgery, St. Jude Children’s Research Hospital, 262 DannyThomas Place, MS 332, Memphis, TN 38105.

3To whom correspondence may be addressed. E-mail: ctmcmurray@lbl.gov or irasnik@physics.emory.edu.

See Author Summary on page 17247.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1105461108/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1105461108 PNAS ∣ October 18, 2011 ∣ vol. 108 ∣ no. 42 ∣ E837–E844

BIOCH

EMISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Dec

embe

r 27

, 202

0

characterized materials, we tested whether there were biochem-ical or conformational differences, which segregated with the re-pair-competent or repair-deficient nature of looped templates.

MSH2/MSH3 Binds Nucleotides with High Affinity at both Repair-Com-petent and Repair-Resistant Templates. We observed little differ-ence in nucleotide affinity when MSH2/MSH3 was prebound torepair-resistant ðCAGÞ7 hairpin and ðCAGÞ13 templates, or to aðCAÞ4 loop, which is a good substrate for MSH2/MSH3 in vitro(22, 24, 26) and in vivo (27, 28). As measured by UV-cross-linking(Fig. S1), the affinity of ADP or ATP to either subunit of DNA-bound MSH2/MSH3 was substantially weaker when MSH2/MSH3 was bound to DNA (Table 1). However, the reduction innucleotide affinity for MSH2/MSH3 did not display striking dif-ferences among ðCAGÞ7, ðCAGÞ13, and the ðCAÞ4 loop templates(Table 1).

MSH2/MSH3 Binds with Similar Affinity to the Repair-Competent ðCAÞ4Loop and to the Repair-Resistant CAG Hairpins. To test whether nu-cleotide binding to MSH2/MSH3 altered its association withDNA, we labeled each DNA template with fluorescein at the5′-end of the bottom strand, and measured the DNA-bindingaffinity by fluorescence anisotropy (FA). In the absence of bound

nucleotide, the apparent affinity of MSH2/MSH3 for both theðCAÞ4 loop and hairpin templates was in the low nanomolarrange (Table 2), and was in good agreement with previous mea-surements (24, 26, 31). The presence of magnesium decreasedthe affinity of ATP-boundMSH2/MSH3 to any template by about10-fold, but, in general, DNA binding of ADP- or ATP-boundMSH2/MSH3 did not distinguish repair-competent ðCAÞ4 loopfrom the repair-resistant ðCAGÞ7 hairpin or ðCAGÞ13 hairpin.

MSH2/MSH3 Stabilizes a High FRET State When Bound to the Repair-Competent ðCAÞ4 Loop. The ðCAÞ4 loop differs structurally fromthe ðCAGÞ13 DNA in that the latter forms a hairpin comprisingG-C hydrogen bonded base pairs and A/A mispaired bases everythird nucleotide in the stem (24, 26). To test for conformationdifferences between the two templates, we measured the protein-induced DNA conformational dynamics using smFRET. We pre-pared DNA substrates, which were identical to those used in thebiochemical DNA-binding experiments, except that the bottomstrand of 18 nucleotides was labeled with Cy3 (on the 5′ end,green ball) and Cy5 (on the 3′ end, red ball) (Fig. 2A,C, E). Thus,for each template used in the smFRET experiments, the localenvironment of the fluorophores was identical. For both tem-plates, the top strand at the 3′ end contains a poly-dT extensionand a biotin tag for immobilization on streptavidin coated coverslips for observation (Fig. 2 A, C, E, blue ball). The extensionwas designed to prevent potential interaction of the fluorophoreswith the streptavidin surface. The smFRET was used to probeproximity between the Cy3 and Cy5 tags and the conformationaldynamics of the DNA.

We determined the FRET efficiency (EFRET) values for hun-dreds of individual molecules. In the absence of protein, the dis-tribution of EFRET for ðCAÞ4 substrates was a single narrow peakat EFRET ∼ 0.31 (Fig. 2A, DNA only; the peak at EFRET ¼ 0represents substrates with an inactive acceptor). In addition tothe EFRET population distribution, we followed the dynamics ofeach individual FRET pair by plotting time traces of donor (Cy3)and acceptor (Cy5) emission. However, there were no observabletransitions within our time resolution. To determine that the tran-sitions were fast but not absent, we measured the recovery rate ofthe acceptor dye intensity from the transitions to nonfluorescentstates in the presence of 2-mercaptoethanol. We compared re-sults for the ðCAÞ4 loop relative to the homoduplex DNA (withan identical local environment of the fluorescent dyes). The recov-ery of the intensity for ðCAÞ4 loop was an order of magnitudefaster then the homoduplex DNA (Fig. S2). The recovery is facili-tated by close proximity (2–3 nm or closer) of the donor andacceptor. Thus, the ends of the CA4 loop substrate came into closeproximity confirming that there were conformational fluctuations,even though they were not observable within our time resolution.

Addition of MSH2/MSH3 to the ðCAÞ4 loop template, in theabsence of nucleotides, led to the appearance of a new EFRETpeak (bound state, “high FRET”) at EFRET ∼ 0.4 (Fig. 2A,þMSH2∕MSH3). The population in the high FRET state in-creased with protein until the entire population had shifted toEFRET ∼ 0.4 (Fig. 2A, þMSH2∕MSH3). Consistent with highaffinity binding (Table 2), the high FRET state saturated at a

Fig. 1. Conformational Integrity of Mismatch Recognition Complexes andDNA templates. (A) Schematic structure of the three-way junction DNA tem-plates. The ðCAÞ4 loop and CAG hairpins are centrally located in the upperunlabeled strand. The duplex base comprising eighteen base pairs, whichwere labeled at the 5′ end with Cy3 and at the 5′ end with Cy5 for smFRETexperiments, or with a 5′ fluorescein for the FA experiments. (B) The purifiedtemplates resolved on native polyacrylamide gels and visualized by ethidiumbromide staining. SS is the 18nuc single strand DNA that is the complemen-tary strand for the looped templates, DS is the 18 bp homoduplex DNA, andthe heteroduplex looped substrates are as labeled. (C) Resolution of purifiedhuman MSH2/MSH3 (middle lane) and MSH2/MSH6 (right lane) proteins bySDS-PAGE. The size markers (right) indicate the molecular weights. (D) SAXSstructure of MSH2/MSH3 protein alone (top, left) or in the presence of theðCAÞ4 loop (top, right) overlaid on the crystal structure of humanMSH2/MSH6bound to a G-T mispaired base (11). View of the MSH2∕MSH3-ðCAÞ4 loopcomplex from front (bottom, right) and side (bottom, left).

Table 1. Nucleotide binding affinities of MSH2/MSH3 subunits determined by cross linking in theabsence or presence of DNA templates, KD in nM

Subunit Ligand No DNA CA4 CAG7 CAG13

MSH2 ADPð−Mg2þÞ 210 ± 90 250 ± 90 190 ± 50 530 ± 120ADPðþMg2þÞ 150 ± 60 150 ± 120 250 ± 50 440 ± 130ATPð−Mg2þÞ 730 ± 240 5,300 ± 700 4,800 ± 500 4,300 ± 800

MSH3 ADPð−Mg2þÞ nq nq nq nqADPðþMg2þÞ 160 ± 90 7,200 ± 1,000 4,000 ± 1,100 3,700 ± 540ATPð−Mg2þÞ 550 ± 190 53,800 ± 15,100 11,400 ± 4,000 59,200 ± 15,100

Nq: not quantifiable—lack of signal.

E838 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1105461108 Lang et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 27

, 202

0

protein concentration in the nanomolar range (Fig. 2A,þMSH2∕MSH3). Thus, MSH2/MSH3 formed a stable complex with therepair-competent ðCAÞ4 loop template in which the two ends ofthe heteroduplex loop were positioned more closely, suggestive ofbending.

To monitor the conformational dynamics of the transitions, wefollowed individual Cy3 (green) and Cy5 (red) emission traces forthe MSH2∕MSH3-ðCAÞ4 loop complex (Fig. 2B, the calculatedFRET efficiency curves are displayed in blue). The observationtime was limited typically to less than 60 s by photodestructionof the acceptor (indicated by black arrows, Fig. 2B). The singlemolecule traces did not vary significantly (Fig. 2B). A few tracescaptured conformational transitions (Fig. 2B, blue trace) consis-tent with detection of a protein-binding event (Fig. 2B, bluetrace). However, there were no observable dynamics withinour time resolution (Fig. 2B, black line), and the lifetime ofthe average transition was longer than our maximum observationtime (Fig. S2). Similar results were obtained when MSH2/MSH3bound to a comparably labeled A2 bulge, which is also a templatefor MSH2/MSH3-dependent repair (Fig. S3).

We evaluated additional control DNA templates. MSH2/MSH3 has weak affinity and rapidly dissociates from homoduplexDNA (24, 26, 31). Consistent with these properties, no high

smFRET population was observable when MSH2/MSH3 wasadded to homoduplex DNA, even at very high protein concentra-tion (Fig. 2 C and D). Both MutS and MSH2/MSH6 bend G-Tmismatched DNA at the site of the mismatch (15, 16, 18). Thus,we purified MSH2/MSH6 and added it to a comparably labeledG-T base mismatch template. Indeed, for an MSH2/MSH6 com-plex, we observed a high FRETshift, (Fig. 2E), which was similarin magnitude to that induced by MSH2/MSH3 on the ðCAÞ4 loop(Fig. 2C). The single molecule traces indicated that the highFRET state was stable (black horizontal line, Fig. 2F). MSH2/MSH6 does not bind to the ðCAÞ4 loop (26, 31), and no highsmFRET population was observable when MSH2/MSH6 wasadded to that template, even at very high protein concentration.Thus, MSH2/MSH6 andMSH2/MSH3 complexes displayed simi-lar transitions when bound to their preferred repair-competenttemplates with an estimated bending angle of 40 to 45° (Fig. S4).

Nucleotide Binding Increases the Dissociation of MSH2/MSH3 from theðCAÞ4 Loop Under Hydrolytic Conditions. MSH2/MSH6 and MSH2/MSH3 couple DNA binding and ATP hydrolysis to initiate down-stream repair (11–13). Thus, we tested the effects of ATP bindingand hydrolysis on the conformational dynamics of the MSH2/MSH3-bound ðCAÞ4 substrate. ATP was added to a complex con-taining MSH2/MSH3-bound ðCAÞ4 loop DNA in the presence(þMg) or absence (−Mg) of magnesium, and distribution ofsmFRET efficiencies was measured under both hydrolyzing andnonhydrolyzing conditions (Fig. 3). Induction of the high FRETstate by DNA-bound MSH2/MSH3 was independent of both

Table 2. DNA-binding affinities of wild-type MSH2/MSH3determined by Fluorescence Anisotropy in the presence andabsence nucleotides, KD in nM

Template CA4 CAG7 CAG13

(−Mg) 2.5 ± 0.5 3.8 ± 0.6 6.4 ± 0.7(+Mg) 2.6 ± 0.7 5.1 ± 0.6 5.3 ± 0.9ADP(−Mg) 1.6 ± 0.06 1.6 ± 0.06 2.8 ± 0.04ADP(+Mg) 41.9 ± 5.3 41.6 ± 6.2 27.7 ± 2.4ATP(−Mg) 3.0 ± 0.05 4.2 ± 0.04 6.6 ± 0.05ATP(+Mg) 39.7 ± 4.3 70.9 ± 11.6 43.0 ± 5.3

Fig. 2. Binding of MSH2/MSH3 and MSH2/MSH6 to their preferred repairsubstrates increases FRET efficiency. The dynamics of different substrate mo-lecules in the presence of or in the absence of addedMSH2/MSH3. (A) smFRETefficiencies for MSH2/MSH3 binding to the ðCAÞ4 loop substrate without(top) or with (bottom) MSH2/MSH3. The schematic of the labeled substrates:green ball is Cy3 label; red ball is Cy5 label (bottom); blue ball is biotin label.The protein concentration is indicated. (B) The time traces of representativedonor fluorescence (green, Cy3) and acceptor fluorescence (red, Cy5). Theblack line indicates the time of the ðCAÞ4 loop in the high FRET state. Timeof acceptor photobleaching is indicated by black arrow. (upper) Blue tracesare the corresponding FRETefficiencies. (C, D) Same as (A, B) for homoduplexsubstrate. (E, F) Same as (A, B) for binding of MSH2/MSH6 to a G/T mis-matched substrate.

Fig. 3. ATP increases dissociation of MSH2/MSH3 from a ðCAÞ4 loop sub-strate. FRET efficiency histograms for MSH2/MSH3 binding to the ðCAÞ4substrate at different protein concentrations in the presence of 100 μM ATP(A) without MgCl2 and (B) with 5 mM MgCl2. The MSH2/MSH3 concentrationis indicated. (C) The dynamics of the ðCAÞ4 substrate upon MSH2/MSH3binding in (B). The time traces of representative donor fluorescence (green,Cy3) and acceptor fluorescence (red, Cy5). The black line indicates the time ofthe ðCAÞ4 loop in the high FRET state. Time of acceptor photobleaching isindicated by black arrow. (upper) Blue traces are the corresponding FRETefficiencies. (D) Time traces like the one shown in (C) analyzed in a two statesystem using a Hidden Markov Model (36) to determine the average transi-tion rates from initial to high and high to initial FRET states. The transitionrates from the initial to high FRET states depends on protein concentration(gray balls), while the transition rates from high to initial FRET state wereindependent of the protein concentration (black balls). (E) Model for confor-mational dynamics observed for MSH2/MSH3 binding to ðCAÞ4 loop sub-strate: MSH2/MSH3 binds to the a high FRET state of the ðCAÞ4 loop, whileaddition of ATP or ADP under hydrolytic conditions increases dissociation ofMSH2/MSH3 from the ðCAÞ4 loop. The high FRET state is indicated as a bentstructure.

Lang et al. PNAS ∣ October 18, 2011 ∣ vol. 108 ∣ no. 42 ∣ E839

BIOCH

EMISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Dec

embe

r 27

, 202

0

magnesium (Fig. S5) and nucleotide binding (Fig. 2A and Fig. 3 Aand B). ATP(þMg) binding weakened the affinity of MSH2/MSH3 for the ðCAÞ4 DNA (Table 2), and more nucleotide-boundMSH2/MSH3 was required to saturate the high FRET shift(Fig. 3B) relative to the absence of nucleotide (Fig. S5). However,ATP binding, under hydrolytic conditions, resulted in a strikingalteration in the dynamics of MSH2/MSH3 binding (compareFig. 2B and Fig. 3C). Multiple transitions between high FRETstates and low FRET states were obvious in the single moleculetraces, and the lifetime of nucleotide-bound MSH2/MSH3 on theðCAÞ4 loop dropped from minutes (Fig. 2B) to seconds (Fig. 3C).Similar results were obtained when ADP(þMg) was the addednucleotide (Fig. S6). Thus, MSH2/MSH3 binding was sufficientto stabilize the high FRETstate, and binding of ATPorADP underhydrolytic conditions increased dissociation of MSH2/MSH3 fromthe ðCAÞ4 loop.

To determine the binding and dissociation kinetics, we applieda hidden Markov model (36) to hundreds of time traces for sev-eral MSH2/MSH3 concentrations to generate robust measures ofthe average transition times (Fig. 3D). We found that the transi-tion rate to the high FRET state increased with protein concen-tration, but the transition rate back to the initial FRET state wasindependent of protein concentration. Collectively, these findingsindicated that the shift to the high FRET state depended onMSH2/MSH3 binding to the ðCAÞ4 loop, while the reverse tran-sition rate arose from MSH2/MSH3 dissociation (Fig. 3E).

Binding of MSH2/MSH3 to the Repair-Deficient ðCAGÞ13 HairpinResults in the Appearance of a Unique Conformational Population.Both the ðCAGÞ13 hairpin and the ðCAÞ4 loops bind well toMSH2/MSH3 (Table 2), but only the ðCAÞ4 loop is accurately ex-cised and repaired in vitro (24, 26) and in vivo (27, 28). Therefore,we tested whether the conformational dynamics of the ðCAGÞ13hairpin might be relevant to its repair-deficient nature.

The ðCAGÞ13 hairpin DNA (Fig. 4A) displayed a relativelybroad distribution of FRET efficiency, around EFRET ∼ 0.3(Fig. 4B, DNA only). Remarkably, binding of MSH2/MSH3 to theðCAGÞ13 hairpin resulted in two new FRET populations (Fig. 4B,þMSH2∕MSH3), one of which was similar to that observed fromthe ðCAÞ4 loop. The high and a low FRET distributions of theðCAGÞ13 hairpin, around EFRET ∼ 0.43 and EFRET ∼ 0.20, respec-tively, had the same dependence on MSH2/MSH3 concentration(Fig. 4B, þMSH2∕MSH3). Each FRET state was stable, withan average lifetime longer than 30 s (Fig. 4 C and D). Thus, incontrast to ðCAÞ4 loops, the MSH2/MSH3-bound ðCAGÞ13 hair-pin adopted two conformational populations in which and themajority of the ends (65%) had moved apart.

Surprisingly, the high FRETstate (35% of ends) largely disap-peared when ðCAGÞ13-bound MSH2/MSH3 was occupied withnucleotide (Fig. 4E, þMSH2∕MSH3). Under hydrolytic condi-tions, addition of ATP to ðCAGÞ13-bound MSH2/MSH3 shiftedthe equilibrium populations towards the low FRET state (com-pare Fig. 4 B and E). Analysis of the single molecule traces in-dicated that ATP(þMg) occupancy of MSH2/MSH3 significantlyshortened the average lifetime for the high FRETstate to around5–10 s (Fig. 4G, horizontal black line). Under the same condi-tions, the low FRET state was stable, and dissociation was rarelyobserved (Fig. 4F). Under hydrolytic conditions, the FRETefficiencies for ATP(þMg) were similar to those of ADP(þMg)(Fig. S7). Thus, the repair-deficient ðCAGÞ13 template differedfrom the repair-competent ðCAÞ4 loop: nucleotide occupancy ofMSH2/MSH3 promoted its dissociation from the high FRETstate and the nucleotide-bound MSH2/MSH3 was, instead,trapped in the low FRET state.

The differential dynamics between the repair-resistantðCAGÞ13 hairpin and the repair-competent ðCAÞ4 loop templateswere striking. The shift to the low FRET state could not be ex-plained by differential affinity of nucleotide-bound MSH2/MSH3: the biochemical data indicated that nucleotide bindingand DNA-binding affinity of MSH2/MSH3 to the ðCAGÞ13 hair-pin and the ðCAÞ4 loop were similar (Table 1). The differences inMSH2/MSH3-induced DNA conformational dynamics could alsonot be explained by oligomerization of MSH2/MSH3 on theDNA templates. We have previously reported that the stoichio-metry of MSH2/MSH3 on both the ðCAÞ4 loop and the ðCAGÞ13hairpin is one heterodimer per DNA molecule as measured bysedimentation equilibrium analysis (31). Thus, models in whichthe low and high FRET states were stabilized by two or moreMSH2/MSH3 heterodimers were unlikely.

However, two models seemed plausible. The low and highFRET states could arise if two distinct nucleotide-bound formsof MSH2/MSH3 were able to bind to the ðCAGÞ13 hairpin, andinduce distinct conformations. Alternatively, the low and highFRET states might arise if ðCAGÞ13 template itself formed twomajor DNA conformations that were able to bind MSH2/MSH3.In either case, the ratio of MSH2/MSH3 and ðCAGÞ13 templatewould be 1∶1. We considered both possibilities.

The High and Low FRET States of the ðCAGÞ13 Hairpin Loop Do NotArise from Binding of Distinct Nucleotide-Bound Forms of MSH2/MSH3. Different from MSH2/MSH6, the MSH2 and MSH3 sub-units of MSH2/MSH3 bind nucleotides stochastically (26), andefficient hydrolysis results in formation of ADP-MSH2/MSH3-empty and empty-MSH2/MSH3-ADP in solution. Only ADP-MSH2/MSH3-empty stably binds to the ðCAÞ4 loop DNA (26).However, it was possible that the altered conformation of theðCAGÞ13 hairpin template permitted binding of both ADP-boundforms of MSH2/MSH3 (Fig. 5A). In such a model, binding of thetwo distinct ADP-bound forms of MSH2/MSH3 to the ðCAGÞ13templates would result in the high and low FRET states.

To test this hypothesis, we created mutants of MSH2/MSH3in which only one of the Walker motifs was competent to bind

Fig. 4. Repair-resistant ðCAGÞ13 template traps nucleotide-bound MSH2/MSH3 in the low FRET state. (A) A fold back loop of CAG DNA formswith A/A mispaired base every three nucleotides. Green ball is Cy3 label; Redball is Cy5 label (bottom). (B) FRET efficiency for MSH2/MSH3 binding tothe ðCAGÞ13 hairpin in the absence of nucleotides. The FRET efficiency histo-grams indicate that MSH2/MSH3 binding induces a high FRET state and alow FRET state relative to the substrate alone. Individual time traces ofthe low (C) and high (D) FRET states. The time traces of representative donorfluorescence (green, Cy3) and acceptor fluorescence (red, Cy5). Black barsindicate the binding event, and the arrows indicate photo bleaching ofthe acceptor dye. (E) Addition of ATP to (B) strongly reduces the relativeabundance of the high FRET state compared to the low FRET state. (F) Theindividual time traces indicate that the low FRET state remains stable and(G) the time in the high FRET state is shorter in the presence of ATP relativeto its absence (D).

E840 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1105461108 Lang et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 27

, 202

0

nucleotides (Fig. 5B). We changed the critical lysine of the Walk-er A sites (GGKST/S) to a methionine in one, the other, or bothof the ATP binding sites. These mutants are referred to as sgl2(mutation in MSH2 only), sgl3 (mutation in MSH3 only), or dbl(both subunits mutated) (Fig. 5B), depending on the site(s) of theamino acid change. The amino acid substitutions had no effect onthe expression of the protein relative to theWT protein, and eachsubunit was expressed at stoichiometric levels (Fig. 5C). Thus, wepurified each mutant MSH2/MSH3 heterodimeric complex andcharacterized its behavior with respect to DNA and nucleotidebinding.

The mutant MSH2/MSH3 proteins had the expected nucleo-tide binding properties. As judged by X-linking, neither ½α-32P�-ADP nor ½α-32P�-ATPðþMgÞ bound to the MSH2/MSH3 dbl(Fig. 5D, lanes 4), while sgl2 and sgl3 bound nucleotides only intheir intact site (Fig. 5D, lanes 2 and 3) and wt MSH2/MSH3bound both sites equally (Fig. 5D, lanes 1), as measured by ani-

sotropy, both sgl2 and sgl3 bound ADP with equivalent affinityas WT MSH2/MSH3 (Table 3). Thus, mutation in one site didnot influence the nucleotide affinity in the other. Consequently,each mutant MSH2/MSH3 heterodimer was able to form a singlenucleotide-bound complex, which varied only in the nucleotide-bound subunit.

We next tested how well the mutant MSH2/MSH3 proteinscould bind to DNA, by measuring FA of labeled DNA substrates(Table 4). With the exception of dbl, wt and mutant MSH2/MSH3had good affinity for the ðCAGÞ13 hairpin in the presence of nu-cleotide (Table 4). We measured nucleotide affinity for the DNA-bound wt and MSH2/MSH3 mutants using fluorescently labeledATP and ADP (Fig. 5 E and F). ATP bound well to the MSH2 orMSH3 subunit of wt, sgl2, or sgl3 as free heterodimers (Fig. 5E,open symbols), but none of these ATP-bound complexes adopteda stable ATP-bound state on the ðCAGÞ13 hairpin DNA (Fig. 5E,closed symbols). ADPðþMgÞ retained high affinity for ðCAGÞ13-bound wt and sgl3, but had little affinity for the ðCAGÞ13-boundsgl2 mutant (Fig. 5F, solid inverted triangles). Thus, nucleotide-bound MSH2/MSH3 associated to the ðCAGÞ13 hairpin onlywhen nucleotide occupied the MSH2 subunit and the MSH3 sub-unit was empty (sgl3), yet both high and low FRET states wereobserved (Fig. 5G). These experiments argued against a modelwhere the high and low FRET states arose from binding oftwo distinct nucleotide-bound MSH2/MSH3 complexes to theðCAGÞ13 hairpin.

The Repair-Resistant ðCAGÞ13 Template Traps MSH2/MSH3 in the LowFRET State. The experimental results raised the possibility that theðCAGÞ13 template adopted more than one DNA conformationfor binding of ADP-MSH2/MSH3-empty. Indeed, in the absence

Fig. 5. A ðCAGÞ13 hairpin binds only one nucleotide-bound MSH2/MSH3complex, but displays both high and low FRET states. (A) Schematic diagramof two possible MSH2∕MSH3-ðCAGÞ13 hairpin complexes with nucleotidebound in either the MSH2 or MSH3 subunit. (B) Sequences of the mutantMSH2 and MSH3 subunits aligned with the canonical Walker A box sequencemotif of the wt MSH2/MSH3. The conserved Lysine residue has been replacedwith a methionine in the mutant proteins to destroy the ATP binding pocket.(C) Resolution of the wild-type and mutant MSH2/MSH3 on denaturing gels.Wt, wild-type MSH2/MSH3 subunits, sgl2, Walker A box mutations in theMSH2 subunit only, sgl3, Walker A box in the MSH3 subunit only, dbl, WalkerA motif mutations in both subunits. (D) Binding of ½α-32P�-ATP to wild-typeand mutant MSH2/MSH3 proteins analyzed by UV-cross-linking followed byresolution on denaturing gel. Only intact nucleotide binding sites bind ATPefficiently. (E) Fluorescence anisotropy measurements of Bodipy-labeled ATPbinding to both wild-type and mutant MSH2∕MSH3-ðCAGÞ13 hairpin com-plexes. (F) Mutation of the Walker (A) box in the MSH2 subunit only inhibitsbinding of Bodipy labeled ADP to a MSH2∕MSH3-ðCAGÞ13 hairpin complex.(G) Association of nucleotide-bound wild-type and mutant MSH2/MSH3 tothe ðCAGÞ13 templates result in high and low FRET states. ATP is retainedpoorly in the MSH3 subunit when bound to DNA (18). Thus, sgl3 and wildtype in the presence of nucleotides are similar, and sgl2 in the presence ofnucleotides is the same as wt without bound nucleotides. ATP is 100 μM.

Table 3. Nucleotide binding affinities of wild-type and mutantMSH2/MSH3 proteins determined by Fluorescence Anisotropy,KD in nM

Ligand WT Sgl2 Sgl3 Dbl

ADP(−Mg) 217 ±12 1,124 ± 164 354 ± 19 >1,300ADP(+Mg) 53.9 ± 3.5 60.8 ± 5.3 110.3 ± 4.1 > 1,200ATP(−Mg) 409 ± 42 373 ± 39 415 ± 65 > 2,700ATP(+Mg) 70.2 ± 5.1 37.9 ± 2.9 212.7 ± 2.0 > 3,000

Table 4. DNA-binding affinities of mutant MSH2/MSH3 proteinsdetermined by Fluorescence Anisotropy, KD in nM

Template CA4 CAG13

Mutant Sgl2(−Mg) 6.3 ± 1.1 4.8 ± 0.6(+Mg) 3.7 ± 0.5 2.8 ± 0.4ADP(−Mg) 4.1 ± 0.6 2.8 ± 0.04ADP(+Mg) 12.1 ± 1.2 20.1 ± 4.6ATP(−Mg) 13.5 ± 2.9 3.9 ± 0.7ATP(+Mg) 45.9 ± 4.5 50.0 ± 3.4

Mutant Sgl3(−Mg) 4.1 ± 0.7 2.3 ± 0.4(+Mg) 3.2 ± 0.4 2.4 ± 0.4ADP(−Mg) 3.0 ± 0.5 3.8 ± 1.0ADP(+Mg) 2.3 ± 0.4 3.4 ± 0.5ATP(−Mg) 14.4 ± 1.0 5.1 ± 0.7ATP(+Mg) 43.2 ± 4.4 59.0 ± 6.4

Double Mutant(−Mg) 16.0 ± 2.2 8.3 ± 1.3(+Mg) 28.0 ± 2.5 33.5 ± 12.9ADP(−Mg) 15.7 ± 1.9 2.8 ± 0.04ADP(+Mg) 118.2 ± 16.1 >5,600ATP(−Mg) 18.4 ± 5.6 No bindingATP(+Mg) No binding No binding

Lang et al. PNAS ∣ October 18, 2011 ∣ vol. 108 ∣ no. 42 ∣ E841

BIOCH

EMISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Dec

embe

r 27

, 202

0

of protein, by increasing MgCl2 concentrations, we could resolvethe broad FRETefficiency peak at EFRET ∼ 0.24 into two closelyspaced DNA populations around EFRET ∼ 0.24 and EFRET ∼ 0.21(Fig. 6A). The single molecule traces indicated that these twoDNA populations were rapidly interconverting (Fig. 6 B and C).The ðCAGÞ13 DNA can be characterized as a three-way DNAjunction with two homoduplex and one heteroduplex arm (theðCAGÞ13 stem). Perfectly paired three-way DNA junctions forma single stable conformation (37). Thus, we hypothesized thatthe unpaired A-A mispaired bases in the stem of a ðCAGÞ7 orðCAGÞ13 loops might allow rearrangement of the junction intotwo major conformational populations of the DNA. If theðCAGÞ13 DNA intrinsically adopted high and low FRET states,then MSH2/MSH3 might preferentially bind to one.

To test this idea, we stabilized the junction by converting thetwo A-A pairs closest to the junction of the ðCAGÞ13 hairpin tem-plate into A-T pairs (AT-ðCAGÞ9) (Fig. 6D). Introduction of thetwo A-T pairs at the base of the junction “locked” it into a singlenarrow distribution, which did not show two populations uponincreasing MgCl2 (Fig. 6E). Moreover, AT-ðCAGÞ9 adopted asingle stable state as shown in the single molecule traces (Fig. 6 Fand G). We next added MSH2/MSH3 to the AT-ðCAGÞ9, DNA,and tested whether MSH2/MSH3 would promote the high andlow FRET conformations. Remarkably, binding of MSH2/MSH3to the AT-ðCAGÞ9, resulted in only a low FRET conformationalpopulation (Fig. 7A). AddingATPandMSH2/MSH3, under hydro-lyzing conditions, lowered the affinity of MSH2/MSH3 to theAT-ðCAGÞ9 hairpin (and reduced the shift) (Fig. 7B) but did notalter the overall conformation of the ðCAGÞ13 hairpins. Further-more, MSH2-MSH3 binding did not increase the dynamics ofAT-ðCAGÞ9; few conformation transitions were observed evenwhen MSH2/MSH3 was in the nucleotide-bound state (Fig. 7 Cand D). Thus, MSH2/MSH3 bound stably to the AT-ðCAGÞ9junction and did not dissociate readily from the low FRETconfor-mation. When bound to MSH2/MSH3, the AT-ðCAGÞ9 hairpin

adopted a single stable low FRETstate, which was not observed forthe ðCAÞ4 loop template under any condition tested.

DiscussionHow insertion and deletion mutations arise in the genome andwhy some loops are repaired better than others are unknown.Here, we show that MSH2/MSH3 discriminates between a repair-competent and a repair-resistant loop by sensing the confor-mational dynamics of their three-way junctions. We propose thatthe conformational properties of the substrate junction governwhether a loop is removed or becomes precursor for mutation.We find the repair-competent ðCAÞ4 substrate is intrinsically aflexible hinge with dynamics that are faster than our time resolu-tion. MSH2/MSH3 binds and stabilizes the bent state (Fig. 7E,bending), presumably to verify the lesion. Upon nucleotide bind-ing, the enzyme undergoes a series of rapid nucleotide-dependentsteps and eventually dissociates to signal downstream repair(Fig. 7E, sliding). Indeed, the smFRET results imply that thesubstrate dynamics induced by nucleotide-bound MSH2/MSH3at the ðCAÞ4 loop have a nonexponential dwell-time distributionconsistent with the presence of more than one kinetic step(Fig. S8). Rapid association and dissociation poises the proteincomplex to verify and move away from the lesion and to initiateinteractions necessary for downstream signaling.

Fig. 6. The ðCAGÞ13 loop intrinsically adopts more than one conformationalstate in the absence of protein. (A) The FRET efficiency histograms of theðCAGÞ13 substrate in the absence of protein. Increasing MgCl2 concentrationresolves the presence of two FRET states. (B) Few transitions are observedwithin the time resolution for concentrations below 5 mM MgCl2, but be-come more apparent at 20 mM MgCl2. (C) Blue trace is the correspondingFRET efficiency. (D) A schematic of the sequence changes at the junctionof the ATðCAGÞ9 template. The two A-A mismatches closest to the junctionin the ðCAGÞ13 substrate have been replaced by A-T pairs. (E) Conformation ofAT-ðCAGÞ9 substrate at different MgCl2 concentrations. In the absence ofprotein, the substrate exists in a single conformation. Addition of MgCl2 in-creases the shift towards higher FRET values, (F, G) but the individual timetraces are not dynamic. (F) The time traces of representative donor fluores-cence (green, Cy3) and acceptor fluorescence (red, Cy5), and (G) the bluetraces are the corresponding efficiencies.

Fig. 7. The junction of the AT-ðCAGÞ9 hairpin adopts only one stable three-way junction, from which MSH2/MSH3 does not dissociate. (A) The FRETefficiencies for binding of MSH2/MSH3 to the AT-ðCAGÞ9 substrate in thepresence of (þMgCl2) results in a single low-FRET population. (B) The FRETefficiency histograms for AT-ðCAGÞ9 in the presence of ATP under hydrolyticconditions. The affinity of nucleotide-bound MSH2/MSH3 for the AT-ðCAGÞ9substrate (þMgCl2) is reduced, but binding results in the same low FRETstate as observed in the absence of nucleotides. (C) The individual time tracesof the AT-ðCAGÞ9 substrate alone (top), AT-ðCAGÞ9 bound to MSH2/MSH3(middle), and AT-ðCAGÞ9 bound to MSH2/MSH3 at the indicated concentra-tions in the presence of ATP (þMg) (100 μM) (bottom). All traces are similarand display no dynamics. (C) The time traces of representative donor fluor-escence (green, Cy3) and acceptor fluorescence (red, Cy5), and (D) the bluetraces are the corresponding FRET efficiencies. (E) Proposed model for con-formational regulation of loop repair byMSH2/MSH3 at three-way DNA junc-tions. The conformational flexibility of the substrate determines the possiblebinding modes of MSH2/MSH3. (F) Binding of MSH2/MSH3 to a ðCAÞ4 loopbinds, bends DNA. Upon downstream nucleotide hydrolysis and exchange,MSH2/MSH3 adopts a doubly bound form which is verify and to leave thelesion to signal downstream repair by the MMR machinery. (G) The straigh-tened ðCAGÞ13 hairpin junction traps nucleotide-bound MSH2/MSH3 in anonproductive complex, which cannot leave the lesion to initiate efficientrepair by theMMR pathway. Successful mismatch repair couples DNA bindingand ATP hydrolysis. Trapping does not allow processing of ATP in the MSH3subunit, and prevent ADP/ATP exchange needed to leave the site. Circleswith 2 and 3 represent the MSH2/MSH3 heterodimer. Red ball are ADP andblue balls are ATP.

E842 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1105461108 Lang et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 27

, 202

0

The repair-resistant ðCAGÞ13 junction intrinsically adopts dis-crete conformational states as indicated by the two-state-FRETdistribution (most noticeable at high MgCl2 concentrations)(Figs. 4 and 6). Unliganded MSH2/MSH3 recognizes both ofthese conformational states with similar affinity and furtherseparates them. MSH2/MSH3 can convert some of the hairpinjunctions into a repair-competent bent state. However, uponnucleotide binding, MSH2/MSH3 dissociates from the bent stateand, instead is trapped by a junction configuration from whichit cannot dissociate (Fig. 7F, trapped). The nucleotide-boundprotein becomes “stuck” on the lesion, and likely cannot carry outthe steps leading to ADP/ATP exchange, which is critical for dis-sociation and downstream repair. These findings imply that therepair-resistant CAG hairpins provide a unique but nonproduc-tive binding site for nucleotide-bound MSH2/MSH3, which failsto effectively couple DNA binding with ATP hydrolysis.

The AT-ðCAGÞ9, junction differs by only two nucleotidesrelative to the ðCAGÞ13 hairpin, but only one intrinsic conforma-tion is available for MSH2/MSH3 binding. Similar to the repair-resistant ðCAGÞ13 hairpin loop, MSH2/MSH3 cannot convert theAT-ðCAGÞ9 into a bent state, rather, the template exists in asingle junction conformation, which captures MSH2/MSH3.The residence of MSH2/MSH3 on the AT-ðCAGÞ9, is long lived,whether or not the protein complex is bound with nucleotides(Fig. 7 C and D). Thus, dynamics of the junction is an activeparticipant in directing loop conformation. We envision that con-formational regulation of small loop repair occurs at the level ofthe junction dynamics.

This mechanism has strong mechanistic implications for asecond class of mismatch repair deficits. Mutations in the MMR(Mismatch Repair) machinery lead to an increase in spontaneousmutation rate, which is typically referred to as a mutator pheno-type (1–3). For example, about 15% of patients with hereditarynonpolyposis colorectal cancer have widespread genome instabil-ity, characterized by single base changes or changes in copy num-ber at repetitive tracts (1–3). The mutational spectrum in thisclass of MMR deficits reflects the inability of the mutated MMRmachinery to correct postreplicative errors throughout the gen-ome (1–3). Our data provide a plausible mechanism for a secondclass of MMR defects in which the lesion itself prevents itsprocessing by the normal repair machinery (32). Defective repairarises when the repair-resistant loops trap the MMR proteinsduring recognition of the lesion and they remain uncorrected.The resulting insertion and deletion mutations, in this case, willbe “site-specific” in that they are limited to particular locationswhere the repair-resistant lesions reside. The properties of trinu-cleotide expansion characterize this type of mutation.

We do not as yet know whether the unusual junction dynamicsprovides a general mechanism underlying all “class two” insertion/deletion mutations or whether the unusual dynamics are restrictedto only some junctions. However, our results provide, at the struc-tural level, a glimpse into why some loops are recognized differ-ently by MSH2/MSH3 and imply that the junction dynamics is atleast one component in a complex process that leads to mutation.

Integration of our biochemical and smFRET data clarifies twokey issues bearing on the expansion mutation. First, the role ofMSH2/MSH3 ATP hydrolysis activity in causing expansion hasbeen unclear. A G674A Walker A site mutation in the MSH2subunit suppresses CTG (Cytosine-thymine-guanine) expansionin mice (38), and prevents GAA (Guanine-adenine-adenine) de-letion in yeast (39), implying that ATP hydrolysis in the MSH2subunit is a requisite step in expansion. However, we observe in

our biochemical measurements that the G674A Walker A sitemutant in the MSH2 subunit binds ATP poorly, if at all, in thecontext of MSH2/MSH3 (Fig. S9). Thus, the G674A Walker Asite mutation does not block hydrolysis per se, but failure to bindATP in the MSH2 subunit prevents formation of ADP-boundMSH2/MSH3, the major lesion-binding form (26).

While ATP hydrolysis is reduced in MSH2/MSH3-bound CAGhairpin (31, 40), the apparent nucleotide affinity and the kcat∕KMfor ATP hydrolysis are similar for MSH2/MSH3 when bound to arepair-competent ðCAÞ4 loop and the repair-resistant ðCAGÞ13hairpin (31, 40). Thus, a second issue is the extent to which therecognition properties of MSH2/MSH3 differ between these twotypes of loops. smFRETresolves discrete populations, and our dataprovide definitive evidence that MSH2/MSH3 captures a distinctconformation of the ðCAGÞ13 hairpin, which significantly length-ens the lifetime of bound protein relative to repair-competentðCAÞ4 loop. Because MSH2/MSH3 binds with equal apparentaffinity to the ðCAÞ4 and the CAG hairpin templates, the kcat∕KMis expected to be similar, but the altered recognition propertiesof MSH2/MSH3 on the low FRET population cannot be resolvedin bulk measurements (40). The time scale of the changes requiressensitive, high-resolution techniques to observe them. BecauseDNA binding inhibits ATP hydrolysis for MSH2/MSH6 (41, 42)and MSH2/MSH3 (31, 40), the longer lifetime of the MSH2/MSH3 on the repair-resistant template implies a reduction of ATPbinding and/or hydrolytic activity in the straightened conformation(31). Collectively, our proposed model provides a basis for howan intact MMR complex can become inefficient when bound toparticular types of loops. The junction dynamics are poised to bea pivot point for coupling DNA loop binding and ATP hydrolysisby an intact MSH2/MSH3 to outcomes of mutation or repair.

MethodsDetailed methods are provided as SI Methods.

Protein Purification. His-tagged human MSH2/MSH3 and MSH2/MSH3 wasoverexpressed in SF9 insect cells using a pFastBac dual expression system(GIBCO-BRL) and purified as described previously (24, 26).

Nucleotides and Oligonucleotides. Oligonucleotides used in the bindingstudies were obtained from Operon or IDT, SI Methods. Fluorescently labeledoligos were labeled at the 5′ end with fluoresceine for single label experi-ments. Nucleotides of the highest grade were purchased from Sigma.½α32P�-ATP was purchased from Perkin Elmer, and ½α32P�-ADP was derivedby incubation of ½α32P�-ATP with hexokinase. All preparations of nucleo-tides used contained less than 1% contamination of other nucleotides.

UV-Cross-Linking, DNA, Nucleotide Binding, Fluorescence Anisotropy. Experi-ments were performed as described previously (26).

SAXS. Data were collected at the SIBYLS (beamline 12.3.1), and analyzed withthe automated pipeline described previously (34). The Atsas program suite(35) and GASBOR (43) were used to extract shapes from SAXS scatteringcurves (see SI Methods for more details).

Single Molecule FRET. The single molecule FRET experiments were performedon a prism-type total internal reflection microscope which features 532 nmexcitation from a Nd:YAG laser (50 mW, CrystaLaser), as previously described(44, 45).

ACKNOWLEDGMENTS. This work was supported by the Mayo Foundation,the National Institutes of Health Grants NS40738 (C.T.M.), GM066359(C.T.M.), NS062384 (to C.T.M.), and CA092584 (C.T.M.), NS060115 (to C.T.M.)and National Science Foundation PHY-0748642 (I.R.).

1. Fishel R (2001) The selection for mismatch repair defects in hereditary nonpolyposis

colorectal cancer: revising the mutator hypothesis. Cancer Res 61:7369–7374.

2. Peltomaki P (2001) Deficient DNA mismatch repair: a common etiologic factor for

colon cancer. Hum Mol Genet 10:735–740.

3. Wei K, Kucherlapati R, Edelmann W (2002) Mouse models for human DNA mismatch-

repair gene defects. Trends Mol Med 8:346–353.

4. Wang Y-H, Griffith J (1991) The effects of bulge composition and flanking sequence

on the kinking of DNA by bulged bases. Biochemistry 30:1359–1363.

5. Wang Y-H, Barker P, Griffith JD (1992) Visualization of diagnostic heteroduplex

DNAs from cystic fibrosis deletion heterozygotes provides an estimate of the kinking

of DNA by bulged bases. J Biol Chem 267:4911–4915.

Lang et al. PNAS ∣ October 18, 2011 ∣ vol. 108 ∣ no. 42 ∣ E843

BIOCH

EMISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Dec

embe

r 27

, 202

0

6. Lee S, Elenbaas B, Levine A, Griffith JD (1995) p53 and its 14 kDa C terminal domainrecognize primary DNA damage in the form of insertion/deletion mismatches. Cell81:1013–1020.

7. Degtyareva N, Subramanian D, Griffith JD (2001) Analysis of the binding of p53 toDNAs containing mismatched and bulged bases. J Biol Chem 276:8778–8784.

8. Wang Y-H, Bortner C, Griffith JD (1993) RecA binding to bulge and mismatch contain-ing DNA: certain single base mismatches provide signals for RecA binding equal tomultiple base bulges. J Biol Chem 268:17571–17577.

9. Subramanian D, Griffith JD (2002) Interactions between p53, hMSH2-hMSH6 and HMGI(Y) on Holliday Junctions and Bulged Bases. Nucleic Acids Res 30:2427–2434.

10. Fishel R, Ewel A, Lee S, Lescoe MK, Griffith JD (1994) Binding of mismatches microsa-tellite DNA sequences by the human MSH2 protein. Science 266:1403–1405.

11. Gradia S, et al. (1999) hMSH2-hMSH6 forms a hydrolysis-independent sliding clamp onmismatched DNA. Mol Cell 3:255–261.

12. Iyer RR, Pluciennik A, Burdett V, Modrich PL (2006) DNA mismatch repair: functionsand mechanisms. Chem Rev 106:302–323.

13. Kunkel TA, Erie DA (2005) DNA mismatch repair. Annu Rev Biochem 74:681–710.14. Schofield MJ, Hsieh P (2003) DNA mismatch repair: molecular mechanisms and

biological function. Annu Rev Microbiol 57:579–608.15. Obmolova G, Ban C, Hsieh P, Yang W (2000) Crystal structures of mismatch repair

protein MutS and its complex with a substrate DNA. Nature 407:703–710.16. Lamers MH, et al. (2000) The crystal structure of DNA mismatch repair protein MutS

binding to a GxT mismatch. Nature 407:711–717.17. Alani E, et al. (2003) Crystal structure and biochemical analysis of the MutS.ADP.

beryllium fluoride complex suggests a conserved mechanism for ATP interactions inmismatch repair. J Biol Chem 278:16088–16094.

18. Warren JJ, et al. (2007) Structure of the human MutSalpha DNA lesion recognitioncomplex. Mol Cell 26:579–592.

19. Tessmer I, et al. (2008) Mechanism of MutS searching for DNA mismatches and signal-ing repair. J Biol Chem 283:36646–36654.

20. Sass LE, Lanyi C, Weninger K, Erie DA (2010) Single-molecule FRET TACKLE revealshighly dynamic mismatched DNA-MutS complexes. Biochemistry 49:3174–3190.

21. Harrington JM, Kolodner RD (2007) Saccharomyces cerevisiaeMsh2-Msh3 acts in repairof base-base mispairs. Mol Cell Biol 27:6546–6554.

22. Palombo F, et al. (1996) hMutSbeta, a heterodimer of hMSH2 and hMSH3, binds toinsertion/deletion loops in DNA. Curr Biol 6:1181–1184.

23. Acharya S, et al. (1996) hMSH2 forms specific mispair-binding complexes with hMSH3and hMSH6. Proc Natl Acad Sci USA 93:13629–13634.

24. Wilson T, Guerrette S, Fishel R (1999) Dissociation of mismatch recognition and ATPaseactivity by hMSH2-hMSH3. J Biol Chem 274:21659–21664.

25. Bill CA, Taghian DG, Duran WA, Nickoloff JA (2001) Repair bias of large loopmismatches during recombination in mammalian cells depends on loop length andstructure. Mutat Res 485:255–265.

26. Owen BA, Lang WH, McMurray CT (2009) The nucleotide binding dynamics of humanMSH2-MSH3 are lesion dependent. Nat Struct Mol Biol 16:550–557.

27. Genschel J, Littman SJ, Drummond JT, Modrich P (1998) Isolation of MutSbeta fromhuman cells and comparison of the mismatch repair specificities of MutSbeta andMutSalpha. J Biol Chem 273:19895–19901.

28. Marsischky GT, Filosi N, Kane MF, Kolodner RD (1996) Redundancy of Saccharomycescerevisiae MSH3 and MSH6 in MSH2-dependent mismatch repair. Genes Dev10:407–420.

29. Miret JJ, Pessoa-Brandao L, Lahue RS (1997) Instability of CAG and CTG trinucleotiderepeats in Saccharomyces cerevisiae. Mol Cell Biol 17:3382–3387.

30. McMurray CT (2010) Trinucelotide repeat instability during human devlopment. NatRev Genet 11:786–799.

31. Owen BA, et al. (2005) (CAG)(n)-hairpin DNA binds to Msh2-Msh3 and changesproperties of mismatch recognition. Nat Struct Mol Biol 12:663–670.

32. McMurray CT (2008) Hijacking of the mismatch repair system to cause CAG expansionand cell death in neurodegenerative disease. DNA Repair 7:1121–1134.

33. Brouwer JR, Willemsen R, Oostra BA (2009) Microsatellite repeat instability andneurological disease. Bioessays 31:71–83.28.

34. Hura GL, et al. (2009) Robust, high-throughput solution structural analyses by smallangle X-ray scattering (SAXS). Nat Methods 6:606–612.

35. Konarev PV, Petoukhov MV, Volkov VV, Svergun DI (2006) ATSAS 2.1, a programpackage for small-angle scattering data analysis. J Appl Crystallogr 39:277–286.

36. McKinney SA, Joo C, Ha T (2006) Analysis of single-molecule FRET trajectories usinghidden Markov modeling. Biophys J 91:1941–1951.

37. Duckett DR, Lilley DM (1990) The three-way DNA junction is a Y-shaped molecule inwhich there is no helix-helix stacking. EMBO J 9:1659–1664.

38. Tome S, et al. (2009) MSH2 ATPase domain mutation affects CTG*CAG repeat instabil-ity in transgenic mice. PLoS Genet 5:e1000482.

39. Kim H-M, et al. (2008) Chromosome fragility at GAA tracts in yeast depends on repeatorientation and requires mismatch repair. EMBO J 27:2896–2906.

40. Tian L, et al. (2009) Mismatch recognition protein MutSbeta does not hijack (CAG)nhairpin repair in vitro. J Biol Chem 284:20452–20456.

41. Mazur DJ, Mendillo ML, Kolodner RD (2006) Inhibition of Msh6 ATPase activity bymispaired DNA induces a Msh2(ATP)-Msh6(ATP) state capable of hydrolysis-indepen-dent movement along DNA. Mol Cell 22:39–49.

42. Antony E, Hingorani MM (2003) Mismatch recognition-coupled stabilization ofMsh2-Msh6 in an ATP-bound state at the initiation of DNA repair. Biochemistry42:7682–7693.

43. Svergun DI, Petoukhov MV, Koch MH (2001) Determination of domain structure ofproteins from X-ray solution scattering. Biophys J 80:2946–2953.

44. Rasnik I, McKinney SA, Ha T (2006) Nonblinking and long-lasting single-moleculefluorescence imaging. Nat Methods 11:891–893.

45. Myong S, Rasnik I, Joo C, Lohman TM, Ha T (2005) Repetitive shuttling of a motorprotein on DNA. Nature 437:1321–1325.

E844 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1105461108 Lang et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 27

, 202

0