Mismatch Repair
A single E. coli cell replicates 4.6 million base pairs. This happens roughly once every 20-40 minutes. DNA polymerase III makes an error only about once every 10^7 nucleotides it adds. That is already remarkable fidelity, but it is not good enough. Left uncorrected, that error rate would produce hundreds of new mutations every generation. Something must scan the freshly made DNA. It must find the rare mistake and fix it after the replication fork has already passed. That "something" is mismatch repair (MMR), a proofreading-of-last-resort system. It raises overall replication fidelity by another 100- to 1,000-fold. The deep puzzle MMR had to solve is not detection. A mismatched base pair distorts the double helix in a way proteins can sense. The real challenge is discrimination. When a repair enzyme finds a G•T mispair, both the G and the T are normal bases. Nothing about the mismatch itself says which one is the newly made error. Nor does it reveal which is the original, correct template strand. Fixing the wrong strand would fix nothing. It would simply make the mutation permanent. MMR is therefore as much a story about how cells mark strand identity as it is about how they recognize errors. This system matters far beyond bacterial genetics. In humans, inherited loss-of-function mutations in MMR genes cause Lynch syndrome. It is one of the most common hereditary cancer predispositions. Somatic loss of MMR function also drives a large fraction of sporadic colorectal and endometrial cancers. The same defect also determines whether a tumor responds to modern immunotherapy. This makes MMR status a frontline clinical biomarker today. This article works through the molecular logic of MMR. It covers how mismatches are found, how the new strand is distinguished from the template, and how the error is excised and correctly resynthesized. It also reviews the classical genetic and biochemical experiments that established the mechanism. These include Meselson-era genetics, in vitro reconstitution by Modrich's lab, and the human genetics of Lynch syndrome.
- Mismatch repair (MMR) corrects base-pair mismatches and small insertion/deletion loops left behind by DNA polymerase after replication, increasing fidelity ~100-1,000-fold.
- In E. coli, MutS recognizes mismatches, MutL coordinates the response, and MutH cleaves the unmethylated (new) strand at a hemimethylated GATC site — this is how the new strand is identified.
- Eukaryotes lack MutH; strand discrimination instead uses pre-existing nicks in the lagging strand and the PCNA clamp, which is loaded with a defined orientation marking the newly synthesized strand.
- MMR proteins are evolutionarily conserved: bacterial MutS/MutL correspond to eukaryotic MSH and MLH/PMS heterodimer families.
- Loss of MMR causes microsatellite instability (MSI) — a mutator phenotype detectable clinically and central to Lynch syndrome and many sporadic cancers.
- MSI/MMR-deficient tumors accumulate neoantigens and respond unusually well to immune checkpoint inhibitors, making MMR status a key precision-oncology biomarker.
Mismatch repair fixes replication errors after the fact by using strand-specific signals — methylation state in bacteria, nicks and clamp polarity in eukaryotes — to tell the correct template strand from the erroneous new one, and then excises and resynthesizes only the new strand.
The core idea
Mismatch repair is a post-replicative surveillance system that detects base-base mismatches (e.g., G•T, A•C) and small insertion-deletion loops (IDLs) generated when DNA polymerase misincorporates a nucleotide or slips on a repetitive template, and corrects them before they become fixed mutations at the next round of replication. This is conceptually distinct from proofreading, which is the 3′→5′ exonuclease activity built into replicative polymerases that removes a misincorporated nucleotide within seconds of its addition, before the polymerase has even moved past it. MMR is the second, independent layer of quality control: it operates after the fork has moved on, over stretches of newly synthesized DNA that may extend for hundreds to thousands of base pairs behind the fork. The central conceptual problem MMR solves is strand discrimination. A mismatch is symmetric with respect to sequence — a G•T mispair looks identical whether you approach it "from the G side" or "from the T side" — so a repair enzyme cannot tell by local chemistry alone which base is the error. The system must therefore read some transient, strand-specific signal that exists only immediately after replication and that marks the daughter strand as "new" (and hence untrustworthy) relative to the parental template strand. In bacteria this signal is the methylation state of GATC sequences; in eukaryotes it is thought to be the nicks and free ends inherent to newly synthesized Okazaki fragments and leading-strand initiation sites, together with the directional loading of the sliding clamp PCNA. Once the new strand is identified, MMR excises a stretch of it spanning the mismatch and resynthesizes it using the (correct) template strand as a guide, restoring the original sequence. The whole pathway is modular: recognition (MutS/MSH family), signal-coupling and coordination (MutL/MLH-PMS family), strand discrimination and incision (MutH in bacteria; PCNA/exonuclease machinery in eukaryotes), excision (exonucleases), and resynthesis/ligation (DNA polymerase and DNA ligase). This modularity is conserved from bacteria to humans, reflecting a single ancient solution to the strand-discrimination problem, elaborated with different molecular signals in different lineages.
Mechanism
Step 1 — Mismatch recognition by MutS (bacteria) / MSH2-MSH6 or MSH2-MSH3 (eukaryotes). MutS is a homodimer (functionally) that clamps onto DNA and kinks it sharply (roughly 60°) at the site of a mismatch, inserting a conserved phenylalanine residue into the helix to probe the distortion. In eukaryotes the equivalent recognition complexes are heterodimers: MutSα (MSH2-MSH6) recognizes single base-base mismatches and 1-2 nucleotide IDLs, while MutSβ (MSH2-MSH3) recognizes larger IDLs (up to ~10-16 nucleotides), the kind generated by polymerase slippage on repetitive microsatellite sequences. MutS/MutSα is an ATPase; ATP binding drives a conformational change that converts MutS from a mismatch-scanning "searcher" into a sliding clamp that can diffuse along DNA away from the mismatch.
Step 2 — Recruitment of MutL / MLH1-PMS2. ATP-bound MutS recruits MutL (a homodimer in bacteria; the heterodimer MutLα = MLH1-PMS2 in humans). MutL is itself an ATPase and acts as a molecular matchmaker, coupling mismatch recognition to the downstream strand-discrimination and excision machinery. This step is where the "search" for the strand-discrimination signal, which may lie far from the mismatch itself, is licensed.
Step 3 — Strand discrimination. This is the step that differs most between bacteria and eukaryotes.
- In E. coli: Newly replicated DNA is transiently hemimethylated. The bacterial Dam methyltransferase adds a methyl group to the adenine in every GATC sequence, but only on the parental strand — the newly synthesized strand remains unmethylated for a window of a few minutes after replication (methylation lags behind the fork). MutL-activated MutH is an endonuclease that recognizes hemimethylated GATC sites and nicks specifically the unmethylated strand — i.e., the new strand — even if that GATC site is far (up to ~1 kb or more) from the mismatch. MutS and MutL, bound at the mismatch, physically loop the DNA to bring the mismatch and the distant GATC site together, licensing MutH's incision only on the correct (new) strand.
- In eukaryotes: There is no Dam-like adenine methylation system and no MutH homolog. Strand discrimination instead exploits the transient nicks intrinsically present in newly made DNA — the 5′ ends of Okazaki fragments on the lagging strand, and, it is thought, similar discontinuities associated with leading-strand synthesis. Biochemical reconstitution shows that a pre-existing nick anywhere on the new strand (even far from the mismatch) is sufficient to direct repair to that strand in vitro. The sliding clamp PCNA, which is loaded onto DNA with defined orientation at these ends, also contributes to marking strand polarity and helps recruit the excision machinery to the correct strand.
Step 4 — Excision. Once the new strand is marked (nicked), an exonuclease is loaded and degrades the DNA from the nick, past the mismatch, and some distance beyond, removing the erroneous segment. In E. coli this involves UvrD (helicase II) unwinding the duplex and one of several exonucleases (ExoI, ExoVII, ExoX, or RecJ, depending on which side of the mismatch the nick lies) degrading the single strand. In eukaryotes, Exonuclease 1 (EXO1) carries out most of the excision, generating a single-stranded gap that can extend from the nick to well past the mismatch, coated and protected by replication protein A (RPA).
Step 5 — Resynthesis and ligation. DNA polymerase (Pol III holoenzyme in bacteria; Pol δ, aided by PCNA, in eukaryotes) fills the resulting gap using the intact template strand, and DNA ligase seals the remaining nick, restoring a continuous, correctly paired duplex. Net effect: the erroneous base introduced during replication is removed and replaced with the correct one, dictated entirely by the original template strand — because the template strand, not the mismatch itself, is what MMR trusts.
The evidence
Genetic identification of the mutator genes (E. coli, 1960s-1970s). The first clues came from classical bacterial genetics: mutations in genes later named mutS, mutL, mutH, and mutU (=uvrD) were isolated as "mutator" strains — mutants with elevated spontaneous mutation rates across the genome, often 100- to 1,000-fold above wild type, without any change in overall DNA damage. This showed that these genes normally suppress mutagenesis rather than cause it — i.e., they encode a repair function, not a source of error. Crucially, mutator phenotypes clustered specific classes of mutation (base substitutions and small frameshifts at repeats), the signature later understood as the fingerprint of failed mismatch correction.
In vitro reconstitution and the hemimethylation experiment — Paul Modrich's laboratory (1980s-1990s). The decisive mechanistic proof came from biochemistry. Modrich and colleagues constructed heteroduplex DNA substrates — plasmids or bacteriophage DNA engineered to contain a single defined mismatch (e.g., a G•T) at a known position, with GATC sites methylated on one strand only (hemimethylated), mimicking the state immediately after replication. When such hemimethylated heteroduplexes were incubated with purified MutS, MutL, MutH, UvrD, an exonuclease, SSB, Pol III, and ligase, repair occurred and — this is the key result — it was strand-specific: the unmethylated strand was always the one repaired, regardless of which strand actually carried the "wrong" base. If the mismatch had been engineered so the unmethylated strand carried the correct sequence and the methylated (parental) strand carried the wrong one, the system still "corrected" the unmethylated strand, converting it to match the methylated strand — i.e., MMR does not know which base is biologically correct, only which strand is methylated. This directly proved that methylation, not sequence correctness, is the strand-discrimination signal, and that MMR blindly trusts the methylated strand. Fully methylated (symmetrically methylated) heteroduplexes were repaired inefficiently and without strand bias, confirming that hemimethylation, present only transiently after replication, is what licenses directional repair in vivo. This work (recognized in Modrich's share of the 2015 Nobel Prize in Chemistry, awarded jointly with Tomas Lindahl and Aziz Sancar for mechanistic studies of DNA repair) is one of the cleanest examples in molecular biology of reconstituting a complex genetic phenomenon entirely from purified components.
Human genetics — Lynch syndrome and microsatellite instability (1990s). Independently, human geneticists studying hereditary non-polyposis colorectal cancer (HNPCC, now called Lynch syndrome) found that tumors from affected families showed pervasive instability in short tandem repeat sequences (microsatellites) — lengths of mono-, di-, and trinucleotide repeats varied from cell to cell and differed from the patient's germline, a phenotype named microsatellite instability (MSI). Linkage analysis mapped the syndrome to loss-of-function mutations in human homologs of the bacterial mismatch repair genes — MSH2, MLH1, MSH6, PMS2 — proving, independently of the bacterial biochemistry, that a conserved mismatch repair pathway operates in humans and that its failure produces exactly the mutator phenotype (especially at repetitive sequences, the classic substrate for polymerase slippage) predicted from the bacterial mechanism. This convergence of bacterial biochemistry and human cancer genetics, arrived at via completely different experimental routes, is strong independent confirmation of the mechanism.
Worked examples
Example 1 — Predicting mutation-rate elevation from an MMR defect. Suppose wild-type E. coli has a proofread replication error rate of 1 in 10^7 nucleotides incorporated (after Pol III's intrinsic 3′→5′ exonuclease proofreading, but before MMR), and functional MMR further reduces the final, observed mutation rate to 1 in 10^9-10^10 nucleotides (a commonly cited combined fidelity figure). This implies MMR itself corrects roughly 99-99.9% of the mismatches that escape proofreading — i.e., MMR provides on the order of a 100- to 1,000-fold fidelity improvement on top of proofreading. Now consider a mutS⁻ strain growing with a genome of 4.6 × 10^6 bp. If MMR normally reduces the error rate from 10^-7 to 10^-9 per bp replicated, loss of MutS returns the rate to ~10^-7 per bp per replication round. Per genome replication:
- With MMR: 4.6 × 10^6 × 10^-9 ≈ 0.0046 new mutations/genome/generation.
- Without MMR (mutS⁻): 4.6 × 10^6 × 10^-7 ≈ 0.46 new mutations/genome/generation.
That is roughly a 100-fold increase in mutation accumulation per generation — consistent with the "mutator" phenotypes originally used to identify mutS, mutL, and mutH, and large enough to be readily detected by classical fluctuation-test-style mutation-rate assays (the same logic Luria and Delbrück used to measure spontaneous mutation rates, applied here to a repair-deficient background).
Example 2 — Reasoning through a Lynch syndrome pedigree and clinical decision. A 38-year-old patient presents with colorectal cancer; family history shows her father and paternal grandmother both had colorectal cancer before age 50, and an aunt had endometrial cancer at 45. This pattern — early-onset, right-sided colorectal cancer clustering across generations, plus an associated endometrial cancer — is classic for Lynch syndrome, caused by a heterozygous germline loss-of-function mutation in one of MLH1, MSH2, MSH6, or PMS2, inherited in an autosomal dominant pattern (one defective allele inherited from an affected parent; cancer arises when the second, wild-type allele is somatically lost in a colonic or endometrial cell — a "two-hit" mechanism analogous to classical tumor-suppressor genetics). Tumor testing shows high microsatellite instability (MSI-H): PCR analysis of several standard mononucleotide/dinucleotide repeat markers shows novel-length alleles in the tumor DNA not present in the patient's normal tissue, reflecting the expected mutator phenotype at repetitive sequences once MMR is inactivated in the tumor cell. Immunohistochemistry on the tumor shows loss of MSH2 and MSH6 protein staining (MSH6 is destabilized when its obligate heterodimer partner MSH2 is lost, since MutSα cannot form or persist without both subunits) while MLH1 and PMS2 stain normally — this pattern points specifically to an underlying MSH2 mutation (rather than MLH1, which would show loss of both MLH1 and its partner PMS2). Clinically, this MSI-H, MMR-deficient status has two major consequences: first, at-risk relatives should be offered germline genetic testing and, if positive, earlier and more frequent colonoscopic surveillance; second, because MMR-deficient tumors carry an unusually high mutational/neoantigen burden, this patient's tumor is a strong candidate for immune checkpoint inhibitor therapy (e.g., anti-PD-1) even in a metastatic setting where such tumors are often unusually responsive compared to MMR-proficient (microsatellite-stable) cancers.
Common misconceptions & traps
| Misconception | Why it's wrong |
|---|---|
| "Mismatch repair and proofreading are the same thing." | Proofreading is the polymerase's own 3′→5′ exonuclease, acting in real time during synthesis; MMR is a separate, later system that scans finished daughter DNA, often long after the fork has passed. |
| "MMR can tell which base is 'correct' by looking at the mismatch." | A mismatch itself carries no information about which base is right — MMR only knows which strand is new (via methylation or nicks) and always repairs that strand, even experimentally forced to "correct" toward the wrong sequence. |
| "All organisms use DNA methylation to mark the new strand." | Only bacteria with Dam methylation (like E. coli) use hemimethylated GATC; eukaryotes use strand discontinuities (nicks) and PCNA orientation instead — there is no known eukaryotic Dam-GATC equivalent. |
| "MMR mainly fixes large insertions or deletions from recombination." | MMR's main substrates are point mismatches and small (1-16 nt) insertion/deletion loops from polymerase slippage, especially at microsatellites — not large structural rearrangements, which are handled by other pathways. |
| "Losing one MMR gene copy has no effect until both copies are lost." | This is true at the cell/tumor level (classic two-hit tumor-suppressor logic) but heterozygous carriers already have sharply elevated lifetime cancer risk because any single somatic cell that loses the remaining wild-type allele becomes a fully MMR-deficient founder clone. |
The scientist behind it
Paul Modrich built his career on reconstructing a subtle, largely invisible genetic phenomenon entirely from purified proteins in a test tube — a style of biochemistry that turns genetics into a mechanism you can watch happen. Working at Duke University from the late 1970s onward, Modrich systematically purified MutS, MutL, MutH, UvrD, and the associated exonucleases and reconstituted methyl-directed mismatch repair using engineered heteroduplex DNA substrates with defined, single mismatches and controlled methylation states. This let him ask precise, quantitative questions — which strand gets repaired, how far can the GATC signal be from the mismatch and still work, what happens if methylation is symmetric — that pure genetics could not answer as cleanly. His work established not just that mismatch repair exists, but exactly how it decides which strand to trust, turning a genetic curiosity (bacterial "mutator" mutants) into a fully defined biochemical pathway.
Origins · 1 minPaul L. Modrich and how this was discovered
Where this matters today
Mismatch repair status has become a routine clinical biomarker. Tumors are now tested for MSI (via PCR panels of repeat markers) or for loss of MMR protein expression (via immunohistochemistry for MLH1, MSH2, MSH6, PMS2) as standard-of-care workup in colorectal and endometrial cancer, both to flag possible Lynch syndrome for genetic counseling of the patient's relatives and to guide treatment. MMR-deficient (MSI-H) tumors accumulate far more somatic mutations than MMR-proficient tumors, generating abundant neoantigens that make them unusually visible to the immune system; this is the biological basis for the FDA approval of PD-1 checkpoint inhibitors (such as pembrolizumab) specifically for MSI-H/MMR-deficient solid tumors, regardless of tissue of origin — one of the first "tumor-agnostic" approvals in oncology, resting directly on the mechanism described here. Beyond human medicine, engineered or exploited mismatch repair variation is used in biotechnology: strains with attenuated MMR are sometimes used to boost mutation rates for directed evolution and strain improvement, while MMR proficiency is deliberately preserved (or restored) in industrial and agricultural microbial strains where genomic stability during large-scale fermentation is essential. In basic research, MMR reporter assays (detecting reversion of frameshift mutations in repeat tracts) remain a standard tool for measuring mutation rates and testing candidate mutagens or DNA-damaging agents, connecting directly back to DNA Damage and Mutation: An Overview and to the broader logic of Proofreading and Replication Fidelity. Understanding strand-discrimination signals in MMR has also informed research into other strand-specific repair and epigenetic-marking systems, including how cells distinguish parental from newly synthesized chromatin more generally — a theme connected to DNA Replication Fork biology and to Epigenetic Inheritance.
Ask the tutor: How does mismatch repair's strand-discrimination problem compare to how cells distinguish parental from newly synthesized histones during chromatin replication?
Is mismatch repair the same as nucleotide excision repair or base excision repair?
No. All three are excision-based repair pathways, but they target different problems. Base excision repair removes single damaged or modified bases (e.g., deaminated cytosine, 8-oxoguanine) via a glycosylase. Nucleotide excision repair removes bulky, helix-distorting lesions like UV-induced pyrimidine dimers, cutting out a longer oligonucleotide on either side. Mismatch repair specifically targets mispaired-but-otherwise-normal bases and small insertion/deletion loops generated during replication, and its defining extra challenge — unique among these systems — is that it must first solve strand discrimination before it can even decide what to excise.
Why doesn't proofreading alone make replication accurate enough, so that mismatch repair is unnecessary?
Proofreading, the intrinsic 3′→5′ exonuclease activity of replicative polymerases, catches most errors immediately, bringing the raw polymerase error rate (roughly 1 in 10^5) down to about 1 in 10^7. But proofreading is imperfect and occasionally the polymerase extends past a mismatch before removing it, or slips on repetitive sequence. Left at 1 in 10^7, a genome the size of a human's (~3×10^9 bp) would accumulate hundreds of new errors every cell division. Mismatch repair supplies an additional, independent 100- to 1,000-fold fidelity boost, bringing the final combined error rate down to roughly 1 in 10^9-10^10, which is what genome stability over many generations actually requires.
How does eukaryotic mismatch repair know which strand is new, if there's no methylation signal?
Eukaryotes lack the Dam methylation system bacteria use, so they rely on the physical discontinuities intrinsic to newly made DNA. Okazaki fragments on the lagging strand are joined by ligase but briefly leave nicks; similar transient nicks or free 3′/5′ ends occur near replication origins and fork-associated processing sites. In vitro, a single strand-specific nick anywhere on the new strand, even hundreds of base pairs from the mismatch, is sufficient to direct EXO1-mediated excision to that strand. The asymmetric loading orientation of the PCNA sliding clamp at these sites is also thought to help orient the repair machinery toward the correct (new) strand.
What is microsatellite instability, and why is it the clinical fingerprint of MMR failure?
Microsatellites are short tandem repeat sequences (like (CA)n or (A)n) scattered throughout the genome. DNA polymerase frequently slips on these repeats during replication, transiently misaligning template and new strand and creating small insertion-deletion loops. Functional MMR (especially MutSβ/MSH2-MSH3 in humans) efficiently corrects these loops back to the original repeat length. When MMR is lost, these slippage errors go uncorrected and accumulate, so a tumor's microsatellite lengths drift and diverge cell-to-cell and from the patient's own germline sequence — a readily PCR-detectable phenotype called microsatellite instability (MSI), used clinically as a proxy for MMR deficiency.
Why is Lynch syndrome inherited as autosomal dominant if MMR loss requires losing both gene copies?
The germline mutation itself is inherited as a single defective allele from one parent, so at the organism level the predisposition behaves dominantly — any carrier has a markedly elevated cancer risk. But at the cellular level, cancer only initiates once a somatic cell loses its one remaining functional allele (a "second hit," e.g., by loss of heterozygosity or a new somatic mutation), fully knocking out MMR in that cell lineage. This is the same two-hit logic classically described for tumor suppressor genes like RB1, applied here to a DNA repair gene rather than a cell-cycle regulator.
Can mismatch repair ever cause harm rather than prevent it?
Yes, in specific contexts. MMR proteins recognizing certain types of damage (rather than simple mismatches) — for example, O6-methylguanine adducts paired with thymine, produced by alkylating agents — can trigger a futile, repeated cycle of recognition and attempted (but unsuccessful) repair that instead signals cell-cycle arrest or apoptosis. This MMR-dependent cytotoxic response is actually exploited clinically: the chemotherapy drug temozolomide relies partly on MMR-dependent signaling to kill glioma cells, and tumors that lose MMR can become resistant to such alkylating agents even as they simultaneously become more susceptible to immunotherapy.
How is MMR related to proofreading and to the overall concept of replication fidelity?
Replication fidelity is built in layers, each independently improving accuracy by roughly 100- to 1,000-fold. Watson-Crick base-pairing geometry and polymerase active-site selectivity provide the first layer (~1 error per 10^4-10^5 nt). The polymerase's intrinsic proofreading exonuclease removes most remaining errors, improving fidelity to about 1 per 10^7 nt. Mismatch repair is the third, independent layer, scanning finished daughter DNA using strand-discrimination signals to catch what proofreading missed, bringing overall fidelity to roughly 1 error per 10^9-10^10 nt — the compounded product of all three layers acting in series, not redundantly.