Lynch syndrome genes: MLH1, MSH2, MSH6, PMS2, and what a DNA file can show

Four repair genes and a neighbour

Copying DNA before a cell divides leaves occasional errors, and mismatch repair is the system that removes them^[2]. The proteins work in pairs. MSH2 joins MSH6 to find the error^[4], and MLH1 joins PMS2 to coordinate the proteins that cut out the faulty stretch and replace it with a corrected sequence^[2].

A person with Lynch syndrome has one faulty copy of a repair gene in every cell and one working copy beside it. The cancers that follow have lost the working copy too, through a second change acquired in the tumour, and with repair gone the tumour's short repeated stretches of DNA change length, the signature called microsatellite instability^[3]. A pathology laboratory can see that signature, or the missing repair protein, in tumour tissue, and the clinical route to a diagnosis starts there.

EPCAM is the odd one. It is not a repair gene at all; it makes a protein that helps cells in the body's linings stick together, and it lies next to MSH2 on chromosome 2. The variants behind Lynch syndrome delete the end of EPCAM, including the signal that tells the cell where the gene stops, so its transcript runs on into MSH2 and the MSH2 promoter is silenced by methylation in the tissues where EPCAM is active^{[5, 6]}. MedlinePlus puts EPCAM variants at up to 3 percent of Lynch syndrome cases^[5]. The four repair genes are on the American College of Medical Genetics and Genomics list of genes that a clinical genome is checked for^[7], which our guide to the ACMG list covers.

The measured risk depends on the gene

Risk estimates usually come from carriers found because they or their relatives had cancer, which may push the figures up^[8]. The Prospective Lynch Syndrome Database (PLSD) counts differently: it follows carriers forward from their first planned colonoscopy and records the cancers that appear during follow-up. Its 2018 report, on 3,119 carriers, already found different patterns by gene and lower risks for PMS2^[9]. The 2020 update pooled 6,350 carriers: 2,607 with MLH1 variants, 2,495 with MSH2, 841 with MSH6 and 407 with PMS2^[10].

MLH1 and MSH2 carry the highest risks, with colorectal cancer by 75 near one in two. MSH6 behaves as what the authors call a sex-limited trait: women carry a high endometrial cancer risk, while colorectal cancer risk is modest in both sexes, 18 percent in men. For PMS2 the study found no increase in cancer before age 50 and only a non-significant one later^[10].

The group behind the database argues that no carrier has an average Lynch syndrome gene, and puts the consequence bluntly.

Two limits travel with these numbers. The carriers were under surveillance, so cancers prevented or caught early by it shape the figures, and the paper has no untested control group^[10]. And they were mostly found through families with cancer. A population study is the check on that. Of 454,756 UK Biobank participants with exome data, 830 carried a pathogenic or likely pathogenic mismatch-repair variant; their colorectal and endometrial cancer incidence by 70 resembled the PLSD estimates, except that PMS2 carriers showed no excess of endometrial cancer^[8].

The milder genes are the commoner ones

Clinical cohorts are mostly MLH1 and MSH2 carriers, about three in four of the 8,500 in the PLSD at its latest count, partly because lower penetrance hides the inheritance pattern in MSH6 and PMS2 families^[8]. The population looks the other way round. Modelling the families of 5,744 people with colorectal cancer, Win and colleagues estimated that 1 in 279 people carries a mismatch-repair variant, and that PMS2 and MSH6 variants are each more common than MLH1 or MSH2 variants^[11].

UK Biobank shows the same inversion in sequenced people: 431 of its 830 carriers had a PMS2 variant and 290 an MSH6 variant, 87 percent between them by our arithmetic, against 62 for MLH1 and 47 for MSH2. The authors point out that each gene's carrier frequency runs opposite to its colorectal cancer penetrance, which is what stronger selection against the more damaging genes would produce^[8].

Finland's two MLH1 founder variants

Finland's Lynch syndrome families are unusually alike. In 1995 Nyström-Lahti and colleagues found that two MLH1 variants together explained 19 of 30 Finnish families meeting the clinical criteria of the time, 63 percent^[12]. Mutation 1 was first seen as 165 letters missing from the MLH1 message, the whole of exon 16, and turned out to be a deletion of about 3.5 kilobases of the gene, most likely from recombination between two Alu repeats. Mutation 2 destroys the splice acceptor site of exon 6^[12].

The deletion is the commoner of the two. A 2024 study from Helsinki and Jyväskylä states that some 50 percent of Finnish Lynch syndrome cases are caused by it^[16]. The national Lynch Syndrome Registry of Finland, set up in 1982, held 1,800 verified carriers from 400 families by 2023^[17]. The ACMG guide tells the 1995 story from the side of the gene list; what follows is about the files.

Why the exon 16 deletion leaves no line in a file

A carrier has the deletion on one copy of chromosome 3 and an intact MLH1 on the other. Across those 3,538 bases there is one copy of DNA instead of two, and the copy that is there reads like any other stretch of the genome. Nothing in it differs from the reference because of the deletion.

A variant caller writes down disagreements with the reference, so a plain VCF has nothing to write here. A gVCF writes reference blocks, genotype 0/0, at about half the depth seen on either side, a clue that a report reading genotypes alone passes over. The reads in a BAM or CRAM hold the same drop in depth, which is what a copy-number caller looks for; our guide to genome files covers the general case.

A chip export is a list of rows, each an identifier, a chromosome, a position and a two-letter genotype. A probe inside the deletion reads only the intact copy and reports what looks like an ordinary homozygous genotype. The authors of the FinnGen study call the deletion "virtually undetectable with standard DNA microarray analysis"; attempts to design probes for it foundered on the near-identical sequence where two Alu repeats had fused, and the variant was too rare for conventional imputation^[16]. Chips are also unreliable for rare single-letter variants: in UK Biobank only 16 percent of chip calls for variants rarer than 1 in 100,000 were confirmed by sequencing^[19].

The same blind spot covers EPCAM, whose Lynch variants are deletions at the end of the gene^[6]. In a cohort of 194 carriers of such a deletion assembled at Radboud University in Nijmegen, the cumulative risk of colorectal cancer before 70 was 75 percent, while the risk of endometrial cancer was 12 percent and seemed confined to deletions reaching close to the MSH2 promoter^[20]. A large sequencing study ran into the same limit: the UK Biobank analysis left out whole-exon deletions and duplications, though they are frequent in Lynch syndrome, and skipped exons 11 to 15 of PMS2, which a nearby pseudogene, PMS2CL, resembles too closely^{[8, 21]}.

The clinical route starts from a tumour

A Finnish study tested the tumour-first route in 1998. Aaltonen and colleagues examined the tumours of 509 consecutive people with colorectal cancer: 63, 12 percent, showed replication errors, the microsatellite instability of mismatch-repair failure, and 10 of those, 2 percent of the 509, carried a germline variant in MLH1 or MSH2^[22]. Most tumours with the signature were not inherited, which is why the tumour result opens a sequence of tests rather than ending in a diagnosis.

At Kaiser Permanente Northwest, which tested all new colorectal and endometrial cancers from 2016, 313 colorectal tumours were stained; 15 stayed positive after the reflex tests, and 6 of those people were diagnosed with Lynch syndrome. Three of the five diagnosed through the programme had a PMS2 variant^[23]. That fits the population data: in UK Biobank, 14.4 percent of PMS2 carriers reported a parent or sibling with colorectal cancer, against 56.5 percent of MLH1 carriers^[8].

What our report shows for each file

MLH1, MSH2, MSH6 and PMS2 sit on our clinical-findings panel, which follows the ACMG list. A variant in a sequenced file is matched to ClinVar by position and letters; two review stars or more on a pathogenic or likely pathogenic classification make a finding, one star a separate lower-confidence lead, and anything else is not shown. Finnish mutation 2, c.454-1G>A, is classified pathogenic by ClinVar's expert panel^[14], so a VCF that lists it gets a finding.

EPCAM has no row of its own on the findings panel, and the report's wider ClinVar screen leaves it out on purpose: its Lynch syndrome variants are deletions, and its small variants in ClinVar are asserted for a different, recessive condition. Its coverage still counts. When a gVCF or the reads prove the genes were read and nothing was found, the row says "Screened, nothing found" and adds a line in the report's words: "Lynch syndrome caused by EPCAM nearly always comes from a large deletion at the end of the gene, which switches off the neighbouring MSH2 gene. A deletion of that kind does not appear in a variant file, so a clear EPCAM result here rests on less than a clear result for the other four genes on this condition." If PMS2 is not read well enough, the row says the condition was screened except PMS2.

Two more limits. The Deep Read deletion screen calls two recurrent deletions, in the carrier genes TYROBP and CLN3, and none in a Lynch gene, so an exon 16 carrier with a clean VCF gets no finding from us. And a variant of uncertain significance in an MMR gene is never shown, though UK Biobank found about 1 participant in 60 carrying one^[8]; our guide to uncertain variants explains how such classifications change over time. Why a chip row in a gene like this misleads is in our piece on raw-data false alarms; BRCA1 and BRCA2 go through the same panel.