INTRODUCTION An intronic germline mutation in the MSH2 gene, A→T at nt942+3, interferes with the exon 5 donor splicing mechanism leading to a mRNA lacking exon 5. This mutation causes typical hereditary non-polyposis colorectal cancer (HNPCC) and has been observed in numerous probands and families world wide. Recurrent mutations either arise repeatedly de novo or emanate from ancestral founding mutational events. The A→T mutation had previously been shown to be enriched in the population of Newfoundland where most families shared a founder mutation. In contrast, in England, haplotypes failed to suggest a founder effect. If the absence of a founder effect could be proven world wide, the frequent de novo occurrence of the mutation would constitute an unexplored predisposition.
METHODS We studied 10 families from England, Italy, Hong Kong, and Japan with a battery of intragenic and flanking polymorphic single nucleotide and microsatellite markers.
RESULTS Haplotype sharing was not apparent, even within the European and Asian kindreds. Our marker panel was sufficient to detect a major mutation arising within the past several thousand generations.
DISCUSSION As a more ancient founder is implausible, we conclude that the A→T mutation at nt942+3 of MSH2 occurs de novo with a relatively high frequency. We hypothesise that it arises as a consequence of misalignment at replication or recombination caused by a repeat of 26 adenines, of which the mutated A is the first. It is by far the most common recurrent de novo germline mutation yet to be detected in a human mismatch repair gene, accounting for 11% of all known pathogenicMSH2 mutations.
- recurrent mutation
- splice donor site of exon 5
- founder mutation
Statistics from Altmetric.com
The large “C” family from Newfoundland that allowed hereditary non-polyposis colorectal cancer (HNPCC) to be mapped to the short arm of chromosome 21 was subsequently shown to have an in frame cDNA deletion of MSH2 comprising nucleotides 265-314.2 Later, the genomic abnormality was identified as an exon 5 splice donor site mutation comprising an A→T change at the third nucleotide of the intron, abbreviatedMSH2 IVS5+3 A→T or A→T at nt942+3, and the same mutation was identified in two further North American HNPCC kindreds.3 Subsequently, when four out of 33 kindreds in eastern England were found to have it,4 a founding mutational event in an Anglo-Saxon ancestor appeared plausible. It was later shown that as many as 11 out of 41 HNPCC kindreds in Newfoundland had the same mutation, constituting 27% of all known HNPCC families.5 In 1999, an in depth study of haplotypes in kindreds with the mutation from Newfoundland, England, and the USA was undertaken to determine its origin.5 That study concluded that in Newfoundland, eight families out of 11 studied had identical haplotypes, suggesting a single founding event. As Newfoundland began to be settled as recently as 1610, this finding was consistent with the observed conserved haplotype extending over a large region of 10 cM. In contrast, no clear evidence of haplotype sharing was observed among three US and five English families, prompting the suggestion that the mutation was unlikely to be the result of a world wide ancestral event.5
The rationale for undertaking the present investigation was that the same mutation began to be reported frequently in Europe and Asia.3 6-14 By early 2000, a total of 114 different germline mutations of MSH2 had been listed in the HNPCC database (www.nfdht.nl), and the A→T at nt942+3 was frequently observed. Indeed, considering both the database and the publications cited above, it is by far the most commonly reportedMSH2 germline mutation. It currently accounts for 11% of all MSH2 germline mutations reported in the HNPCC database. If this were an ancestral mutation that occurred on several continents including Asia, its origin would have to be quite ancient. It follows that, as a result, the putative conserved and shared haplotype would only comprise a very small chromosomal region. Well known examples of conserved ancestral haplotypes occurring world wide include the region around the β globin gene on chromosomes with the Glu→Val missense mutation in codon 6 leading to sickle cell disease,15 and the region of the CFTR gene on chromosomes with the ΔF508 mutation responsible for cystic fibrosis.16 In both cases the mutation occurs world wide and accounts for a major proportion of the respective disease chromosomes. Allele sharing at flanking marker loci can be used to estimate the number of generations since founding, provided that the genetic or physical distance between the mutation and marker loci is known.17 18 Indeed, using this strategy, the age of the ΔF508 mutation inCFTR was tentatively determined as some 54 000 years or over 2000 generations.16 By analogy, if the A→T at nt942+3 mutation of MSH2 were an ancient founding mutation that occurs world wide, the shared haplotype would be predicted to be quite small.
If it could be convincingly shown that the A→T at nt942+3 mutation was not inherited from one or a few common ancestors, it would constitute an interesting repeated de novo occurrence in need of an explanation. The present study was designed to settle the question by revisiting previous marker and haplotype data by studying the inheritance of several polymorphic markers inside and immediately adjacent to the MSH2 gene in patients and families with the mutation.
Materials and methods
DNA from the blood of 42 members of 10 families with the A→T at nt942+3 mutation was obtained through clinics in Europe and Asia (three Italian, four English, two Hong Kong, and one Japanese). The Italian families from Milan were ascertained through the Hereditary Colorectal Tumor Registry of the National Cancer Institute of Milan, Italy. The characteristics of the kindreds are shown in table 1. All families satisfied the Amsterdam criteria for HNPCC families.
MARKER IDENTIFICATION AND MAPPING
Intragenic single nucleotide polymorphisms (SNPs) were identified using the web site of the International Collaborative Group on Hereditary Non-Polyposis Colon Cancer atwww.nfdht.nl/database/msh2-poly.htm. Three SNPs were identified as likely to be informative, each with allele frequencies at the lesser allele of greater than 20% (c→g at 211+9, t→a at 1511-9, g→a at 1661+12). In addition, six microsatellites within or nearMSH2 were identified. Of these, the three tetranucleotide repeat microsatellites are intragenic and mapped relative to the exons, while the remaining three dinucleotide repeat microsatellites (CA) have not been precisely mapped, but have been sequenced on a BAC containing the MSH2 gene. These three CA repeat sequences are within 70 kb ofMSH2. Parametric methods for assessing linkage disequilibrium depend on genomic sequence position, but many introns of MSH2 have not yet been fully sequenced. We used cross_match version 0.990315 (P Green, unpublished data, www.phrap.org) to compare MSH2 mRNA and DNA (accession number AC009600), and to obtain relative genomic positions for the markers where possible. These markers were typed in all available family members. Another set of 16 SNPs were identified for study, but based on published reports are thought to be relatively uninformative. Nonetheless, a fortunate pairing of a disease allele with a low frequency marker allele may provide powerful evidence for linkage disequilibrium. These SNPs were thus typed for two subjects in each kindred, in either parent child pairs (seven kindreds) or sib pairs (three kindreds) to determine if further typing was warranted. All 16 markers were homozygous for the major allele in allele typed subjects and thus were not informative for this study.
MOLECULAR GENETIC ANALYSIS
Single nucleotide polymorphisms were genotyped by direct sequencing of genomic PCR products. PCR reactions were done in 25 μl volumes with 100 nmol/l of each of the respective PCR primers (table2), 25 ng of genomic DNA, 100 μmol/l of each dNTP, 1.0 U Amplitaq Gold DNA polymerase (Perkin-Elmer, Norwalk, CT), 10 mmol/l pH 8.3 Tris-HCl, 50 mmol/l KCl, and 2 mmol/l MgCl2. PCR fragments were purified using the Exonuclease I/Shrimp Alkaline Phosphatase PCR Product Presequencing Kit (USB-Amersham Life Science). After purification according to the manufacturer's protocol, 2 μl of the PCR products were sequenced using the BigDye Terminator AmpliTaq FS Cycle Sequencing Kit (PE Biosystems, Foster City, CA). The method for sequencing has been previously described.19
The primers used are listed in table 2. Amplifications were done in 15 μl PCR reaction volumes. Concentrations of the following reagents were used: 1 μl of each 8 μmol/l primer (the 5′ primer is fluorescently labelled), 10 ng of genomic DNA, and 8 μl of HotStarTaq Master Mix (Qiagen). The following thermal cycling profile was used: one cycle of 95°C for 12 minutes, followed by 35 cycles of 95°C for 10 seconds, 55°C for 15 seconds, and 72°C for 30 seconds, followed by one final extension of 72°C for 30 minutes, followed by a soak at 4°C. Respective PCR reactions for each marker were pooled together and loaded onto the PE377 automated sequencer. Allele sizing and calling was done using Genotyper software (PE Biosystems).
HAPLOTYPE ANALYSIS AND CONTROL HAPLOTYPES
Haplotype analysis was performed in each kindred using the GENEHUNTER linkage program,20 augmented by inspection and direct Bayesian calculation. In each kindred the haplotype associated with disease was identified and any additional unambiguous haplotypes were used to form a control sample. The use of control haplotypes within the same kindreds can be important to reduce biases owing to population stratification,21 and the use of external control genotypes would require the assumption of linkage equilibrium in order to construct haplotype frequencies.
Markers were examined individually for evidence of allelic association with disease. In most instances, a unique allele could be identified as in phase with the disease mutation. Otherwise, both alleles in a genotype were included as disease alleles, which does not affect the type I error in the analysis. Allelic association with disease was assessed by Fisher's exact test. Additional multipoint linkage disequilibrium analyses were performed, using the SNP data alone and in combination with the microsatellite data. One of these analyses was a non-parametric test in which haplotypes of varying widths were examined for association with disease. For each haplotype of a fixed width, a chi-square test was performed on the contingency table of haplotype by disease status,22 and the maximum value of the statistic over all haplotype widths was recorded. Overall significance was assessed by generating 1000 random permutations of the status (disease v control) of the haplotypes to create an empirical null distribution for the overall statistic. In addition, the program DISMULT23 was used to assess the evidence for linkage disequilibrium. Because three of the markers are not precisely mapped and because SNPs are thought to have a much lower mutation rate than microsatellites, five separate multipoint analyses were performed: (1) all markers used,unmapped markers placed 5′ of MSH2, (2) all markers used, unmapped markers placed 3′ ofMSH2, (3) using the six mapped markers only, (4) using the unmapped markers only, and (5) using SNPs only.
In the 10 families, as we expected, there was at most one recombination in the markers in any of the kindreds, based on a 5′ placement of the unmapped markers to MSH2. Table 3 presents the disease allelic association evidence at each marker locus. None of the markers shows even marginally significant association with disease. Allelic association with disease can be detected more powerfully based on identification of ancestral founder haplotypes, rather than relying on individual marker genotypes.24 In seven of the kindreds the identification of the disease haplotype was unambiguous, because at least two generations of probands were typed or a sufficient number of affected sibs were typed. In two of the remaining kindreds (England 2 and Hong Kong 1), the phase of one of the mapped markers could not be known with certainty, and the disease haplotype was chosen according to an approximation to the maximum posterior probability.20 In another kindred (England 1), the phases of two of the mapped markers were uncertain, and no reconstruction was attempted. However, this kindred was informative for haplotype analysis using SNPs alone. Analyses performed without these last three kindreds do not alter the conclusions of our study.
Table 4 describes the haplotypes observed in the kindreds. Ambiguous haplotypes for the unmapped markers are left blank. None of the five multipoint linkage disequilibrium analyses described in Methods was significant on these data. Using our non-parametric analysis, empirical p values ranged from 0.21 to 0.96. Both the DISMULT program and the p value approximation suggested by Terwilliger23 yielded p values all in excess of 0.9. The lack of association with disease is apparent from tables 3 and 4. Using simple formulae describing decay of linkage disequilibrium,24 we can explore the plausibility of a world wide founding event. From tables 3 and 4, we can rule out founding events within the past few thousand generations as responsible for the majority of A→T at nt942+3 mutations. For example, three of the markers are within about 20 kb of the mutation site, and historical recombinations over such an interval would occur at a rate of only about 10% in 500 generations, assuming the average correspondence 1 cM = 1 Mb. Thus, such a major haplotype would be apparent in at least the European kindreds. A major ancestral haplotype arising 2000 generations (approximately 50 000 years) ago would still be apparent in 45% of chromosomes at a distance of up to 40 kb. A single even more ancient founding event might fail to be recognised as such, because of recombinations between the mutation and the nearest markers. However, we deem it likely that such an ancient single ancestral mutation would still give rise to more recent founding haplotypes that would be apparent in subpopulations of more recent lineage. Again, we see no such evidence from our data, but we cannot entirely rule out an ancient founding event giving rise to multiple derived haplotypes.
The mutation rates of SNPs are thought to be much lower than of microsatellites, in the order of 10-9 per meiosis,25 and we analysed the data using the SNP haplotypes alone. The second and third SNPs are separated by only 160 base pairs and are thus sensitive for detecting linkage disequilibrium, because recombinations between the two SNPs will be exceedingly rare. Thus these SNPs may be treated as a single, more informative locus. Table 5 shows the SNP haplotypes alone and the corresponding families. Again there is no overall evidence of association with disease.
The HNPCC Mutation Database (www.nfdht .nl) currently lists a total of 281 different germline mutations believed to be responsible for HNPCC. The listing is not complete but does reflect our present understanding of the mutation spectrum. Nevertheless, there are several good reasons to predict that the spectrum will change. First, the methods used to detect mutations have different sensitivities for different types of mutations. For example, hard to detect deletions are not uncommon.26 27 Also, the absence of gene transcript owing to genetic changes not detectable by current routine methods may account for a sizeable proportion of all mutations.27Additionally, HNPCC may result from mutations in genes that have not yet been identified. Finally, as population based studies are becoming feasible, mismatch repair gene mutations are being detected in affected subjects that do not belong to large families or are entirely “sporadic”.28 Mutations ascertained in this way may have a different spectrum.
Despite the above shortcomings, the mutations listed in the database and relevant publications can be relied on to some extent to assess the existence of recurrent mutations. Among a total of 114 different germline mutations reported in MSH2, the great majority have been seen only once (in one patient or kindred). Some 16 have been seen in two to four patients or kindreds, often from the same country. Most likely, all of these represent relatively close genealogical kinships so that the mutation is derived from a not too distant common ancestor. The same may or may not be true of the AAT deletion at codon 596 in exon 12 that has been reported on at least seven different occasions.
Among the 114 mutations of MSH2 is the A→T change at nt942+3 of the splice donor site of exon 5, which appears world wide. It has been reported in over 20 patients or families not counting the ones in Newfoundland.4 5 As it appears to occur in virtually all populations that have been extensively studied so far,6-9 we wished to exclude the possibility that present day mutation carriers might descend from a distant common ancestor. The alternative hypothesis that the mutation arises de novo with a relatively high frequency appeared more likely in view of our previous studies.4 5 If this hypothesis could be proven correct, it would suggest the existence of a hitherto unexplored mechanism that predisposes to this particular change.
Our results lend full support to the hypothesis that the mutation has arisen de novo on multiple occasions. This does not by any means invalidate our previous findings in Newfoundland patients. In the Newfoundland study, eight of 11 families share a mutation associated haplotype of some 10 cM in length. Such an extensive shared haplotype is fully compatible with a single major founding event in Newfoundland, the mutation having been introduced by an early settler some time after 1610, that is, less than 15 to 20 generations ago. In contrast, we found no evidence of a shared haplotype in English or Italian families even with a battery of intragenic SNP and microsatellite markers that we calculate would not sustain more than a few historical recombinations in 500 to 2000 generations. Moreover, three Asian families also did not show evidence of haplotype sharing. It should perhaps be mentioned that even with several intragenic markers studied in 10 different kindreds, formal exclusion of haplotype sharing in occasional kindreds cannot be claimed, because many marker alleles in mutation associated haplotypes are also the most common alleles in control haplotypes. For example, focusing on the three intragenic SNPs, the haplotype 111 on the disease chromosome is shared by as many as six families, whereas other haplotypes are seen in the remaining four families. However, in control chromosomes, the 111 haplotype is seen 13 times whereas other haplotypes occur nine times (table 5). Thus, a p value of 0.38 suggests no difference between these two distributions. This analysis therefore strongly suggests the absence of a shared ancestral haplotype, but obviously cannot prove it. Nevertheless, we consider our data to be strong because we were able to analyse not only these three intragenic SNPs, which have only moderately informative allele frequencies, but also six microsatellite polymorphisms which are highly informative. All our results suggest that the mutation recurs de novo.
What is the mechanism that strongly predisposes the A nucleotide in position 942+3 to become a T instead? We assume that the change seen in germline tissue occurs in meiosis, which makes it relatively unlikely to be the result of environmental influence other than ionising radiation. We have no reason to implicate ionising radiation; however, to exclude such an effect may be difficult. Instead, we are inclined to consider whether the DNA sequence itself in the immediate vicinity of the affected nucleotide might contribute to the risk. That a high risk is indeed present not only in meiosis but also in mitosis is evident from one pertinent previous observation. One of us found the same A→T change as a somatic mutation (“second hit”) in an endometrial cancer that developed in a germline carrier of anotherMSH2 mutation in an HNPCC family (M Miyaki, unpublished observations). Of note, somatic mutations ofMSH2 have been described relatively rarely,3 so a high relative frequency of the A→T change as a somatic event is a possibility. This would considerably strengthen the likelihood of it being because of a sequence peculiarity that affects meiosis as well as mitosis.
The intron between exons 5 and 6 has not been fully sequenced, but the BAT-26 locus contains a 26 adenine repeat beginning with the third nucleotide of the intron, that is, the A that is replaced by a T in the mutation. Thus, the A→T mutation occurs at the first position in this highly mutable sequence. BAT-26 was initially thought to be quasi-monomorphic ((A)26) in the population, displaying only extremely rare alleles being deleted for one ((A)25) or two ((A)24) nucleotides.29 However, in Africans and African-Americans, it displays outright widespread polymorphism with alleles containing repeat lengths as short as ((A)20) to ((A)10).30 31 A distinctive feature of this repeat is its extreme susceptibility to deletion in mismatch repair deficient tumours.33 It is virtually always deleted in tumours that show a high degree of microsatellite instability and is, therefore, widely used as a marker for mismatch repair deficiency.32 33 Mutation mechanisms resulting from DNA replication errors occur by both base mispairing and strand slippage that leads to either base pair substitution or insertion/deletion respectively. This study indicates that it is a de novo mutational hot spot for base pair substitution. In a somewhat analogous case, Laken et al 34have reported a mutation (T to A at APC nucleotide 3920) found in 6% of Ashkenazi Jews and about 28% of Ashkenazim with a family history of CRC. This mutation creates an (A)8 repeat that constitutes a small hypermutable region of the gene, indirectly causing cancer predisposition.
Do mutational hotspots occur in MLH1 as well? Among 161 different mutations described in the database, two stand out as being highly recurrent; however, both have so far been seen exclusively in ethnic Finns. Extensive haplotype analyses have shown that both represent ancient founder mutations enriched in the Finnish population.24 35 Two further mutations have been seen more than just a few times. A change of C→T in codon 117 of exon 4 has been seen at least seven times, and a deletion of AAG in codon 616 of exon 16 has been reported in some 13 families world wide. If these turn out to arise recurrently de novo, they may be additional examples, albeit not as prevalent, of true mutational hotspots of unknown causation.
The overall significance of recurring mutations is at least twofold.36 First, in populations where certain mutations are so enriched (by a founder affect) that they account for a high proportion of all mutations, they have obvious diagnostic implications. This is the case with the present mutation in Newfoundland (27% of all HNPCC) and the two prevalent MLH1 mutations in Finland (>50% of all HNPCC). Second, mutations such as the present one that arise spontaneously de novo probably result from a predisposition of either environmental or genetic nature or both. Currently these cannot be distinguished, but by eventually elucidating the mechanisms in detail, clues to their prevention may emerge.
We thank Dr Bo Yuan for analysis of theMSH2 sequence and Dr Natalia Pellagata for assistance with molecular genetics analysis. This study was supported in part by grant P30CA16058, the National Cancer Institute, Bethesda, Maryland. PR was supported by grants from the Italian Association and Foundation for Cancer Research (AIRC/FIRC). FAW was supported in part by NIH GM58934. AdlC was supported by grants CT940676 from the European Union and CA67941 from the National Institutes of Health.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.