Multiple rare variants in different genes account for multifactorial inherited susceptibility to colorectal adenomas

Abstract

Clear-cut inherited Mendelian traits, such as familial adenomatous polyposis or hereditary nonpolyposis colorectal cancer, account for <4% of colorectal cancers. Another 20% of all colorectal cancers are thought to occur in individuals with a significant inherited multifactorial susceptibility to colorectal cancer that is not obviously familial. Incompletely penetrant, comparatively rare missense variants in the adenomatous polyposis coli gene, which is responsible for familial adenomatous polyposis, have been described in patients with multiple colorectal adenomas. These variants represent a category of variation that has been suggested, quite generally, to account for a substantial fraction of such multifactorial inherited susceptibility. The aim of this study was to explore this rare variant hypothesis for multifactorial inheritance by using multiple colorectal adenomas as the model. Patients with multiple adenomas were screened for germ-line variants in a panel of candidate genes. Germ-line DNA was obtained from 124 patients with between 3 and 100 histologically proven synchronous or metachronous adenomatous polyps. All patients were tested for the adenomatous polyposis coli variants I1307K and E1317Q, and variants were also sought in AXIN1 (axin), CTNNB1 (β-catenin), and the mismatch repair genes hMLH1 and hMSH2. The control group consisted of 483 random controls. Thirty of 124 (24.9%) patients carried potentially pathogenic germ-line variants as compared with 55 (≈12%) of the controls. This overall difference is highly significant, suggesting that many rare variants collectively contribute to the inherited susceptibility to colorectal adenomas.

Familial adenomatous polyposis (FAP), autosomal recessive MYH-associated polyposis (MAP), hereditary nonpolyposis colorectal cancer (HNPCC), and other rare inherited cancer syndromes probably account for <4% of the colorectal cancer burden in a typical European population. Another 15–20% of the population are thought to have some form of hereditary multifactorial predisposition to colorectal cancer (1). Adenomas are considered to be the precursors of most colorectal cancers (2). As in essentially all cancers, a series of genetic and epigenetic events is responsible for the transition from normal mucosa to adenoma and then carcinoma. The presence of multiple adenomas in an individual is therefore a strong indicator of an increased risk of developing colorectal cancer, although not all adenomas will become cancers.

Missense germ-line variants in the adenomatous polyposis coli (APC) gene have been described in patients with multiple adenomatous polyps (adenomas), but at a level much lower than that for the clear-cut familial polpyoses, or in patients who have presented with colorectal cancer at a young age (3, 4). The APC I1307K missense variant, for example, which is found in Ashkenazi Jewish populations with an incidence of ≈6% (5), confers a significantly increased risk of developing multiple adenomas and colorectal cancer (3, 5–9). The APC E1317Q missense variant, which is found in non-Jewish Caucasian populations at a very low frequency, similarly appears to confer a significantly increased risk of multiple adenomatous polyps (3, 4). The APC I1307K and E1317Q missense variants both are single amino acid substitutions in a region of the APC protein where it interacts with several other proteins (10). They may, therefore, have mild dominant negative effects on APC function, which could account for the associated increased risk of adenomatous polyposis (11). This result suggested the general hypothesis that such comparatively rare, mostly “subpolymorphic,” variants may collectively represent a category of variation that could account for a substantial fraction of the type of multifactorial inherited susceptibility seen for multiple adenomas (3, 11–13). The aim of this study was to explore this rare variant hypothesis of multifactorial inheritance by using multiple colorectal adenomas as the model. Possibly similar variants have been described in a number of other candidate genes implicated in colorectal tumorigenesis, including, for example, cyclin D1 (CCND1) (14), E-cadherin (CDH1) (15), and BMPR1A, a member of the transforming growth factor β family of signaling molecules (16).

Our approach was to analyze DNA from 124 individuals with multiple adenomatous polyps (between 3 and 100) and to screen for germ-line variants in a variety of genes involved in Wnt signaling (APC, AXIN1, and CTNNB1) and mismatch repair (hMLH1 and hMSH2) on the assumption that such variants could give rise to inherited susceptibility to colorectal adenomas. These candidate genes were selected because of their previously known involvement in inherited susceptibility to colorectal adenomas and colorectal cancer, in somatic genetic or epigenetic changes in colorectal tumors, or in related functions.

Methods

Subjects. Peripheral venous blood samples were obtained from 124 United Kingdom patients with between 3 and 100 histologically proven synchronous or metachronous adenomatous polyps. Local ethical committee approvals were obtained, and informed consent was obtained from all participants. Clinicopathological information was derived from hospital records, and any family history of cancer or polyps was extracted from hospital records or retrospective questionnaires. No patient fulfilled the criteria for familial adenomatous polyposis, autosomal recessive MYH-associated polyposis, or hereditary nonpolyposis colorectal cancer on clinical grounds. Some of these patients had been screened already for germ-line mutations in the APC gene during previous studies (3, 4).

There were 483 random controls used for comparing variant frequencies with those in the patients. Most of the controls (397) consisted of genomic DNA samples provided by the European Collection of Cell Cultures (Centre for Applied Microbiology and Research, Porton Down, Salisbury, Wiltshire, England). These samples had originally been collected in Birmingham as controls for a population study on diabetes, with permission for future use as random controls. The remaining controls (86) consisted of genomic DNA derived from lymphoblastoid cell lines created from the peripheral blood of United Kingdom random controls originally collected as part of an HLA allele frequency study (supplied by the Cancer Research UK Cancer and Immunogenetics Laboratory with the permission of the late Julia Bodmer).

DNA Extraction and Primers. Genomic DNA was extracted from blood samples by using standard techniques. Primers were chosen to encompass the entire coding sequences of CTNNB1 (β-catenin), AXIN1 (axin), and the mismatch repair genes hMLH1 and hMSH2, including the intron/exon boundaries and 5′ and 3′ UTRs (Table 5, which is published as supporting information on the PNAS web site). The primers for CTNNB1 were based on those described in ref. 17, supplemented by additional oligonucleotides designed specifically for this study. The AXIN1 primers were largely based on those described in refs. 18 and 19, with minor modifications. The primers for the mismatch repair genes were based on those previously described and verified for use on the WAVE machine (Transgenomic, Omaha, NE) (20).

PCR Amplification. All PCRs were carried out for each primer set with 60–90 ng of template DNA in a total volume of 50 μl by using final reaction concentrations of 100 μM dNTP, 1× standard Tris·HCl/potassium chloride buffer (Promega), 1.5 mM magnesium chloride, and 1 μM each primer. Two units of Applied Biosystems AmpliTaq Gold and 0.2 unit of Stratagene Pyrococcus furiosus (Pfu) polymerase were added to each reaction. The PCR reactions for AXIN1 exons 5 and 6 were carried out by using 60 ng of template DNA in a total volume of 30 μl with final reaction concentrations of 200 μM dNTP, 1× Qiagen (Valencia, CA) buffer, 1× Q solution, 1.5 mM magnesium chloride, 1 μM each primer, 2 units of Qiagen HotStarTaq, and 0.1 unit of Stratagene Pfu polymerase. PCR was performed in Advanced Biotechnologies (Columbia, MD) 96-well skirted plates with minicaps. Positive and negative DNA controls were included on all plates. Amplification was performed with a PTC-225 thermal cycler (MJ Research, Cambridge, MA) with a cycling protocol of 95°C for 12 min, 35 cycles of 92°C for 30 sec, annealing temperature (Table 5) for 30 sec, 72°C for 90 sec, and 72°C for 2 min. PCR products of the predicted length were confirmed by running 2% agarose gels.

Mutation Analysis. Mutation detection was carried out on a Transgenomic WAVE HPLC machine after hybridization of PCR products. Temperature gradients were designed by using wavemaker (Transgenomic) software to give melting profiles, allowing evaluation of the entire length of each PCR product. Most PCR products were then run at one or two different temperatures, chosen on the basis of this profile (Table 5). Buffer A consisted of 0.1 M triethylammonium acetate (TEAA) solution at pH 7.0, and buffer B consisted of 0.1 M TEAA solution containing 25% acetonitrile at pH 7.0. Any abnormal chromatographs were noted. Positive mutational controls for CTNNB1 mutations included DNA from the colorectal cancer cell lines Colo201, Colo205, and Colo206, all of which are derived from the same tumor and carry the CTNNB1 N287S mutation in exon 6 (21), as well as Colo678 and CX1, which have polymorphism rs2953 in the 3′ UTR of exon 16. Similarly, the colorectal cancer cell lines DLD1, HCT8, and HCT15 have an L445M mutation in AXIN1 (19). DNA from colorectal cancer cell lines GP2D and GP5D provided a positive mutational control for the K618A mutation in exon 16 of hMLH1 (22).

DNA Sequencing. The nucleotide sequences of all PCR products with abnormal chromatograms were determined by direct sequencing. Each PCR product was purified by using a QIAquick (Qiagen) PCR purification column. Five microliters of purified PCR product was used in a sequencing reaction with 1 μl of either the forward or reverse PCR primer and 4 μl of BigDye ready reaction mix (Applied Biosystems). Polyacrylamide gel electrophoresis was performed under standard conditions on a PRISM 377 sequencer (Applied Biosystems). All sequences were obtained with both forward and reverse primers and were performed in duplicate to eliminate sequencing errors. Sequences were analyzed by using sequencher (Gene Codes, Ann Arbor, MI) software and compared with published sequences. The CTNNB1 sequence was derived from the GenBank entries detailed in Table 5 and GenBank entry NM_001904 (23). Published sequences of AXIN1 were obtained from GenBank entries AF009674 and XM_027520. Codons in AXIN1 were numbered according to GenBank entry XM_027520. GenBank entries NM_000249, U17839–U17856, U40960–U40977, and U49078–U07418 were used for hMLH1, and entries NM_000251 and U41206–U41221 were used for hMSH2. Variants in the mismatch repair genes were checked against the International Collaborative Group (ICG) web site (www.insight-group.org) and references.

Amplification Refractory Mutation System (ARMS)–PCR Analysis for APC I1307K and E1317Q Variants. ARMS–PCR primers were designed with the aid of oligolite (Molecular Biology Insights, Cascade, CO) and macvector (Accelrys, San Diego) software. Two forward allele-specific primers were used for each variant. The common primer for I1307K was CTAATACCCTGCAAATAGCAGAAAT, and the variant primer was CTAATACCCTGCAAATAGCAGAAAA. The common primer for E1317Q was AGAAAAGATTGGAACTAGGTCAGCTG, and the variant primer was AGAAAAGATTGGAACTAGGTCAGCTC. The reverse primer TGAGTGGGGTCTCCTGAACATA for both PCRs was nonspecific. ARMS–PCRs were carried out by using ≈25 ng of template DNA in a total volume of 7.5 μl with final reaction concentrations of 0.2 μM dNTP, 1× buffer (67 mM Tris·HCl/16.6 mM ammonium sulfate/0.01% Tween 20) (24), 1.5 mM magnesium chloride, and 1 μM each of the forward and reverse primers. AmpliTaq Gold (0.75 unit) was added to each reaction. Amplification was performed on an MJ Research PTC-225 thermal cycler with a cycling protocol of 95°C for 15 min, 5 cycles of 95°C for 25 sec, 70°C for 45 sec, and 72°C for 30 sec, 30 cycles of 95°C for 25 sec, 56°C for 45 sec, and 72°C for 30 sec, and 72°C for 10 min. Alkaline-mediated differential interaction (24), a pH-dependent gelless technique utilizing fluorescent dye, was used to screen for positive ARMS–PCR products. Heterozygote-positive controls consisted of a known APC I1307K carrier and the colorectal cancer cell line SKCO1 that carries the APC E1317Q variant.

Microsatellite Analysis. Microsatellite analysis was carried out to determine whether there was a particular haplotype associated with the CTNNB1 N287S and the hMLH1 K618A variants. Microsatellite markers were chosen to encompass the regions flanking these genes. D3S3522, D3S3527, D3S3521, D3S3658, D3S3559, D3S3685, D3S3564, D3S3678, and D3S3647 (heterozygosity, 0.56–0.85) all lie within 2.4 Mb of the CTNNB1 gene. D3S1609, D3S1619, D3S1612, D3S1611, D3S3623, D3S1260, and D3S1298 (heterozygosity, 0.66–0.89) lie within 6.7 Mb of hMLH1. One primer of each pair was marked with a fluorescent tag. Microsatellite PCR reactions were carried out on 10 ng of DNA in a total volume of 10 μl with final reaction concentrations of 0.2 μM dNTP, 1× standard Tris·HCl/potassium chloride buffer (Promega), 3 mM magnesium chloride, and 0.3 μM each primer, with 0.2 unit of AmpliTaq Gold. Amplification was performed on an MJ Research PTC-225 thermal cycler with a protocol of 96°C for 10 min, 35 cycles of 96°C for 30 sec, published annealing temperature for 30 sec, and 72°C for 60 sec, and 72°C for 10 min. The PCR product for each microsatellite was diluted by a factor of 10 with water. One microliter of the dilution was then added to 10 μl of Hi-Di Formamide and Rox size standard (10 μl of standard in 1 ml of Hi-Di) (Applied Biosystems) and denatured for 3 min at 96°C. All samples were run on a PRISM 3100 genetic analyser (Applied Biosystems), and results were analyzed in genescan 3.7 and genotyper 3.6 (Applied Biosystems).

Controls. Control DNAs were examined for any germ-line variant found that it was thought might have a functional effect. Controls were tested for APC variants by using ARMS–PCR and alkaline-mediated differential interaction and for variants in other genes by using the Transgenomic WAVE machine and confirmed by sequence analysis.

Statistical Analysis. For each variant identified, differences between the case and control populations in the proportion of individuals with the rarer variant were tested by using Fisher's exact test. When more than one variant was identified at a locus, heterogeneity between variable sites was tested for by using a Mantel–Haenszel test (25). If no heterogeneity was detected, a Mantel–Haenszel estimator of the overall odds ratio was computed and tested for significance (25). A continuity correction of 0.5 was used, when appropriate, for all data in Mantel–Haenszel tests. The process of testing for heterogeneity and computation of the Mantel–Haenszel estimator of the odds ratio was also applied to the complete data set.

Results

Patient Characteristics. The clinicopathological data are summarized in Table 1. The 124 multiple adenoma patients from the United Kingdom had a mean age of 55 (range 27–77) years. No specific information on ethnic origin was available. All patients had histological confirmation of adenomatous polyps. The number of polyps ranged between 3 and 96 (mean 12.1) in the 104 patients who had the precise number of polyps determined. The remaining 20 patients all had multiple adenomatous polyps but were simply recorded as “multiple” in hospital records. In addition to adenomatous polyps, there were 12 cancers in 9 (7.2%) patients. One patient had three cancers and another had two.

Table 1. Clinicopathological data of patients with multiple adenomatous polyps.

No. of patients	124
Male/female ratio	85:36 (3 not recorded)
Mean age, years	56.0
Age range, years	27-77
Mean no. of polyps	12.05
Range in polyp no.	3-96
No. of CRCs	12 cancers in 9 patients
Family history of CRC	36% (70 not recorded)

Family history refers to a first- or second-degree relative of any age with a colorectal cancer or adenoma. CRC, colorectal cancer.

Mutational Analysis. APC. Three of 124 United Kingdom individuals with multiple adenomatous polyps and 6 of 480 random controls had the APC E1317Q missense variant (Table 2). None of the multiple adenoma patients had the APC I1307K variant that usually is found only in Ashkenazi Jews.

Table 2. Coding missense changes found in the APC, CTNNB1, and AXIN1 genes in patients with multiple adenomatous polyps and controls.

Gene	Exon	GenBank accession no.	Nucleotide position	Codon	Allele 1	Allele 2	Mutation	No. of patients	No. of controls	Reference
APC	15	NM42_000038	3949	1317	GAA	CAA	E1317Q	3	6/480	3, 4
CTNNB1	6	NM42_001904	1074	287	AAC	AGC	N287S	1	3/483	21*
AXIN1	2	AF009674	899	312	CCC	ACC	P312T	1	0/481	This work
AXIN1	4	AF009674	1158	398	CGT	CAT	R398H	1	0/481	This work
AXIN1	5	AF009674	1186	445	CTG	ATG	L445M	1	0/483	19*
AXIN1	6	AF009674	1600	545	GAC	GAG	D545E	2	6/469	36
AXIN1	7	AF009674	2063	700	GGC	AGC	G700S	6	19/480	38
AXIN1	10	AF009674	2637	891	CGA	CAA	R891Q	7	14/478	36

The number of patients tested in each case was 124, whereas the number of controls with definitive results varied from 469 to 483 as indicated. Allele 1 is the common allele, and Allele 2 is the variant. The altered nucleotide giving rise to the variant allele is underlined.

^*Reference refers to report of somatic, not germ-line, change.

CTNNB1. As shown in Table 2, 1 of 124 (0.8%) United Kingdom patients with multiple adenomatous polyps had a heterozygous N287S missense variant in CTNNB1. This variant was also found in 3 of 483 (0.6%) random controls. A silent, and so not functionally effective, change due to a single base-pair substitution was found in exon 15 in 2 of 124 (1.6%) patients and 9 of 483 (1.9%) random controls. Three further variants, also likely to be nonpathogenic, were detected with variable frequencies (Table 6, which is published as supporting information on the PNAS web site).

AXIN1. The AXIN1 missense variants P312T, R398H, and L445M were each detected in 1 of 124 patients with multiple adenomatous polyps but not found in any controls (Table 2). Three other missense mutations, D545E, G700S, and R891Q, were found at variable frequencies in patients and at lower frequencies in the controls. All of these mutations were heterozygous in both patients and controls, as expected from their low frequencies. The overall frequency of the rare variants was significantly higher in the patients as compared with the controls (14.5% vs. 7.7%, P = 0.012, Table 4), with no evidence of heterogeneity between the results for the different variants. Several additional silent, presumed polymorphic, variants were detected, including three that had been previously uncharacterized (Table 6).

Table 4. Summary of variants found in germ-line samples from 124 patients with multiple adenomas and 483 random controls.

Gene	Variants	Patients (%)	Controls (%)	Odds ratio	P
APC	E1317Q	3/124 (2.4)	6/480 (1.3)	2.0	0.400
CTNNB1	N287S	1/124 (0.8)	3/483 (0.6)	1.32	1.000
AXIN1	P312T, R398H, L445M, D545E, G700S, R891Q	18/124 (14.5)	37/479.3* (7.7)	2.0	0.012
hMLH1	G22A, K618A	5/124 (4.0)	11/482.5* (2.3)	2.0	0.175
hMSH2	H46Q, E808X, ex4SDS	3/124 (2.4)	0/479.3 (0.0)	11.7	0.001
	Combined	30/124 (24.9)	55/479.8*† (11.5)	2.2	0.0001

SDS, splice donor site variant.

^*Average number of controls typed for all variants.

^†Two individuals had two variants.

Mismatch Repair Genes. Five of 124 individuals with multiple adenomatous polyps carried presumed functional missense germ-line variants in hMLH1 (Table 3). Four patients had the same K618A missense variant due to a heterozygous double nucleotide substitution of AA to GC at positions 1873/1874 (GenBank entry NM_000249) (26). Three of 124 patients had presumed pathogenic germ-line mutations in hMSH2, one of which was a nonsense mutation (Table 3). Both hMLH1 mutations were found in the control group, whereas none of the hMSH2 mutations was found in controls. The difference between cases and controls in the overall frequency of the hMSH2 variants was significant (P = 0.001, Table 4). One missense variant, hMLH1 F626V in exon 16, was found in 1 of 483 random controls but not in the patients. Details of silent variants are listed in Table 6.

Table 3. Mutational analysis of the hMLH1 and hMSH2 genes in patients with multiple adenomatous polyps and controls.

Gene	Exon/Intron	accession no.	Nucleotide position	Codon	Allele 1	Allele 2	Mutation	No. of patients	No. of controls	Reference
hMLH1	Exon 1	NM42_000249	86	22	GGG	GCG	G22A	1	1/483	49
hMLH1	Exon16	NM42_000249	1873/1874	618	AAG	GCG	K618A	4	10/482	26
hMSH2	Exon 1	U03911	141	46	CAC	CAG	H46Q	1	0/483	This work
hMSH2	Exon 4	U03911	795i + 5	Exon 4 SDS	A	G	IVS4 + 5	1	0/474	This work
hMSH2	Exon14	U03911	2425	808	GAA	TAA	E808X	1	0/481	This work

The number of patients tested in each case was 124, whereas the number of controls tested varied from 474 to 483 as indicated. The numbering of intronic sequence variants is determined from the nearest coding nucleotide (i.e., 795i + 5 is 5 base pairs 3′ of the end of exon 4 in hMSH2). Changes referred as “This work” have not been previously reported in the www.insight-group.org database or referenced. Allele 1 is the common allele, and Allele 2 is the variant. The altered nucleotide giving rise to the variant allele is underlined. SDS, splice donor site.

Microsatellite Analysis. Microsatellite analysis showed that the affected patient and the three controls carrying the CTNNB1 N287S variant shared an allele for each of the two markers immediately on either side of the CTNNB1 gene (D3S3658 and D3S3564) and for the more distant marker, D3S3678. These markers cover an interval of ≈2.4 Mb around the gene CTNNB1, and so the data strongly suggest that the CTNNB1 N287S variants in these four individuals have a common origin. The colorectal cancer cell line Colo206, which also carries the CTNNB1 N287S variant, was homozygous for all of the microsatellite markers in the vicinity of the CTNNB1 gene showing loss of heterozygosity at this position, as expected from previous results and its known homozygosity for this variant (21). The variant in this line, however, did not share the same alleles found in the germ-line samples. This finding shows that, as expected from its presumed somatic origin, the variant CTNNB1 N287S in the cell line Colo206 was an independent mutation.

Analysis of the microsatellites flanking hMLH1 showed very strong linkage disequilibrium between the hMLH1 K618A variant and alleles of five microsatellite markers (D3S1609, D3S1612, D3S1611, D3S3623, and D3S1260) flanking this variant over a distance of ≈6.7 Mb. Thus, of the 14 carriers of this variant, 7 shared an allele from each of the five markers, whereas the remaining 7 had at most one allele that was not shared. The GP5D colorectal cell line, on the other hand, shared only two of five alleles with the K618A variant. These data strongly suggest a common origin for all of the germ-line K618A variants, which is thus, in effect, a “founder allele,” whereas the presumed somatic mutation in the GP5D cell line clearly has an independent origin.

Discussion

The overall association found between the rare alleles at the five loci tested and the multiple adenomas, as compared with controls, is highly significant with an odds ratio of 2.2. A summary of the associations is given in Table 4. This summary is based on the data described in detail in Tables 2 and 3. Individually, only the data for the AXIN1 and hMSH2 loci are significant, and none of the results for the individual alleles is significant. This is hardly surprising, because much larger numbers of both cases and controls would be needed to achieve significance for individual alleles, giving rise to odds ratios ≈2 and with frequencies of, at most, a few percent. The overall significance of the results is supported by the fact that there is no significant heterogeneity among the data for the different alleles, thus justifying the pooling of the data.

Additional evidence in favor of a probable contribution of the variants to the multiple adenoma risk comes from a consideration of their possible functional effects. Thus, as has been discussed before (3, 4, 10), the APC germ-line missense variant E1317Q leads to a charge change in the critical area of 20 amino acid repeats responsible for β-catenin binding. It is therefore, in heterozygotes, likely to have a dominant negative effect on APC function analogous to, but much less severe than, the truncating APC mutations responsible for the vast majority of overt familial adenomatous polyposis cases. This result would be consistent with the variant's effect on multiple adenoma risk, as has been shown before.

The CTNNB1 N287S variant targets the fourth Armadillo repeat in the β-catenin protein, a highly conserved region responsible for complexing with APC and E-cadherin. The variant has previously been described in a homozygous state in the colorectal carcinoma-derived cell lines, Colo201, Colo205, and Colo206, which are all from the same patient's tumor (21), strongly suggesting a homozygous functional effect. The variant has not previously been described in the germ line. The microsatellite analysis shows that the four germ-line variants we have described, all in heterozygotes, most probably have a common origin and are independent of the presumed somatic mutation, followed by loss of heterozygosity, in the Colo201 and other cell lines. The overall data clearly support the possibility of a functional effect of heterozygosity for the CTNNB1 N287S consistent with a modest increased risk of multiple adenomas.

The AXIN1 P312T missense variant substitutes a polar threonine for a neutral proline residue in the region of the axin protein between the APC and GSK3β binding sites (27–30). The AXIN1 R398H variant causes a charge change in the region of the GSK3β binding domain (31) that also mediates binding with I-mfa (32). AXIN1 L445M is a nonconservative substitution that also occurs in the GSK3β binding domain. This variant has previously been described in the cell lines DLD-1, HCT8, and HCT15 (19), which are all derived from the same cancer and have the same APC mutation (33). A previous study had not found AXIN1 L445M in any of 274 unrelated random controls (19). An analogous murine mutation has been shown to interfere strongly with axin-GS3β binding (34).

The three other missense variants found in AXIN1 all occurred at marginally polymorphic frequencies in both multiple adenoma patients and controls. The AXIN1 D545E variant is a relatively conservative substitution but is in a highly conserved region of the axin protein that binds β-catenin (35). This substitution has previously been described in patients with sporadic medulloblastomas (36). The AXIN1 G700S substitution of neutral glycine by polar serine also occurs within a functionally important domain that binds the catalytic subunit of protein phosphatase 2A (37) and has been described in patients with both hepatoblastoma and hepatocellular carcinoma (38). The AXIN1 R891Q variant causes a charge change within the Dax domain responsible for homodimerization and Dishevelled binding at the carboxyl terminus of the axin protein (37, 39). This variant has also been described in hepatoblastoma patients (38). The overall data on the AXIN1 missense variants are certainly consistent with functional effects that could explain associated increases in relative risk of ≈2 for multiple adenomas.

The hMLH1 K618A missense variant causes a charge to neutral change in a region of four charged amino acids that is highly conserved in the hMLH1 protein of mammals and is even found in Drosophila melanogaster. The microsatellite analysis shows that all of the cases of the variant, both in patients and controls, most probably have a common origin. A number of clearly pathogenic mutations have been reported in this region of hMLH1 (40–42), and studies in yeast have shown that mutations similar to the K618A variant at the 618 codon (K618T and K618del) have a functional effect (43). The hMLH1 K618A mutation affects the hPMS2-binding domain of hMLH1 (44–46) and has been shown to cause a >85% loss of interaction between hMLH1 and hPMS2 (45). We have previously found the K618A variant in the colorectal cancer-derived cell lines GP2D and GP5D, which came from the same tumor but from our microsatellite analysis is clearly an independent, presumably somatic, mutation (22). These cell lines are mismatch defective but do not have hMLH1 promoter methylation or loss of heterozygosity at the hMLH1 locus and do express hMLH1 protein (22). Seven adenomas from one of the patients carrying the K618A variant were shown, by using the microsatellite BAT26 marker, to be microsatellite stable and so not mismatch repair defective (data not shown). This finding suggests a possible heterozygous effect of the K618A variant in the absence of any mismatch repair defect, something we have argued for previously (22). It also implies that the GP2D and GP5D cell lines must have one or more changes in addition to the K618A mutation to account for being mismatch repair defective. There seems little doubt that the K618A variant is most likely to have a functional effect, which could account for its association with an increased risk of multiple adenomas.

The hMLH1 G22A variant is a relatively conservative change, but it occurs within the amino-terminal ATP-binding site of hMLH1 (47, 48). It has also been reported in a single hereditary nonpolyposis colorectal cancer kindred from Newfoundland, where it was not found in controls.¶ This variant, therefore, is also likely to have a functional effect.

The hMSH2 H46Q missense variant is a nonconservative change in the amino-terminal region of the protein. The hMSH2 E808X nonsense variant, which is in the nucleotide binding region of hMSH2 (49), occurred in a woman with a previous history of both colorectal and endometrial cancers under the age of 50 and also has some family history. This hMSH2 E808X variant may well, therefore, represent a classical HNPCC mutation. The splice donor variant in exon 4 of hMSH2 is clearly likely to have an effect on the protein produced. It should be emphasized that the overall association of rare variants with multiple adenomas remains highly significant even if one or both of these latter two hMSH2 variants are omitted.

The analysis above shows that all of the rare variants we have described most probably have functional effects that are consistent with a modest dominantly determined increased risk of multiple adenomas. The functional analysis adds substantially to the epidemiological evidence of a statistically very significant overall association between multiple adenomas and the occurrence of the rare variants. Most polymorphic variants, where polymorphism is classically defined as a frequency of 1% or more for the least common variant (50), are neutral in the sense that they occur in introns or intergenic regions or are synonymous, occurring in third base pair positions (12). Thus, it would be highly unlikely to find a collection of potentially functional variants, such as we have found in the multiple adenoma cases, that are in fact neutral polymorphisms. The microsatellite analysis of two of the variants confirms that they are effectively “founder” variants; namely, mutations that have increased in the population by chance. Such variants are likely to be quite population-specific. Not surprisingly, none of the variants APC I1307K, APC E1317Q, CTNNB1 N287S, hMLH1 K618A, and hMLH1 F626V was found in a further cohort of 53 Korean patients with multiple adenomas who fulfilled similar inclusion criteria to the United Kingdom patients (unpublished observations in collaboration with J. C. Kim, University of Ulsan College of Medicine, Seoul, South Korea). A few examples of similar associations between increased risk of colorectal cancer and rare founder variants in specific populations have been reported, including the hMLH1 D132H polymorphism in Israelis of any denomination (51), the hMSH2 A636P polymorphism in Ashkenazi Jews (52), and the CDH1 T340A and L599V variants in Koreans (15).

The unequivocal demonstration of an increased risk associated with a particular rare variant requires larger samples than we have so far been able to analyze. It will also be sensitive to the choice of controls, especially given the likely population specificity of such rare variants. Our controls were not selected to be free from multiple adenomas, which increase in incidence with age. A single adenoma may be found in 12–43% of asymptomatic individuals, with three or more adenomas present in 0.6–11.4%, depending on age (53–56). A number of our controls may therefore be expected to have had colorectal adenomas with, so far, no detectable clinical effect. This hypothesis effectively reduces the observed odds ratio between rare variants and multiple adenomas, suggesting that our overall estimate of an odds ratio of ≈2 may well be an underestimate.

Even testing a limited number of candidate genes has begun to account for possibly up to ≈25% of the multifactorial inherited susceptibility to multiple colorectal adenomas. Furthermore, it is quite possible that some of the variants increase the risk of developing cancers other than colorectal, and this should be examined. Testing more candidate genes may eventually reveal a relatively large collection of rare variants that together explain the majority of the multifactorial inherited susceptibility to multiple adenomas. The key to identifying candidate genes and variants is to look in genes where there is an apriori basis for expecting an effect or where mutations with a severe effect on protein function have been shown to have analogous but much more severe clinical effects, as in the case of familial adenomatous polyposis and the APC gene. The variants will be most likely to lead to amino acid changes that affect protein–protein interactions and to have dominant or dominant negative effects. Variants in promoter regions may have dominant effects on levels of gene expression, and so they may also be important in this context. The selective disadvantage of such variants will be minimal, especially in relation to the common chronic diseases of the present, which were hardly of any significance until the recent past. Thus, such variants will from time to time drift up in frequency by chance, and it is those that we now see as founder variants such as CTNNB1 N287S or hMLH1 K618A.

The rare variant hypothesis, which we believe our data on multiple adenomas strongly support, proposes that it is rare, mainly missense, variants of the sort we have described that collectively can explain a substantial proportion of multifactorial inherited susceptibility to a variety of chronic diseases, including heart disease, cancer, and mental diseases such as schizophrenia and Alzheimer's disease. This explanation for susceptibility might well account for the difficulty in finding such genetic variation by case-control studies of relatively frequent polymorphic markers. No individual rare variant is likely to contribute enough to the inherited susceptibility of any, even well defined, disease phenotype to be found in this way. There may be no other way to identify this major contribution to multifactorial inheritance than to look systematically for variants of candidate genes by screening for mutations in the germ-line DNA from appropriate patient groups.