Is TGFBR1*6A a Susceptibility Allele for Nonsyndromic Familial Colorectal Neoplasia?

Introduction

Several case-control studies have reported an association between a common variant of the type I transforming growth factor-β receptor (TGFBR1*6A) and human colorectal cancer (1-5). This gene colocalizes to the same chromosomal region (9q22.2-31.2) identified by the Colon Neoplasia Sibling Study (6) and recently validated in studies from Sweden and United Kingdom as a putative susceptibility locus for colorectal neoplasia (7, 8). To determine whether the TGFBR1*6A allele is responsible for the observed linkage at 9q22.2-31.2, we genotyped 53 kindreds from the Colon Neoplasia Sibling Study with colon cancer or advanced colon adenomas for the presence of the TGFBR1*6A or TGFBR1*9A allele. We then examined the evidence for linkage and allelic association with the TGFBR1*6A allele.

Ascertainment and Phenotyping

Ascertainment of individuals and family members and exclusion criteria were as previously described (6). From this sample we selected the 53 original kindreds used to identify the linkage signal on chromosome 9 (6). We utilized all siblings within a sibship and classified the individuals as affected, unaffected, or unknown as previously described (6).

Statistical Methods

Genotypes from the TGFBR1 locus (genotyping details available from the authors) along with the markers from the original genome scan (6) were used to obtain multipoint probabilities that sib-pairs shared zero, one, or two alleles identical by descent. These were used to determine whether study subjects were “linked” or “unlinked” to chromosome 9q22.2-31.2. In the absence of linkage, all sibling pairs are expected to share one allele identical by descent; in the presence of linkage, we expect increased allele sharing between concordant sib-pairs (i.e., >1) and decreased allele sharing between discordant sib-pairs (i.e., <1). We considered affected individuals as linked if they shared two alleles identical by descent with another affected sibling with probability of ≥0.85 or shared zero or one allele with an unaffected sibling with probability of 1 (n = 36). We classified affected individuals as unlinked, if they shared zero or one allele identical by descent with another affected sibling with probability of ≥0.95 or if they shared two alleles identical by descent with an unaffected sibling with a probability of ≥0.70 (n = 16). A final group comprised the individuals for whom identity by descent could not be so clearly distinguished, and these individuals were classified as unknown (n = 63).

We refined our Haseman-Elston regression analysis (9-11) by including the number of TGFBR1*6A alleles (0, 1, 2, 3, or 4) for each pair of sibs as a covariate in a multiple regression equation to determine if the TGFBR1*6A genotypes could account for the observed linkage signal. This provides a direct means to test for allelic association that, if there is linkage, is attributable to TGFBR1*6A or an allele in linkage disequilibrium with it. Reported P values were confirmed by comparison to a Monte Carlo sample of the permutation distribution created by permuting the allele sharing values relative to the pair labels (concordant or discordant). The number of permutations done, both across sibships of the same size and within sibships (always between 2,308 and 126,580), was sufficient to assure with 95% confidence that the estimated P value was within 5% of the true P value. Family-based tests of association (12-14) were also done.

Results and Discussion

The TGFBR1*6A allele was identified in 21% of affected siblings and in 28% of unaffected siblings (Table 1 ). The distribution of the TGFBR1*6A genotypes in the linked and unlinked affected individuals showed that proportionately more 9A/6A heterozygotes (4 of 16, 25%) were in the affected unlinked group than in the affected linked group (4 of 36, 11%; Table 1). This proportion changed little when individuals with small adenomas or late onset colon cancer were classified as affected (25% unlinked group versus 16% linked; see Table 1). Among all affected and unaffected individuals (parents and offspring) there is no significant difference in the percentage of subjects with one or more TGFBR1*6A alleles (Table 1). In fact, more siblings who had normal colon screening examinations had one or more TGFBR1*6A alleles than the siblings affected with colorectal cancer or advanced adenomas (28% versus 21%, respectively).

Inclusion of the TGFBR1 genotypes into the Haseman-Elston regression analysis resulted in a reduction of the linkage signal at D17S1786, which is the most significant marker from the original genome scan and is within 2cM of TGFBR1 (Table 2a and b ) rather than an increase, as would be expected for a susceptibility variant. To determine whether including individuals with small adenomas (n = 27) or colon cancer after the age of 65 (n = 2) would change our results, we repeated the analysis with them as affected. This resulted in further loss of signal at the TGFBR1*6A site (P = 0.16). The results of adding the number of TGFBR1*6A alleles as a pair-specific covariate in the Haseman-Elston regression analysis are shown in Table 2c. The analysis encapsulates the evidence for association/linkage disequilibrium being provided by the TGFBR1 genotypes: when adding the pair-specific covariate, the β estimate for allele sharing reflects the remainder of linkage that is not accounted for by the covariate. If the TGFBR1*6A allele were the sole disease-causing variant in our sample, all the evidence for linkage would be captured by the covariate as association/linkage disequilibrium, resulting in a loss of linkage significance at D9S1786 but significance of the covariate TGFBR1. Rather than this scenario, the Haseman-Elston regression coefficient for the allele sharing at D9S1786 remained almost the same after inclusion of the covariate (Table 2c). Furthermore, relaxing the definition of unlinked (sharing zero or one allele identical by descent with another affected sibling with probability of ≥0.70 or sharing two alleles identical by descent with an unaffected sibling with a probability of ≥0.65) to increase their number made little difference. Thus, the evidence for linkage to this chromosomal region is provided by genotypes at a location other than that of the TGFBR1*6A/9A genotypes. All family-based tests of association were also negative (data not shown). One might question the power of the CNNS study to exclude the TGFBR1*6A allele, given that a recent metaanalysis case yielded an estimated odds ratio of 1.2 (4). From a pure association point of view (ignoring linkage), we have little power to exclude this locus. However, Blackwelder and Elston (15) showed that only 70 concordant sib-pairs are necessary for 95% power, given a dominant model with a population prevalence of 1% when 𝛉 = 0 (i.e., directly at the disease locus, as is our hypothesis for the TGFBR1*6A allele). With 93 concordant sibling pairs we have 99% power to detect linkage at the disease locus. This assumes that TGFBR1*6A is the sole disease-causing variant in our sample, which is the hypothesis being tested. Although our analysis cannot exclude the possibility of the TGFBR1*6A allele increasing the risk of colorectal neoplasia in some populations, it can definitely be excluded as a susceptibility allele at the nearby linked locus we found on chromosome 9. A recent study validating linkage of colorectal cancer to chromosome 9q also excluded the TGFBR1*6A allele as a disease-causing variant in that sample (8). We conclude that there remains an unidentified susceptibility locus in the region 9q22.2-31.2.