METHODS

Data collection

In total, 221 patients (124 male, 97 female) with classical FAP (>100 colorectal adenomas) who had undergone prophylactic colectomies were entered into this study, having been systematically ascertained through the St Mark's Hospital Polyposis Registry, Harrow, UK. The germline APC mutation was known for 175 patients who were therefore employed for the analysis of genotype-phenotype associations. A total of 166 individuals from 55 families were available for intrafamilial analysis of modifying genes; in 128 of these patients, an APC mutation had been identified. Median age of patients at colectomy was 21.1 years and 45% were aged 14–21 years at the time of colectomy (interquartile range 17.8–29.7 years). Polyp counts on colectomy specimens were performed routinely, as described previously,¹⁶ and corrected for differences in colon length. Although adenoma number appears to depend on patient age—for example, very young children with APC mutations rarely have macroscopic polyps—we have previously shown that age is unlikely to be a major determinant of severity in the patients analysed in this study because polyp counts change little over time once patients are in their teens and older.¹⁶ Nevertheless, we reassessed the influence of age on the number of polyps using regression analysis in the group of patients studied here, some of whom had been studied in the previous analysis.

Relationship between genotype and phenotype

Germline APC mutations were denoted by their nucleotide and/or codon position using the APC cDNA reference sequence (Genbank NM_000038). To analyse the relationship between genotype and phenotype, APC mutations were grouped according to their position within the functional domains of the gene (fig 1). The presumption was that truncating mutations in the same region of the gene are likely to result in similar phenotypic effects. The domains defined were: (i) pre-armadillo region (codons 168–453); (ii) armadillo repeat region plus (codons 454–1019); (iii) β-catenin binding region (codons 1020–1168); (iv) post-β-catenin binding region (codons 1169–1250); (v) central mutation cluster region (MCR, codons 1250–1400); and (vi) post-MCR (codons 1400–1580). Affected members of the same family were all assumed to have an identical APC mutation, whether or not the mutation had been established within all affected family members. Data were log transformed on the basis of the observed distribution, thereby improving the normality of the different groups and reducing the differences in regional variances (details not shown). Differences in the severity of polyposis (number of adenomas at colectomy) between regions were assessed using F and t tests. In order to adjust for multiple testing, the Bonferonni correction was applied to significance levels.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Box and whisker plot and aligned table show the relationship between severity of familial adenomatous polyposis and APC protein domain. The horizontal lines correspond to the 5th, 25th, 50th, 75th, and 95th centiles. The APC domains are arranged in their order on the gene 5` to 3`, left to right. The data shown are untransformed—that is, raw polyp counts—while calculations used in this paper were on log transformed data values. Summary statistics are shown in the table. Three severity groups were distinguished using multiple unpaired t tests: severe (mutation cluster region (MCR) region), moderate (pre-armadillo and post-MCR), and mild (all other groups). **p<0.01 compared with MCR, armadillo, β-catenin, and post-β-catenin (unpaired t test); †††p<0.001 compared with all other regions (unpaired t test).

Evidence for modifying genes

In general, closer relatives from a FAP kindred would tend to share alleles at a modifier locus and thus have more similar phenotypes. To examine intrafamilial effects independent of germline APC mutation, we first computed correlation coefficients between different types of relative pairs within families (siblings, parent-offspring, and second degree relatives) using log transformed data which had been corrected for germline mutations. All standard statistical analyses were undertaken using the program STATA (version 7.0; Stata Corporation, 702 University Drive East, College Station, Texas 77840, USA). To assess the evidence for a major gene component against polygenic effects, we performed a segregation analysis under a mixed model¹⁷ using the program PAP, version 4.0.¹⁸ Evidently, as only individuals with FAP could be scored for number of polyps, unaffected individuals were treated as unknown. Significance tests were based on standard likelihood ratio tests (that is, to test competing hypotheses, minus twice the log likelihood calculated under the general model was subtracted from the likelihood when one or more parameters were held constant, the difference being distributed as a χ² with a number of degrees of freedom equal to the number of free parameters).

RESULTS

There was no significant relationship between age and number of adenomas (r²=0.09, p>0.3) after patients were stratified by mutation, implying that polyp number changes little around the time that colectomies are undertaken. These data are therefore in agreement with our previous analysis using paired data which showed no significant effect of age on polyp count in patients having prophylactic colectomy.¹⁶ Therefore, for the purpose of this specific analysis, age correction was not applied. Figure 1 shows the severity of colonic polyposis according to mutations in each of the six functional domains of APC. The most severe disease was associated with mutations within the MCR (mean polyp count 3459). Mutations within the pre-armadillo region and 3` to the MCR were associated with moderate polyposis (mean polyp counts 1087 (937.4) and 1454 (1551), respectively), while mutations in the remaining domains were associated with mild colonic polyposis (armadillo(+), mean 646 (766.3); β-catenin, mean 703 (1024); post-β-catenin, mean 550 (870.6)).

Twenty two of the affected individuals harboured mutations within the MCR of APC. Fourteen of these were carriers of the 1309 del5bp mutation. There was no difference in the number of colorectal adenomas in these patients compared with those who possessed other MCR mutations (p=0.66). Similarly, there was no difference between the number of adenomas in the 29 patients with the common mutation at codon 1061 in the β-catenin binding domain compared with individuals with other mutations in this domain (p=0.98).

To compute correlations between different groups of relatives and combine data across the different APC domain groups, trait values were Z transformed:where x_i is log transformed polyp count; μ_g is the mean for the APC domain group; and s_g is the standard deviate for the APC domain group.

There was a highly significant correlation (r=0.42) in the severity of disease in sibling pairs (table 1). The correlation between parent-offspring pairs (r=0.29) was also significant but weaker whereas the correlations between more distantly related relatives were much weaker and non-significant (table 1). This pattern of familial correlation is consistent with a genetic component influencing FAP severity but not related to APC mutation.

Table 2 shows the results of segregation analysis. Single gene models fitted the data better than a simple polygenic model. Of the two single gene models, the dominant model fitted the observed data better than the recessive model. There was however evidence from the segregation analysis to suggest that a single gene genetic model is not adequate to explain the distribution of polyps in families and that a more complex model is required. Both single gene models fitted the data less well than the mixed model (χ²=7.7, comparing a dominant model with a mixed model; χ²=9.2, comparing a recessive model with the mixed model).

DISCUSSION

There are a number of possible sources of variation in the expression of the FAP phenotype. In our study of classical FAP, we have found clear evidence of a relationship between the site of the germline APC mutation and the severity of the colonic phenotype, as measured by the number of adenomatous polyps at prophylactic colectomy. While previous data emphasised the different mutation spectrum in classical and attenuated FAP,³ the common codon 1309 mutation was the only genotype associated with more severe colonic polyposis.¹ We have shown that severe disease is associated not only with mutations at codon 1309 but that other MCR mutations confer comparable disease. In addition, the data indicate that patients with mutations in codons 168–453 and 1400–1465 of APC have moderate polyposis and mutations outside these regions confer a milder phenotype.

Although there is evidence of a relationship between genotype and phenotype, there is also evidence of intrafamilial variability in FAP.¹⁹ The pattern of variation within families provides some indication as to the aetiology of this intrafamilial variation. The higher phenotypic correlations between first degree than second degree relatives would certainly be expected for effects produced by modifying genes at unlinked loci but might also result from environmental factors that are more likely to be shared among closer relatives. Determining the role of any environmental familial factor can, in principle, be addressed by examining concordance in monozygotic twins. There were no monozygotic twins available for analysis in our study and, given the comparative rarity of FAP, accruing sufficient numbers of monozygotic twins to make this a viable proposition is unlikely. It is however difficult to envisage environmental influences such as dietary carcinogens having a profound effect on the severity of FAP as the disease manifests early in life and patients appear to differ in absolute adenoma number rather than in the rate of progression from microscopic to macroscopic lesions.¹⁶

Although both types of first degree relative pairs were more similar than all other relative pairs, we found greater phenotypic similarity between sibs than between parents and offspring. Both sibling pairs and parent-offspring pairs are first degree relatives and are therefore expected to share the same proportion of non-allelic modifier alleles. Any differences detected in correlation might result from differences in environmental sharing or temporal trends in data collection but might alternatively result from genetic variation linked to APC, such as an effect of the ostensibly “wild-type” APC allele. Affected siblings share the same wild-type allele by descent from the unaffected parent for 50% of the time in contrast with parent-offspring pairs who will more rarely do so. Functional polymorphisms of the normal APC allele might therefore provide a genetic explanation for a higher correlation between siblings than among parent-offspring pairs. There is evidence for such a mechanism in some mendelian diseases.^20,21 While to date there is no direct evidence for such an effect in FAP, functional polymorphisms such as APCI1307K²² do exist.

Since the identification of Mom1 in the Min mouse,^9,10 the existence of modifier genes for FAP has been the subject of intense speculation. We have provided evidence that the clinical phenotype in human colonic FAP is consistent with the action of modifier genes. Direct confirmation of the hypothesis will require identification of the responsible genes. In the absence of a clear model of inheritance of FAP severity in our data, it is likely that identification of FAP modifier genes will rely on testing candidate loci. These candidates include genes implicated in colorectal carcinogenesis a priori such as COX2, MTHFR, and DNA repair genes, and genes located within chromosomal regions syntenic to susceptibility loci in mouse models. Although there may be some limited benefits for clinical practice in FAP, identification of FAP modifier genes is most likely to be useful in providing insight into mechanisms of colorectal tumorigenesis, indicating possible therapeutic targets and suggesting excellent candidates for colon cancer susceptibility genes in general.