86
haplotype compared with our previous data (paper I), further supporting the view that genetic variation in OX40L contributes to the development of CAD.
With respect to the OX40 receptor, our studies (paper II) have shown that the OX40 gene is highly conserved between species, indicating that OX40 fulfils vital functions in mammals203, 273. One of the few genetic variants reported in public databases caused a synonymous substitution in exon 5 and was found to be associated with HDL cholesterol levels but not with MI. Recent evidence indicates that even synonymous substitutions are subject to constraint, often because they affect splicing and/or mRNA stability314, resulting, at least in inferior organisms, in a biased usage of synonymous codons intended to maximize the rate of protein synthesis315-317. Therefore, keeping in mind that the HDL cholesterol concentration might be influenced by a linked polymorphism located elsewhere, the role of such silent changes should be carefully considered.
5.1 METHODOLOGICAL CONSIDERATIONS
87 successfully used a combination of purely positional and candidate gene
approaches. This strategy has often been proved to be optimal because it relies at the same time on both freedom from pre-formulated hypotheses and on existing knowledge of gene properties or mechanisms and their expected involvement in the disease. Specifically, the mouse model enabled us to identify the gene and study its expression and lesion size in mice deficient for candidate genes, while the human population with its different haplotype blocks allowed us to test all genes in the region in a hypothesis-driven association study which confirmed that only OX40L was associated with MI.
Such an approach is more powerful statistically than the typical association study that tests multiple genes with no a priori hypothesis. In fact, this strategy allowed us to test the human population accepting a conventional threshold of significance instead of the much smaller one obtained by correcting for multiple testing (Bonferroni), as we would have had to do if we had tested each of the nine genes in the Ath1 region with no prior hypothesis. Thus, using the animal model to generate a hypothesis allows for a more efficient study of human populations.
5.1.2 Mouse models of atherosclerosis
Many issues of experimental design, including specific model, strain, gender, atherogenic stimulus and selection of the area for lesion size quantification, may influence the outcome and interpretation of mouse model studies.
Therefore, a few aspects of animal models need to be further discussed. First, it is noteworthy that a quantitative trait by definition is characterized by measurable phenotypic variation. This variation can consist of discrete values or can be continuous, and sometimes a threshold must be crossed for the quantitative trait to be expressed, as it is often the case among complex diseases. Therefore, it is relevant for the success of a QTL analysis to both carefully define and measure the phenotypes and set the threshold.
Measurement of lesions in the aortic root is the most frequently used mode of quantifying diet-induced atherosclerotic lesions in mice318-320, and it was previously used to find the phenotype of Ath1321-323. Therefore, we used it also in this study according to established procedures for measurements.
Quantification was done in the aortic root because this is the only area where atherosclerosis is consistently present in all models naturally susceptible to diet-induced atherosclerosis318.
Another issue is the choice of the animal model. Models used for studying atherosclerosis and identifying related QTLs are essentially divided in two groups on the basis of their natural propensity to develop the disease when fed
88
a special diet, or as a consequence of genetic manipulation. The ability of mice to develop diet-induced atherosclerosis in a strain-dependent manner has been known for over two decades319-321, and considerable variation in proneness to atherosclerosis has been demonstrated among inbred strains of mice such as C57BL/6319, 324, the strain we used in our experiments.
It is clear that gender could exert a major effect on the outcome of atherosclerosis studies325, 326. Female mice are generally considered to develop a greater extent of atherosclerosis327, but this has not been a consistent finding326,
328. However, we only studied female mice because the Ath1 phenotype was found in females319-321.
Finally, there has been some concern about the different diets used to induce and accelerate lesion formation. Despite the absence of extensive systematic studies of the role of different fats and cholesterol contents on the development of atherosclerosis in the most commonly used mouse models, evidence obtained so far is controversial. Some studies have shown that inclusion of saturated fat and cholesterol in the diet promotes the same atherogenic process as occurs during feeding with a normal laboratory diet329. However, Paigen and colleagues have shown that a diet containing cholesterol and sodium cholate is absolutely necessary for developing detectable aortic lesions in mice, including C57BL/6249. Thus, what is known as the “Western diet” (saturated fat 21% wt/wt and cholesterol 1.25%) has to be modified by adding sodium cholate (0.5%). On the other hand, feeding mice such a diet leads to enlarged livers, engorged with fat, and gallstone formation as early as after 3 weeks, due to the presence of cholesterol and cholate330; the role of cholate has been questioned also for its propensity to initiate inflammatory processes331. Therefore, Paigen and colleagues manipulated the atherogenic diet by decreasing the levels of these two dietary components rather than excluding them totally, to produce a diet that minimized lipid accumulation in the liver249.
5.1.3 Linkage disequilibrium
Our results reported in papers I and III showed that the variation in OX40L is organized in two LD blocks across the gene. This consideration relies on the D′ measurements. As already mentioned, D′ values can be affected by allele frequency and sample size. Concerning the first issue, some analyses have suggested that D′ is relatively robust to variation in allele frequency332, 333, and even if sometimes a significant relationship between D′ and mean marker heterozygosity has been detected, it does not influence the results334. Perhapsof greater concern is the possibility that D′ can be skewed when one or both markers contain rare alleles335, but this should not be the case with the allele
89 frequencies of the variants we analyzed. A further complication when using D'
to measure LD is the influenceof sample size. For small sample sizes D' is upwardly biased, leading to overestimation of LD. Simulations based on334 suggest that the SCARF sample size may have led to an overestimation of D' by 0.02. It is notable that several studies of LD in human populationshave relied on considerably smaller samples than the ones describedhere336-338