The study population included 3644 participants, 1780 cases and 1864 controls, with available data regarding epidemiological interviews and genetic analysis. Serum lipid levels 1 year prior to CRC diagnosis were available only for 272 cases and 625 controls. A description of the subjects’ characteristics is in . Cases had larger reported weight and BMI 1 year before the diagnosis. Also had more often family history of CRC, lower physical activity and vegetables intake. Long-term regular use of aspirin and statins were also inversely associated with CRC risk.
Baseline characteristics of cases and controls from the study population
Though energy intake was similar for cases and controls, total fat intake and the proportion of energy derived from fat (nutrient density) was larger for cases than controls. Serum lipid levels [high-density lipoprotein (HDL) and low-density lipoprotein (LDL)] measured 1 year prior to diagnosis were similar in cases and controls.
Fatty acid SNPs and CRC risk
Thirty SNPs in eight fatty acid biosynthesis and metabolism genes were genotyped in CRC cases and controls. A detailed description of the allele and genotype frequencies for the studied SNPs is in supplementary Table I
(available at Mutagenesis
Online). Two SNPs (rs16940302 of LIPC
and rs1057910 of CYP2C9
) were not in Hardy–Weinberg equilibrium. Nevertheless, none of these two SNPs showed significant associations with the variables included in this study, CRC risk or serum lipid levels. The P
-values for main-effect associations assessed between the SNPs and CRC risk, serum lipid levels, BMI and dietary fat intake are summarized in . Except for those indicated in the table, the reported P
-values correspond to the log-additive model and are adjusted for age and sex.
We first considered whether genetic variation in these genes was associated with CRC risk. An association between selected SNPs in LIPC gene and CRC risk was observed. These associations remained statistically significant when BMI, serum lipid levels (HDL and LDL) and dietary fat intake were also included in the model as potential confounders. Weak associations were also observed between polymorphisms in PPARG gene and CRC risk. Details for the analysis of these SNPs are shown in . The association for LIPC rs9652472 with CRC risk was significant after Bonferroni correction. The G allele of this SNP was associated with a per-allele OR of 1.52 (95% CI 1.20–1.92, P-value = 0.0005).
Association between polymorphisms in fatty acid metabolism genes and CRC risk
Since multiple SNPs were assessed for LIPC
and three of them showed significant associations, a haplotype analysis was performed. However, this was not more informative than the analysis of individual SNPs since they were chosen as haplotype tagging SNPs. This large gene contains four haplotype blocks (supplementary Figure 1
, available at Mutagenesis
Online). The frequencies were under 0.15 and no associations between LIPC
haplotypes and CRC risk were observed (data not shown).
Path analysis using SEM in R. The continuous arrows show the associations that remain significant in the SEM model (P > 0.02). Discontinuous arrows show associations that were no longer significant and could be considered indirect effects.
Fatty acid SNPs and serum lipid levels
Serum lipid levels 1 year before diagnosis were only available for a limited number of subjects (272 cases and 625 controls). Thus, the power to detect associations with SNPs was reduced. Only the SNP rs1800588 in LIPC was marginally associated to HDL levels in MECC study participants (). The T allele of this SNP was associated with higher HDL levels; mean differences were increased in 1.61 points per allele (95% CI 0.20–3.01, P-value = 0.025). No SNP was related to LDL serum levels.
Association between polymorphisms in fatty acid metabolism genes and serum lipid levels 1 year before CRC diagnosis.
Fatty acid SNPs and BMI
Selected polymorphism in the LIPC gene, specifically the SNPs rs16940302, rs7174210, rs4775053, and rs4775072, were found to be associated with changes in the median of BMI of the MECC study participants. Moreover, an association between rs2470890 in CYP1A2 gene and rs1801282 in PPARG gene and BMI were also observed (, for details see ). None of these associations remained statistically significant after Bonferroni correction.
Association between polymorphisms in fatty acid metabolism genes and BMI
Fatty acid SNPs and diet
The potential association of the SNPs with energy intake, fat intake and physical activity was also explored to exclude unexpected findings. None of the SNPs analysed was related to these behavioural variables ().
Path analysis of SNPs, dietary fat intake, lipid levels, BMI, physical activity, energy and CRC
The main aim of this analysis was to perform a combined analysis of all players in the lipid metabolism. Path analysis using SEMs provide an interesting tool for this since a simultaneous estimate of associations can be done for an a priori proposed model. We built the model from the observed associations in the multiple univariate analyses previously described and tried to fit SEMs to identify the associations resistant to adjustment for indirect effects.
Figure 1 shows the proposed model reduced to relevant associations. The continuous arrows show the associations that remain significant in the SEM model. Discontinuous arrows show associations that were no longer significant and could be considered as indirect effects. This analysis confirms the direct effect of LIPC rs9652472 polymorphism on CRC and the indirect effect of several other LIPC polymorphisms through serum lipid levels. It is remarkable that some of these associations were weak and not significant in the univariate analysis, but the path analysis that accounts for global confounding has revealed the associations. Other behavioural variables like total energy intake, fat intake and physical exercise also remain related among each other and associated to colon cancer, showing the relevance of these factors as mediator variables. SNPs of PPARG in the path analysis lose the direct effect on CRC risk, but rs1801282 maintains its association to BMI. (available at Mutagenesis Online) shows the coefficients and P-values with the final model of the path analysis.