In this study, in which type 2 diabetes cases and non-diabetic subjects did not differ in BMI, we found no statistically significant association between the FTO rs9939609 polymorphism and type 2 diabetes when analyzing the population as a whole. This result agrees with some previous studies in which no association with type 2 diabetes was reported [6, 28–30]. However, in other investigations higher type 2 diabetes risk in carriers of the minor allele (obesity-risk allele) has been reported [13–17, 20, 21, 23–29]. Among them there were many studies [1, 23, 27–29], including the first GWAs that detected the association between the FTO rs9939609 polymorphism and obesity risk , in which such association with type 2 diabetes disappears after adjusting for BMI, leading the authors to conclude that as the association between the FTO polymorphism and type 2 diabetes was mediated by BMI, the FTO is a susceptibility locus for obesity, but not for type 2 diabetes. However, in other reports [13–17, 20, 21] the association of the FTO minor allele with type 2 diabetes risk persisted even after adjustment for BMI increasing the evidence that the FTO can also be considered a diabetes-prone gene. However, some of these studies have been criticized for analyzing prevalent cases of type 2 diabetes and for differences of BMI between diabetic and non-diabetic subjects , recommending future case-control studies paired by BMI in order to better examine the independent effects. In the PREDIMED study, we fulfilled this requirement of having no differences in BMI between groups. This is a strength of our study and we were able to better analyze the effects of the FTO on type 2 diabetes more specifically.
Accordingly, the main finding and novelty of our results is that we have found that the association between the FTO rs9939609 polymorphism and type 2 diabetes depends on the diet consumed. Thus, when the dietary pattern departed from the traditional MedDiet (low-adherence to the MedDiet), the FTO rs9939609 was significantly associated with higher type 2 diabetes risk, while a good adherence to the MedDiet blunted this association. This gene-diet interaction was robust regardless of adjustment for BMI. Our results are supported by studies in mice in which a modulation by diet on the association of the fto gene with glucose intolerance has been reported . As far as we know this is the first time that a significant interaction between the FTO rs9939609 and diet in determining type 2 diabetes has been reported in humans. Another study  concomitantly examined the effects of physical activity and caloric intake on the association between the FTO rs8050136 and diabetes in U.S. women, but found no statistically significant interaction. The small number of type 2 diabetes cases in that study  was a limitation.
There is increasing evidence that the MedDiet protects against type 2 diabetes [36, 37], so it is not surprising that high adherence to this dietary pattern cancels the effects of greater genetic susceptibility to diabetes in FTO risk allele carriers. Such an interaction with diet might help explain the discrepancies in the published studies if those that did not find an association between the FTO rs9939609 polymorphism and diabetes risk [6, 28–30] were enriched in subjects following a diet similar to high adherence to the MedDiet pattern, while studies that detected this association [13–17, 21] dealt with populations with a less healthy dietary pattern, compatible with low adherence to the MedDiet.
One limitation of our study is that we analyzed prevalent cases of type 2 diabetes, as the incidence of diabetes in our cohort is still being compiled. Nevertheless, the dietary pattern of our study subjects was quite stable over time [44, 54] and we did not detect differences in the adherence to the MedDiet depending on the duration of diabetes in this analysis. Similar results of no differences in diet were found in another study in Spain . Thus, diabetes diagnosis did not change significantly the overall adherence to the MedDiet minimizing the reverse causation bias. Moreover, we have observed a similar protective effect of the MedDiet on type 2 diabetes risk when analyzing prevalent or incident type 2 diabetes cases in sub-samples of the PREDIMED study [37, 54]. Likewise, in a Scandinavian population , the FTO rs9939609 was associated with both prevalent type 2 diabetes (OR 1.13; P<0.001) and the risk of developing incident type 2 diabetes (OR 1.16; P<0.001) having comparable results. Thus, although it is necessary to investigate the effect of the interaction between the level of adherence to the MedDiet and the FTO rs9939609 on incident type 2 diabetes cases in future studies, it is foreseeable that the results would be similar.
Just as for the FTO rs9939609, we found no associations of the MC4R rs17782313 with type 2 diabetes for the whole cohort despite a recent meta-analysis identifying the MC4R loci as a new loci related to type 2 diabetes in European populations . Again, prior results from genetic association studies regarding this polymorphism and type 2 diabetes are discordant and sometimes vary after adjustment for BMI [2, 18, 19, 22, 25, 26]. Although Qi et al  described a higher risk of type 2 diabetes in carriers of the minor allele, supporting preliminary data of Loos et al , Thomsen et al  in a large sample of Danish subjects did not find such association. Noticeably, we detected, for the MC4R rs17782313, a similar interaction with adherence to the MedDiet as for the FTO rs9939609, and this is also a relevant and novel finding of the present investigation. Moreover, when we analyzed the aggregate genetic score of the FTO and MC4R polymorphisms, we also observed an additive effect of these polymorphisms on the gene-diet interaction, thus strengthening our results.
Besides examining the genetic interactions with adherence to the MedDiet, we analyzed interactions with various macronutrients and food groups, but found none (not shown). This strengthens the notion that for dietary modulation the contribution of one food is not crucial, but it is rather the overall dietary pattern with various foods or nutrients synergizing among them that is important. Considering the significant gene-diet interaction results that we have obtained for the FTO and MC4R loci, it would be interesting in future studies to analyze this interaction for other polymorphisms previously associated with obesity and/or diabetes [55–58].
Finally regarding our secondary objective aimed on studying the role of folate intake in this gene-diet interaction, given that recent literature is highlighting the importance of epigenetics in insulin resistance and type 2 diabetes [42, 59, 60], we found interesting preliminary results that require confirmation in future studies. Although no interaction of folate intake with the genetic variants on type 2 diabetes was observed, we examined fasting glucose concentrations as a more dynamic diabetes-related trait and found a statistically significant interaction between the FTO rs9939609 polymorphism and folate intake in non-diabetic subjects. Thus, the FTO variant allele tended to be associated with higher fasting glucose concentrations when folate intake was low, but not when it was high. Currently, the FTO gene has been outlined as an important gene in which effects may be mediated through epigenetics . A study reported that the CpG site in the first intron of the FTO gene was hypomethylated in type 2 diabetes cases relative to controls . Folate is required for the synthesis of S-adenosyl methionine, which serves as a methyl donor for DNA methylation events; thereby folate availability may be crucial in the DNA methylation status . The MedDiet is rich in folate and so one of the mechanisms underlying its protective effect against type 2 diabetes could be the influence of folate on DNA-methylation and fasting glucose. Some clinical trials have shown folic acid supplementation reduces insulin resistance . Although it could be one of the mechanisms that may contribute to explaining the observed gene-diet interaction, we believe that it is not the only one and that more research has to be undertaken on this point. Although for MC4R we found no significant interaction with folate, there was a similar trend and, when analyzing the aggregate variable of both polymorphisms, the interaction term reached statistical significance, supporting additive effects. In this regard, there is a study in mice showing that diet might have an effect on the methylation status of the Mc4r gene . However, our statistically significant results should be accepted with caution as folate intake may simply reflect a healthy dietary pattern and not a causal association with that micronutrient, given that, in our study, we did not carry out a methylation analysis to test this hypothesis. Moreover, although we have found a nominally significant interaction between the FTO polymorphism and the aggregate score and folate intake in determining fasting plasma glucose concentrations, we cannot rule out the possibility that, as this is a secondary hypothesis and we have not corrected the P-values for multiple comparisons, the association obtained represents a false positive result.