Effect of linoleic acid on ischemic heart disease and its risk factors: a Mendelian randomization study

Cardiovascular disease (CVD) is the leading cause of mortality globally [1]. Lowering intake of saturated fat and replacing it with unsaturated fat has been a mainstay of heart health dietary advice for the last 30 years [2]. Randomized controlled trials (RCTs) have largely substantiated the cardiovascular benefits of monounsaturated fats, such as olive oil [3]. In contrast, controversy has arisen about the effects of n-6 polyunsaturated fatty acids (PUFAs), particularly linoleic acid (LA) which is a major constituent of widely used polyunsaturated vegetable oils, such as sunflower, corn, soybean, and cottonseed oils [4]. Over the last half century in the USA, with the shift of dietary fat sources toward polyunsaturated seed oils, mainly soybean oil [5], dietary LA intake has increased dramatically [5], and adipose tissue LA has more than doubled [6].

Observationally, dietary LA intake or serum LA is usually inversely associated with ischemic heart disease (IHD) [7] and its risk factors, including diabetes [8], low-density lipoprotein (LDL) cholesterol, total cholesterol, and blood pressure [9]. However, observational studies are difficult to interpret because of potential confounding by factors, such as socioeconomic position, lifestyle, and health status, as well as the difficulty of distinguishing between co-occurring dietary elements. The cholesterol-lowering effect of LA is well-established [10]; however, the effect of LA on IHD is inconsistent in meta-analyses of RCTs, showing a beneficial effect [4], or neutral effects [11, 12], as new analyses of RCTs from many years ago, such as the Sydney Diet Heart Study and the Minnesota Coronary Experiment [10, 11], have come to light. Recommendations for LA intake in dietary guidelines vary, from less than 4% to up to 10% of energy intake [8].

Comparing events according to different levels of genetically predicted LA, i.e., Mendelian randomization (MR), can provide unconfounded estimates in an observational setting, because the genetic variants are determined at conception and thus are not affected by key confounders in conventional observational studies, such as health status, socioeconomic position, and lifestyle [13]. MR has been applied to examine the effect of LA on colorectal cancer [14], but to our knowledge, no MR study has assessed the effect of LA on IHD. Using genetically instrumented LA from a genome-wide association study (GWAS) of n-6 PUFA [16], and a very large case-control study of IHD with extensive genotyping, we conducted an MR study to examine the independent role of LA in IHD. We similarly assessed the associations with its risk factors, including diabetes, lipids, and blood pressure, to identify if any associations with IHD were independent of these risk factors. We also examined its role in reticulocyte count, the red blood cell precursor, which has been recently identified as a possible causal factor for IHD [15].

Using MR to obtain unconfounded estimates, our novel study shows an inverse association of LA with diabetes and lipids. Our study, together with a previous cohort study [8], suggests a benefit of LA for diabetes; the inverse association with lipids is also consistent with the cholesterol-lowering effect of LA. However, the benefit for IHD remains to be confirmed, consistent with the mixed findings from RCTs and meta-analysis of RCTs [4, 11, 12]. The associations with blood pressure and reticulocyte count are less clear; however, a positive association cannot be excluded.

To our knowledge, our study is the first MR study examining the effect of LA on IHD, diabetes, lipids, blood pressure, and reticulocyte count. Using genetic variants as proxies for LA, MR is less likely to be affected by the residual confounding and reverse causality inherent in observational studies. Moreover, the metabolism of n-6 PUFA interacts with that of n-3 PUFA [2], and our study applying MR to large publicly available data enables us to examine the independent effect of LA on IHD and its risk factors, in a cost-efficient way [39]. The IHD case-control study with over 70,000 cases and 260,000 controls, at an approximated R² of 0.15 (percentage of variance explained by the three SNPs with top significance) in the GWAS of LA [16], has 0.8 power to detect an odds ratio (OR) of about 0.96 per percentage in total fatty acid increase in LA [40].

Nevertheless, several limitations exist. MR is based on three assumptions, i.e., the genetic instruments are associated with the exposure, no confounders of the associations of the genetic instruments with the outcomes exist, and the genetic variants are not linked with the outcomes other than via the relevant exposure (no pleiotropy) [13]. To satisfy these assumptions, we only selected SNPs strongly associated with LA, in functionally relevant genes, and also SNPs with top significance. We used PCA to use all the SNPs, a method that does not suffer from numerical instabilities arising from the potentially arbitrary SNP selection [20]. In addition, the sample for genetic variants on LA and for IHD and its risk factors only slightly (~ 1%) overlaps. As such, any correlation of the genetic variants with unmeasured confounders in the sample with LA is unlikely to be replicated in the samples with IHD and its risk factors, due to the different data structures [41]. We checked for known pleiotropy and used MR Egger to detect unknown pleiotropy. Given that population stratification might affect MR estimates, we obtained all the genetic associations from studies in people of European ancestry and with genomic control. As such, the associations might not apply to other populations. However, causal effects are not expected to vary by setting [42], although they would not be detectable or relevant in populations that do not consume vegetable oils. In addition, we could not assess whether associations varied by baseline level of LA. However, meta-analyses of the associations of LA with IHD do not suggest heterogeneity due to geographic locations with different baseline levels of LA [12]. A potential nonlinear association of serum LA with IHD could not be evaluated using summary data; to assess non-linearity needs individual-level data [43]. The influence of genetic predictors might be damped or buffered by compensatory developmental processes or feedback mechanisms [13]. However, such feedback mechanisms would be expected to mitigate the genetic effects, thus biasing toward the null, which would not explain the inverse associations of LA with diabetes and lipids in our study.

The effect of endogenous LA could differ from the effect of dietary LA. However, our findings have some consistency with RCTs specifically using LA as a dietary intervention, such as the Sydney Diet Heart Study and the Minnesota Coronary Experiment [10, 11], and their meta-analysis [11], as well as with the well-established cholesterol-lowering effect of dietary LA [10]. Using genetic instruments of objective biomarkers, i.e., serum LA, also minimizes measurement error arising from self-reported dietary consumption in nutrition studies. Our study could be affected by survivor bias (selection bias); however, the samples for LA and outcomes were not in the age range (75+ years) where survivor bias is thought to have a substantial effect [44]. The small effect size may not be clinically significant and represents the effect of lifetime exposure. MR estimates are less precise although less confounded than those from conventional observational studies [45]. However, relatively small effects of causal factors may still be an important determinant of population health, particularly for dietary factors which are modifiable and commonly consumed. We cannot exclude the possibility of reverse causality; however, it is unlikely that SNPs in genes related to the metabolism of LA affect LA via cardiovascular disease risk factors. A bi-directional MR is not feasible due to the lack of availability of genetic variants associated with LA. In addition, we could not assess whether the effect of LA on IHD and its risk factors varies by sex or age using the data freely available. As such, our estimates are likely to be conservative as some associations could be sex-specific, but the directions should not be reversed.

Notably, our novel study suggests a benefit of LA in diabetes and lipids, but that the benefit may not translate into a clear benefit for IHD. These paradoxical findings might also just be chance, which will be clarified by the use of stronger instruments and larger studies of IHD. However, a similar pattern has also occurred for other interventions targeting lipids, such as the cholesteryl ester transfer proteins, which lowered cholesterol and diabetes, but had collectively an unexpectedly null effect on IHD and an off-target hypertensive effect [46], possibly mediated by steroidogenesis [46, 47]. As such, a more general explanation with a mechanism underlying such a paradox might exist, given androgens are increasingly realized to affect cardiovascular risk [48, 49] as well as to improve glucose metabolism. Clarifying the underlying pathway would be beneficial for identifying effective new interventions for both diabetes and IHD.

Our novel study suggests a benefit of LA for diabetes, LDL cholesterol, and total cholesterol, but no benefit for blood pressure or reticulocyte count. The benefit for IHD remains to be confirmed. Clarifying the role of LA and its underlying pathways would be worthwhile, with relevance to dietary recommendations and the identification of effective new interventions for both diabetes and IHD.