Using genetic variation to disentangle the complex relationship between food intake and health outcomes.

Diet is considered as one of the most important modifiable factors influencing human health, but efforts to identify foods or dietary patterns associated with health outcomes often suffer from biases, confounding, and reverse causation. Applying Mendelian randomization in this context may provide evidence to strengthen causality in nutrition research. To this end, we first identified 283 genetic markers associated with dietary intake in 445,779 UK Biobank participants. We then converted these associations into direct genetic effects on food exposures by adjusting them for effects mediated via other traits. The SNPs which did not show evidence of mediation were then used for MR, assessing the association between genetically predicted food choices and other risk factors, health outcomes. We show that using all associated SNPs without omitting those which show evidence of mediation, leads to biases in downstream analyses (genetic correlations, causal inference), similar to those present in observational studies. However, MR analyses using SNPs which have only a direct effect on the exposure on food exposures provided unequivocal evidence of causal associations between specific eating patterns and obesity, blood lipid status, and several other risk factors and health outcomes.

Introduction

Given its impact on human well-being, diet is one of the most studied human behaviours. Quality, quantity, and patterns of consumed foods are associated with a wide range of medical conditions such as metabolic, inflammatory, or mental health diseases.[1] However, despite the growing number of studies reporting associations between diet and health outcomes, it has been challenging to establish causal relationships due methodological limitations such as measurement error, confounding, and reverse causation [2,3]. To date, several approaches have been devised to try to account for intrinsic limitations in nutritional studies such as the use of methods to calibrate food records [4] through the use of 24h recalls [5], biomarkers [6] and doubly labelled water, [7] or the implementation of domiciled feeding studies.[8] Although the implementation of these methods or study designs have helped in addressing some of the limitations of nutrition research, difficulties remain especially when it comes to estimate the causal effect of diet on health outcomes.

In this context genetics may represent an alternative approach by the use of Mendelian Randomization (MR). MR is a methodological approach in which genetic variants associated with an exposure of interest are used as instrumental variables to investigate the causal association between this exposure and an outcome.[9] To date, several MR studies have been designed to investigate the associations between the consumption of single food groups, such as alcoholic beverages [10], coffee [11], milk [12-14] and specific health outcomes, but a systematic study investigating the overall role of diet on multiple health outcomes is missing. Previous MR studies have not accounted for the fact that genetic variants associated with reported dietary intake may be primarily associated with other risk factors, reporting characteristics or social determinants of health which may confound the causal estimates.

The present study was designed to initially identify the genetic variants associated with reported food consumption, and then to leverage a causal inference statistical framework to systematically investigate the causal effects of dietary factors on health outcomes while accounting for the reverse causal effects that health determinants have on habitual dietary intake reporting.

Methods

Given large number of analyses conducted for this study and their complexity Fig 1 summarises the main analyses.

Fig 1

Overall study design.

Study population and genome-wide association for dietary intake

The UK Biobank [15] is a large population-based cohort including 500 000 adults aged between 40 and 69 years at baseline across 22 assessments centres in the United Kingdom. Data were collected based on clinical examinations, assays of biological samples, detailed information on self-reported health characteristics, and genome-wide genotyping. Dietary intake in UK Biobank was assessed using a touchscreen dietary frequency questionnaire which included questions about the frequency of consumption specific foods and beverages over the past year. The number of samples used for each trait can be found in Table A in S1 Table while a detailed description of the phenotypes, can be found in the in the S1 Note 1.2 and Table B in S1 Table. For alcohol consumption traits analyses were limited to people drinking at least one glass of the alcoholic beverage a week. This choice was due to very high number of people who reported to drink 0 glasses of the specific alcoholic beverage per week (between 58% for red wine to 94% for fortified wine) which would have biases the analyses, due to reverse causation [16]. A similar issue applies to coffee consumption traits where stratifying for coffee type required to restrict the analysis to coffee drinkers. Finally, we have excluded people who reported eating certain foods (e.g. beef) less than once a week due to the very large range of different consumptions which this response corresponds to. The proportion of people used for the analysis compared to the overall UK biobank participants can be found in Table A in S1 Table. Validity of the food consumption measures has been evaluated by Bradbury and others [17] which concluded that “the dietary touchscreen variables, available on the full cohort, reliably rank participants according to intakes of the main food groups”.

We used the BOLT-LMM software [18] to assess the association between the genetic variants across the human genome and 29 food phenotypes. Analyses were conducted on genetic data release version 3 imputed to the HRC panel [19], as provided by the UK Biobank (http://www.ukbiobank.ac.uk/wp-content/uploads/2014/04/UKBiobank_genotyping_QC_documentation-web.pdf). Population stratification was assessed using LD-score regression as implemented in ldsc [20,21] using the LD scores provided with the software which refer to the HapMap [22] v3 SNPs. Table Q in S1 Table reports for each food trait the LD regression intercept and heritability estimation using the ldsc software [20]. Cluster analysis conducted on the foods identified five main independent groups of traits (see additional online methods paragraph 1.8 and 2.2 for details of group definition) and we thus set the genome-wide significance threshold at 1x10-8 (5x10-8/5). This work was conducted using the UK Biobank resource (application 19655). Participants enrolled in UK Biobank have signed informed consent forms. Replication analyses for the identified signals associated with food phenotypes were conducted independently by using genetic and dietary data from the EPIC-Norfolk Study [23] and the Fenland Study [24]. Details additional online methods 1.4.

Investigating the causal effect of health outcomes on reported food intake

Univariable MR analyses were initially conducted to measure the causal effect of health outcomes on food consumption using the TwoSampleMR [25] R package. Exposures of interest were selected amongst those for which nutritional advice is given and included body mass index (BMI), low density lipoprotein cholesterol (LDLc), high density lipoprotein cholesterol (HDLc), Total cholesterol, Triglycerides, Diastolic and Systolic blood pressure, Type 2 diabetes, and coronary artery disease. In addition, we included educational attainment amongst the exposure traits for the multivariable MR, as a proxy of socio-economic status which is likely to affect food consumption. The full list of studies from which the summary statistics were derived is detailed in Table F in S1 Table. For each exposure we selected all SNPs with p<5 x 10−8 and r2<0.001 to be used as instruments in the MR analysis. After performing stepwise heterogeneity pruning to remove SNPs which showed evidence of heterogeneity in the causal effect estimate, we performed MR analysis using the inverse variance method [26]. We then tested if the intercept from the MR-Egger [27] regression was different from zero (p<0.05). If this was the case, MR-Egger was used for the analysis instead.

Identification of genetic variants with predominantly direct effects on diet

One of the most important assumptions in MR is that the effect of the instrument on the outcome must be mediated only through the exposure of interest (sometimes referred as exclusion restriction criteria) [28]. In this light, genetic instruments whose effect on food is mediated through the health outcomes or through educational attainment may violate this assumption acting as confounders in the relationship between the exposure and the outcome. Moreover, if the mediating trait is acting on the reporting of food consumption and not food consumption itself it would mean that the genetic variant is not truly associated to food consumption, and it would thus not be a valid instrument. It is thus important to estimate the direct effect (i.e., the effect that acts directly on food intake rather than is mediated through other factors see Fig 2) the SNPs are exerting on actual food consumption in order to properly select the genetic variants to be used as instrumental variables.

Fig 2

Direct and indirect SNP effects.

The plot shows the causal path of exemplar genes identified for cheese consumption. In the multivariable MR model cheese consumption is causally influenced by educational attainment (EDU), low density lipoprotein cholesterol levels (LDL) and systolic blood pressure (SBP). The effect of PDCH17 and is mediated through educational attainment, while SIX3 has a direct effect on cheese consumption. The mediated effects cannot be used reliably as MR instruments as they could be affecting either consumption or its reporting. Moreover, they could act as confounders in the MR analysis and thus they need to be identified.

To this end we use a modified version of bGWAS [29], in which corrected estimates for genetic variants are obtained after accounting for the effect of other phenotypes on these genetic variants. Further details about bGWAS can be found in S1 Note 1.6. We applied bGWAS to all 29 food phenotypes. As potential mediators, we used the same cardiometabolic phenotypes as before except total cholesterol to avoid collinearity issues with LDL and HDL cholesterol, and we added summary statistics from Crohn’s disease and ulcerative colitis and smoking as they are likely to affect dietary patterns. A full list of the traits used as exposures and their sources can be found in Table F in S1 Table. We identified genetic variants with only a direct effect on diet based on the corrected to uncorrected ratio (CUR) as the ratio between the corrected and the uncorrected effects (see additional methods 1.7 for a detailed explanation). The threshold to define genetic variants with non-mediated effects (CUR = 1±0.05) is based on simulations provided in the S1 Note 2.1 and on the genetic variants with known biological function (i.e. bitter taste receptors). We defined as “non- mediated” those SNPs whose CUR fell within the defined ranges while “uncertain” the others.

Genome-wide genetic correlations between corrected dietary intake and health outcomes

We used LD-score regression implemented in LD Hub [20,21] to estimate genome-wide genetic correlations between dietary intake phenotypes and 844 health outcomes and intermediary phenotypes. Genetic correlations were estimated both with the corrected and uncorrected GWAS summary statistics using the bivariate LD-score regression model. Stratified LD-score regression [30] analyses were implemented using ldsc and the annotation files available on the ldsc website.

Definition of food group variables

In order to define measures of dietary patterns we first performed cluster analysis of the 29 food items applying iCLUST [31] to the corrected genetic correlation matrix between the different foods. iCLUST clusters items in different groups based on a hierarchical structure (Details additional methods 1.8). Fig 3 shows the resulting dendrogram and its comparison with the genetic correlation matrix.

Fig 3

Clustering of the food traits and definition of measures of dietary patterns.

The plot reports the genetic correlation plot amongst the food traits after applying the correction. The stars report the Bonferroni-corrected significant correlations. The dendrogram and the boxes represent the clustering according to the ICLUST algorithm. The labels on the dendrogram branches show the traits used to define each measure of dietary pattern. The dashed line represents the traits excluded from the estimation of the dietary pattern traits. The “Vegetarian” trait was excluded from the “Meat PC” trait but was included in the overall dietary pattern measure (All PC).

We then defined based on the resulting structure several measures of dietary pattern at different levels of the dendrogram as shown in Fig 3. For each group we performed principal component analysis of the items of that group. The rotation matrix was derived from the eigen decomposition of the genetic correlation matrix of the foods in the PC trait of interest. For example, for the Coffee PC measure we performed principal component analysis of “Ground Coffee”, “Instant Coffee” and “Decaf Coffee”. Once the rotation matrix was estimated for each SNP its effect on the new measure was estimated as the linear combination of the effect on each food trait using as weights the loadings on each PC. This method has been described before in Tsepilov et al. 2020[32]. For each group of traits only the first component which explained the greatest amount of genetic variance was retained for further analyses. A correlation plot of the loadings of each item onto the PC traits can be found in Fig G in S1 Note.

MR analyses to assess causal relationships between food intake and health outcomes

MR analyses were conducted to estimate the effects of the food phenotypes on 64 health related phenotypes (see Table S in S1 Table for details) available in MR-base.[25] Genetic instruments for each exposure of interest included independent genetic variants (p<5x10-8 and pruning for LD (r2<0.001)). For dietary patterns exposures SNPs were selected as outlined in additional methods 1.9. Briefly once each defined group of traits we estimated the loadings of each index item as the eigen decomposition of the corresponding correlation matrix as outlined in the previous section. This procedure was repeatedly applied to both the original and corrected effects which allowed us to estimate for each SNP on each PC trait effect size, standard error, p-value and CUR. The PC traits where then treated as any other trait applying the same p-value threshold to the projected traits.

For the main analysis we restricted the genetic instruments to those that had evidence of purely direct effect (i.e., not affecting the main exposure through a different pathway; CUR 1±0.05). Discussion of the relationship with other methods can be found in S1 Note 2.7. Weights for the genetic instruments were based on the uncorrected effects. To verify the effects of using only direct effect only SNPs on MR, all the analyses were also conducted without applying the CUR filtering.

After selecting the genetic instruments, exposure and outcome data were harmonised. The MR estimates were tested for heterogeneity and outliers were removed using the MR-Radial method.[33] MR analyses were based on the inverse variance weighted method, which estimates the causal effect of an exposure on an outcome by combining ratio estimates using each variant. A random effect model was used if significant heterogeneity between the different estimates was detected. We then tested for the presence of directional pleiotropy using the intercept from the MR-Egger regression. MR median and MR-Raps were used as sensitivity analyses. All results have been made available through an online app (https://npirastu.shinyapps.io/Food_MR/) and can be found in additional Table T in S1 Table.

Results

Genetic variants associated with food intake

In a GWAS of 29 food phenotypes we identified 412 genetic associations in 256 independent loci (Fig 4 and additional Table D in S1 Table) at Bonferroni corrected level of significance (P< 1×10−8). The principal component association analysis revealed 160 additional SNP-trait associations with additional 27 loci for a total of 572 genetic associations in 283 distinct loci.

Fig 4

302 independent genomic loci associate with food choices.

Results for both univariate (256 loci) and PC traits (additional 27 loci see paragraph S2.3) analyses are included. For each SNP the lowest uncorrected p-value for all traits was plotted. The upper panel represents the unadjusted GWAS associations while the lower panel represents the association with food choices, after adjustment for mediating traits, such as health status for the same snp-trait pair used for the upper panel.

Replication was sought in two additional UK-based cohorts including up to 32,779 participants. Despite relatively limited power in replication cohorts, concordant direction of effect was observed for 82% of the signals (p = 7.82x10-35, Binomial test; Table E in S1 Table), and nominal significance was achieved by 32% of the signals (p = 9.47x10-54). Gene prioritization is described in S1 Note 1.10 while biological annotation, network analysis and tissue enrichment analysis are discussed in additional paragraphs 1.11, 2.4 and 2.5. Several of the identified loci have been previously associated with BMI. However, contrary to our expectations, the BMI-raising allele was consistently associated with lower reported consumption of energy-dense foods such as meat or fat, and higher reported intake of low-calorie foods.

Genetic variants associated with food intake are strongly influenced by other phenotypes

In univariable MR we identified 81 instances in which health-related traits significantly influencing food intake (Fig 5 additional Table G in S1 Table). For example, BMI and Educational attainment influenced more than 50% of the food traits. Similar effects extend to a broad range of traits, for example LDL and triglycerides influenced 15 and 18 traits respectively.

Fig 5

Health status influences reported food choices.

The plot reports only the univariable MR results which were significant at FDR<0.05. For each food outcome the effect estimate (β) is reported in standard deviations of the exposure trait, together with 95% confidence intervals. Each colour represents a different exposure. BMI, body mass index; CHD, coronary heart disease; DBP, diastolic blood pressure; HDL, high density lipoprotein cholesterol; LDL, low density lipoprotein cholesterol; TotalC, total cholesterol. Champ/Wh wine, champagne, white wine. Temp, temperature.

Higher genetically predicted CAD associates with higher consumption of fish and red wine, and lower consumption of whole milk, salt and lamb. These findings suggest that some of the signals identified in GWAS for reported food phenotypes are not directly associated with food intake but are mediated through a wide range of potential confounders.

The Multivariable MR confirmed the univariable MR results (Fig B in S1 Note panel B and Table H in S1 Table). The percentage of genetic variance for the reported food phenotypes explained by health determinants ranged from 42% for cheese to ~0% for fortified wine and white wine/champagne (Fig B in S1 Note panel A and Table H in S1 Table). We systematically compared the estimated effect sizes of each genetic variants influencing food consumption before and after correcting for the effect of health determinants and showed that in many loci the variant initially identified for food phenotypes changed dramatically after considering the effect of health factors (Fig 4). For example, the effect size of the lead FTO variant (rs55872725, p = 2x10-29) on milk fat percentage chosen decreased three-fold after accounting for the mediated effects. To further explore the magnitude of this indirect effect on food intake phenotypes, we compared the correlation patterns between the 29 food phenotypes and 832 phenotypes present in the LD hub [21] database identifying great differences. For example, low fat milk intake was correlated with a beneficial effect on body fat percentage (rG = -0.43) but this association diminished to near zero (rG = -0.04) after accounting for indirect effects (S1 Note 2.2 and additional Table L in S1 Table). The effects of the correction procedure on the genetic correlation amongst the traits and with the 844 health traits are discussed in S1 Note 2.2 while full results can be found at in Table L in S1 Table and browsed at https://npirastu.shinyapps.io/rg_plotter_2/. These findings highlight the relevance of biases and confounding in genetic correlation studies and we provide a framework to mitigate these problems and to reliably study complex physiological relationships.

Causal inference analyses for diet phenotypes and health outcomes

A total of 245 out of 572 genetic variants initially associated with food phenotypes were categorized as “non-mediated” associations (Table C in S1 Table). Most loci contained either non-mediated (146/283 loci) or uncertain associations (92/283 loci), while the remaining 45 contained a mixture of the two.

The fraction of uncertain genetic associations varied by food group, ranging from mostly direct effect for tea, decaffeinated coffee, poultry and processed meat, to mostly uncertain for percentage fat in milk and adding spread to bread (Table C in S1 Table).

In two-sample MR analyses we found 52 significant associations between food phenotypes and health outcomes after multiple test correction (q-value < 0.05, Table T in S1 Table). None of them showed sign of heterogeneity amongst the estimates (heterogeneity test q-value >0.05). Fig 6 reports full results for all significant food exposure trait outcome pairs.

Fig 6

Significant effects of food choice on disease related traits.

The heatmap reports the results for all significant food trait exposure trait outcome. Only dietary pattern exposures summarising the overall group consumption (PC1) have been reported. All exposures have been aligned to have a positive loading onto the “overall unhealthy diet” measure. Significant food/trait association are indicated with *. Abbreviations: BMI Body Mass Index, WHR Waist to Hip Ratio, TRY triglycerides, TC total cholesterol, HDL HDL cholesterol, LDL LDL cholesterol, Hb% Haemoglobin percentage, PLT Platelet count, Edu Educational attainment, CD Chron’s Disease, IBD Inflammatory Bowel Disease. Panel has been divided in two to separate quantitative traits where effect size is in SDoutcome per SDexposure (higher effect equals red colour) from qualitative traits where effect sizes are expressed in log(ORoutcome) per SDexposure (higher effect equals green colour).

Overall, we found that the “overall unhealthy diet” measure did not show significant associations with many traits except for BMI where higher values of this measure corresponded to higher obesity. However, we showed that specific components had effects on different traits. For example, BMI was significantly associated with genetically determined meat consumption, particularly pork and processed meat, but also with a general tendency to heating less healthy foods. In contrast, other measures of adiposity such as waist-to-hip ratio were not associated with meat consumption.

We identified 13 instances in which we would have not detected significant associations without filtering out the non-direct effect instruments such as the effect of increased fruit consumption on triglycerides levels (estimated uncorrected effect = -0.09 (SE = 0.04) vs. estimated corrected effect = -0.17 (SE = 0.05)) or the effect of increased healthy foods consumption on BMI (uncorrected effect = 0.004 (-0.13, 0.14) vs corrected effect = -0.16 (CI 0.07–0.26). In addition, we found 109 food/trait relationships that were not significant after applying CUR filtering, showing that either confounding effects or reduced power explain the lack of association (see additional note 2.6). For example, Psychoactive drinks consumption was initially associated with increased lung cancer (uncorrected effect = 0.27 (CI 0.09–0.45)), but there was little evidence of an association after filtering out the instruments not directly influencing Psychoactive drinks (corrected effect 0.02 (CI -0.17–0.19)). On the flip side, we showed that the effect of alcohol consumption on mean corpuscular volume remains substantially unchanged when applying the filtering approach (beta 0.05 (SE 0.02) uncorrected and 0.05 (SE 0.06) corrected), suggesting that our approach could precisely identify relevant biological relationships.

A full description of our findings is found in Table T in S1 Table and have been made available through an online app (https://npirastu.shinyapps.io/Food_MR/).

Discussion

In this study we have quantified the complex interplay between diet and health outcomes showing that the causal path from food intake to adverse health outcomes is not unidirectional and may be influenced by reverse causation and confounding even when MR is used. We showed that genetic correlations and causal inference can be improved by leveraging statistical approaches that consider these mediated effects and identify genetic variants that have predominantly direct effect on the exposure of interest. This information allowed us to perform causal inference analyses that helped identifying more reliable potential causal effects of food on health outcomes.

Results in context

Previous MR studies have mainly focused on specific food groups such as coffee, alcohol and milk consumption while none has comprehensively investigated the role of different food groups on health outcomes. Here we have expanded this approach to encompass a wide range of specific foods and dietary patterns allowing to compare results across the different traits and giving us more insight in the interpretation of the results. Findings from this study suggest that the biases affecting measures of food consumption (reporting bias, confounding and reverse causation) are propagated to genetic associations. We have shown that these issues extend beyond obesity and socio-economic status, and revealed a broad range of intermediate factors. For example, we showed that LDL and triglycerides concentration influence a wide variety of food traits, implying these phenotypes should be considered as potential sources of bias in future MR studies. For our analyses we have used UK biobank in which participants were aged between 40 and 69 at the time of the questionnaire, it is likely that a younger cohort will suffer less from some of these biases (i.e. altered food consumption due to elevated LDL cholesterol or blood pressure) as it is unlikely that they will display pathological level of these traits.

Our results showing that genetic variations associated with food phenotypes could be influenced by reverse causation and confounding are in contradiction to some previous studies, in which no evidence of reverse causation was reported.[34,35] We believe that this difference is due to our novel approach, which does not correct for potential mediators based on their correlation (through linear regression), but rather based on their causal effect (through MR), which should be able to distinguish the forward and reverse effects when the causal relationship is bidirectional. Our study suggests that it is possible to disentangle these different colliding effects, and identify genetic instruments with a non-mediated effect. This particularity of our approach enables the use of MR for the assessment of causal relationships between food and health.

Many studies have looked at the relationship between nutritional composition and health outcomes. One of the most salient examples is the relationship between saturated fat intake and cardiovascular disease and all-cause mortality, in which recent studies suggest that food sources of saturated fatty acids are more important than saturated fat content per se [36]. Our study provides a new angle on the importance of food sources by showing that foods with similar nutrient profile, for example cheese and meat, have opposite effects on some metabolic risk factors such as BMI but there is no difference in other phenotypes such as blood lipids. Similarly, food with relatively different nutritional composition such as fruit, vegetables, and fish had the beneficial effect on triglycerides. While these findings require further investigations, our genetic evidence lends support for the importance of studying foods in their complexity and not as a mere mixture of nutrients. This approach, in fact, does not consider that the sources of the nutrients are not equal due to the food matrix, the different preparations and that foods are seldom consumed by themselves but in patterns which are likely to modify the effects on health.

Our findings illustrate that the effect of diet on health outcomes is complex, and components of specific food groups have a differential association with health. In this case, although fish and fruit and vegetables have a very different macronutrient composition it was impossible to separate their effect on triglyceride concentrations. This suggests that at least in this case the macronutrient composition is not as important as an overall tendency to eat certain foods and it highlights the importance of always including the assessment of dietary patterns before claiming health effects of single foods or nutrients. This example also highlights one of the limitations of MR. We can study only specific food exposures for which valid and relatively specific instrument exists and the resolution stops at the point where genes influence food groups/patterns rather than specific items. Better powered studies may enable the identification of genetic markers associated with specific food items, enabling more refined analyses.

Our study also provide evidence that the overall unhealthy diet is almost exclusively associated with BMI, with no evidence of associations with any other outcome. This means that it may be possible to design dietary interventions by modulating only specific food depending on the effect we want to obtain for everyone. For example, if we consider obesity related traits, meat seems to have an effect specifically on BMI while consumption of healthy foods, fish, fruit and vegetables influence fat distribution as indicated by the association with WHR and WHR adjusted for BMI. Although thus there is heterogeneity in effects both lowering BMI and WHR are desirable outcomes and thus and overall healthy diet is still the desirable intervention if we aim at maximising the beneficial health effects across all outcomes.

Our study has several potential limitations. First, the number of items available in the dietary questionnaire in the UK biobank is limited, and therefore it limited our ability to capture overall diet or specific food groups not available. The inclusion of white and relatively healthy and educated participants from UK Biobank may have limited the generalisability of our findings. The self-reported nature of the diet questionnaire is prone to measurement error and bias, and the use of a short food frequency survey could have further reduced the resolution of dietary data collected. More accurate dietary intake assessment methods such as the use of dietary intake biomarkers (doubly labelled water, urinary nitrogen, minerals, and vitamins) for calibration purposes would be valuable in future studies, specially to obtain more precise estimates of the causal effect sizes, however these are challenging to implement in large-scale cohort. Another source of bias may be due to the missing samples due to either nonresponse or to removal due to the phenotype definition which may induce spurious correlation similarly to sampling bias. Moreover estimated effect sizes could be inflated because of the underestimation of the SNP effects on the actual food trait consumption, rather than its self-report, if so, this will have inflated our estimates of the effects of food on health, due to the noise in the questionnaire responses, and warrants further statistical investigations. Even so, our method should not have falsely identified a causal effect or reversed its direction, but further studies are needed to assess the precise effect sizes. Finally, we must consider the possibility of residual confounding effects through variable either not included in our models or imperfectly measured. For example, we have used educational attainment as a proxy for socio economic status, so we cannot exclude that more precise measures would result in even better estimates. Despite these limitations, our methodological approach offers a possibility to improve our understanding of the genetics of diet and strengthen causality in nutrition research.

In conclusion, our findings show that overall, what is generally considered a healthy diet leads to many favourable health outcomes and to reducing a wide range of risk factors broadly agreeing with current guidelines aimed at reducing meat and alcohol consumption while increasing fruit vegetables and fish. We also show that some of these effects are mostly attributable to specific food or group of foods which however are not characterized by common nutrient composition thus adding granularity to our knowledge on the effect of diet on health. This information can be useful to inform the design and implementation of future studies to reduce the burden of diet-related diseases.

Supplementary Methods and results.

Fig A. Directed acyclic graph explaining the two possible scenarios for the effect of a SNP on the trait of interest Y. (a) The SNP has a direct effect on Y not mediated through X. Then the estimated effect of SNP on X will be normally distributed around 0, thus the corrected and uncorrected effects will be similar and their CUR will be close to 1. (b) The SNP effect is mediated through X, thus the corrected effect will deviate from the observed one and CUR will deviate from 1. Fig B. Results for the Mutlivariable MR. Panel A. The plot represents the proportion of genetic variance which is explained by the effect of the health related traits on the food traits. Clearly some of the food traits are extremely biased having up to 25% of genetic variance due to the mediation of the health related traits. Panel B The heatmap represents the effect of the health related traits on each food trait using from the multivariable model. The color is proportional to the effect size. Fig C. Diagram describing the relationships between the simulated traits and their relative parameters. Gy refers to the genetic variants which directly affect Y before any influence of confounding or other mediated traits (Yt). Gu represents the genetic component of a confounder trait U which causally affects both Y and X. Gx represents the genetic component of the outcome trait X which is in turn causally affecting the trait Yi. Yo represents the actual observed trait to which we add noise to reflect the test-retest correlation in FFQ data. Fig D. Scatterplot of the CUR values for Gx (in red), Gy (in green) and Gu (in blue) at the different values of the effect of Y on X and of X on Y. Fig E. Corrected-to-uncorreected ratio (CUR) successfully distinguishes mediated and non-mediated associations. (a) Graph showing mediated and non-mediated pathways. The values of CUR that different types of simulated SNPs (Gx, Gy, Gu) assume at different explained variances (σ2) of X->Y when σ(Y->X)≠0, i.e. presence of reverse causality (b). The values we used for defining a “non-mediated” variant are highlighted in purple. (c) The proportion of variants that are truly Gy, that is directly associated with the trait of interest, across a range of CUR. (d) The overall proportion of variants directly associated with the trait (SNPy) whose CUR falls inside the specified ranges, i.e., the probability of detecting SNPy over all possible scenarios.When the effect of Y->X is equal to zero, Gy is clearly distinguishable from Gx and Gu using CUR (Fig E), however, when β(Y->X) increases, values of CUR for both Gy and Gx start varying and overlapping (Fig Eb). We thus determined which values of CUR would maximise the probability of correctly selecting Gy under all scenarios. Clearly the parameters we have chosen for defining a “non-mediated” SNPs maximise both the probability of correctly selecting a SNPy. Fig F. Clustering of food consumption traits before and after correction. Comparison between the hierarchical clustering of the food traits based on the uncorrected (on the left) and corrected (on the right) genetic correlations. Black lines connect the same traits for which the clustering has changed. Dendrograms connect the items in each case with the boldness of the line representing the strength of support for the tree nodes. Unique nodes are represented with a dashed line while shared nodes with a bold one. The thickness of the line is thicker for conserved higher level nodes. Fig G. Corrplot of the loadings of each food item onto the measures of dietary pattern. All items have been aligned to the “Overall unhealthy diet” measure. The items that have been flipped are noted as “Less” to clarify the direction of the relationship. Fig H. Heatmap of tissue and functional enrichments. The colour is proportional for the enrichment revealed by stratified LD-score regression. Only correlations with FDR<0.05 are reported. (a) Enrichment among different classes of functional annotation. (b) Tissue enrichment from Gtex expression. (c) Tissue enrichment from ROADMAP epigenetics. (d) Tissue enrichment from the Franke lab dataset. Fig I. Dotplot of the differential expression analysis run on the prioritised genes from the non-mediated loci and the overrepresentation analysis performed with MAGMA. The overexpressed tissue involved by the two methods were highly overlapping with the analysis performed on the prioritised genes showing the tissues in which there is evidence of underexpression. Fig L. STRING network of genes in non mediated loci. Network plot of the genes in the non-mediated loci. After performing community detection we identified ten different clusters of genes each with its particular set of functions and expression patterns (see additional paragraph 2.6 for details). Nodes have been colored according to community membership. Fig M. Tissues which overexpress the genes in each community. Fig N. Overlap in Go-terms between different communities. The Fig shows that there is no overlap (with the exception of 2 terms) between the terms enriched in each community. The labels have been removed as the plot is meant to only show the overlaps. Fig O-AD show the enriched terms for each community separately. Fig O. Enriched GO-Terms for community 1. Fig P. Enriched GO-Terms for community 2. Fig Q. Enriched GO-Terms for community 3. Fig R. Enriched GO-Terms for community 4. Fig S. Enriched GO-Terms for community 5. Fig T. Enriched GO-Terms for community 6. Fig U. Enriched GO-Terms for community 7. Fig V. Enriched GO-Terms for community 8. Fig Z. Enriched GO-Terms for community 9. Fig AA. Enriched GO-Terms for community 10. Fig AB. Enriched GO-Terms for community 11. Fig AC. Enriched GO-Terms for community 12. Fig AD. Enriched GO-Terms for community 13. Fig AE. Selected forest plots of MR-estimated effect sizes. A Forest plot of the effect of Cheese and Meat consumption on lipid and obesity measures. Despite both foods have a high protein and high fat content their effects on lipid levels and BMI are different. Abbreviations BMI Body Mass Index, TRY Triglycerides, TC Total Cholesterol, LDL Low Density Lipoprotein. B Effect of several foods related to healthy foods on blood tryglicerides levels. Effect for all foods are very similar and make it impossible to distinguish the contribution of each food. Fig AF. Effect of food on obesity related measures. The forest plot compares the effect of each food trait on four obesity related measures: BMI, Body Fat, Waist to Hip Ratio (WHR) and BMI adjusted WHR (WHR|BMI). Each color and shape represents a different obesity related measure. Fig AG. Forest plots of the exposure/outcome pairs significant at the uncorrected analysis. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AH. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AI. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AL. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AM. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AO. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AP. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AQ. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AR. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AS. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AT. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AU. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AV. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig AZ. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BA. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BB. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BC. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BD. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BE. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BF. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BG. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BH. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BI. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BL. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BM. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BN. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits. Fig BO. The forest plots represent the estimated effect sizes for all the non CUR filtered MR analyses. The squares represent the point estimates while the bars the 95% confidence intervals. Results from the uncorrected analysis (raw) and CUR filtered IVs (CUR) are reported. The exposure trait is indicated in the header of the plots while the row labels refer to the outcomes. Beta’s always refer to standard deviations for the exposure while for the outcomes it is standard deviations for the quantitative traits and log(OR) for the disease traits.

(DOCX)

Click here for additional data file.

Supplementary tables.

Table A: Number of samples used for the GWAS for each trait. The questionnaire column indicates from which questionnaire the item was taken. Proportion of sample indicates the proportion of samples used compared to the number of people who participated in the UK biobank at baseline (501,520). Table B: Description of the phenotypes: the table reports the description and coding of the phenotypes used for the GWAS analyisis. Phenotype, name of the phenotype; Question, question asked in the tochscreen questionnaire; Answer—Converted to, coding of the phenotype; Covariates, covariates used for the analysis, standard = age+sex; Transformation, transformation applied for normalising the trait. Table C: Significant loci per trait and type. Table D. Summary statistics of top SNP per trait per locus. Locus N, genomic locus number. Chr, chromosome; Start, starting position of the locus; End ending position of the locus; rsid, per trait top SNP in the locus; a1 a0 Effect allele, Other allele; N number of samples; Beta effect of the coded allele; SE Standard error of the uncorrected effect; p p-value of the initial association; Corr.Beta, effect of a1 after the GW mediation correction procedure;Corr.SE Standard error of the corrected effect; Corr.p p-value of the corrected association; CUR corrected to raw ratio; Gene prioritised genes, if the top-SNP was between genes and the closet one was chosen, the distance from the gene was indicated; type category of the association based on the CRR values. Table E. Replication results. The table reports the replication results for the traits/SNP for which replication was availble. Locus.N locus number; rsid SNP use for replication; a0 a1, Other allele effect allele; Beta effect of the effect allele in the discovery cohort; p p-value in the discovery cohort; n number of samples used in the discovery cohort; rep_effect effect of the a1 allele in the replication cohorts; rep_p p-value in the replication cohorts; N_rep, number of samples used for replication. Table F. a: GWAS used for the Multivariate MR analysis and GW mediation analysis. Name, name of the trait; Consortium, consortium that performae the GWAS; Reference link to paper describing the GWAS; Download; source of the summary statistics. b: GWAS used for the Multivariate MR analysis and GW mediation analysis. Name, name of the trait; Consortium, consortium that performae the GWAS; Reference link to paper describing the GWAS; Download; source of the summary statistics. Table G: Full results of univariate MR. Outcome, outcome trait; Exposure, exposure trait. Method used for the MR analysis; nsnp, number of SNPs used for th IV; b effect of the exposure on the outcome; se standard error of the effect; pval, p-value; results_adjusted, Multiple test corrected pvalues. Table H: Multivariable MR results: for each food trait (otucome column) the best model used for the prior estimation is reported. Exposure column indicates the exposure used; Beta the multivariable effect of the exposure on the outcome; Se standard error of the effect; P-value p-value of the effects; Cor^2 the squared correlation between the prior and the original z-scores. Table I: Full Genetic correlation amongst food results: The table reports the full genetic correlations between the food traits using both corrected and uncorrected results. p1 trait 1; p2 trait 2; rg genetic correlation; se standard error of the genetic correlation estimate; z z-score of the genetic correlation; p p-value; the extension Raw and Corrected refer to the uncorrected and corrected results respectively. Table L: Genetic correlations between the food traits and common traits. Both uncorrected and corrected (the prefix.Corrected is used to indicate corrected results) results are reported. rg, genetic correlation, se standard error of the genetic correlation, p p.value. Table M: Interaction network results analysis performed with STRING. Table N: Comunity membership of the genes in the STRING interaction network. Table O: GO term enrichment for the 13 Detected communities. Table P. Results for tissue enrichment analysis performed on each community. Only tissues with FDR<0.05 are reported. Table Q. LD score regression intercept and heritability estimates. Table R. Proportion of genetic variance explained of each food trait by the causal health related traits. Trait, name of food trait, rg2 proportion of genetic variance explained, rg2se standard error of the proportion of explained genetic variance. Table S: Information on the outcomes used for MR. Table T. Full results of the Food Mendelian Randomization: The table reports the full resutls of the mendellian randomization of food on all the selected outcomes. Exposure: exposure trait, Outcome: outcome trait; Method: method for the main MR analysis: IVW (FE) inverse variance method fixed effect, IVW (RE) inverse variance method random effect, Wald ratio: Wald ratiomethod; nsnp, number of SNPs used for the MR; b effect size relative to the method columns; se standard error relative to the method column; pval p-value of relative to the method column; type exposure betas used (Uncorrected, Corrected); outliers number of SNPs excluded because they were foudn to be outliers at the MR radial analysis; egger.pleio p-value relative to the mr egger test for directlional pleiotropy; mr.raps.beta/me.raps.se beta and standard error relative to the MR-RAPS method; beta_egger/se_egger beta and standard error realtive to the MR-Egger analysis; beta_median/se_median beta and standard error realtive to the Median analysis; het.p Heterogeneity test p-value; qval Storeys qvalues, the qvalues have been estimated only on the CRR Uncorrected analysis; sens.het Heterogeneity test run on the estimates coming from the different methods.

(XLSX)

Click here for additional data file.

18 May 2021

Dear Dr Pirastu,

Thank you very much for submitting your Research Article entitled 'Using genetic variation to disentangle the complex relationship between food intake and health outcomes.' to PLOS Genetics.

The manuscript was fully evaluated at the editorial level and by independent peer reviewers. The reviewers appreciated the attention to an important problem, but raised some substantial concerns about the current manuscript. Based on the reviews, we will not be able to accept this version of the manuscript, but we would be willing to review a much-revised version. We cannot, of course, promise publication at that time.

Should you decide to revise the manuscript for further consideration here, your revisions should address the specific points made by each reviewer. We will also require a detailed list of your responses to the review comments and a description of the changes you have made in the manuscript.

If you decide to revise the manuscript for further consideration at PLOS Genetics, please aim to resubmit within the next 60 days, unless it will take extra time to address the concerns of the reviewers, in which case we would appreciate an expected resubmission date by email to plosgenetics@plos.org.

If present, accompanying reviewer attachments are included with this email; please notify the journal office if any appear to be missing. They will also be available for download from the link below. You can use this link to log into the system when you are ready to submit a revised version, having first consulted our Submission Checklist.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Please be aware that our data availability policy requires that all numerical data underlying graphs or summary statistics are included with the submission, and you will need to provide this upon resubmission if not already present. In addition, we do not permit the inclusion of phrases such as "data not shown" or "unpublished results" in manuscripts. All points should be backed up by data provided with the submission.

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool.  PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org.

PLOS has incorporated Similarity Check, powered by iThenticate, into its journal-wide submission system in order to screen submitted content for originality before publication. Each PLOS journal undertakes screening on a proportion of submitted articles. You will be contacted if needed following the screening process.

To resubmit, use the link below and 'Revise Submission' in the 'Submissions Needing Revision' folder.

[LINK]

We are sorry that we cannot be more positive about your manuscript at this stage. Please do not hesitate to contact us if you have any concerns or questions.

Yours sincerely,

Ulrike Peters

Guest Editor

PLOS Genetics

Hua Tang

Section Editor: Natural Variation

PLOS Genetics

We hope the authors find the thorough review by the reviewers helpful. To be considered responsive it will be critical that the authors respond to all comments. The authors need to describe the assessment toll in detail in the methods (including how the 29 food phenotypes have been assessed) so that the reader can better judge the work. The authors should give particular attention to the issue that they start off with an assessment tool that is known to be biased and error prone. This is likely particularly the case for the UKB given its limited dietary assessment tool that is not a food frequency questionnaire. As part of this a comparison with the 24 hr recall assessment is an excellent idea and needs to be added to the paper for being considered responsive to the reviewers’ comments. Accordingly, the authors should tone down their statements about the superiority of their approach compared to other work.

The authors also need to add a comprehensive discussion to the manuscript about the limitations raised by the different reviewers, particularly about the issues of weak instrumental variable bias.

The authors also missed to point out the excellent work conducted to find objective biomarkers for dietary intake, such as doubly labeled water, urinary nitrogen, minerals, and vitamins that can be used in calibration studies.

Why does the # of participants per phenotype varies so dramatically? Does this raising concerns about the reliability of the data?

As pointed out by the reviewers the authors need to reconsider the choices of “exposure” variables or causal health related traits they adjusted in the multivariable MR analysis. Related to this it is not clear how it is possible to investigate the association between foods and BMI if as part of the correction all SNPs that show associations between BMI and the foods were excluded. The same question holds for other “exposure” variables or causal health related traits they adjusted in the multivariable MR analysis.

The paper does not describe well what results are derived from the UKB vs the other 2 UK cohorts included. This needs to be clearly described in the main text (not only in one of the many supplemental sections)

As data on smoking are available the authors should investigate if the association between unhealthy diet and lung cancer is truly confounded by smoking or not.

Reviewer's Responses to Questions

Comments to the Authors:

Please note here if the review is uploaded as an attachment.

Reviewer #1: The authors write: "we develop a statistical genetics framework that enables us, to directly assess the impact of food choices on health outcomes." No genetic analysis can ever do this, let alone the one implemented in this study.

The authors write in the Abstract that: "that the biases which affect observational studies extend also to GWAS, genetic correlations and causal inference through genetics, which can be corrected by applying our methods." This is not actually demonstrated in the study. The authors simply apply some methods (which are mix of previously described approaches) and generate some results, but we don't know whether they correct any previous biases.

The authors are overly critical of previous studies in the introduction without any supporting citations or solid evidence for their claims, yet then rely on self-reported data themselves.

In essence, the paper boils down to multivariable Mendelian randomization. However, the description is incredibly confusing and the authors seem to constantly mix up mediation and genetic confounding (which cannot be distinguished through statistical analysis alone - biological insight is required).

The choice of outcomes is poorly defined - why not also look at osteoarthritis, stroke, peripheral arterial disease, kidney disease, liver disease etc, which are also affected by dietary choices?

No attempt is made to address weak instrument bias in a multivariable Mendelian randomization setting. This will likely be a major limitation, as the variants explain a small proportion of the variance in the exposures and their associations with different dietary choices are correlated.

Neither does the paper seem to have been properly proofread. For example, on line 426 it says "[citation error]".

Reviewer #2: This is an exciting piece of research that aims to disentangle the association between diet and health outcomes. The authors have build a thorough analytical pipeline to GWAS dietary factors in UK Biobank, replicate findings in independent cohorts and apply novel design ideas within the Mendelian randomization (MR) framework with an aim to achieve unbiased MR estimates for diet and health outcomes. There is dearth of MR findings for dietary intake and health outcomes, as it is difficult to identify robust and non-pleiotropic genetic instruments for dietary intake (except for alcohol, coffee and dairy intake), and the authors have tried to address this limitation of previous research. They should be commended for their approach, but there are some issues I would like to raise that could partially invalidate some of the approaches.

1) The UKBB study has not used a FFQ for dietary assessment, instead there was a touchscreen diet questionnaire offered to the totality of participants. This small but important misconception should be corrected in the Introduction. As a consequence, the likelihood of measurement error in the dietary assessment is higher compared to the use of a more established method of dietary assessment like the FFQ. In addition, only a relatively small range of dietary intakes were assessed with the current questionnaire that might bias the definition of healthy or unhealthy diet pattern that the authors used. Some further discussion on these issues is required. In addition, ~140K participants have completed at least 1 of the 4 web-based 24h dietary assessment post-recruitment, and it would be nice to see a sensitivity analysis with these data.

2) The authors have performed MR analyses of several "exposures" (i.e. BMI, lipids, BP, education, T2D and CHD) in relation to dietary intakes and they have used these results via MR mediation analyses to estimate the association of diet to health outcomes not mediated via one of these exposures in an attempt to limit horizontal pleiotropy. This is a smart idea! However, there are couple of additional exposures that may be important to consider further such as smoking and physical activity. In addition, it was not clear in the manuscript whether the authors have performed bidirectional MR analyses between the dietary intakes and the "exposures". In the presence of a bidirectional association, the approach would likely lead to collider bias or adjust out the complete genetic variants to diet association. More formal evaluation and discussion of the assumptions of mediation analysis is required.

3) The authors used a stepwise pruning for heterogeneity for their definition of instrumental variables, which is an interesting approach but I would be hesitant to base this process only on statistics (i.e. the p-value for heterogeneity) and would like to see biological knowledge of gene function to take a central role in the process. The authors should check the association of their final selected IVs with any secondary traits and comment on residual horizontal pleiotropy.

4) The authors defined locus definition and independence based on physical distance (e.g. 250kb). Did they consider also LD structure in different genomic areas?

5) The final MR results of dietary intakes in relation to health outcomes are impressive, and some but not all associations seem to coincide with the hypothesized directions for the associations. However, most of the associations do not survive the multiple comparison p-value threshold and even less survive the sensitivity MR analyses, which the authors should comment clearer upon in the Results text. Could the authors discuss the potential reasons for this? Does their method lead to lower statistical power or to bias in some circumstances when adjusting for exposures acting as colliders?

Reviewer #3: General comments:

This is an interesting work on a very relevant topic, whether Mendellian Randomization (MR) analysis can help circumvent some of the chronic limitations of nutritional epidemiology when relating the intake(s) of specific foods, as well as patterns of foods, to health outcomes. The study used genetic and dietary data from the UKB, as well as data from several available datasets, to identify genetic variants to be used as instrumental variables in etiological models.

Nevertheless, several important aspects of the current approach are not fully convincing, nor several details of the employed methodology are sufficiently clearly described. As a result, the description of the methodology reads very confusing.

In addition, although the study attempts to address weaknesses of nutritional epidemiology like confounding, reverse causation and exposure measurement errors, within the genetic framework employed, genetic instrumental variables for food intakes are estimates in models where dietary assessments by means of food frequency questionnaires were used. It is well known that these measurements are prone to exposure misclassification, potentially leading to biased identification of relevant genetic variants. These limitations are not even addressed in the current version of the manuscript.

Detailed comments:

Abstract, line 48: “determinants of dietary intake” – what dietary intake, how many food groups, how were they chosen?

Line 50-51: the text reads very optimistic in describing the approach but no details whatsoever are provided to enable the reader understand the novelty and the relevance of the work. In line 53, the text mentions “our work” but again no elements are given to capture the importance of the study.

Overall, the abstract is improvable. It reads rather vague and does not release elements necessary to judge the relevance of the study.

Line 56, the text reads “…and several other risk factors”. Should other morbid condition or health outcomes be mentioned instead?

Line 114: the text is very technical, though some details related to the relevance of the use of the “ldsc” command should be disclosed.

Lines 116-117: the reasoning leading to the choice of using genome-wise (statistical) significance threshold equal to 10xE-8 in the study is not explained clearly.

Lines 120-121: The text does not read appropriate.

Lines 123-124: MR analysis is NOT conducted to measure (estimate) the effect of health outcomes on food consumption. Rather the other way round.

Lines 124-125: “foods for which nutritional advice is given”. This is an important information, but it does not clarify what criteria were used to pick up the chosen list of food items this study focused on.

Line 128, “we included educational attainment” – Where?

Line 131: SNP were included if p<5 x 10E-8 and r2<0.001. Up above the criteria seemed to be 10E-8, not 5 x 10E-8. Also SNP should be selected if r2 was greater than 0.001, not lower.

Lines 139-149: The text does not read correct, nor the MR description is appropriate.

Line 150: what is bGWAS?

Lines 150-159: The description of the MR methodology used in the study is very confusing.

Lines 120-121: The text does not read appropriate throughout page 7. Statements like the ones in lines 152-154 are really obscure. The CUR approach should be referenced. The analysis moved from univariate to multivariable methodology, but te various steps are not clearly outlined.

There is a massive amount of information provided as a supplementary material and Tables. It is not ideal to exhaustively capture several nuances of the study.

Reviewer #4: This is a comprehensive analysis and interesting paper that assesses the utility of using genetic data to unravel the complex relationship between dietary intakes and health outcomes. Some comments:

• Smoking seems like an omission from the analysis as it is well established that smoking influences dietary intakes. The authors included educational attainment as it is likely to affect food consumption. What was the rationale for not including smoking?

• Within the UK Biobank, dietary intake has been assessed 4 times during the follow-up period for a subset of participants. How do the genetic variants related to dietary intake identified using the baseline/enrolment (2006-2010) data associate with self-reported intake during the follow-up period assessments (2012-2013; 2014+; 2019+)?

• The authors mention in the results text that contrary to their expectations, the BMI-raising allele was consistently associated with lower reported consumption of energy-dense foods such as meat or fat, and higher reported intake of low-calorie foods. This observation should be discussed further. Do the authors have a possible explanation for this finding?

• Did the authors consider multivariable MR analyses for the full MR results (Table S18)? One of the key findings was a positive effect estimate for alcohol consumption on lung adenocarcinoma risk. It would be good to see multivariable MR analyses adjusted for lifetime smoking behaviour.

• Why did the authors use education as a proxy of SES when Townsend deprivation index may be a more appropriate indicator?

• Did the authors consider validation of the 2-sample MR results in other studies/populations (e.g., FinnGen, Biobank Japan)?

• The paper although generally clearly written has several grammatical errors and inconsistent formatting throughout. A thorough edit of the paper is required.

**********

Have all data underlying the figures and results presented in the manuscript been provided?

Large-scale datasets should be made available via a public repository as described in the PLOS Genetics data availability policy, and numerical data that underlies graphs or summary statistics should be provided in spreadsheet form as supporting information.

Reviewer #1: None

Reviewer #2: Yes

Reviewer #3: Yes

Reviewer #4: Yes

**********

PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

Reviewer #4: No

17 Dec 2021

Submitted filename: responses to reviewers.docx

Click here for additional data file.

20 Jan 2022

Dear Dr Pirastu,

Thank you very much for submitting your Research Article entitled 'Using genetic variation to disentangle the complex relationship between food intake and health outcomes.' to PLOS Genetics.

The manuscript was fully evaluated at the editorial level and by independent peer reviewers. The reviewers appreciated the attention to an important problem, but raised some substantial concerns about the current manuscript. Based on the reviews, we will not be able to accept this version of the manuscript, but we would be willing to review a much-revised version. We cannot, of course, promise publication at that time.

Should you decide to revise the manuscript for further consideration here, your revisions should address the specific points made by each reviewer. We will also require a detailed list of your responses to the review comments and a description of the changes you have made in the manuscript.

If you decide to revise the manuscript for further consideration at PLOS Genetics, please aim to resubmit within the next 60 days, unless it will take extra time to address the concerns of the reviewers, in which case we would appreciate an expected resubmission date by email to plosgenetics@plos.org.

If present, accompanying reviewer attachments are included with this email; please notify the journal office if any appear to be missing. They will also be available for download from the link below. You can use this link to log into the system when you are ready to submit a revised version, having first consulted our Submission Checklist.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Please be aware that our data availability policy requires that all numerical data underlying graphs or summary statistics are included with the submission, and you will need to provide this upon resubmission if not already present. In addition, we do not permit the inclusion of phrases such as "data not shown" or "unpublished results" in manuscripts. All points should be backed up by data provided with the submission.

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool.  PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org.

PLOS has incorporated Similarity Check, powered by iThenticate, into its journal-wide submission system in order to screen submitted content for originality before publication. Each PLOS journal undertakes screening on a proportion of submitted articles. You will be contacted if needed following the screening process.

To resubmit, use the link below and 'Revise Submission' in the 'Submissions Needing Revision' folder.

[LINK]

We are sorry that we cannot be more positive about your manuscript at this stage. Please do not hesitate to contact us if you have any concerns or questions.

Yours sincerely,

Ulrike Peters

Guest Editor

PLOS Genetics

Hua Tang

Section Editor: Natural Variation

PLOS Genetics

A detailed response to the remaining comments from the reviewers is needed.

Furthermore, the response to the first comment of reviewer 3 is not sufficient. Comparing one dietary assessment tool (FFQ) relying on self-report with another dietary assessment tool (24-hr recall) relying on self-report is not proof that the assessment in valid. It would be different if measurements like doubly labeled water or 24 hr urinary nitrogen would be available. Accordingly, it is critical that the authors address these weaknesses and the potential that the exposure misclassification can lead to biased identification of relevant genetic variants and the possible consequences in the discussion. As part of this the authors need to add a discussion on the use of objective biomarkers for dietary intake, such as doubly labelled water, urinary nitrogen, minerals, and vitamins that can be used in calibration studies.

It seems like a strong assumption that the impact of the sizable missingness in the response to the dietary questionnaire is likely modest. This statement needs to be revised and the range of % missing should be added.

Reviewer's Responses to Questions

Comments to the Authors:

Please note here if the review is uploaded as an attachment.

Reviewer #2: The authors have adequately addressed my comments/suggestions.

Reviewer #4: This is a really interesting study with a clever approach to study the role of dietary intakes on health outcomes. The authors are to be commended for the extensive work undertaken for the revised manuscript. I have the following comments:

• The study design is quite complex. Please could the main text include a figure summarising the different aspects of the study design.

• Why was smoking not considered as a health related trait to be assessed in relation with food intake (while education/SES is)? The rationale for the list of exposures could be better described. Smoking is often associated with unhealthy diet pattern but also with weight loss. Could the authors conduct a sensitivity analysis to examine the effect of smoking on food intake? It is very plausible that smoking would influence the intake of food traits and warrant inclusion.

• One of the authors recently published a nice editorial (PMID: 34980908) on the challenges of using genetics in nutritional epidemiological studies. One of the challenges rightly mentioned is taking into account how dietary intakes can change over the lifecourse. The authors should undertake a sensitivity analysis on this using UK Biobank data. Approximately 49,000 UKB participants had repeat diet assessment during an imaging visit ~8+ years after the baseline information was collected (https://biobank.ndph.ox.ac.uk/showcase/field.cgi?id=1369). How stable/consistent were participants dietary intakes between the baseline and imaging visits? Are a similar pattern of genetic signals found when a GWAS of food intake measured during the imaging visit diet is conducted (n=~49,000 participants)?

• There remain several typos throughout the paper and also in the Shiny app. For example, line 72 – “Mendellian”; line 91 – “Hell et al”; throughout – words capitalized mid-sentence. Please ensure that a careful editorial review is undertaken.

**********

Have all data underlying the figures and results presented in the manuscript been provided?

Large-scale datasets should be made available via a public repository as described in the PLOS Genetics data availability policy, and numerical data that underlies graphs or summary statistics should be provided in spreadsheet form as supporting information.

Reviewer #2: None

Reviewer #4: Yes

**********

PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: No

Reviewer #4: No

15 Feb 2022

Submitted filename: Response to reviewers.docx

Click here for additional data file.

21 Feb 2022

Dear Dr Pirastu,

Thank you very much for submitting your Research Article entitled 'Using genetic variation to disentangle the complex relationship between food intake and health outcomes.' to PLOS Genetics.

The manuscript was fully evaluated at the editorial level and by independent peer reviewers. The reviewers appreciated the attention to an important topic but identified some concerns that we ask you address in a revised manuscript

We therefore ask you to modify the manuscript according to the review recommendations. Your revisions should address the specific points made by each reviewer.

In addition we ask that you:

1) Provide a detailed list of your responses to the review comments and a description of the changes you have made in the manuscript.

2) Upload a Striking Image with a corresponding caption to accompany your manuscript if one is available (either a new image or an existing one from within your manuscript). If this image is judged to be suitable, it may be featured on our website. Images should ideally be high resolution, eye-catching, single panel square images. For examples, please browse our archive. If your image is from someone other than yourself, please ensure that the artist has read and agreed to the terms and conditions of the Creative Commons Attribution License. Note: we cannot publish copyrighted images.

We hope to receive your revised manuscript within the next 30 days. If you anticipate any delay in its return, we would ask you to let us know the expected resubmission date by email to plosgenetics@plos.org.

If present, accompanying reviewer attachments should be included with this email; please notify the journal office if any appear to be missing. They will also be available for download from the link below. You can use this link to log into the system when you are ready to submit a revised version, having first consulted our Submission Checklist.

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org.

Please be aware that our data availability policy requires that all numerical data underlying graphs or summary statistics are included with the submission, and you will need to provide this upon resubmission if not already present. In addition, we do not permit the inclusion of phrases such as "data not shown" or "unpublished results" in manuscripts. All points should be backed up by data provided with the submission.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

PLOS has incorporated Similarity Check, powered by iThenticate, into its journal-wide submission system in order to screen submitted content for originality before publication. Each PLOS journal undertakes screening on a proportion of submitted articles. You will be contacted if needed following the screening process.

To resubmit, you will need to go to the link below and 'Revise Submission' in the 'Submissions Needing Revision' folder.

[LINK]

Please let us know if you have any questions while making these revisions.

Yours sincerely,

Ulrike Peters

Guest Editor

PLOS Genetics

Hua Tang

Section Editor: Human Variation

PLOS Genetics

It is not clear why the authors only described missingness for alcohol and coffee intake. To be responsive to the editor's previous comment please add in a supplemental table the % missing for each dietary variable that was assessed and refer to this table in the methods or result section so that readers can judge for themselves about the potential magnitude of this problem. The very short and vague sentence in the discussion about "informative missingness" should also refer to this table.

2 Mar 2022

Submitted filename: Response to reviewers.docx

Click here for additional data file.

8 Mar 2022

Dear Dr Pirastu,

Thank you very much for submitting your Research Article entitled 'Using genetic variation to disentangle the complex relationship between food intake and health outcomes.' to PLOS Genetics.

The manuscript was fully evaluated at the editorial level and by independent peer reviewers. The reviewers appreciated the attention to an important topic but identified some concerns that we ask you address in a revised manuscript

We therefore ask you to modify the manuscript according to the review recommendations. Your revisions should address the specific points made by each reviewer.

In addition we ask that you:

1) Provide a detailed list of your responses to the review comments and a description of the changes you have made in the manuscript.

2) Upload a Striking Image with a corresponding caption to accompany your manuscript if one is available (either a new image or an existing one from within your manuscript). If this image is judged to be suitable, it may be featured on our website. Images should ideally be high resolution, eye-catching, single panel square images. For examples, please browse our archive. If your image is from someone other than yourself, please ensure that the artist has read and agreed to the terms and conditions of the Creative Commons Attribution License. Note: we cannot publish copyrighted images.

We hope to receive your revised manuscript within the next 30 days. If you anticipate any delay in its return, we would ask you to let us know the expected resubmission date by email to plosgenetics@plos.org.

If present, accompanying reviewer attachments should be included with this email; please notify the journal office if any appear to be missing. They will also be available for download from the link below. You can use this link to log into the system when you are ready to submit a revised version, having first consulted our Submission Checklist.

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org.

Please be aware that our data availability policy requires that all numerical data underlying graphs or summary statistics are included with the submission, and you will need to provide this upon resubmission if not already present. In addition, we do not permit the inclusion of phrases such as "data not shown" or "unpublished results" in manuscripts. All points should be backed up by data provided with the submission.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

PLOS has incorporated Similarity Check, powered by iThenticate, into its journal-wide submission system in order to screen submitted content for originality before publication. Each PLOS journal undertakes screening on a proportion of submitted articles. You will be contacted if needed following the screening process.

To resubmit, you will need to go to the link below and 'Revise Submission' in the 'Submissions Needing Revision' folder.

[LINK]

Please let us know if you have any questions while making these revisions.

Yours sincerely,

Ulrike Peters

Guest Editor

PLOS Genetics

Hua Tang

Section Editor: Human Variation

PLOS Genetics

The authors state that they “have added the proportion of samples used compared to the number of people who have responded to the questionnaire.” It is not clear where this information was added. The table that shows these results need to be specified in the paper and the table needs to be added to the main text or supplement. Furthermore, the authors should add the % missing not only compared to the number of people who have responded to the questionnaire but who was asked to respond to the questionnaire as this is the truly critical denominator to judge missingness.

The authors added “Finally, we have excluded people who reported eating certain foods (e.g. beef) less than once a week due to the very large range of different consumptions which this response corresponds to” It is not clear that this is a good approach. It is very unusual to exclude an entire response group from the analysis as this will limited the range of exposure, which in this case is even not that large as it goes from <1 per week to 0 times per week. There are many good approaches to deal with these type of situations, such as evaluate national dietary assessments to estimate the average intake in this group (<1/wk) and evaluate factors that impact these, such as race, ethnicity, sex, BMI,… so that a value can be assigned within population substrata.

The authors added Line 481: “Another source of bias may be due to the missing samples due to either nonresponse or to removal due to the phenotype definition which may induce spurious correlation similarly to sampling bias. Our method however being agnostic to the source of bias, should be able to include these bias and thus select out the SNPs which are associated because of these biases.”

What evidence exist to make the broad claim that the 2nd sentence is true? It seems that this sentence should be deleted.

16 Mar 2022

Submitted filename: Reply to editor.docx

Click here for additional data file.

22 Mar 2022

Dear Dr Pirastu,

We are pleased to inform you that your manuscript entitled "Using genetic variation to disentangle the complex relationship between food intake and health outcomes." has been editorially accepted for publication in PLOS Genetics. Congratulations!

Before your submission can be formally accepted and sent to production you will need to complete our formatting changes, which you will receive in a follow up email. Please be aware that it may take several days for you to receive this email; during this time no action is required by you. Please note: the accept date on your published article will reflect the date of this provisional acceptance, but your manuscript will not be scheduled for publication until the required changes have been made.

Once your paper is formally accepted, an uncorrected proof of your manuscript will be published online ahead of the final version, unless you’ve already opted out via the online submission form. If, for any reason, you do not want an earlier version of your manuscript published online or are unsure if you have already indicated as such, please let the journal staff know immediately at plosgenetics@plos.org.

In the meantime, please log into Editorial Manager at https://www.editorialmanager.com/pgenetics/, click the "Update My Information" link at the top of the page, and update your user information to ensure an efficient production and billing process. Note that PLOS requires an ORCID iD for all corresponding authors. Therefore, please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field.  This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager.

If you have a press-related query, or would like to know about making your underlying data available (as you will be aware, this is required for publication), please see the end of this email. If your institution or institutions have a press office, please notify them about your upcoming article at this point, to enable them to help maximise its impact. Inform journal staff as soon as possible if you are preparing a press release for your article and need a publication date.

Thank you again for supporting open-access publishing; we are looking forward to publishing your work in PLOS Genetics!

Yours sincerely,

Ulrike Peters

Guest Editor

PLOS Genetics

Hua Tang

Section Editor: Human Variation

PLOS Genetics

www.plosgenetics.org

Twitter: @PLOSGenetics

----------------------------------------------------

Comments from the reviewers (if applicable):

----------------------------------------------------

Data Deposition

If you have submitted a Research Article or Front Matter that has associated data that are not suitable for deposition in a subject-specific public repository (such as GenBank or ArrayExpress), one way to make that data available is to deposit it in the Dryad Digital Repository. As you may recall, we ask all authors to agree to make data available; this is one way to achieve that. A full list of recommended repositories can be found on our website.

The following link will take you to the Dryad record for your article, so you won't have to re‐enter its bibliographic information, and can upload your files directly:

http://datadryad.org/submit?journalID=pgenetics&manu=PGENETICS-D-21-00378R4

More information about depositing data in Dryad is available at http://www.datadryad.org/depositing. If you experience any difficulties in submitting your data, please contact help@datadryad.org for support.

Additionally, please be aware that our data availability policy requires that all numerical data underlying display items are included with the submission, and you will need to provide this before we can formally accept your manuscript, if not already present.

----------------------------------------------------

Press Queries

If you or your institution will be preparing press materials for this manuscript, or if you need to know your paper's publication date for media purposes, please inform the journal staff as soon as possible so that your submission can be scheduled accordingly. Your manuscript will remain under a strict press embargo until the publication date and time. This means an early version of your manuscript will not be published ahead of your final version. PLOS Genetics may also choose to issue a press release for your article. If there's anything the journal should know or you'd like more information, please get in touch via plosgenetics@plos.org.

3 May 2022

PGENETICS-D-21-00378R4

Using genetic variation to disentangle the complex relationship between food intake and health outcomes.

Dear Dr Pirastu,

We are pleased to inform you that your manuscript entitled "Using genetic variation to disentangle the complex relationship between food intake and health outcomes." has been formally accepted for publication in PLOS Genetics! Your manuscript is now with our production department and you will be notified of the publication date in due course.

The corresponding author will soon be receiving a typeset proof for review, to ensure errors have not been introduced during production. Please review the PDF proof of your manuscript carefully, as this is the last chance to correct any errors. Please note that major changes, or those which affect the scientific understanding of the work, will likely cause delays to the publication date of your manuscript.

Soon after your final files are uploaded, unless you have opted out or your manuscript is a front-matter piece, the early version of your manuscript will be published online. The date of the early version will be your article's publication date. The final article will be published to the same URL, and all versions of the paper will be accessible to readers.

Thank you again for supporting PLOS Genetics and open-access publishing. We are looking forward to publishing your work!

With kind regards,

Zita Barta

PLOS Genetics

On behalf of:

The PLOS Genetics Team

Carlyle House, Carlyle Road, Cambridge CB4 3DN | United Kingdom

plosgenetics@plos.org | +44 (0) 1223-442823

plosgenetics.org | Twitter: @PLOSGenetics