Precision Nutrition for Cardiovascular Disease Prevention

Cardiovascular diseases (CVDs) are the leading cause of death, making them a major public health issue [1]. It is estimated that more than 75% of early CVD can be prevented by control of risk factors making CVD prevention the best strategy to improve the quality of life of individuals and to reduce the burden on the healthcare system [2].

Over the past decades, several lifestyle habits (e.g., smoking, physical inactivity, unhealthy diet), metabolic conditions (e.g., dyslipidemia, hypertension, obesity, diabetes), and psychosocial elements (e.g., lower education, lack of access to healthcare) have been identified as CVD risk factors [3]. Consequently, an important part of CVD prevention is achieved by improving lifestyle habits (e.g., stop smoking, exercise more, eating healthier) [2]. To this end, health authorities issued various nutritional recommendations such as reducing saturated fatty acids (SFAs) and sodium intake while increasing fruit and vegetable and whole grain consumption [1]. However, modifications of nutritional and lifestyle habits are not always sufficient and CVD prevention frequently includes pharmacological treatments aimed at reducing low-density lipoprotein cholesterol (LDL-C) (e.g., statins) and blood pressure (e.g., ACE inhibitors) [2].

Although this strategy contributed to reduce CVD mortality in recent years, the large interindividual variability observed in the response to treatment merits further investigations [2, 4]. Non-modifiable risk factors such as age, gender, and genetics are known to influence both efficacy and toxicity of treatments [4]. The Human Genome Project demonstrated that 99.9% of the genome sequence is identical between individuals, the remainder being sequence variations of which single-nucleotide polymorphisms (SNPs) are the most common [5, 6]. Such variants are of particular interest in genome-wide association studies (GWAS) to establish genotype-phenotype associations [7].

Modern medicine’s focus on treating specific symptoms with pharmacological agents often relegates nutrition to the background, whereas it has been widely demonstrated that chronic diseases such as CVD have multifactorial causes and that nutrition is the cornerstone of prevention [1, 8]. One reason for this is the difficulty in establishing a consensus on nutritional recommendations. Unlike drugs, which often contain a single active compound, foods have a complex composition and therefore multiple biological effects that can be difficult to assess [6]. Moreover, the response to food is subject to the same interindividual variability as drug response [8].

Precision nutrition is based on principles like those of precision medicine, namely, that one diet does not fit all [5, 8]. Moreover, it is not limited to the impact of genes on nutrient response; this relationship is reciprocal, and foods impact genome stability, the gut microbiome, and gene expression through epigenetic marks and other transcriptional regulatory mechanisms influencing both the proteome and the metabolome (shown in Fig. 1) [5, 6].

Despite the importance of this topic, health professionals still have little knowledge on precision nutrition [9]. However, surveys show an interest of the general population in personalized treatments when guided by health professionals such as nutritionists and physicians [10].

Therefore, this review aimed to provide an overview of precision nutrition, its key concepts as well as its current and future applications. This review is not intended to be exhaustive of all aspects of precision nutrition and will therefore focus on some of the sources of the interindividual variability as well as discuss their current applications and applicability for CVD prevention.

SNPs can be found in either coding or non-coding regions and can, depending on their position, be silent or functional and influence transcription (at the promoter), splicing (at the intron/exon boundary), or alter the protein structure by changing an amino acid in its sequence (shown in Fig. 2) [5, 11]. Functional SNPs can affect a wide range of cellular functions depending on whether the protein is an enzyme, a transporter, a receptor, or a transcription factor [5, 6]. Although many SNPs are found in non-coding intergenic regions, they can still be associated with a phenotype through genetic linkage and serve as markers for variants of unidentified genes [6, 12].

GWAS and the classical gene candidate approach have identified several SNPs associated with CVD risk factors, allowing targeted studies to include nutritional data [13, 14]. For example, Corella et al. 15‒17 demonstrated that individuals homozygous for the C allele of a polymorphism located on the promoter of the apolipoprotein A2 gene (APOA2 rs5082) had a significantly higher BMI than T allele carriers while consuming a high-SFA diet. More recently, Noorshahi et al. [18] showed that the same gene-diet interaction was also associated with higher LDL-C levels and LDL-C/high-density lipoprotein-cholesterol (HDL-C) ratio in individuals with diabetes. Consequently, these results suggest that individuals with the CC genotype for this APOA2 SNP will benefit from a low-SFA diet to prevent hypercholesterolemia and obesity.

These studies focused on single macronutrients, but it is increasingly common to compare dietary patterns to account for the synergistic effect of foods [1]. The CORDIOPREV intervention study compared a group consuming a Mediterranean diet containing 35% fat, including 22% monounsaturated fatty acids, with another consuming a “low-fat” diet containing 28% fat (12% monounsaturated fatty acids) for 12 months [19]. Findings showed that individuals within the Mediterranean diet group carrying the T allele for a SNP on the cholesterol ester transfer protein gene promoter (CETP rs3764261) had higher HDL-C and lower triglyceride (TG) levels than those with the GG genotype [19]. This observation of gene-diet interactions helps explain why some individuals respond differently than others to the same diet.

Previous examples described gene-diet interactions based on a SNP investigation. However, risk factors such as dyslipidemia or obesity are multifactorial, and their etiology will likely be polygenic [20, 21]. The genetic risk score (GRS) approach assigns a score to individuals based on their number of risk alleles known to be associated with the phenotype of interest [21]. Goni et al. [22] demonstrated the predictive power of a GRS of 16 SNPs associated with obesity. Individuals in the high-risk group (score >7) had higher BMI and body fat index values of 0.93 kg/m² and 1.69%, respectively, and a 1.94-cm larger waist circumference compared with those in the low-risk group (score =7). In addition, results included variations between the two groups for many food categories such as an increase in acute myocardial infarction associated with polyunsaturated fatty acid (PUFA) consumption for the high-risk group and the opposite for the low-risk group [22]. This demonstrates how such predictive tools could serve as a basis for stratification allowing high-risk individuals to modulate their consumption of certain foods based on their genotype. Recently, Vallée Marcotte et al. [23] used fine mapping to refine a previously published GRS of 10 SNPs. Imputation from the 1,000 Genomes Project database allowed the creation of a 31-SNP GRS predicting the risk of being a non-responder to omega-3 PUFA supplementation aimed at lowering TG levels [23].

These examples present genetic testing as a powerful predictive tool enabling development of targeted nutritional recommendations for CVD prevention. Furthermore, because the genetic code remains relatively constant throughout life, these predictions are invariant over time [5]. Still, few studies associating SNPs with CVD risk factors will include nutritional data and even fewer are intervention studies which constitute an essential step in establishing causality of gene-diet interactions [8, 24]]. To this end, large cohort studies including genetic and nutritional data such as the Canadian Longitudinal Study on Aging and recent intervention studies will help advance the field 25‒27. However, most epidemiological studies such as those detailed above assess dietary intakes using self-reported methods with known limitations such as food frequency questionnaires (FFQs) [28]. It would therefore be essential, in the long run, to complement these tools with state-of-the-art analytical methods to establish a reliable and bias-free picture of dietary intake.

Democratization of sequencing in the last decade has made whole genome sequencing very popular, allowing the discovery of rare SNPs [29]. Yet, very few GWAS address other types of variants such as variations in the copy number of a gene, which can play a significative part in multifactorial diseases such as CVD [29]. Furthermore, most GWAS have been performed in populations of European descent and these associations may be more difficult to generalize to other populations [30]. It is therefore imperative for the reliability and generalizability of genetic testing that more GWAS are performed in populations of different ethnic origins. However, identification of novel SNPs also requires more functional studies as well as new phenotyping methods to establish causality and understand the mechanisms underlying these associations [29, 31]. Moreover, most GRSs do not account for complex gene-gene interactions (epistasis) [32, 33]. Therefore, an ideal prediction tool for precision nutrition will take these interactions into account. Finally, while genotyping can be an effective tool on its own, its main limitation comes from the fact that the genetic code alone does not provide information on gene expression, itself influenced by nutrients [34, 35]. Consequently, complementing genetic testing with other methods would be beneficial for the implementation of precision nutrition in CVD prevention.

One reason behind the limited phenotype prediction capabilities of genotyping is the existence of epigenetic marks regulating gene expression [34]. Thus, epigenetics is concerned with mechanisms that modify gene function in a heritable and reversible manner without changing the DNA sequence [36]. This phenomenon, which begins during embryogenesis, is the basis of cellular differentiation leading to the formation of various organs and tissues [37]. Moreover, some of these epigenetic marks will change throughout life, influenced by age and environmental factors such as nutrition and medication, and are therefore sources of interindividual variability [38]. This reciprocal interaction is modulated by three main mechanisms working in concert to activate or inactivate genes: DNA methylation, post-translational modifications of histones, and non-coding RNAs (shown in Fig. 2) [36].

DNA methylation is the most studied and the best understood epigenetic mechanism [34]. In mammals, the addition of a methyl group on the cytosines of cytosine-phosphate-guanine dinucleotides located within the promoter of a gene will inactivate it, whereas being located within its coding sequence will influence splicing during transcription [38]. Some nutrients act as methyl group donors modulating the methylation of several CVD-associated genes [39]. For example, a genome-wide DNA methylation study by Adaikalakoteswari et al. [39] associated vitamin B12 deficiency with reduced methylation of the sterol regulatory element (SREBF1) and LDL receptor genes, thus increasing their expression as well as cholesterol biosynthesis. Similarly, a reciprocal relationship has been shown between vitamin D levels and DNA methylation of genes of the cytochrome P450 family involved in its metabolism [40]. These mechanisms warrant further investigation as interindividual variability could help explain why neither vitamin D nor B12 supplementation have shown to be beneficial for CVD prevention in individuals without deficiencies 41‒43.

Histones are structural proteins on which DNA wraps to form nucleosomes, the basic unit of chromatin [5]. Histones can be modified by addition of chemical groups on their tails (e.g., acetyl, methyl, phosphoryl, ubiquitin) to modulate DNA strand accessibility for transcription [36]. For instance, acetylation of histones will generally allow expression of a gene by opening the chromatin structure, making DNA accessible for transcription [44]. Curcumin is a bioactive compound with cardioprotective properties shown to inhibit acetylation of certain histones resulting in downregulation of CVD-related pro-inflammatory genes such as the receptor for myeloid cell expression (TREM-1) [44, 45]. MicroRNAs (miRNAs) are non-coding RNAs whose function is to bind to messenger RNAs (mRNAs) triggering their degradation and preventing their translation into proteins [34]. For example, it has been reported that omega-3 PUFAs regulate the expression of genes involved in lipid metabolism through miRNAs [46].

The reciprocal relationship between the epigenome and nutrition has the potential to provide complementary information to genotyping, allowing development of targeted nutritional recommendations. However, epigenetic studies including nutritional interventions are scarce and generally focus on DNA methylation, whereas most mechanisms involving histone modifications and miRNAs have been studied in cellular or animal models [34]. Therefore, more intervention studies are needed to establish causality between specific foods and epigenetic marks associated with CVD risk. This will, of course, be complicated by the fact that some of these marks may change over time and that various periods of life appear to be more favorable to such changes [36]. Moreover, epigenetic marks will generally be tissue-specific, thus necessitating the use of blood or oral cells as markers of epigenetic events in hard-to-access tissues, an approach that has some limitations [34]. In addition, the varied chemical composition of the different types of epigenetic markers (e.g., DNA, RNA, proteins) requires different sequencing techniques, which increases the complexity and cost [47, 48]. Therefore, in the short term, epigenomics is still too new to contribute significantly to precision nutrition for CVD prevention.

Some of the points previously outlined in this paper highlighted the importance of using methods complementary to the FFQ to obtain accurate and precise estimates of the dietary intake of study participants. To this end, the metabolome, which represents the set of metabolites contained in a biological system (e.g., cell, organ, organism), is of great interest [49]. Metabolomics involves the analysis of small molecules (<1,500 Da) using nuclear magnetic resonance as well as a combination of different chromatography and mass spectrometry techniques [49]. These analyses can be targeted for better identification and quantification of metabolites or non-targeted to enable discovery of new compounds [50]. Samples can come from specific tissues as well as from blood or urine [50]. This allows for the profiling of metabolomic markers specific to a food or diet to measure compliance in an intervention study or corroborate FFQs of an observational study [51]. For example, Gibbons et al. [52] recently developed a calibration curve capable of predicting orange juice consumption using proline betaine as a citrus biomarker, with which they were able to establish good agreement between consumption predicted by the biomarker measure and FFQs in a cohort of 565 individuals. Similarly, biomarkers exist for several foods as well as for dietary patterns [50]. Thus, this method holds great promise for improving the assessment of dietary intake.

It is also possible to stratify individuals based on metabolomic phenotyping (metabotyping) to provide targeted nutritional recommendations [50]. A recent study by Fiamoncini et al. [53] stratified 70 healthy individuals into two metabotypes (A and B) based on metabolomic profiling consisting of plasma markers of lipid catabolism during a meal-mixture tolerance test. Subjects with both metabotypes were assigned into an intervention group of 40 participants on a hypocaloric diet (by 20%) or a control group to measure weight loss for 12 weeks [53]. After an average weight loss of 5.6 kg, only individuals with metabotype B, who had the least favorable baseline values, showed improvement in glycemic response and markers of metabolic diseases [53]. Thus, metabotyping would be a useful tool to predict the outcome of a nutritional intervention for the prevention of the metabolic syndrome, which comprise several CVD risk factors.

Metabolomics can provide a myriad of information relevant to precision nutrition. However, although obtaining blood and urine samples is relatively straightforward, the information obtained remains quite limited, as a good proportion of blood metabolites are transitory and those in urine are destined for elimination [31]. Moreover, this reduces the time window for sample collection which is less than 24 h for urine or between postprandial and 12 h of fasting for serum or plasma [50, 54]. This is an important limitation, especially since not all metabolites have the same excretion kinetics and only blood contains fat-soluble metabolites [54]. Consequently, a combination of both methods is needed to get a complete picture. Although many databases exist for dietary metabolites, very few biomarkers are validated by intervention studies, an essential step because compounds, once ingested, are metabolized into various intermediates whose origin is particularly difficult to determine within a complex dietary model [51]. Moreover, interindividual variability requires finding metabolomic markers reproducible across populations. Despite these limitations, metabolomics has the potential to significantly advance the implementation of precision nutrition for CVD prevention.

The gut microbiota plays an essential role in nutrient digestion and is an integral part of the innate immune system [55]. This symbiosis provides the organism with various bioactive compounds such as certain vitamins and neurotransmitters in addition to deriving energy from soluble fiber in the form of short-chain fatty acids (e.g., propionate, butyrate) as well as protecting from infection [50]. Everyone has a unique microbiota corresponding to a unique microbiome composed of a large diversity of species of microorganisms with varied genetic makeups [56]. Therefore, the microbiome is an important source of interindividual variability and its composition is influenced by various external elements such as nutrition, stress, or medication [55]. Moreover, it has been widely demonstrated that alterations in the gut microbiota are linked to several diseases, including CVD [57]. Furthermore, it has been reported that such disruptions happening in infancy can still have a significant impact on metabolic health later in life [58].

The microbiome is mainly investigated by two complementary methods using fecal samples [59]. Sequencing of the gene coding for 16s rRNA is the method of choice for identifying species making up a microbiota, and shotgun metagenomic sequencing (SMS) allows the sequencing and cataloguing of all genes in a microbiome [59]. The PREDICT 1 intervention study recently examined the diet-microbiome-cardiometabolic interactions of 1,098 individuals by methods including 16s rRNA, SMS, and machine learning [60]. Results included a strong association between gut microbiome composition and postprandial TG, insulin, and C-peptide levels, but only a weak correlation with glucose levels [60]. However, Zeevi et al. [61] had great success using machine learning to make microbiome phenotyping an accurate tool for predicting postprandial glucose response. Another important feature presented by PREDICT 1 is that a subpopulation of 480 twins (monozygotic and dizygotic) showed that the influence of host genetics on microbiome composition could be limited [60]. Yet, this is only a glimpse of the wealth of information provided by the microbiome regarding host health as well as its predictive capabilities of dietary response.

Microbiome phenotyping could become a very effective predictive tool in addition to those already presented as fecal sampling is quite simple and analytical methods are very mature [59]. However, although 16s rRNA gene sequencing is reliable and inexpensive, it cannot detect eukaryotic microorganisms such as yeast and does not provide information about other genes present [62]. Therefore, it must be complemented by SMS, which generate large amounts of data requiring complex and expensive bioinformatics analysis [62]. In addition, environmental factors such as diet and medication can modify the microbiome composition and influence the results of such phenotyping [56]. Consequently, clear guidelines must be provided to users of such services. Nevertheless, these limitations can likely be overcome to make microbiome phenotyping an integral part of CVD prevention through precision nutrition.

Although all methods for predicting the response to nutritional intervention presented so far have various limitations, some of them are already being put into practice. Genetic testing and gut microbiota profiling have been offered by private companies for several years, and some offer personalized advice through nutritionists [5, 63]. However, some authors raised several ethical considerations regarding the information provided by these tests [5, 63]. Recently, Zeisel highlighted various issues, including interpretation of results by the patient, validity of the data, difficulty of obtaining an informed consent for genetic information, confidentiality of genetic information, and the need to inform the patient of the risks of incidentally found diseases [5]. These issues need to be addressed to find solutions.

This review has shown precision nutrition as the foundation of an approach offering targeted nutritional recommendations for CVD prevention. The main strengths and areas of improvement have been provided for the major sources of interindividual variability. We propose that three elements are essential for a systemic implementation of precision nutrition for CVD prevention: (1) development of data validation standards and clinical practice guidelines (CPGs); (2) establishment of a certified training program; (3) integration into the healthcare system.

The first point consists in creating validated databases to establish CPGs for nutrigenetics, as this method is the most mature and best implemented to date. Keathley et al. [64] recently developed the first nutrigenomic CPGs for three gene-diet associations, including the previously discussed 31 SNP GRS for the plasma TG response to omega-3 fatty acids, using the GRADE methodology and AGREE II approach. This process needs to be applied to other gene-diet interactions allowing targeted nutritional recommendations for different CVD risk factors. Subsequently, this approach should also be applied to metabotyping and microbiome phenotyping so that they can all complement each other. In addition, clear protocols must be established regarding ethical considerations associated with these data. The second point is the development of a certified training program in precision nutrition for nutritionists and physicians, both at the university level and for professionals in clinical practice. We consider essential that this information be used by trained professionals to ensure appropriate guidance to patients. Finally, precision nutrition for the prevention of CVD must be part of the healthcare system to be accessible to all individuals.

Once these steps are completed, addition of other components such as precision lifestyle, including the impact of genes on dietary behaviors and on the response to physical activity, will allow to propose effective and sustainable solutions not only for CVD, but also for other chronic diseases. Consequently, precision nutrition could soon take its rightful place as the cornerstone of preventive medicine.

Figures were made from L.-C.D.’s design by graphic designer Constance Koziej-Lévesque.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

L.-C.D. is funded by graduate studentships from the Chaire de nutrition de l’Université Laval and the Fonds Jean-Paul-Houle. M.-C.V. holds a Canada Research Chair in Genomics Applied to Nutrition and Metabolic Health.

M.-C.V. and L.-C.D. conceived the idea for the manuscript. L.-C.D. performed the literature review and wrote the manuscript. M.-C.V. edited and critically reviewed the manuscript. All authors approved the final version of the manuscript.