CPT1A Missense Mutation Associated With Fatty Acid Metabolism and Reduced Height in GreenlandersCLINICAL PERSPECTIVE
This article requires a subscription to view the full text. If you have a subscription you may use the login form below to view the article. Access to this article can also be purchased.
Background—Inuit have lived for thousands of years in an extremely cold environment on a diet dominated by marine-derived fat. To investigate how this selective pressure has affected the genetic regulation of fatty acid metabolism, we assessed 233 serum metabolic phenotypes in a population-based sample of 1570 Greenlanders.
Methods and Results—Using array-based and targeted genotyping, we found that rs80356779, a p.Pro479Leu variant in CPT1A, was strongly associated with markers of n-3 fatty acid metabolism, including degree of unsaturation (P=1.16×10−34), levels of polyunsaturated fatty acids, n-3 fatty acids, and docosahexaoenic acid relative to total fatty acid levels (P=2.35×10−15, P=4.02×10−19, and P=7.92×10−27). The derived allele (L479) occurred at a frequency of 76.2% in our sample while being absent in most other populations, and we found strong signatures of positive selection at the locus. Furthermore, we found that each copy of L479 reduced height by an average of 2.1 cm (P=1.04×10−9). In exome sequencing data from a sister population, the Nunavik Inuit, we found no other likely causal candidate variant than rs80356779.
Conclusion—Our study shows that a common CPT1A missense mutation is strongly associated with a range of metabolic phenotypes and reduced height in Greenlanders. These findings are important from a public health perspective and highlight the usefulness of complex trait genetic studies in isolated populations.
Over thousands of years, the combined influences of geographical isolation, small population size, the harsh arctic climate, and an almost purely carnivorous diet have shaped the genomes of Inuit. The traditional Inuit diet, consisting predominantly of marine mammals and fish, is highly enriched in fat and protein, but has low levels of carbohydrates.1–3 Relative to Western foods, the dietary fat is particularly rich in omega-3 polyunsaturated fatty acids and long-chained monounsaturated fatty acids.4 Physiologically, Inuit living on a traditional diet have relied on fatty acids and ketone bodies rather than glucose as the main source of energy,5 and any mutation favoring this form of metabolism will have had a strong selective advantage. Indeed, recent studies searching for signatures of selection in genomes of present-day Inuit from Greenland and indigenous Northeast Siberian coastal populations, respectively, provide evidence for positive selection at a cluster of fatty acid desaturase genes and at CPT1A, a gene that regulates mitochondrial long-chained fatty acid oxidation.6–8
See Editorial by McGarrah
Little is known about specific genetic associations with direct measures of fatty acid metabolism in Inuit. Here, we address this question by assessing >200 serum metabolite measures using a high-throughput nuclear magnetic resonance spectroscopy (NMR) metabolomics platform9 and performing genetic association scans in a population-based sample of 1570 Greenlanders.
In 2013, we recruited individuals from 7 towns in Greenland (Figure 1). Eligible participants were >16 years and born in Greenland. Individuals were identified through the Greenlandic Civil Registration System and received a letter inviting them to participate. Written and informed consent was given by all participants and by parents for participants <18 years. The study was approved by the Commission for Scientific Research in Greenland (approval no 2013-17) and by the Danish Data Protection Agency.
On the basis of power calculations done in R, we found that a sample size of ≈1500 is large enough to have adequate power (>80%) to detect common variants (minor allele frequency > 0.05%) with modest to high effect sizes (>0.6 SD) at genome-wide significance level (α=5×10−10) corresponding to Bonferroni correction for ≈100 independent traits at conventional single-trait genome-wide significance (α=5×10−8). We randomly selected 1630 of the recruited individuals for genotyping.
Array Genotyping and Sample Quality Control
DNA was extracted from buffy coats using a Chemagic STAR DNA Buffy Coat 200 kit (PerkinElmer). All samples were genotyped on the OmniExpressExome chip (HumanOmniExpressExome-8v1-2_A; Illumina), which is a genotyping array of 964 043 variants, including ≈700 000 tag variants optimized for genome-wide association studies (GWAS) and >250 000 functional exonic markers. Genotyping was performed using the HiScan System (Illumina), and genotypes were called using the GenCall module of the GenomeStudio Software (Illumina) using default cluster data. Before analysis, we used the software PLINK9 to identify and remove samples with sex discordance, high rates of missing data, duplicates. This led to the removal of 60 samples, leaving 1570 genotyped samples for analysis.
To investigate population structure, estimate admixture proportions, and calculate population branch statistics, we included a set of 209 individuals from the HapMap Consortium10 for joint analysis with the Greenlandic data. The HapMap samples had been genotyped by Illumina using the same genotyping array (HumanOmniExpressExome-8v1-2_A; Illumina) as the Greenlandic samples. The set of HapMap samples used in our analyses came from 44 unrelated Han Chinese individuals from Beijing, China (CHB), 45 unrelated Japanese individuals from Tokyo, Japan (JPT), 60 unrelated Utah residents with ancestry from northern and western Europe (CEU), and 60 unrelated Yoruba individuals from Ibadan, Nigeria (YRI).
Serum Metabolomics and Quality Control
All samples were analyzed using a high-throughput serum NMR metabolomics platform.11 This methodology provides a standard output of 233 direct and derived serum measures,12 including quantitative molecular data on 14 lipoprotein subclasses, their lipid concentrations and composition, apolipoprotein A-I and B, multiple cholesterol and triglyceride measures, various fatty acids, as well as on numerous low-molecular-weight metabolites, including amino acids, glycolysis-related measures, and ketone bodies (a complete list is provided in Table I in the Data Supplement). Samples with low protein content, low glutamine/high glutamate, high lactate, high ethanol, or presence of polysaccharides were flagged during NMR analysis. For flagged samples, metabolite measures were set to missing as recommended by the laboratory, for example, glutamine and pyruvate values were set to missing for samples flagged as having low glutamine/high glutamate concentrations. In addition, we tested all other metabolite measure for correlation with any of the flags. For correlated metabolite measures (P<10−3), the value was set to missing in samples with the corresponding flag.
Genome-Wide Association Analysis
To avoid increased type I error rates and decreased statistical power caused by population structure,13 we used the linear mixed effects model implemented in Genome-Wide Efficient Mixed Model Association14 to test for association. For each variant and each metabolite measure, we modeled the effects of the genotype and additional covariates as fixed effects and admixture and relatedness as random effects. The random effects were assumed to follow a multivariate normal distribution with variance proportional to a relatedness matrix estimated from standardized genotypes from all autosomal variants with an minor allele frequency >5% and <1% missing genotypes.
Before testing for association, we filtered out all variants with minor allele frequencies <1% or >5% missing genotypes and imputed remaining missing genotypes within the sample using Eagle.15 All traits were quantile-transformed, within sexes, to a standard normal distribution. Analysis were then performed using an additive genetic model, that is, with genotypes coded as 0, 1, and 2 denoting the number of minor alleles. The association tests were corrected for sex and age by inclusion of these variables as additional covariates in the mixed-effects model. We performed conditional analysis by further including the genotypes of the variants, which we conditioned on, as additional covariates in this linear mixed model.
To obtain a threshold for significant association, taking into account the number of metabolite measures tested, as well as their intercorrelations, the threshold value for single-trait genome-wide association was divided by the number of independent tests in the NMR data as estimated using the method of Li and Ji.16 We adopted the conventional P value of 5.0×10−8 as the threshold for single-trait genome-wide association. The effective number of phenotypes was estimated to be equal to 83, resulting in a P value threshold for genome-wide significance in the present study of 5.0×10−8/83=6.02×10−10.
Genotyping of Candidate Causal Variant
To follow-up on the main association signal, we performed targeted genotyping of the candidate causal variant rs80356779 using competitive allele-specific polymerase chain reaction chemistry at LGC Genomics (Hoddesdon, United Kingdom).
Anthropometric/Physiological Measurements and Clinical Biochemistry
During the consultation at the sampling location, we recorded the height and weight of the participants, as well as their systolic and diastolic blood pressure. The blood pressure measurements were taken 3 times per participant, and we used the median value in the analysis. We calculated the body mass index (BMI) of the participants based on their measured height and weight. We further performed clinically relevant routine laboratory measurements of 39 additional traits. The complete list of the anthropometric/physiological and clinical traits is shown in Table II in the Data Supplement. We performed single-variant association tests between the candidate causal variant and these 44 additional traits, adopting a P value of 0.05/44=0.0011 as our threshold for trait-wide association.
Admixture and Relatedness Estimation
We estimated admixture proportions for the genotyped samples using the software ADMIXTURE17 with K=2, that is, we assume that the ancestry of all individuals can be explained by 2 distinct populations, the ancestral Inuit population and the European population. We included 60 unrelated CEU samples described above as a population reference. We ran the ADMIXTURE program 30 times with different random starting points to ensure that convergence was reached. The estimated admixture proportions were used as input for RelateAdmix,18 a program estimating relatedness between pairs of individuals while taking admixture into account.
On the basis of the estimated admixture proportions and relatedness described above, we identified all individuals with <2% European ancestry and chose for selection analyses 157 of these such that none were closely related. That is, no pair had a relatedness coefficient r=0.5×k1+k2 larger than 0.125, where k1(k2) is the proportion of the genome where 1(2) chromosome(s) is identical by descent between the 2 individuals. We combined the genotyping data for the 157 minimally related Greenlandic Inuit (GI) individuals with data from 60 unrelated CEU and 44 unrelated CHB samples described above. We then computed per-variant values of Fst for all pairwise combinations of populations (GI, CEU, and CHB) taking into account the sample sizes.19 On the basis of these estimates, we calculated variant-specific, the population branch statistics (originally proposed by Yi et al20) for GI using CHB as sister population and CEU as an out-group.
Exome Sequencing Data
We combined exome sequencing data21 from 104 unadmixed Nunavik Inuit (NUI) individuals from 10 villages of Nunavik (Northern Quebec, Canada) with whole-genome sequencing data from 99 Europeans and 103 Chinese Han individuals from phase III of the 1000 Genomes Project (1KGP). We extracted all high confidence variants (with mean coverage >10× in the NUI samples and within exome targets of 1KGP samples) in a 15-Mb region on chromosome 11 (58–74 Mb) that were polymorphic in the merged data. We considered all single nucleotide variations and 1-bp INDELs within a consensus coding sequence (CCDS release 14) gene plus 6-bp flanking regions, resulting in 1403 known and novel variants, across the 3 populations. For these variants, we calculated allele frequencies in NUI and in the 1KGP European and Chinese samples.
We recruited study participants from 7 regions in Greenland (Figure 1A) and successfully conducted array-based genotyping of 1570 individuals with the Illumina Human OmniExpressExome chip. Demographics of the genotyped participants can be found in Table II in the Data Supplement. To prioritize individuals of Inuit ancestry, we primarily invited individuals born in Greenland to parents also born in Greenland to participate. Even so, we generally observed a high degree of European admixture with a sample average of 27% European ancestry, although all participants had at least 20% Inuit ancestry and many participants had no European ancestry (Figure 1B). To identify genetic variants regulating metabolism in Greenlanders, we conducted genome-wide association scans for 233 serum NMR metabolic measures (summarized in Table I in the Data Supplement ) using a linear mixed effects model to account for relatedness and admixture.14
GWAS Associations With Serum NMR Fatty Acid Measures
The genome-wide association scans identified an extended region on chromosome 11q12.2 to 13.3 strongly associated with several measures of fatty acid metabolism. The most significant association results were seen for fatty acid degree of unsaturation (Figure 2A and 2B).
Variants with association test P values lower than our genome-wide significance threshold of 6.02×10−10 spanned a region wider than 10 Mb (Figure 2C), led by the variant rs1017640 (P=2.85×10−27; Table IV in the Data Supplement). Extended, but weak linkage disequilibrium characterizes this region, and we explored the independence of the signals through conditional analyses, conditioning on the top variant from 4 peaks in the region (Figure I in the Data Supplement; Table V in the Data Supplement). Although the association signal for degree of unsaturation at lead variant rs1017640 remained genome-wide significant when conditioning on the other 3 top variants (rs174570, rs3741395, and rs377432; Figure IA–IC in the Data Supplement; Table V in the Data Supplement), conditioning on rs1017640 left no variant in the region associated (P<10–6; Figure ID in the Data Supplement; Table V in the Data Supplement).
Strong associations were also seen between rs1017640 and several additional measures related to fatty acid metabolism (Table IV in the Data Supplement; Figure II in the Data Supplement). Again, multiple additional association peaks were seen across a ≈10-Mb region, but could be accounted for by rs1017640 (Table V in the Data Supplement). Thus, for the examined phenotypes, we saw little evidence for multiple independent signals in this region.
Genomic inflation factors ranged from 0.993 to 1.087 and showed the highest inflation for the most strongly associated metabolic measures (Figure 2B; Figure II in the Data Supplement). Recalculating the genomic inflation factors while omitting the variants in the region harboring the association signal (chr11:58 Mb–74 Mb) showed no inflation (range, 0.987–1.012 for the most strongly associated metabolic measures; Figure 2B; Figure II in the Data Supplement). Thus, the extended LD in the region could account for the inflation observed initially.
Candidate Causal Variant rs80356779
The lead variant, rs1017640, is intronic in CPT1A, which encodes the liver isoform of carnitine palmitoyltransferase I, a key regulator of mitochondrial long-chained fatty acid oxidation. Previous studies have found that rs80356779, a p.Pro479Leu variant in CPT1A, occurs at derived allele frequencies of >70% in indigenous arctic populations, including Southwest Alaska Yup’ik, Inuit from Greenland, and Canadian Nunavut Inuit.22–24 An exome sequencing study of 100 NUI individuals found that the derived allele of rs80356779 (L479) occurred at a frequency of 95.5% and identified only 2 other coding variants in CPT1A, both of which were synonymous.21 In other populations, L479 is essentially absent; for example, it is seen only in 2 heterozygotes (1 Latino and 1 non-Finnish European) among the 60 706 individuals in the Exome Aggregation Consortium data.25
Because rs80356779 is not present on the OmniExpressExome array, we performed targeted genotyping of this variant in our study sample. We found that L479 was strongly associated with a lower degree of unsaturation (P=1.16×10−34; Figure 3A), a lower concentration of n-3 fatty acids (P=3.37×10−14), a lower concentration of docosahexaoenic acid (DHA) (P=9.22×10−21), lower ratios of polyunsaturated fatty acids, n-3 fatty acids, and DHA to total fatty acids (P=2.35×10−15, P=4.02×10−19, and P=7.92×10−27, respectively), and a higher ratio of monounsaturated fatty acids to total fatty acids (P=9.02×10−16; Table 1). Other fatty acid concentrations and ratios showed less significance or no association (Figure III in the Data Supplement; Table VI in the Data Supplement), and no other NMR measures were associated with genome-wide significance (Table VI in the Data Supplement).
Conditional analyses showed that rs80356779 was still significantly associated with degree of unsaturation when conditioning on the top variants from the 3 other peaks in the region (P<10−17) and strongly associated (P=4.8×10−9) when conditioning on the GWAS lead variant rs1017640 (Table V in the Data Supplement). Conversely, the GWAS lead variant and the top variants from the 3 other peaks were not associated with degree of unsaturation when conditioning on rs80356779 (P>0.02; Figure 3A and 3B; Table V in the Data Supplement). Indeed, no variant in the 13-Mb surrounding region was associated (P<10−4) when conditioning on rs80356779 (Figure 3B). We found similar results for conditional analysis for concentration of n-3 fatty acids, DHA, and for the ratios of PUFAs, monounsaturated fatty acids, n-3 fatty acids, and DHA to total fatty acids (Table V in the Data Supplement). These findings suggest that rs80356779 may indeed be the causal variant underlying the association signal in the locus.
Additional Associated Traits
Next, we tested for association between rs80356779 and 44 additional anthropometric/physiological and clinical traits that were available for the study participants (summarized in Table II in the Data Supplement). We found that L479 was strongly associated with decreased height (P=1.0×10−9), and each copy of L479 corresponded to a 2.1-cm decrease in height (Table 2). We further found that L479 was associated with decreased levels of thyroid-stimulating hormone and with higher levels of alkaline phosphatase and transferrin (P=1.0×10−9, 8.7×10−8, and 2.8×10−7, respectively; Table 2). No other anthropometric/physiological and clinical traits were associated with trait-wide significance, P<0.0011 (Table VII in the Data Supplement). Also for these traits, the GWAS lead variant and the 3 other top variants were not associated when conditioning on rs80356779 (Table VIII in the Data Supplement).
To verify that the identified associations were not artifacts because of population stratification, we divided our study population into different strata of Inuit ancestry according to the estimated admixture proportions. Within each stratum, we then assessed the effect of the variant on the trait. We performed the stratified association analysis for degree of unsaturation and for height. Besides demonstrating that increasing proportion of Inuit ancestry is associated with lower height, the association analysis stratified by Inuit admixture proportion demonstrated that the effect of rs80356779 on height is observed across all strata (Figure 4A). Similarly, the effect of the variant was present across the different strata of samples for degree of unsaturation (Figure 4B).
Signatures of Positive Selection
Whole-genome sequencing of 25 individuals from indigenous Northeast Siberian coastal populations recently identified rs80356779 as the likely driver of a strong signal of positive selection.8 To investigate whether the observed association signal was in concordance with the causative variant being under strong selection, we calculated population branch statistics for the GWAS data. We identified a subset of 157 minimally related Inuit individuals with no European ancestry, whose genotypes we analyzed in combination with genotypes from 60 unrelated individuals of European ancestry and 44 unrelated Chinese Han. We found highly elevated levels of population branch statistics for CPT1A variants, supporting the notion of a functional variant in the region under strong positive selection (Figure 5; Table IX in the Data Supplement). The selection statistics pattern closely matched the association signal: the top associated variants were also top population branch statistics variants (Figure 3A versus Figure 5B). The frequency of L479 was 100% among the 157 GI with no European ancestry (76.3% in the total study population), consistent with possible drive of selection seen in the locus.
Exome Sequencing Data
It is conceivable that an unknown neighboring variant with a similar frequency pattern as rs80356779 in the Inuit and European populations might be able to cause the observed association signal, mediated by linkage disequilibrium. To address this possibility, we consulted exome sequencing data from 104 NUI (northern Quebec, Canada) described in the Methods section. The NUI population is assumed to be a sister population to the GI,26 although some genetic differentiation is expected because of the small population sizes and geographical challenges to migration in the region. As reported previously,21 the frequency of L479 was 95.7% in the NUI and L479 was not found in the 1KGP CEU and CHB individuals. Among all 1403 coding variants in the 58 to 74 Mb region of chromosome 11, we found none that were as frequent in the NUI and at the same time as rare in the CEU as rs80356779. In Table X in the Data Supplement , we list all coding variants with European minor allele frequency <35% that occur at 65% or higher frequency in the NUI.
Circulating metabolites have key roles in many disease-related biological pathways, and understanding the genetic basis of their regulation may translate into improved prevention and clinical care. This study found that rs80356779, a p.Pro479Leu variant in CPT1A, was highly significantly associated with a range of fatty acid metabolism measures in a population-based sample from Greenland. We found strong signatures of positive selection at the locus, most extremely for rs80356779, where the derived allele was fixed in individuals of pure Inuit ancestry within our sample, and absent in the CEU and CHB HapMap populations. Furthermore, we found that each copy of L479 was associated with a 2.1-cm reduction in average height.
In exome sequencing data from a sister population, the NUI, we found no other likely causal candidate variant than rs80356779. Conditional analyses indicated that the associations observed for other variants in the region (including rs174570 in the FADS2 gene) could be accounted for by rs80356779 for all reported significant NMR measures and anthropometric/physiological and clinical traits. These findings support rs80356779 as the likely causative variant, although it remains possible that the association could be caused by a noncoding variant in high LD not captured by the exome sequencing.
The effect of rs80356779 on various traits in Arctic populations has been addressed before,22,23 but with a study design that did not allow correction for population structure. Studies in Inuit from Greenland and Yu’pik Eskimos have reported that L479 was associated with elevated fasting high-density lipoprotein-cholesterol and ApoA1 levels22,23 and with lowered BMI and other obesity measures.22 We were not able to replicate these results. BMI showed no association, and although we found nominally significant associations between rs80356779 and high-density lipoprotein-cholesterol and ApoA1 (Table VI in the Data Supplement; Table VII in the Data Supplement), L479 corresponded to lower rather than higher levels of these metabolites. However, high-density lipoprotein-cholesterol and ApoA1 showed strong positive correlation with overall Inuit ancestry in our sample (P=7.92×10−17 and P=1.01×10−13, respectively), whereas BMI showed negative correlation (P=2.39×10−7). Because the L479 is a proxy for Inuit ancestry, the estimates of the variant’s effects can be highly confounded if the statistical model does not correct for population structure. Thus, uncorrected analyses of our data showed that L479 corresponded to higher levels of high-density lipoprotein and ApoA1 and lower levels of BMI (Table XI in the Data Supplement). Such examples of confounding underline the importance of proper modeling in studies of admixed populations.
Interestingly, in the final stages of the preparation of this article, a GWAS on erythrocyte membrane fatty acid composition among Greenlanders was published.27 The study found that the CPT1A variant rs80356779 was associated with the levels of a range of fatty acids in the phospholipid fraction of erythrocyte membranes. These findings complement our serum metabolomics results in understanding the role of the p.Pro479Leu variant in CPT1A in regulating fatty acid levels in different blood components. Similar to our results, the authors report that L479 was associated with measures of smaller body size.
Our study further highlights the usefulness of complex trait genetic studies in historically isolated, but recently admixed populations. The demographic history of the Greenlandic population makes it possible to investigate the effects of a variant like rs80356779, which reached fixation in the ancestral Inuit population. Not only are the observed effects not an artifact of population structure it would also not have been possible to detect them in a purely Inuit population because all individuals would be homozygous for L479.
CPT1A is located in the mitochondrial outer membrane and is the rate-limiting enzyme responsible for importing long-chain fatty acids into the mitochondria of liver cells.28 During fasting, CPT1A activity increases, allowing fatty acids to enter the mitochondria for energy production by β-oxidation. As fasting continues, the acetyl-CoA resulting from β-oxidation is increasingly used to synthesize ketone bodies (ketogenesis) that are circulated to provide energy for the brain and other tissues.29 In the fed state, CPT1A activity is inhibited by the presence of malonyl-CoA, an intermediate product in fatty acid synthesis. Allosteric binding of malonyl-CoA to the regulatory domain of CPT1A prevents fatty acid entry to mitochondria and thereby effectively inhibits liver β-oxidation and ketogenesis.29 Given the central importance of CPT1A for fatty acid metabolism, deficiencies in CPT1A can have severe consequences.30 It is therefore intriguing that a p.Pro479Leu missense mutation in CPT1A exists in high frequencies in our study sample and in other indigenous Arctic populations. This suggests that the effects of the mutation have historically provided a selective advantage for individuals in these populations.
The variant affects the activity of CPT1A in 2 different ways, as illustrated in Figure IV in the Data Supplement. First, in vitro studies of cultured fibroblasts from homozygous carriers of L479 found that the activity of the enzyme was greatly reduced compared with control cells.24,31 Thus, under low malonyl-CoA concentrations (in the fasting state), the enzymatic activity of CPT1A is reduced, although overall β-oxidation seems to be only moderately decreased.24 Second, however, the mutation also markedly decreases the inhibitory effect of malonyl-CoA on liver β-oxidation,31 apparently by affecting a predicted binding site for malonyl-CoA, formed by Pro-479 together with neighboring amino acids.32,33 Thus, under high malonyl-CoA concentrations (in the fed state), the residual CPT1A activity has been shown to be 3 to 4 times higher in homozygous L479 fibroblasts compared with wild-type cells.31
This ability to rely on fatty acids and ketone bodies for energy even in the nonfasting state is predicted to be highly advantageous in individuals living on a traditional Inuit diet.24,29 In the keto-adapted state, glucose is spared for cells with no or few mitochondria (eg, erythrocytes), whereas the brain is fueled by ketone bodies and most other tissues by fatty acids.5 It is thought that carriers of L479 have had a distinct survival advantage by being able to rapidly attain the keto-adapted state and remain in that state also in periods of lower dietary fat and higher protein intake, where ketosis would be switched off in noncarriers.24 It has also been hypothesized that the lower activity of the mutant enzyme might convey a selective advantage by preventing overproduction of ketone bodies.24 Whatever the specific fitness benefits might be, the selection analysis results presented here add to earlier evidence from Northeast Siberian coastal populations8 that L479 has been the likely target of strong positive selection in indigenous peoples of the Arctic region.
At a time where diet and lifestyles become increasingly westernized in arctic communities, genetic variants that were advantageous historically may now be neutral or even associated with adverse health outcomes. Understanding their effects therefore represents an important public health question. Some reports have found that L479 is associated with increased childhood mortality, but whether this represents a causal effect remains to be investigated.34
Strengths of our study include the comprehensive analysis of a wide range of metabolic, anthropometric/physiological, and clinical traits in a GWAS setting, allowing for careful control of potential population structure biases. Finding an independent arctic replication population with genome-wide genotypes, as well as metabolite measurements, available has not been possible and is a limitation of the study. Also, although our results point to rs80356779 as the likely causal variant, further studies involving, for example, cell line experiments would be needed to establish causality at the locus and to illuminate the molecular mechanisms underlying the observed associations.
In conclusion, we found that the p.Pro479Leu mutation was associated with lower degree of unsaturation, lower levels of DHA and n-3 fatty acids in general, lower ratios of polyunsaturated fatty acids, n-3 fatty acids, and DHA to total fatty acids, and a higher ratio of monounsaturated fatty acids to total fatty acids. It was also strongly associated with lower attained height, possibly relating to the effects of fatty acid metabolism on growth hormone secretion.35 To illuminate the biological mechanisms behind the associations reported here, further functional studies are required. In particular, studies including complete characterization of fatty acid and lipid profiles and direct measurements of CPT1A enzymatic activity together with genotypic data might be revealing. Our findings illustrate how carefully designed studies of populations adapted to extreme dietary or environmental conditions can provide important knowledge about basic human physiology.
We thank all volunteers for their generous participation in this study. We are grateful to Henrik Hjalgrim, Karen Bjørn-Mortensen, Emilie Birch, Anne Christine Nordholm, Marie Mila Broby Johansen, Helle Møller, and Lisbeth Carstensen for assistance with the collection and handling of data and samples. We are also thankful to Dr Allan Meldgaard Lund for helpful discussion. Web Resources: R software, version 3.2.0, http://www.r-project.org/. Exome Aggregation Consortium Browser, http://exac.broadinstitute.org/; accessed April 21, 2016.
Sources of Funding
The study was supported by grants from the Danish Medical Research Council, The Greenlandic Ministry of Education, Church, Culture and Gender Equality, the Maersk Foundation (Fonden til Lægevidenskabens Fremme), and the Aase and Ejnar Danielsens Foundation. The Danish National Biobank is supported by the Novo Nordisk Foundation. Dr Feenstra is supported by an Oak Foundation fellowship. Dr Skotte is supported by a Carlsberg Foundation postdoctoral fellowship.
The Data Supplement is available at http://circgenetics.ahajournals.org/lookup/suppl/doi:10.1161/CIRCGENETICS.116.001618/-/DC1.
- Received September 9, 2016.
- Accepted April 6, 2017.
- © 2017 American Heart Association, Inc.
- Krogh A,
- Krogh M
- Bang HO,
- Dyerberg J,
- Sinclair HM
- Fumagalli M,
- Moltke I,
- Grarup N,
- Racimo F,
- Bjerregaard P,
- Jørgensen ME,
- et al
- Soininen P,
- Kangas AJ,
- Würtz P,
- Suna T,
- Ala-Korpela M
- Loh PR,
- Palamara PF,
- Price AL
- Li J,
- Ji L
- Alexander DH,
- Novembre J,
- Lange K
- Moltke I,
- Albrechtsen A
- Reynolds J,
- Weir BS,
- Cockerham CC
- Yi X,
- Liang Y,
- Huerta-Sanchez E,
- Jin X,
- Cuo ZX,
- Pool JE,
- et al
- Lemas DJ,
- Wiener HW,
- O’Brien DM,
- Hopkins S,
- Stanhope KL,
- Havel PJ,
- et al
- Rajakumar C,
- Ban MR,
- Cao H,
- Young TK,
- Bjerregaard P,
- Hegele RA
- Raghavan M,
- DeGiorgio M,
- Albrechtsen A,
- Moltke I,
- Skoglund P,
- Korneliussen TS,
- et al
- Andersen MK,
- Jørsboe E,
- Sandholt CH,
- Grarup N,
- Jørgensen ME,
- Færgeman NJ,
- et al
- Prasad C,
- Johnson JP,
- Bonnefont JP,
- Dilling LA,
- Innes AM,
- Haworth JC,
- et al
- Brown NF,
- Mullur RS,
- Subramanian I,
- Esser V,
- Bennett MJ,
- Saudubray JM,
- et al
- Morillas M,
- Gómez-Puertas P,
- Rubí B,
- Clotet J,
- Ariño J,
- Valencia A,
- et al
- Morillas M,
- Gómez-Puertas P,
- Bentebibel A,
- Sellés E,
- Casals N,
- Valencia A,
- et al
- Gessner BD,
- Wood T,
- Johnson MA,
- Richards CS,
- Koeller DM
Variation in fatty acid levels is strongly associated with cardiovascular and metabolic health outcomes. An improved understanding of the combined effects of diet and genetics on fatty acid levels is therefore of considerable medical and physiological interest. Indigenous people of the Arctic represent an intriguing study population, having lived for thousands of years in an extremely cold environment relying on a diet highly enriched for marine-derived fat and depleted of carbohydrates. Here, we address the question of how this selective pressure is reflected in the genetic regulation of metabolism in present-day Greenlanders, by combining detailed metabolomics and genomics data in a population-based sample. Using the genome-wide association study approach, we identified a CPT1A missense mutation (rs80356779, p.Pro479Leu) strongly associated with a range of measures related to fatty acid metabolism, including lower degree of fatty acid unsaturation and lower concentrations of docosahexaoenic acid and n-3 fatty acids. CPT1A encodes the liver isoform of carnitine palmitoyltransferase I, a key regulator of mitochondrial long-chained fatty acid oxidation. The p.Pro479Leu variant is common in Arctic populations but virtually absent elsewhere, and it constitutes one of the most pronounced examples of human genetic adaptation. As diet and lifestyle become increasingly westernized in Arctic communities, genetic variants that were historically advantageous may now be associated with adverse health outcomes. Therefore, understanding the effects of p.Pro479Leu represents an important public health question. Furthermore, deeper insights into the molecular-level effects of the variant form of the CPT1A enzyme may provide leads for novel therapeutic approaches for cardiovascular disease.