Greenlandic Inuit show genetic signatures of diet and climate adaptation

See allHide authors and affiliations

Science  18 Sep 2015:
Vol. 349, Issue 6254, pp. 1343-1347
DOI: 10.1126/science.aab2319

Greenlanders' genomes signal a fatty diet

The evolutionary consequences of inhabiting a challenging environment can be seen within the genomes of Greenland Inuit. Fumagalli et al. have found signs of selection for genetic variants in fat metabolism, not just for promoting heat-producing brown fat cells but also for coping with the large amounts of polyunsaturated fatty acids found in their seafood diet (see the Perspective by Tishkoff). Genes under selection in these populations have a strong effect on height and weight of up to 2 cm and 4 kg, respectively, as well as a protective effect on cholesterol and triglyceride levels.

Science, this issue p. 1343; see also p. 1282


The indigenous people of Greenland, the Inuit, have lived for a long time in the extreme conditions of the Arctic, including low annual temperatures, and with a specialized diet rich in protein and fatty acids, particularly omega-3 polyunsaturated fatty acids (PUFAs). A scan of Inuit genomes for signatures of adaptation revealed signals at several loci, with the strongest signal located in a cluster of fatty acid desaturases that determine PUFA levels. The selected alleles are associated with multiple metabolic and anthropometric phenotypes and have large effect sizes for weight and height, with the effect on height replicated in Europeans. By analyzing membrane lipids, we found that the selected alleles modulate fatty acid composition, which may affect the regulation of growth hormones. Thus, the Inuit have genetic and physiological adaptations to a diet rich in PUFAs.

Previous studies have attempted to understand the genetic basis of human adaptation to local environments, including cold climates and a lipid-rich diet (1). A recent study found evidence that a coding variant in CPT1A, a gene involved in the regulation of long-chain fatty acid, has been the target of strong positive selection in native Siberians, possibly driven by adaptation to a cold climate or to a high-fat diet (2). Another study found evidence that adaptation to the traditional hypoglycemic diet of Greenlandic Inuit may have favored a mutation in TBC1D4 that affects glucose uptake and occurs at high frequency only among the Inuit (3). However, knowledge about the genetic basis of human adaptation to cold climates and lipid-rich diets remains limited.

Motivated by this, we performed a scan for signatures of genetic adaptation in the population of Greenland. The Inuit ancestors of this population arrived in Greenland less than 1000 years ago (4), but they lived in the Arctic for thousands of years before that (5). As such, they have probably adapted to the cold Arctic climate and to their traditional diet, which has a high content of omega-3 polyunsaturated fatty acids (PUFAs) derived from seafood (6) and a content of omega-6 PUFAs that is lower than in Danish controls (7).

We analyzed data from previously genotyped Greenlandic individuals (3) by using the Illumina MetaboChip (8), which is an array enriched with single-nucleotide polymorphisms (SNPs) identified in genome-wide association studies (GWASs) associated with cardiometabolic phenotypes. As a result of recent admixture, modern Greenlanders have, on average, 25% genetic European ancestry (9). To get a representative sample of the indigenous Greenlandic Inuit (GI), we analyzed the subset of 191 individuals that had less than 5% estimated European ancestry per individual (0.5% on average) (9). We combined the data from these individuals with the MetaboChip data from 60 individuals of European ancestry (CEU) and 44 Han Chinese individuals (CHB) from the HapMap Consortium (fig. S1) (10).

To detect signals of positive selection, we used the population branch statistic (PBS) (11), which identifies alleles that have experienced strong changes in frequency in one population (GI) relative to two reference populations (CEU and CHB) (5). A sliding window analysis identified several SNP windows with high PBS values, indicative of selection (Fig. 1 and table S1).

Fig. 1 Results from a genome-wide scan for positive selection.

(A) PBS values in windows of 20 SNPs, using a step size of 5 SNPs. The 99.5th and 99.9th percentiles of the empirical distribution are shown as red dashed horizontal lines. Names of genes associated with the highest peaks are shown. (B) Evolutionary trees underlying the strongest signal of selection. The bottom panel shows genomic-average branch lengths based on FST (fixation index, a measure of population genetic differentiation) for GI, CEU and CHB branches (bottom); the top panel shows branch lengths for the SNPs in the window with the highest PBS values, indicating substantial changes in allele frequencies along the GI branch.

The strongest signal of selection is located within a region on chromosome 11 (Fig. 1A) and encompasses five genes: two open reading frames, C11orf10 (TMEM258) and C11orf9 (MYRF); and three fatty acid desaturases, FADS1, FADS2, and FADS3. The SNP with the highest PBS value falls within FADS2. The function of FADS3 is not known; FADS1 and FADS2 encode delta-5 and delta-6 desaturases, which are the rate-limiting steps in the conversion of linoleic acid (omega-6) and α-linolenic acid (omega-3) to the longer, more unsaturated and biologically active eicosapentaenoic acid (EPA, omega-3), docosahexaenoic acid (DHA, omega-3), and arachidonic acid (omega-6). Polymorphisms in FADS1 and FADS2 are associated with increased levels of plasma and erythrocyte delta-5 desaturases in Alaskan Inuit (12) as well as with levels of PUFA in blood and breast milk (13, 14).

We also found signals of selection in a region on chromosome 1 (Fig. 1A), which encompasses WARS2, a mitochondrial tryptophanyl-tRNA synthetase, and TBX15, a transcription factor member of the T-box family. Within this region, the SNP with the highest PBS value is located upstream of WARS2. Polymorphisms in or near WARS2 and TBX15 have been shown to be associated with numerous phenotypes among individuals of European descent, including waist-hip ratio (15). Based on linkage disequilibrium (LD) patterns in Greenlandic Inuit, the results from (15) suggest that the allele that occurs frequently in Greenlandic Inuit may decrease the waist-hip ratio. TBX15 plays a role in the differentiation of brown (subcutaneous) and brite (typically inguinal) adipocytes (16). The latter, upon stimulation by exposure to cold, can differentiate into cells capable of expressing UCP1 (uncoupling protein 1), which produces heat by lipid oxidation. Therefore, TBX15 may be associated with adaptation to cold in Inuit.

FN3KRP shows evidence of selection as well (Fig. 1A). FN3KRP encodes an enzyme that catalyzes fructosamines, psicosamines, and ribulosamines. This protein protects against nonenzymatic glycation, an oxidative process that is associated with various pathophysiologies (17). A high intake of PUFAs is associated with increased oxidative stress (18); it is possible that the alleles affected by selection in FN3KRP counteract the negative fitness caused by a PUFA-rich diet. A list of additional candidate regions under positive selection is presented in tables S2 and S3.

To corroborate our results from the SNP chip–based analysis described above, we also calculated PBS values (table S4) for exome sequencing data from 18 unrelated GI individuals (3), combined with data from 85 CEU individuals and 97 CHB individuals from the 1000 Genomes Project (fig. S1) (19).

These analyses identified two high-scoring genes (table S5): DSP, a gene associated with cardiomyopathy (20), and ANGPTL6, a gene that counteracts high-fat diet–induced obesity and related insulin resistance through increased energy expenditure (21). Gene ontology enrichment analyses of genes under selection revealed enriched muscle- and heart-development categories, similar to those positively selected in polar bears (table S6) (5, 22).

In addition, these analyses reproduced the strong signal observed in the FADS1-FADS2-FADS3 region, even though the SNPs with the highest PBS values are not detected by the system used for exome capture (Agilent SureSelect; fig. S2), and this region has the SNP with the strongest signal of selection (i.e., highest PBS value) in any of the data analyzed. We therefore focused on this region for the rest of this study. On the basis of an inferred demographic model (5), we estimated a divergence time between CHB and GI of 23,250 years before the present (yr B.P.), unidirectional gene flow from GI to CHB at some point in the history of these populations, and a reduced effective population size of GI (effective population size = 1550). The estimated model (fig. S3A) fits the observed joint site frequency spectrum (fig. S4), and the PBS value for the FADS region is a strong outlier, corroborating the idea that selection probably has affected this region (fig. S5).

Using an approximate Bayesian computation approach, we also estimated the starting time and intensity of selection, s (5). Because of the high LD within the region and the fact that our data were from SNP chip (fig. S6), we could not pinpoint the causative SNP(s) by means of population genetic analyses; we therefore used the SNP with the highest PBS value (reference SNP identification number rs74771917) as a proxy. This SNP has a derived allele frequency of 0.98 in GI, 0.025 in CEU, and 0.16 in CHB. Our analyses produced maximum a posteriori probability (MAP) estimates of the selection starting time, 19,751 yr B.P. [95% Bayesian credible interval (BCI): 2499 to 22,771 yr B.P.] (figs. S3B and S7), and of s, 3.13% (95% BCI: 0.98 to 19.49 %) (fig. S3C). These results suggest that selection began to act on these genes long before the earliest settlement of Inuit in Greenland (4). In population samples from the HGDP-CEPH (Human Genome Diversity Project–Centre d’Etude du Polymorphisme Humain) database, the selected allele of rs74771917 has much higher frequencies among Native Americans than it does among East Asians (fig. S8) (23), suggesting that selection began to act before the Inuit split from the Native Americans, when their common ancestors lived in or around Beringia (24).

Six SNPs in the FADS region (Table 1) have PBS values above 2, suggesting that they have been subjected to strong selection. One of these SNPs, rs174570, is associated with circulating high-density lipoprotein, low-density lipoprotein (LDL), and total cholesterol levels in Europeans (25). We therefore tested for associations between the top six SNPs and 13 metabolic and anthropometric phenotypes in Greenlanders by analyzing data from the Greenlandic cohorts IHIT (Inuit Health in Transition) and B99 (Greenland Population Study 1999), which include 2733 and 1331 genotyped individuals, respectively (3). We analyzed the cohorts separately, combined the results in a meta-analysis (5), and found marginally significant associations with multiple phenotypes, including body-mass index, fasting serum insulin, and fasting serum LDL cholesterol (tables S7 to S12). In all cases, the derived (selected) allele was associated with a reduction in the phenotypic value. The strongest association was with body weight (P = 1.1 × 10−6; rs7115739) and height (P = 0.00012; rs7115739) (table S10). Both of these associations remained significant after Bonferroni correction for testing for association between 13 phenotypes and six SNPs. To further validate the association with height, we genotyped an additional Greenlandic cohort, known as BBH, consisting of 541 Greenlandic individuals who live in Denmark and for whom height information is available. When we added these data to the meta-analysis of height, the association signal for rs7115739 became even stronger (P = 4.6 × 10−7). Moreover, the per-allele effect size estimates for the derived allele for height and weight are –0.66 cm and –2.2 kg in IHIT and –1.2 cm and –2.4 kg in B99 (Fig. 2, A and B, and table S10). As mentioned, the statistical method that we used accounts for admixture. Furthermore, we observed an effect both in Greenlanders with little or no European ancestry and in Greenlanders with more than 40% European ancestry when we stratified the data on the basis of ancestry proportions, which we would not expect if the association signal was caused by admixture in our data (fig. S9). These observations indicate that our association results are not caused by insufficient correction for admixture.

Table 1

Annotation for the top six SNPs under positive selection in Greenlandic Inuit. DAFs for each population (CEU, CHB, and GI) and PBS values are reported, along with the genomic position for each SNP.

View this table:
Fig. 2 The effect of rs7115739 and rs174570 on weight and height.

(A) The effect of rs7115739 on weight in the Greenlandic cohorts IHIT and B99. Shown is the mean value stratified by the genotypes of rs7115739 (top) and the estimated effect of carrying one and two copies of the derived allele, respectively (bottom). The effect-size estimates are adjusted for admixture and other confounding factors and were obtained using a linear mixed model applied to untransformed phenotype measurements. Unlike the estimates in the text and table S10, the estimates shown here were obtained without assuming an additive effect. Error bars, ±1 SE. (B) As in (A), but for height. (C) Effect sizes for height for the derived allele of rs7115739 in three Greenlandic cohorts and seven European cohorts (SDC, Steno Diabetes Center). Point estimates are shown as points and 95% confidence intervals are shown as horizontal bars. For each of the two geographic regions, the results from a meta-analysis of all the cohorts from the region are also shown. N indicates the number of individuals analyzed; DAF, derived allele frequency. For the Greenlandic cohorts, the effect sizes were estimated from height measurements that were quantile-transformed to a standard normal distribution. For the European cohorts, height was analyzed as sex-specific standard (z) scores. Hence, the effect sizes from the two geographic regions are not directly comparable. (D) As in (C), but for rs174570.

The six SNPs with the highest PBS values are also polymorphic in Europeans (Table 1). However, because most of the identified SNPs have low allele frequencies in Europeans, they may have been missed by GWAS studies. When combining seven European cohorts, including GIANT (Genetic Investigation of Anthropometric Traits; 26), we found associations with lower height in carriers of the derived T-allele for rs7115739 (n = 207,300; P = 0.000741) and rs174570 (n = 263,451; P = 1.24 × 10−5) (Fig. 2, C and D, and table S13). The meta-analysis–based effect sizes are equivalent to –0.35 and –0.12 cm for rs7115739 and rs174570, respectively. In contrast, we found no evidence that the six SNPs are associated with weight in Europeans. These results are consistent with results that we obtained when we explicitly tested for differences in effect sizes between Europeans and Greenlandic Inuit (table S14): We found no evidence of a difference in effect size for height for rs7115739 (P = 0.44), but we found significant evidence for a difference in effect size for weight (P = 0.025 and P = 0.012 for rs7115739 and rs174570, respectively), with little or no effect on weight in Europeans. The associations with height in Europeans are unexpected, because this locus was not found to be significant genome-wide in the recent GIANT study of the height of more than 170,000 Europeans (26). In addition to the associations with height, we also found known associations with low fasting serum levels of insulin, total cholesterol, and LDL cholesterol for European carriers of low-frequency–derived alleles of FADS1 variation, suggesting that there may be a protective effect of these variants on cardiometabolic phenotypes (table S13).

To further elucidate the possible functional effects of the alleles of rs7115739 and rs174570, we investigated associations with red blood cell–membrane lipid composition, which reflects fatty-acid intake from the preceding 2 to 4 months and which has previously been measured in IHIT, the largest of our Greenlandic cohorts (27). We found significant associations with multiple different fatty acids (fig. S10 and tables S15 and S16). Particularly, we found that the selected alleles are significantly associated with an increase in the concentration of eicosatetraenoic acid (ETA, 20:4n-3) and other omega-3 fatty acids upstream in the omega-3 synthesis pathway, before conversion to EPA (20:5n-3), but a decrease in the concentration of both EPA and omega-3 docosapentaenoic acid (DPA, 22:5n-3), with no significant effect on DHA (22:6n-3) (Fig. 3). These results are consistent with previous observations of linked alleles in Europeans (28). The conversion of ETA to EPA is catalyzed by delta-5 desaturases encoded by FADS1, and EPA is a major dietary omega-3 fatty acid in the traditional Inuit diet (18). Hence, these results suggest that selection affecting the fatty acid desaturases may have compensated for a high dietary intake of EPA.

Fig. 3 Results of testing for association between the fatty acids in the omega-3 and omega-6 synthesis pathways and each of two SNPs, rs7115739 and rs174570.

The omega-6 (top) and omega-3 (bottom) synthesis pathways are depicted with circles for each fatty acid and arrows for each synthesis step. For each fatty acid, P values for tests of association and effect directions for the derived allele are illustrated by the colors on the left (rs7115739) and right (rs174570) halves of the circle. Green text indicates in which of the synthesis steps FADS1 and FADS2 play a role. Arrows outside the boxes are simplified indications of where different types of diet enter the two pathways.

The changes in the concentration of omega-6 fatty acids mirror those of omega-3 fatty acids (Fig. 3). This might be expected, given that the same enzymes (encoded by FADS1 and FADS2) are involved in both the omega-3 and omega-6 biosynthesis pathways. The similar changes in concentration could therefore be a side effect of selection, driven by a omega-3 PUFA–rich diet. However, selection may also have worked directly on omega-6 fatty acid concentrations early in the ancestral history of Inuit and Native Americans, in the context of a late Paleolithic diet rich in meat from land mammals.

Both rs7115739 and rs174570 show strongly significant associations in conditional analyses where we adjusted for the effects of the other SNP and of rs174602. The remaining three highest-PBS SNPs are in strong LD with rs7115739 in IHIT and would produce similar results. This suggests that there are either multiple causative SNPs or that both rs7115739 and rs174570 are in strong LD with the causal SNP(s).

The challenging environmental conditions of the Arctic have probably imposed strong selective pressures on the Inuit and their ancestors. In all the data that we analyzed, the most pronounced allele-frequency difference between Inuit and other populations was found in a cluster of fatty acid desaturases—FADS1, FADS2, and FADS3—although it is possible that even more extreme differences are present in noncoding regions not covered by our exome data. The FADS region has probably been under selection, driven by a diet high in PUFAs. The FADS genes have previously been hypothesized to be under selection in other populations in response to dietary changes (28, 29), suggesting that these genes in general play an important role in human adaptation to dietary regimes. Our results also show that genetic variants in fatty acid desaturases have a strong effect on height, probably because of the effect of fatty acid composition and concentration on the regulation of growth hormones (30). Previous studies (31) have shown that fish oil supplementation is associated with increased concentrations of plasma insulin-like growth factor–1. This study illustrates the utility of evolutionary studies of locally adapted populations for understanding the genetic basis of phenotypic variation among humans.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S14

Tables S1 to S17

References (3267)


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.Materials and methods are available as supplementary materials on Science Online.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
  32. 32.
  33. 33.
  34. 34.
  35. 35.
  36. 36.
  37. 37.
  38. 38.
  39. 39.
  40. 40.
  41. 41.
  42. 42.
  43. 43.
  44. 44.
  45. 45.
  46. 46.
  47. 47.
  48. 48.
  49. 49.
  50. 50.
  51. 51.
  52. 52.
  53. 53.
  54. 54.
  55. 55.
  56. 56.
  57. 57.
  58. 58.
  59. 59.
  60. 60.
  61. 61.
  62. 62.
  63. 63.
  64. 64.
  65. 65.
  66. 66.
  67. ACKNOWLEDGMENTS: We thank the Greenlandic participants and the funding agencies and research centers that made this study possible: the Human Frontiers in Science Program Organization (grant LT00320/2014); the Danish Council for Independent Research (grant DFF-YDUN); the Villum Foundation; the Steno Diabetes Center; NIH (grant R01-HG003229); the Leverhulme Programme Grant (grant RP2011-R-045); the University of California–Merced startup funds; Karen Elise Jensen’s Foundation and NunaFonden, which supported the collection of data from the Greenlandic cohorts; and the Novo Nordisk Foundation Center for Basic Metabolic Research, which is an independent research center at the University of Copenhagen and is partially funded by an unrestricted donation from the Novo Nordisk Foundation ( We also thank T. Lauritzen and A. Sandbæk for the use of the ADDITION (Anglo-Danish-Dutch Study of Intensive Treatment In People with Screen Detected Diabetes in Primary Care) cohort. The Vejle Diabetes Biobank was funded by the Danish Medical Research Council and Vejle Hospital. The genotyping and exome sequencing data from this project are available to researchers who have received ethics approval from the Greenland Research Ethics Committee ( and can be obtained by contacting T.H.
View Abstract

Navigate This Article