Selection in Europeans on Fatty Acid Desaturases Associated
with Dietary Changes
Matthew T. Buckley,
1
Fernando Racimo,
1
Morten E. Allentoft,
2
Majken K. Jensen,
3
Anna Jonsson,
4
Hongyan Huang,
5
Farhad Hormozdiari,
6
Martin Sikora,
2
Davide Marnetto,
7
Eleazar Eskin,
6,8
Marit E. Jørgensen,
9,10
Niels Grarup,
4
Oluf Pedersen,
4
Torben Hansen,
4
Peter Kraft,
5
Eske Willerslev,
2
and Rasmus Nielsen*
,1,2
1
Departments of Integrative Biology and Statistics, University of California Berkeley, Berkeley, CA
2
Natural History Museum of Denmark, Univer sity of Copenhagen, Copenhagen K, Denmark
3
Department of Nutrition, Harvard T.H. Chan School of Public Health & Channing Division of Network Medicine, Brigham and
Women’s Hospital, Harvard Medical School, Boston, MA
4
The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, Faculty of Health and Medical
Sciences, University of Copenhagen, Copenhagen Ø, Denmark
5
Program in Genetic Epidemiology and Statistical Genetics, Harvard T.H. Chan School of Public Health, Boston, MA
6
Department of Computer Science, University of California, Los Angeles, CA, USA
7
Department of Molecular Biotechnology and Health Sciences, University of Torino, Turin, Italy
8
Department of Human Genetics, University of California, Los Angeles, CA, USA
9
National Institute of Public Health, University of Southern Denmark, Copenhagen, Denmark
10
Steno Diabetes Center Copenhagen, Gentofte, Denmark
*Corresponding author: E-mail: rasmus_nielsen@berkeley.edu.
Associate editor: John Novembre
Abstract
FADS genes encode fatty acid desaturases that are important for the conversion of short chain polyunsaturated fatty
acids (PUFAs) to long chain fatty acids. Prior studies indicate that the FADS genes have been subjected to strong positive
selection in Africa, South Asia, Greenland, and Europe. By comparing FADS sequencing data from present-day and
Bronze Age (5–3k years ago) Europeans, we identify possible targets of selection in the European population, which
suggest that selection has targeted different alleles in the FADS genes in Europe than it has in South Asia or Greenland.
The alleles showing the strongest changes in allele frequency since the Bronze Age show associations with expression
changes and multiple lipid-related phenotypes. Furthermore, the selected alleles are associated with a decrease in linoleic
acidandanincreaseinarachidonicandeicosapentaenoicacidsamongEuropeans;thisisanoppositeeffectofthat
observed for selected alleles in Inuit from Greenland. We show that multiple SNPs in the region affect expression levels
and PUFA synthesis. Additionally, we find evidence for a gene–environment interaction influencing low-density lipo-
protein (LDL) levels between alleles affecting PUFA synthesis and PUFA dietary intake: carriers of the derived allele
display lower LDL cholesterol levels with a higher intake of PUFAs. We hypothesize that the selective patterns observed in
Europeans were driven by a change in dietary composition of fatty acids following the transition to agriculture, resulting
in a lower intake of arachidonic acid and eicosapentaenoic acid, but a higher intake of linoleic acid and a-linolenic acid.
Key words: selection, evolution, human, genetics, FADS.
Introduction
Long-chain polyunsaturated fatty acids (LC-PUFAs) are im-
portant components of mammalian tissue and are crucial for
a variety of biological processes. They are bioactive elements
of cell membranes and have an important role in neuronal
membrane development (
Marszalek and Lodish 2005; Darios
and Davletov 2006
). With large brains composed mostly of
lipids, humans have a particularly strong requirement for
these fatty acids (Mathias et al. 2012). LC-PUFAs also serve
as precursors for cell signaling molecules including eicosa-
noids, such as prostaglandins, which act as messengers in
the central nervous system and exert control over many
bodily systems including inflammation (
Hester et al. 2014).
LC-PUFA concentration has been linked to infant visual and
brain development (
McCann and Ames 2005)andtorisksof
cardiovascular and coronary heart disease and mortality
(
Patel et al. 2010). The physiologically most important LC-
PUFAs include x-6 (or n-6) PUFA arachidonic acid (ARA;
20:4n-6), and x-3 (or n-3) PUFA s eicosapentaenoic acid
Article
Fast Track
ß The Author 2017. Published by Oxford University Press on behalf of the Society for Molecular Biology and Evolution.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License
(http://creativecommons.org/licenses/by-nc/4.0/), which perm its non-commercial re-use, distribution, and rep roduction in any
medium, provided the original work is properly cited. For commercial re-use, please contact journa ls.permissions@o up.com
Open Access
Mol. Biol. Evol. 34(6):1307–1318 doi:10.1093/molbev/msx103 Advance Access publication March 16, 2017 1307
Downloaded from https://academic.oup.com/mbe/article/34/6/1307/3062804 by guest on 21 August 2022
(EPA; 20:5n-3) and docosahexaenoic acid (DHA; 22:6n-3).
Additionally, x-3 LC-PUFAs are common dietary supple-
ments, despite ongoing debate regarding their potential pre-
ventative role against cancer or heart disease (
MacLean et al.
2006
; Rizos et al. 2012).
Humans can obtain LC-PUFAs directly, particularly by the
consumption of meat, fish, and marine mammals.
Alternatively, these compounds can be synthesized endoge-
nously from the short chain x-6 and x-3 polyunsaturated
fatty acids (SC-PUFAs) linoleic acid (LA; 18:2n-6) and a-lino-
lenic acid (ALA; 18:3n-3) (
fig. 1). These SC-PUFAs are consid-
ered essential and are obtained primarily through the
consumption of vegetab le oils. The rate-limiting steps in
the synthesis of long chain PUFAs from short chain PUFAs
are catalysed by two fatty acid desaturases: delta-5 desaturase
(D5D) and delta-6 desaturase (D6D) (
Cho et al. 1999),
encoded by fatty acid desaturase 1 (FADS1)andfatty acid
desaturase 2 (FADS2), respectively. FADS1 and FADS2 are lo-
cated adjacent to each other on chromosome 11 (11q12-
13.1), and next to another gene encoding a third fatty acid
desaturase (FADS3). The function of FADS3 has yet to be
elucidated, but the gene shares 52% and 62% sequence ho-
mology with FADS1 and FADS2, respectively, and likely re-
sulted from gene duplication (
Marquardt et al. 2000).
Previous studies have suggested that the FADS genes are
targets of natural selection in different human populations.
Mathias et al. (2011, 2012) compared allele frequencies of 80
FADS SNPs across African and non-African populations,
showing that African populations had substantially higher
frequencies of derived alleles associated with more efficient
desaturases (
Mathias et al. 2011). In a subsequent study,
Ameur et al. (2012) showed that the FADS genes span two
distinct linkage d isequilibrium (LD) blocks, one of which con-
tains an ancestral haplotype largely present in Eurasia, and a
derived haplotype that is prevalent in Africa. Both these stud-
ies argued that a selectivesweepbeganintheFADS genes
prior to the human migration out of Africa, ro ughly 85 thou-
sand years ago (kya). Humans migrating out of Africa puta-
tively carried mostly the ancestral ha plotype, which remai ned
in high frequency in non-African populations, while the de-
rived haplotype came close to fixation in Africa. It is unclear
why positive selection for the derived haplotype appears to be
restricted to Africa.
Mathias et al. (2012) suggested that the
emergence of regular hunting of large animals, dated to
50 kya, might have diminished the pressure for humans
to endogenously synthesize LC-PUFAs.
In a subsequent genome-wide scan for positive selection in
the Greenlandic Inuit, the FADS region was found to be the
strongest outlier region based on pattern s of allele frequency
differentiation relative to other populations (
Fumagalli et al.
2015
). The two most highly differentiated SNPs (rs7115739
and rs174570), both located in FADS2, were associated with
decreased concentrations of LC-PUFAs but increased concen-
trations of SC-PUFAs (
Fumagalli et al. 2015). In their tradi-
tional diet, Inuit consume extremely high levels of LC-PUFAs
from fish and marine mammals, ostensibly diminishing their
need to endogenously synthesize these LC-PUFAs, but have a
lowintakeofsomeSC-PUFAssuchaslinoleicacid.The
mutations in the Inuit appear to compensate for a decreased
intake of SC-PUFAs and an increased intake of LC-PUFAs.
Fumagalli et al. (2015) also demonstrated that the selected
alleles were associated with a decrease in weight, height, fasting
serum insulin, and fasting serum LDL cholesterol. The FADS
genes have been associated with metabolic traits in multiple
previous studies (
Bokor et al. 2010; Glaser et al. 2011), but the
strong association with heighthadnotbeennotedpreviously,
presumable because these SNPs segregate at very low frequen-
cies in Europeans. Nevertheless, the effect on height was rep-
licated in European cohorts (
Fumagalli et al. 2015).
More recently,
Kothapalli et al. (2016) studied genomes
from populations in South Asia and showed strong signs of
positive selection for an indel (rs66698963, mislabeled as
rs373263659 in 1000 Genomes Project Phase III data) in
FADS2, with the insertion allele putatively endowing South
Asians with the ability to more efficiently synthesize LC-
PUFAs, possibly as an adaptation to a more vegetarian diet.
Mathieson et al. (2015) carried out a genome-wide scan for
positive selection comparing DNA microarray data from an-
cient and present-day European genomes. They carried out
the selection scan using a linear model aimed at predicting
present-day allele frequencies from the allele frequencies in
ancient source populations from the European Neolithic and
Bronze Ages. They then identified SNPs with allele frequencies
that strongly deviated from those predicted by the genome-
wide pattern. One of these SNPs was rs174546, the derived
allele (C) of which tags the derived haplotype in Africans
(
Ameur et al. 2012) and is in high LD (r
2
¼ 0.978; CEU)
with the derived allele of rs174537, a SNP located in the
middle of the selection peak in
Mathias et al. (2012).The
Mathieson et al. (2015) study provides strong evidence of
selection in the FADS region in Europe over the past
4,000 years, in addition to the patterns of selection already
reported in Africans, South Asians, and the Inuit.
Transitioning from selection-oriented studies to targeted
functional studies,
Pan et al. (2017) recently used a reporter
assay to produce clear evidence that rs174557 is involved in
the regulation of FADS1. The variant also lies within Ameur’s
(2012) haplotype and within a PATZ1 transcription factor
binding site.
FIG.1.x-3 and x-6 polyunsaturated fatty acid synthesis pathway.
Dietary intakes and fatty acid desaturases (FADS1 and FADS2) are
shown in context of the fatty acids that they directly affect.
Buckley et al.
.
doi:10.1093/molbev/msx103 MBE
1308
Downloaded from https://academic.oup.com/mbe/article/34/6/1307/3062804 by guest on 21 August 2022
The aim of this study is to further investigate the selection
signal in Europeans and the functional effects of the selected
alleles. Taking advantage of genome-wide sequencing data
from 101 Bronze Age individuals by
Allentoft et al. (2015)
as well as data from the 1000 Genomes Project (Genomes
Project et al. 2015
), we identify novel potential targets of
selection in the FADS region. Using expression data and
data from multiple GWASs, we investigate the functional
effects of the allelic variants that have increased in frequen cy
in Europe over the past 5,000 years. We show that SNPs
associated with increased expression of FADS1 and increased
production of arachidonic acid and eicosapentaenoic acid
have been favored in Europeans since the Bronze Age. Our
results suggest that selection in the FADS region is complex
and has targeted several loci across different populations.
Results
Origin and Structure of Human Haplotypes in the
FADS Region
As previously observed (Ameur et al. 2012), two distinct LD
blocks span the FADS gene cluster. These blocks are especially
evident when focusing on Europeans (supplementary fig. 1,
Supplementary Material online) The first block (Block 1:
chr11:61547000–61625000) overlaps FADS1 and half of
FADS2, while the second block (Block 2: chr11:61627000–
61673000) overlaps with the rest of FADS2 and all of FADS3
(
fig. 2).
We set out to analyse the selective processes in Europe in
detail, across both space and time. First, we inspected the hap-
lotype structure at this locus using present-day human ge-
nomes from phase 3 of the 1000 Genomes Project (
Genomes
Projectetal.2015). Aspreviously noted, theFADS cluster shows
high amounts of haplotypic variation in present-day humans
(Mathias et al. 2012). In Block 1, this variation is largely attrib-
utable to high differentiation between two haplotype clusters:
a cluster widespread in Africa, largely containing derived alleles
and possibly subject to a selective sweep (
Mathias et al. 2011,
2012
), and an ancestral cluster, which is present across Eurasia.
Interestingly, two archaic human genomes (Altai Neanderthal
and Denisova) that are sister groups to each other genome-
wide, actually cluster with different clusters in this region (
fig. 3
and supplementary figs. 2 and 3, Supplementary Material on-
line): the Denisovan genome clusters with the ancestral
(Eurasian-specific) cluster,while the Altai Neanderthal genome
clusters with the derived cluster, which is prevalent in
Africa. Mathieson (http://mathii. github. io/ research/ 2015/ 12/
14/fads1-selection- and-diet; last accessed September 5, 2016)
argues this pattern could be e xplained by a re-introduction of
the ancestral cluster into Eurasians via introgression from ar-
chaic humans, followed by a second selective event in Eurasia.
We observed, however, that this locus does not show
signatures that could be consistent with a simple model of
introgression, at least directly from the populations to which
the sequenced archaic genomes belong. Unlike other docu-
mented cases of adaptive introgression (
Racimo, Gokhman,
et al. 2017
) inspecting of the haplotype network reveals that
none of the branches connecting the archaic haplotypes to
their most similar present-day human haplotypes are less
than half as large as the branches connecting those same
present-day haplotypes to other present-day human haplo-
types (fig. 3). Additionally, we observe no archaic alleles in this
region that are at low frequency (<1%) in Africans but at high
(>20 %) frequency in any particular non-African population
or continental panel from the 1000 Genomes data—a signa-
ture of archaic adaptive introgression (
Racimo, Marnetto,
et al. 2017
). Finally, this region does not appear as a significant
candidate in recent scans for archaic adaptive introgression in
Eurasians (
Sankararaman et al. 2014, 2016; Vernot and Akey
2014
). Another possibility is th at balancing selection has
maintained the haplotype, however, the haplotype is rela-
tively long, and standard models of long-term balancing
FIG.2.Allele frequency changes across FADS region. Three SNPs exhibiting the greatest allele frequency change are labeled in blue. Prominent SNPs
from other studies are also labeled: rs174546 (
Alexander et al. 2009; Mathieson et al. 2015) in red, SNPs from Fumagalli et al. (2015) in green, and
rs66698963 from Kothapalli et al. (2016) in orange (location only). Red blocks indicate the locations and orientations of FADS1, FADS2 , and FADS3.
Blue blocks indicate the locations of major LD blocks.
Selection in Europeans on Fatty Acid Desaturases Associated with Dietary Changes
.
doi:10.1093/molbev/msx103 MBE
1309
Downloaded from https://academic.oup.com/mbe/article/34/6/1307/3062804 by guest on 21 August 2022
selection predict a narrow peak of increased variability.
Accordingly, standard methods for detecting balancing selec-
tion, which take the spatial pattern along the chromosome
into account, do not find evidence for balancing selection
(supplementary fig. 6, Supplementary Material online)
(
DeGiorgio et al. 2014). This suggests a more complicated
scenario, for example, with balancing selection affecting mul-
tiple sites or with multiple introgression events and/or selec-
tive sweep events in both Africa and Eurasia. At the present
we cannot distinguish between these more complicated
scenarios.
We also aimed to determine where in the haplotype net-
work each of the derived alleles from the SNPs discussed in
this study were located (
fig. 3 and supplementary fig. 3,
Supplementary Material online). We find that the derived
allele (T; 16%, CEU) of rs174570—which showed the strongest
signs of selection in the Greenlandic Inuit—is present in a
non-African haplotype cluster in Block 1, but not present in
Africans or the archaic humans. The derived allele (C; 64%,
CEU) from the putatively selected SNP from
Mathieson et al.
(2015), rs174546, is in the other haplotype cluster, which in-
cludes several African and non-African haplotypes, as well as
the Altai Neanderthal genome. The Denisovan genome and
the other non-African haplotypes carry the ancestral allele at
this site (
fig. 3 and supplementary fig. 2, Supplementary
Material online). The derived allele of rs174594 is present in
a cluster of non-African haplotypes in Block 1, which are
closer to certain African haplotypes. Finally, the derived allele
of rs174455 (A; 65 %, CEU) is present in a mostly non-African
haplotype cluster in Block 2.
Allele Frequency Changes in Europe
We then set out to compare patterns of temporal allele fre-
quency differentiation in Europe, by comparing the allele fre-
quencies of FADS SNPs in present-day CEU from the 1000
Genomes Project (
Genomes Project et al. 2015)and54
Bronze Age Europeans (
Allentoftetal.2015) (supplementary
table 7, Supplementary Material online). The vertical axis of
figure 2 shows the absolute value of allele frequency changes
across the FADS cluster. Below, we refer to the four top SNPs
(rs174594, rs97384, 174455 and rs174465, see
table 1), each
with >17% c hange in allele frequency, as highly differentiated
SNPs (HDSs) and they are labeled in blue in
figure 2. rs174594
and rs97384 are located in FADS2, towards the end of LD
Block 1, and both are in high LD (r
2
¼ 0.917) in the 1000
Genomes CEU panel. rs174455 and rs174465 are located
2.7 kb apart in FADS3 on LD Block 2, and are also in relatively
high LD (r
2
¼ 0.776, CEU) with each other, but less so with
FIG.3.Haplotype network plots of the two LD blocks in the FADS region, using the phase 3 1000 Genomes data, the Altai Neanderthal (Pru¨fer et al.
2014) and Denisova (Meyer et al. 2012) genomes and the inferred human–chimpanzee ancestral haplotype (Paten et al. 2008), built using pegas
(Paradis et al. 2010). Block 1: chr11:61547000–61625000. Block 2: chr11:61627000–61673000. The colors denote the continental populations to
which each haplotype belongs. The size of the pie charts is proportional to log2(n) where n is the number of individuals carrying the haplotype. The
black dots on each connecting line denote the number of differences separating each haplotype from its neighbors. The dotted lines denote the
haplotype clusters in which each of the putatively derived alleles from four interesting SNPs are located: rs174570 (
Fumagalli et al. 2015), rs174546
(Mathieson et al. 2015), rs174594 (HDS from this study) and rs174455 (HDS from this study).
Buckley et al.
.
doi:10.1093/molbev/msx103 MBE
1310
Downloaded from https://academic.oup.com/mbe/article/34/6/1307/3062804 by guest on 21 August 2022
rs174594 and rs97384 (r
2
0.4). The lead SNPs from our
study were not included in the SNP typing platform used
by (
Mathieson et al. 2015) and may therefore have been
missed in that study.
In addition to CEU, we also compared two other groups
comprised of panels from the 1000 Genomes Project
(
Genomes Project et al. 2015), Northern European (FIN,
GBR, CEU) and Southern European (TSI, IBS), to the Bronze
Age data from (
Allentoft et al. 2015)(fig. 4). The ancestral
haplotype in LD Block 1 appears to be more prevalent in
Northern Europeans and Bronze Age Europeans than in
Southern Europeans, resulting in lower allele frequency
changes in the comparison between Northern and Bronze
Age Europeans (
fig. 4A) than in the comparison between
Southern and Bronze Age Europeans (
fig. 4B). The lead SNP
from (
Mathieson et al. 2015) (rs174546) in LD Block 1 is one of
most highly differentiated SNPs between present-day
Northern and Southern Europeans (
fig. 4C), whereas our
HDSs show less differentiation between the present-day pop-
ulations. The method used by
Mathieson et al. (2015) to de-
tect selecti on captured information regarding allele frequency
distributions and changes in allele frequencies from the
Neolithic, through the Bronze age, and today. However, it
also models the geographic distribution of allele frequencies.
The relatively high differentiation between Northern and
Southern Europeans may be one of the signals detected by
the method of
Mathieson et al. (2015). However, we also no-
tice that the Singleton Density Score, a selection statistic de-
signed by
Field et al. (2016) expressly to detect recent (<2kya)
selection (supplementary fig. 5, Supplementary Material on-
line), provide additional evidence of selection, confirming the
claims of recent selection from
Mathieson et al. (2015).
Comparison of Targets of Selection in South Asia,
Europe, and Greenland
In figure 2, we also show lead SNPs and indels from other
previous populatio n genetic studies of the FADS region:
rs174546 from (
Alexander et al. 2009; Mathieson et al.
2015
) (red), two SNPs from (Fumagalli et al. 2015) (green),
and rs66698963 from (
Kothapalli et al. 2016)(orange).Note
that the Allentoft et al. (2015) low-coverage aDNA dataset
does not include called indels so we did not plot the allele
frequency change for rs66698963. rs174546, which also tags
the (Ameur et al. 2012) derived haplotype in Africans is ad-
ditionally in very high LD (r
2
¼0.978, CEU) with the lead SNP
from (
Mathias et al. 2012) (rs174537). These two SNPs and
the two HDSs on LD Block 1 (rs174594, rs97384) are tightly
linked to each other in Europe (supplementary table 1,
Supplementary Material online). The indel found to be under
selection in South Asians (rs66698963) (
Kothapalli et al. 2016),
the top hit in the Greenlander Inuit scan (rs174570)
(
Fumagalli et al. 2015), and HDS rs174455 are not in excep-
tionally high LD with each other or with rs174546 (supple-
mentary table 1, Supplementary Material online).
We observed substantial differences in LD patterns among
populations, as can be seen in supplementary table 1,
Supplementary Material online. For example, rs66698963 is
in relatively strong LD with most other SNPs in the Bengali
from Bangladesh (BEB), but much less so in other populations
such as CEU, presumably due to the selective pressures re-
stricted to South Asia and described by (Kothapalli et al.
A
B
C
FIG.4.FADS Allele Frequency Differences Between Bronze Age
Europe, present-day Northern Europe (CEU, FIN, GBR), and
present-day Southern Europe (IBS, TSI). Note the string of SNPs in
Southern Europeans that show greater allele frequency change. The
vertical line in both plots represents the approximate transition point
between LD Block 1 and LD Block 2.
Table 1. SNP Selected Allele Frequencies Reference.
SNP Ancestral
Allele
Derived
Allele
Allele Under
Putative Positive
Selection in Europe
Frequency of
Selected Allele in
Bronze Age
Europeans
Frequency of
Selected Allele
in 1KGP CEU
rs174546 T C C 51% 64%
rs66698963 þ No data 47%
rs174594 C A A 39% 62%
rs97384 T C C 40% 61%
rs174455 G A A 45% 65%
rs174465 C T T 52% 70%
Selection in Europeans on Fatty Acid Desaturases Associated with Dietary Changes
.
doi:10.1093/molbev/msx103 MBE
1311
Downloaded from https://academic.oup.com/mbe/article/34/6/1307/3062804 by guest on 21 August 2022