Content uploaded by Matthew T Webster
Author content
All content in this area was uploaded by Matthew T Webster on Feb 13, 2015
Content may be subject to copyright.
LETTER doi:10.1038/nature11837
The genomic signature of dog domestication reveals
adaptation to a starch-rich diet
Erik Axelsson
1
, Abhirami Ratnakumar
1
, Maja-Louise Arendt
1
, Khurram Maqbool
1
, Matthew T. Webster
1
, Michele Perloski
2
,
Olof Liberg
3
, Jon M. Arnemo
4,5
,A
˚ke Hedhammar
6
& Kerstin Lindblad-Toh
1,2
The domestication of dogs was an important episode in the
development of human civilization. The precise timing and loca-
tion of this event is debated
1–5
and little is known about the genetic
changes that accompanied the transformation of ancient wolves
into domestic dogs. Here we conduct whole-genome resequencing
of dogs and wolves to identify 3.8 million genetic variants used to
identify 36 genomic regions that probably represent targets for
selection during dog domestication. Nineteen of these regions con-
tain genes important in brain function, eight of which belong to
nervous system development pathways and potentially underlie
behavioural changes central to dog domestication
6
. Ten genes with
key roles in starch digestion and fat metabolism also show signals
of selection. We identify candidate mutations in key genes and
provide functional support for an increased starch digestion in
dogs relative to wolves. Our results indicate that novel adaptations
allowing the early ancestors of modern dogs to thrive on a diet rich
in starch, relative to the carnivorous diet of wolves, constituted a
crucial step in the early domestication of dogs.
Domestic animals are crucial to modern human society, and it is likely
that the first animal to be domesticated was the dog. Claims of early,
fossilised dog remains include a 33,000-year-old doglike canid from
the Altai Mountains in Siberia
1
, whereas fossils dating from 12,000–
11,000 years BP found buried together with humans in Israel
2
could rep-
resent the earliest verified dog remains. Patterns of genomic variation
indicate that dog domestication started at least 10,000 years BP
3,4
in south-
ern East Asia
4
ortheMiddleEast
5
. Dog domestication may however have
been more complex, involving multiple source populations and/or back-
crossing with wolves.
It is unclear why and how dogs were domesticated. Humans may
have captured wolf pups for use in guarding or hunting, resulting in
selection for traits of importance for these new roles. Alternatively, as
humans changed from a nomadic to sedentary lifestyle during the
dawn of the agricultural revolution, wolves may themselves have been
attracted to dumps near early human settlements to scavenge
6
. Natural
selection for traits allowing for efficient use of this new resource may
have led to the evolution of a variety of scavenger wolves that con-
stituted the ancestors of modern dogs. Regardless of how dog domesti-
cation started, several characteristics separating modern dogs from
wolves, including reduced aggressiveness and altered social cognition
capabilities
7
, suggest that behavioural changes were early targets of this
process
6
. Dogs also differ morphologically from wolves, showing
reduced skull, teeth and brain sizes
6
. Artificial selection for tameness
in silver foxes indicates that selection on genetic variation in develop-
mental genes may underlie both behavioural and morphological
changes, potentially representing an important mechanism through-
out animal domestication
7,8
.
At present, only a handful of genes separating wild from domestic
forms have been identified in any domestic animals, including coat
colour variants in MC1R in pig
9
and a mutation in TSHR likely to affect
seasonal reproduction in chicken
10
, but to our knowledge in dogs no
genome-wide sequence-based searches have been performed until
now. To identify genomic regions under selection during dog domesti-
cation we performed pooled whole-genome resequencing of dogs and
wolves followed by functional characterization of candidate genes.
Uniquely placed sequence reads from pooled DNA representing
12 wolves of worldwide distribution and 60 dogs from 14 diverse breeds
(Supplementary Table 1) covered 91.6% and 94.6%, respectively, of
the 2,385 megabases (Mb) of autosomal sequence in the CanFam 2.0
genome assembly
11
. The aligned coverage depth was 29.83for all dog
pools combined and 6.23for the single wolf pool (Supplemen-
tary Table 1 and Supplementary Fig. 1). We identified 3,786,655 putat-
ive single nucleotide polymorphisms (SNPs) in the combined dog and
wolf data, 1,770,909 (46.8%) of which were only segregating in the dog
pools, whereas 140,818 (3.7%) were private to wolves (Supplementary
Table 2). Similarly we detected 506,148 short indels and 26,619 copy-
number variations (CNVs) (Supplementary Files 1 and 2). We were
able to experimentally validate 113 out of 114 tested SNPs (Sup-
plementary Table 3 and Supplementary Discussion, section 1).
To detect signals of strong recent selection we searched the dog
genome for regions with reduced pooled heterozygosity (H
P
)
10
and/
or increased genetic distance to wolf (F
ST
). As evident from the skewed
distribution of heterozygosity scores in dog relative to wolf (Fig. 1a and
Supplementary Fig. 2), a major challenge to this approach is to sepa-
rate true signals of selection from those caused by random fixation of
large genomic regions during the formation of dog breeds
11
. We alle-
viate this problem by combining sequence data from all dog pools
before selection analyses and require that detected signals span at least
200 kilobases (kb; Methodsand Supplementary Discussion, sections 2
and 3). Given the complex and partly unknown demographic history
of dogs, it is furthermore difficult to assign strict thresholds that dis-
tinguish selection and drift. We propose that the best way to validate
regions detected here is to study genetic data from additional indivi-
duals and provide evidence for functional change associated with
putatively selected regions. Eventually, indications that similar path-
ways changed during independent domestication events may provide
conclusive evidence for selection. Here we Z-transform the autosomal
H
P
(Z(H
P
)) and F
ST
(Z(F
ST
)) distributions (see Supplementary Dis-
cussion, section 4 for an analysis of the X chromosome) and focus
our description of putatively selected regions to those that fall at least
five standard deviations away from the mean (Z(H
P
),25 and
Z(F
ST
).5), as these represent the extreme ends of the distributions.
By applying these thresholds we identified 14 regions in the dog
genome with extremely low levels of heterozygosity (average length 5
400 kb, average H
P
50.036 (range 0.015–0.056), average autosomal
H
P
50.331) (Fig. 1c and Supplementary Table 4) and 35 regions
with strongly elevated F
ST
values (average length 5340 kb, average
1
Science for Life Laboratory, Department of Medical Biochemistry and Microbiology, Uppsala University, 75237 Uppsala, Sweden.
2
Broad Institute of Massachusetts Institute of Technology and Harvard,
Cambridge, Massachusetts 02139, USA.
3
Grimso
¨Wildlife Research Station, Department of Ecology, Swedish University of Agricultural Sciences, 73091 Riddarhyttan, Sweden.
4
Department of Forestry and
Wildlife Management, Faculty of Applied Ecology and Agricultural Sciences, Hedmark University College, Campus Evenstad, NO-2418 Elverum, Norway.
5
Department of Wildlife, Fish and Environmental
Studies, Faculty of Forest Sciences, Swedish University of Agricultural Sciences, 901 83 Umea
˚, Sweden.
6
Science for Life Laboratory, Department of Clinical Sciences, Swedish University of Agricultural
Sciences, 75651 Uppsala, Sweden.
360 | NATURE | VOL 495 | 21 MARCH 2013
Macmillan Publishers Limited. All rights reserved
©2013
F
ST
50.734 (range 0.654–0.903), average autosomal F
ST
50.223)
(Fig. 1b and Supplementary Table 5). All F
ST
regions are characterized
by low levels of heterozygosity in eitherdog or wolf (although all do not
pass the Z(H
P
),25 threshold), indicating that the two statistics
detect the same events (Methods and Supplementary Discussion,
sections 2 and 3). In total, 36 unique autosomal candidate domestica-
tion regions (CDRs) containing 122 genes were identified by the
two approaches combined (Supplementary Table 6 and Fig. 1b, c).
None of these regions overlaps those of a previous genotype-based
study
5
(Supplementary Discussion, section 3), stressing the im-
portance of identifying domestication regions directly by sequencing
or by comprehensively ascertaining SNPs in wild ancestors before
genotyping.
We searched for significantly overrepresented gene ontology terms
among genes in autosomal CDRs and identified 25 categories, repre-
senting several groups of interrelated terms (Table 1 and Supplemen-
tary Table 7), none of which was indicated in a separate analysis of
selection in wolf (Supplementary Discussion, section 8). The most
conspicuous cluster (11 terms) relates to the term ‘nervous system
development’. The eight genes belonging to this category (Supplemen-
tary Tables 7 and 8) include MBP,VWC2,SMO,TLX3,CYFIP1 and
SH3GL2, of which several affect developmental signalling and synaptic
strength and plasticity
12–16
. We surveyed published literature and iden-
tified 11 additional CDR genes with central nervous system function
(Supplementary Table 9), adding to a total of 19 CDRs that contain
brain genes. These findings support the hypothesis that selection for
altered behaviour was important during dog domestication and that
mutations affecting developmental genes may underlie these changes
7
.
The gene ontology analysis also pinpointstwo genes involved in the
binding of sperm and egg: ZPBP encodes the zona pellucida binding
protein that mediates binding of sperm to the zona pellucida glycopro-
tein layer (ZP) of the egg, and ZP2 codes for one of the proteins that
make up ZP itself. In addition, a CDR on chromosome 6 encompasses
PDILT that also affects binding of sperm to ZP
17
, altogether indicating
that sperm competition may have been an important evolutionary
force during dog domestication
18
.
Overrepresented terms ‘starch metabolic process’, ‘digestion’ and
‘fatty acid metabolism’ include genes involved in starch digestion
(MGAM)andglucoseuptake(SGLT1), as well as a candidate gene for
−8
Z(HP)DOG
−8 −4 0 4
0
500
1,000
1,500
μ = 0
σ = 1
Z(HP)WOLF
−8 −4 0 4
0
500
1,000
1,500
μ = 0
σ = 1
Z(FST)
−4 0 4 8
0
500
1,000
1,500
μ = 0
σ = 1
Number of 200-kb windows
a
0
2
4
6
8
Z(F
ST
)
−6
−4
−2
0
Z(H
P
)
DOG
b
c
Figure 1
|
Selection analyses identified 36 candidate domestication regions.
a, Distribution of Z-transformed average pooled heterozygosity in dog
(Z(H
P
)
DOG
) and wolf (Z(H
P
)
WOLF
) respectively, as well as average fixation
index (Z(F
ST
)), for autosomal 200 kb windows (s, standard deviation; m,
average). b, The positive end of the Z(F
ST
) distribution plotted along dog
autosomes 1–38 (chromosomes are separated by colour). A dashed horizontal
line indicates thecut-off (Z.5) used for extracting outliers. c, The negative end
of the Z(H
P
) distribution plotted along dog autosomes 1–38. A dashed
horizontal line indicates the cut-off (Z,25) used for extracting outliers.
Table 1
|
Enriched gene ontology terms among CDR genes
Gene ontology term
PFDR value Gene count
Regulation of neuron differentiation 0.005 3 (26)
Multicellular organismal process 0.005 21 (3,822)
Digestion 0.008 4 (95)
Neuron differentiation 0.010 5 (210)
Regulation of molecular function 0.011 8 (671)
Central nervous system development 0.013 5 (235)
Regulation of developmental process 0.013 5 (236)
Generation of neurons 0.013 5 (242)
Nervous system development 0.013 8 (716)
Binding of sperm to zona pellucida 0.015 2 (12)
Sperm–egg recognition 0.015 2 (12)
Neurogenesis 0.015 5 (262)
Cell–cell recognition 0.019 2 (14)
Regulation of catalytic activity 0.020 7 (605)
Regulation of hydrolase activity 0.026 5 (307)
Fatty acid metabolic process 0.031 4 (191)
System development 0.034 11 (1,605)
Regulation of GTPase activity 0.039 4 (211)
Anatomical structure development 0.039 12 (2,005)
Intramembranous ossication 0.039 1 (1)
Quinolinate metabolic process 0.039 1 (1)
Starch metabolic process 0.039 1 (1)
Starch catabolic process 0.039 1 (1)
Glucocorticoid catabolic process 0.039 1 (1)
Cell development 0.039 9 (1,242)
Enriched terms are colour-coded to reflect relatedness in the ontology or functional proximity. Blue,
nervous system development; green, sperm–egg recognition; grey, regulation of molecular function;
orange, digestion. For each term, gene count shows number of genes in CDRsrelative to total number of
annotated genes (in parentheses).
LETTER RESEARCH
21 MARCH 2013 | VOL 495 | NATURE | 361
Macmillan Publishers Limited. All rights reserved
©2013
insulin resistance (ACSM2A) that initiates the fatty acid metabolism
19
.
A total of 6 CDRs harbour 10 genes with functions related to starch
and fat metabolism (Supplementary Table 10). We propose that gene-
tic variants within these genes may have been selected to aid adapta-
tion from a mainly carnivorous diet to a more starch rich diet during
dog domestication.
The breakdown of starch in dogs proceeds in three stages: (1) starch
is first cleaved to maltose and other oligosaccharides by alpha-amylase
in the intestine; (2) the oligosaccharides are subsequently hydrolysed
by maltase-glucoamylase
20
, sucrase and isomaltase to form glucose;
and (3) finally, glucose is transported across the plasma membrane
by brush border protein SGLT1
21
. Here we present evidence for selec-
tion on all three stages of starch digestion during dog domestication.
Whereas humans have acquired amylase activity in the saliva
22
via
an ancient duplication of the pancreatic amylase gene, dogs only
express amylase in the pancreas
23
. In dogs the AMY2B gene, encoding
the alpha-2B-amylase, resides in a 600-kb CDR on chromosome 6 with
Z(H
P
)andZ(F
ST
) scores of 24.60 and 7.16, respectively(Figs 1 and 2a).
Interestingly, an 8-kb sequence spanning the AMY2B locus showed
a several-fold increase in aligned read depth in dog relative to wolf
(Fig. 2b), suggestive of a copy number change. Formal comparisons of
regional and local pool coverage, and wolf and dog coverage (Methods),
respectively, also suggest a substantial increase in copy numbers in all
dog pools compared to wolf at this locus (Supplementary Discussion,
section 5).
We confirmed this CNV by quantifying AMY2B copy numbers in
136 dogs and 35 wolves (Supplementary Table 11) using real-time
quantitative PCR (qPCR). Whereas all wolves tested carried only
2 copies (2N52), diploid copy numbers in dog ranged from 4 to 30
(P,0.001, Wilcoxon) (Fig. 2c), corresponding to a remarkable
7.4-fold average increase in dog AMY2B copy numbers. To assess
whether this change correspond to a difference in amylase activity,
we first compared AMY2B gene expression in pancreas from dog
(n59) and wolf (n512) and noted a 28-fold higher average expres-
sion in dog (P,0.001, Wilcoxon, Fig. 2d). We then quantified
amylase activity in frozen serum (Fig. 2e) and found a 4.7-fold
higher activity in dog (9.6–18.4mkat l
21
(n512)) relative to wolf
(1.4–4.3 mkat l
21
(n513)) (P,0.001, Wilcoxon). Similar results were
obtained in comparisons of a limited number of fresh samples
(Supplementary Tables 12 and 13). The change in AMY2B gene copy
number together with a correlated increase in both expression
level (rho 50.84, P,0.0001, Spearman) (Supplementary Fig. 3) and
enzyme activity (rho 50.63, P,0.01, Spearman) (Supplementary
Fig. 4) indicates that duplications of the alpha-amylase locus conferred
a selective advantage to early dogs by causing an increase in amylase
activity.
Maltase-glucoamylase is responsible for the second step in the
breakdown of starch, catalysing the hydrolysis of maltose to glucose
20
.
No copy number changes were observed in the MGAM locus so we
decided to study haplotype diversity across the region to facilitate the
identification of causal variants. We genotyped 47 randomly selected
SNPs in 71 dogs representing 38 diverse breeds and 19 wolves of
worldwide distribution (referred to as ‘the reference panel’, Sup-
plementary Table 14). Sixty-eight of the seventy-one dogs tested car-
ried at least one copy of a 124-kb long haplotype spanning the entire
MGAM and a small neighbouring locus encoding the bittertaste medi-
ating taste receptor 2 member 38 (TASR38) (Fig. 3a–c). Whereas none
of the wolves carried the selected haplotype, 55 dogs were homozygous
for it, 13 were heterozygous and only three dogs lacked it (2 West
Highland White Terriers and 1 Chinese Crested Dog). This high
degree of haplotype differentiation between dog and wolf (average
F
ST
for genotyped SNPs 50.75) indicates that this haplotype may
harbour genetic variation of selective advantage to dogs (Supplemen-
tary Discussion, sections 3 and 6).
We identified several candidate mutations within MGAM that may
have been targeted by selection in this region (Supplementary Table 15).
First a conservative amino acid substitution located in the duplicated
trefoil domain of MGAM (residue 1001) is nearly fixed for isoleucine in
wolf and for valinein dogs. Eleven out of fourteenmammals have valine
at this position, whereas the omnivorous rat, and the insectivorous
hedgehog and short-tailed opossum, carry isoleucine like the wolf
(Supplementary Table 16). Second, another conservative substitution,
methionine to valine, located in the beta sheet of the maltase enzyme
(residue 797), is segregating in wolf but fixed formethionine in dog. The
insectivorous hedgehog and common shrew are the only mammals
without methionine at this evolutionarily conserved position (Sup-
plementary Table 17) and in silico modelling using the SDM-server
indicates that a change from methionine to valine at this residue is
destabilizing
24
. Third, a fixed two-base-pair deletion in dog disrupts
the stop codon, thereby extending the carboxy-terminal end of dog
0
5
10
15
20
25
30
35
0
Number of individuals
Diploid copy number
220151052530
0
5
10
15
20
Wolf Dog
Relative expression
0
5
10
15
20
Amylase activity (μkat l–1)
Wolf Dog
c
ed
34 40 50 75
0
0.2
0.4
0.6
0.8
1.0
HP/FST
0
0.2
0.4
0.6
0.8
1.0
49.4 51
H/FST/rC
Mb
a
b
AMY2B COL11A1RNPC3
Figure 2
|
Selection for increased amylase activity. a, Pooled heterozygosity,
H
P
(blue), and average fixation index,F
ST
(orange), plotted for 200-kb windows
across a chromosome 6 region harbouring AMY2B.b, Heterozygosity,H(blue),
and fixation index, F
ST
(orange), for single SNPs in the selected region. Dog
relative to wolf coverage, rC (green line), indicates increase in AMY2B copy
number in dog. Genes in the region are shown below panel b.c, Histogram
showing the distribution of diploid amylase copy number in wolf (n535)
(blue) and dog (n5136) (red). d, Amylase messenger RNA expressionlevels in
pancreas of wolf (n512) and dog (n59). e, Amylase activity in serum from
wolf (n513) and dog (n512).
RESEARCH LETTER
362 | NATURE | VOL 495 | 21 MARCH 2013
Macmillan Publishers Limited. All rights reserved
©2013
MGAM by two aminoacids: asparagineand phenylalanine. In32 mam-
mals studied only herbivores (rabbit, pika, alpaca and cow) and omni-
vores (mouse lemur and rat) share an extension like that seen in dog
(Supplementary Table 18). A fourth candidate mutation in intron 37
affects a predicted binding site for the glucose metabolism regulator
NR4A2 protein
25
by shifting the wolf sequence away from the canonical
NR4A2-binding motif. Three out of four mammals with the wolf allele
at this site rely heavily on insects or fish for their nutritional require-
ments (Supplementary Table 19).
To decipher whether the candidate mutations act primarily on expres-
sion or protein activity we examined MGAM expression in pancreas
and the resulting enzymatic activity in serum. Dogs showed a ,12-fold
higher expression (P,0.001, Wilcoxon, n
DOG
59, n
WOLF
58) (Fig. 3d)
and a ,twofold increase in maltose to glucose turnover compared to
wolves (average glucose produced in dogs: 0.94 DA
570 nm
(0.64–1.23,
n57) and wolves: 0.52 DA
570 nm
(0.44–0.66, n58), P50.0012,
Wilcoxon) (Fig. 3e). Although we cannot rule out that diet-induced
plasticity contributed to this difference
26
, our results indicate that the
mutation affecting a NR4A2-binding site or another unknown variant
probably affect the expression of MGAM. Selection may thus clearly have
led to increased MGAM expression, but we cannot rule out that the
strong selection affecting this locus may have favoured the accumulation
of protein-coding changes on the same haplotype. Similar scenarios have
been seen for white coat colour in dogs and pigs, where repeated selection
for additional mutations has resulted in an allelic series of white spotting
at the MITF and KIT loci, respectively
27
.
Once starch has been digested to glucose it is absorbed through
the luminal plasma membrane of the small intestine by the sodium/
glucose cotransporter 1 (SGLT1)
21
. To benefit from an increased
capacity to digest starch, dogs would therefore be expected to show a
parallel increase in glucose uptake. A CDR on chromosome 26 (Sup-
plementary Fig. 5a, b) encompasses SGLT1 and a gene (SGLT3) encod-
ing the glucose-sensing sodium/glucose cotransporter 3 protein
28
.To
characterize the haplotype diversity we genotyped 48 randomly chosen
SNPs across this CDR in the reference panel and identified a 50.5-kb
region, spanning the 39section of SGLT1 as well as the 39end of SGLT3,
that is highly divergent between dog and wolf (Supplementary Fig. 5c).
In this region all dogs tested were carriers of a particular haplotype,
for which 63 were homozygous and eight heterozygous. This con-
trasts to 19 wolves where a single individual carried one copy of the
haplotype. Based on the high haplotype differentiation (average F
ST
for 18 SNPs in 50.5-kb haplotype 50.81) it is likely that SGLT1 and
its 39region represents an additional dog domestication locus.
The 50.5-kb region includes a conservative isoleucine to valine
substitution in SGLT1 (residue 244) that affects a loop facing the
extracellular side of the luminal membrane (Supplementary Table 15).
Heterologous expression analysis
29
shows that glycosylation at a nearby
site (residue 248) affects glucose transport, indicating that it is possible
that dogs acquired improved glucose uptake as a result of the observed
substitution. In addition, we see only non-significant differences in SGLT1
expression in pancreas of dog (n59) and wolf (n54) (P50.39,
Wilcoxon) (Supplementary Fig. 6), indicating that selection primarily
targeted a structural rather than regulatory mutation in SGLT1.
In conclusion, we have presented evidence that dog domestication
was accompanied by selection at three genes with key roles in starch
digestion: AMY2B,MGAM and SGLT1. Our results show that adapta-
tions that allowed the early ancestorsof modern dogs to thrive on a diet
rich in starch, relative to the carnivorous diet of wolves, constituted a
crucial step in early dog domestication. This may suggest that a change
of ecological niche could have been the driving force behind the
domestication process, and that scavenging in waste dumps near the
increasingly common human settlements during the dawn of the agri-
cultural revolution may have constituted this new niche
6
. In light of
previous results describing the timing and location of dog domestica-
tion, our findings may suggest that the development of agriculture
catalysed the domestication of dogs.
The results presented here demonstrate a striking case of parallel
evolution whereby the benefits of coping with an increasingly starch-
rich diet during the agricultural revolution caused similar adaptive
responses in dog and human
30
. This emphasizes how insights from
dog domestication may benefit our understanding of human recent
evolution and disease. Finally, by understanding the genetic basis of
adaptive traits in dogs we have come closer to unlocking the potential
in dog and wolf comparisons to decipher the genetics of behaviour.
METHODS SUMMARY
Sequencing. We pooled genomic DNA from 12 individuals before mate-pair
library construction and sequencing on the AB SOLiD system, version 3, accord-
ing to standard manufacturer protocols. Sequencing reads were aligned to the
CanFam 2.0 reference sequence using the Bioscope 1.1 software.
Selection analyses. We identified variable sites in data combined from all pools
and required a minimum of three reads supporting an alternative allele to call a
SNP. We used allele countsat variable sites to identify signals of selection in 200-kb
windows using two approaches: for each window we calculated (1) the average
pooled heterozygosity, H
P
(ref. 10), and (2) the average fixation index, F
ST
,
between dog and wolf. Putatively selected regions were located by extracting
d
0
3
6
9
Wolf Dog
Relative expression
0
0.5
1.0
1.5
Wolf Dog
MGAM activity
e
310 20 30 40 50 60
0
0.2
0.4
0.6
0.8
1.0
HP/FST
0
0.2
0.4
0.6
0.8
1.0
9.2 10.1 11.0
H/FST
10.09 10.23
Wolves
Dogs
Mb
<<<<< >>>
MGAM TASR38 CLEC5A
a
b
c
Figure 3
|
Selection is associated with increased maltase activity. a, Pooled
heterozygosity, H
P
(blue), and average fixation index, F
ST
(orange), plotted for
200-kb windows across a chromosome 16 region harbouring MGAM.
b, Heterozygosity, H(blue), andfixation index, F
ST
(orange), for single SNPs in
the selected region. c, Haplotypes inferred from genotyping of 47 SNPs across
the MGAM locus in 71 dogs and 19 wolves (red and blue colour are major and
minor dog allele, respectively). Genes in the genotyped region are shown below
panel c.d,MGAM mRNA expression levels in pancreas of wolf (n58) and dog
(n59). e, MGAM activity in serum from wolf (n58) and dog (n57).
LETTER RESEARCH
21 MARCH 2013 | VOL 495 | NATURE | 363
Macmillan Publishers Limited. All rights reserved
©2013
windows from the extreme tails of the Z-transformed H
P
and F
ST
distributions by
applying a threshold of 5 standard deviations.
Functional assays. We used multiplex TaqMan assays and SYBR Green real-time
PCR to quantify CNVs and gene expression, respectively. Serum amylase activity
was analysed using an Architect e400 instrument and serum maltase activity was
quantified based on the amount of maltose to glucose turnover.
Full Methods and any associated references are available in the online version of
the paper.
Received 1 July; accepted 11 December 2012.
Published online 23 January 2013.
1. Ovodov, N. D. et al. A 33,000-year-old incipient dog from the Altai mountains of
Siberia: evidence of the earliest domestication disrupted by the last glacial
maximum. PLoS ONE 6, e22821 (2011).
2. Davis, S. J. M.& Valla, F. R. Evidencefor domestication of the dog 12,000years ago
in the Natufian of Israel. Nature 276, 608–610 (1978).
3. Skoglund,P., Gotherstrom,A. & Jakobsson, M. Estimationof population divergence
times from non-overlapping genomic sequences: examples from dogs and
wolves. Mol. Biol. Evol. 28, 1505–1517 (2011).
4. Pang,J. F. et al. mtDNA data indicate a single originfor dogs south of Yangtze River,
less than 16,300years ago, from numerous wolves.Mol. Biol. Evol. 26, 2849–2864
(2009).
5. vonHoldt, B. M. et al. Genome-wide SNP and haplotype analyses reveal a rich
history underlying dog domestication. Nature 464, 898–902 (2010).
6. Coppinger,R. & Coppinger, L. Dogs: a StartlingNew Understanding of Canine Origin,
Behaviour and Evolution (Scribner, 2001).
7. Hare, B., Wobber, V. & Wrangham,R. The self-domesticationhypothesis: evolution
of bonobo psychology is due to selection against aggression. Anim. Behav. 83,
573–585 (2012).
8. Belyaev, D. K. Destabilizing selection as a factor in domestication. J. Hered. 70,
301–308 (1979).
9. Fang, M., Larson, G., Ribeiro, H. S., Li, N. & Andersson, L. Contrasting mode of
evolutionat a coat color locus in wild and domesticpigs. PLoS Genet. 5, e1000341
(2009).
10. Rubin, C. J. et al. Whole-genome resequencing reveals loci under selection during
chicken domestication. Nature 464, 587–591 (2010).
11. Lindblad-Toh, K. et al. Genome sequence, comparative analysis and haplotype
structure of the domestic dog. Nature 438, 803–819 (2005).
12. Koike, N. et al. Brorin, a novel secreted bone morphogenetic protein antagonist,
promotes neurogenesis in mouse neural precursor cells. J. Biol. Chem. 282,
15843–15850 (2007).
13. Cheng, L. et al. Tlx3 and Tlx1 are post-mitotic selector genes determining
glutamatergic over GABAergic cell fates. Nature Neurosci. 7, 510–517 (2004).
14. Napoli, I. et al. The fragile X syndrome protein represses activity-dependent
translation through CYFIP1, a new 4E-BP. Cell 134, 1042–1054 (2008).
15. Weston, M. C., Nehring, R. B., Wojcik, S. M. & Rosenmund, C. Interplay between
VGLUT isoformsand endophilin A1 regulates neurotransmitter release and short-
term plasticity. Neuron 69, 1147–1159 (2011).
16. Varga, Z. M. et al. Zebrafish smoothened functions in ventral neural tube
specification and axon tract formation. Development 128, 3497–3509 (2001).
17. Tokuhiro, K., Ikawa, M., Benham, A. M. & Okabe, M. Protein disulfide isomerase
homologPDILT is required for quality controlof sperm membrane protein ADAM3
and male fertility. Proc. Natl Acad. Sci. USA 109, 3850–3855 (2012).
18. Gardner, A. J. & Evans, J. P. Mammalian membrane block to polyspermy: new
insightsinto how mammalian eggs preventfertilisation by multiplesperm. Reprod.
Fertil. Dev. 18, 53–61 (2006).
19. Boomgaarden, I., Vock,C., Klapper, M. & Doring,F. Comparative analysesof disease
risk genes belonging to the acyl-CoA synthetase medium-chain (ACSM) family in
human liver and cell lines. Biochem. Genet. 47, 739–748 (2009).
20. Nichols, B. L. et al. The maltase-glucoamylase gene: commonancestry to sucrase-
isomaltase with complementary starch digestion activities. Proc. Natl Acad. Sci.
USA 100, 1432–1437 (2003) .
21. Wright, E. M., Loo, D. D. F. & Hirayama, B. A. Biology of human sodium glucose
transporters. Physiol. Rev. 91, 733–794 (2011).
22. Meisler, M. H. & Ting, C. N. The remarkable evolutionary history of the human
amylase genes. Crit. Rev. Oral Biol. Med. 4, 503–509 (1993).
23. Simpson, J. W., Doxey, D. L. & Brown, R. Serum isoamylase values in normal dogs
and dogs with exocrine pancreatic insufficiency. Vet. Res. Commun. 8, 303–308
(1984).
24. Worth, C. L., Preissner, R. & Blundell, T. L. SDM-a server for predicting effects of
mutations on protein stability and malfunction. Nucleic Acids Res. 39,
W215–W222 (2011).
25. Pei, L. et al. NR4A orphan nuclear receptors are transcriptional regulators of
hepatic glucose metabolism. Nature Med. 12, 1048–1055 (2006).
26. Mochizuki, K., Honma, K., Shimada, M. & Goda, T. The regulation of jejunal
inductionof the maltase-glucoamylase gene by a high-starch/low-fat diet in mice.
Mol. Nutr. Food Res. 54, 1445–1451 (2010).
27. Andersson, L. Studying phenotypic evolution in domestic animals: a walk in the
footsteps of Charles Darwin. Cold Spring Harb. Symp. Quant. Biol. 74, 319–325
(2009).
28. Diez-Sampedro, A. et al. A glucose sensor hiding in a family of transporters. Proc.
Natl Acad. Sci. USA 100, 11753–11758 (2003).
29. Hediger, M. A., Mendlein, J., Lee, H. S. & Wright, E. M. Biosynthesis of the cloned
intestinal Na
1
glucose cotransporter. Biochim. Biophys. Acta 1064, 360–364
(1991).
30. Perry, G. H. et al. Diet and the evolution of human amylase gene copy number
variation. Nature Genet. 39, 1256–1260 (2007).
Supplementary Information is available in the online version of the paper.
Acknowledgements We thank Ja
¨rvzoo, Nordens ark and the Canine Biobank at
Uppsala University and the Swedish University of Agricultural Sciences for providing
samples, Uppsala Genomics Platform at SciLifeLab Uppsala for generating the
resequencing data, the UPPNEX platform for assisting with computational
infrastructure for data analysis and the Broad Institute Genomics Platform for
validation genotyping. The project was funded by the SSF, the Swedish Research
Council, the Swedish Research Council Formas, Uppsala University and a EURYI to
K.L.-T.funded by the ESF supportingalso E.A.; K.M. was funded by the HigherEducation
Commission, Pakistan.
Author Contributions K.L.-T. and A
˚.H. designed the study. K.L.-T. and E.A. oversaw the
study. M.-L.A.coordinated and performedthe majority of the samplecollecting and O.L.
and J.M.A. provided samples of critical importance. E.A. performed the SNP detection
and selection analyses; A.R. identified candidate causative mutations and analysed
haplotypes in CDRs; K.M. detected CNVs bioinformatically; M.T.W. performed
phylogenetic analysis and analysed the Canine HD-array data; A.R. performed the
maltase activity assay; M.-L.A. validated CNVs and quantified mRNA expression of
candidate genes; M.P. performed validation SNP genotyping; E.A., A.R., M.-L.A. and
K.L-T. interpreted the data; E.A. and K.L.-T. wrote the paper with input from the other
authors.
Author Information Sequence reads are available under the accession number
SRA061854 (NCBI Sequence Read Archive). Reprints and permissions information is
available at www.nature.com/reprints. The authors declare no competing financial
interests. Readers are welcome to comment on the online version of the paper.
Correspondence and requests for materials should be addressed to E.A.
(Erik.Axelsson@imbim.uu.se) and K.L-T. (kersli@broadinstitute.org).
RESEARCH LETTER
364 | NATURE | VOL 495 | 21 MARCH 2013
Macmillan Publishers Limited. All rights reserved
©2013
METHODS
DNA extraction. DNA was extracted from tissue using Qiagen tissue DNA
extraction kits or from EDTA blood using either manual salt precipitation or
the QIASymphony DNA Midikit (Qiagen) on the QIASymphony robot (Qiagen).
Sequencing. We pooled DNA from 12 individuals per pool before mate-pair
library construction and sequencing on the AB SOLiD system, version 3, accord-
ing to standard manufacturer protocols (Applied Biosystems). Sequencing reads
were aligned to the CanFam 2.0 reference sequence using the Bioscope 1.1 soft-
ware. We removed duplicated (http://picard.sourceforge.net) and poorly mapped
reads (mapping quality ,20 in Samtools)
31
and retained only uniquely mapped
reads for further analyses.
SNP detection. We searched for variable sites in data combined from all pools
(including wolf) to increase sensitivity to rare alleles. We required a minimum of
three reads supporting an alternative allele to call a SNP, and applied a further
filtering step implemented in samtools.pl varFilter (settings: –Q25 –q10 –d3
–D120 –G25–w10 –N2 –l30) to ensure a high callaccuracy that is largely unaffected
by, for example, paralogous sequence variants. We called genotypesfor all SNPs in
all dog pools and the single wolf pool by counting sequencing reads supporting the
reference and variant allele, given a minimum base quality of 20, to estimate allele
frequencies in the dog and wolf populations. A random selection representing
25% of the sequencing reads from pools 4 and 5 were included in this process to
achieve unbiased allele frequency estimates.
Selection analyses. Allele counts and allele frequencies at all identified variable
sites were used to search the dog genome for regions that may have been affected
by selection duringthe early phase of dog domesticationusing two complementary
approaches. First we calculated the average pooled heterozygosity (H
P
) in 200-kb
windows sliding 100kb at a time, for all five dog pools combined, and in the single
wolf pool separately, following the methodology described in ref. 10. Briefly, this
method sums all minor and major allele counts, respectively, at all variable sites
within a window, and estimates the heterozygosity based on the combined allele
counts for the entire window. The advantage of this method over calculating a
simple arithmeticmean of all single-site heterozygosity estimatesis that it accounts
for variable sequence coverage across the window. To avoid spurious selection
signals we discarded 49 out of 21,927 windows containing fewer than 10 inform-
ative sites from both this and the subsequent F
ST
analysis. We Z-transformed the
resultant distribution of H
P
scores and extracted putatively selected windows in
the extreme tail of the distribution by applying a Z(H
P
),25 cut-off.
Second we calculated F
ST
values between dog and wolf for individual SNPs
using a method that adjusts for sample size differences
32
. We averaged F
ST
values
across 200-kb windows, sliding 100 kb at a time and Z-transformed the resultant
distribution. Putative selection targets were extracted from the extreme tail of the
distribution by applying a Z(F
ST
).5 cut-off, and attributed to selection in dog if
the corresponding Z(H
P
)
DOG
,Z(H
P
)
WOLF
, to selection in wolf if Z(H
P
)
WOLF
,
Z(H
P
)
DOG
(three regions), and to selection in both taxa if Z(H
P
)
WOLF
,24 and
Z(H
P
)
WOLF
,24.
Gene ontology analysis. We used the Ensembl gene annotationsto identify genes
residing within regions extending 100kb up- and downstream of CDRs to include
potential effects of regulatory changes on loci at some distance, and to reduce the
risk of excluding the outermostportions of the selected haplotypes by using sliding
windows of fixed size. We tested for enrichment of gene ontology terms (GOa-
human) assigned to the subset of these CDR genes for which human orthology
could be established (79 out of 122) using the GOstat program
33
.
Genotyping validation. We designed an iPLEX assay targeting 124 SNPs located
in CDRs showing a high degree of homozygosity or population differentiation. A
total of 71 dogs, representing 38 different breeds, and 19 wolves (Supplementary
Table 14) were genotyped using standard protocols provided by the manufacturer
(Sequenome). Haplotypes were phased using fastPHASE
34
.
qPCR CNV detection. We quantified DNA copy number variation using Multi-
plex TaqMan assays containing primers and probes (Supplementary Table 20)
matching both the target and reference sequence (housekeeping gene C7orf28b)
according to the manufacturer’s protocol. All reactions were run in triplicate and
data was analysed using the CopyCaller software (Applied Biosystems). Copy
numbers for each target were normalized to the same wolf to account for inter-
plate variability.
qPCR expression analyses. Pancreatic tissue samples from dogs and wolves
where collected postmortem, stored in RNAlater at 4 uC for 24 h and subsequently
freeze-stored at 280uC. We used TRIzol to isolate RNA from these samples,
followed by complementary DNA synthesis using the Advantage RT for PCR
kit according to the manufacturers’ protocols (Life Technologies and Clontech,
respectively). We designed exonic primers (Supplementary Table 21) and quan-
tified the amount of cDNA using SYBR Green real-time PCR (Applied bio-
sciences) on a 7900HT Fast real time PCR system (Applied Biosystems) and
analysed the data using the qbasePLUS (Biogazelle) software according to the
DDC
T
method. All reactions where run in triplicate and normalized by compari-
sons to housekeeping genes RPL32 and RPL13A.
Amylase activity. Peripheral EDTA and serum blood samples where collected
from dogs and both captive and free-ranging wolves. Serum amylase activity was
analysed at the Clinical Pathology service (Swedish Agricultural University) using
an Architect e400 instrument (Abott Laboratories), except for 8 serum samples
(Supplementary Table 13) which were run on a VetScan instrument (Abaxis).
Maltase activity. Maltase activity was assayed according to the principle outlined
in ref. 35, whereby a known amount of maltose substrate is added to serum and
the resultant glucose produced is measured as the change in absorbance after
five minutes (DA
570nm
). We used reagents from the ab83388 Maltose assay kit
(Abcam) and serum sampled as described above. For each individual, glu-
cose residuals were measured in duplicate and maltase assays were performed
in triplicate.
Indel calling. We used Bioscope 1.1 to call small insertions and deletions in each
pool separately. We then combined the results of all pools and extracted a set of
high confident indels by requiring that indels were supported by at least three
sequencing reads.
CNV detection. Four methods were used to detect structural variation in the dog
genome. We searched for deviations in insert size using the large indel tool imple-
mented in Bioscope1.1. We compared the coverage depth between the pooled
samples using CNVseq
36
and the Fixed deletions method
10
and finally identified
regions in which the coverage depth deviated from the pool average using
CNVnator
37
. Methods relying on comparisons of sequence coverage between
pools always used the wolf as reference pool.
Ethics. All animals contributingtissue samples to this study died for other reasons
than participating in this study. All dog samples were taken with the owners
consent. The sampling conformed to the decision of the Swedish Animal
Ethical Committee (no. C62/10) and the Swedish Animal Welfare Agency
(no.31-1711/10).
31. Li, H. et al. The sequence alignment/map format and SAMtools. Bioinformatics
25, 2078–2079 (2009).
32. Weir, B. S. & Cockerham, C. C. Estimating F-statistics for the analysis of
population-structure. Evolution 38, 1358–1370 (1984).
33. Beissbarth, T. & Speed, T. P. GOstat: find statistically overrepresented Gene
Ontologies within a group of genes. Bioinformatics 20, 1464–1465 (2004).
34. Scheet, P. & Stephens, M. A fast and flexible statistical model for large-scale
population genotype data: applications to inferring missing genotypes and
haplotypic phase. Am. J. Hum. Genet. 78, 629–644 (2006).
35. Dahlqvist, A. Method for assay of intestinal disaccharidases. Anal. Biochem. 7,
18–25 (1964).
36. Xie, C. & Tammi, M. T. CNV-seq, a new method to detect copy number variation
using high-throughput sequencing. BMC Bioinformatics 10, 80 (2009).
37. Abyzov, A., Urban, A. E., Snyder, M. & Gerstein, M. CNVnator: An approach to
discover, genotype, and characterize typical and atypical CNVs from family and
population genome sequencing. Genome Res. 21, 974–984 (2011).
LETTER RESEARCH
Macmillan Publishers Limited. All rights reserved
©2013
