Revealing the genetic structure of a trait by sequencing a population under selection.

Abstract

One approach to understanding the genetic basis of traits is to study their pattern of inheritance among offspring of phenotypically different parents. Previously, such analysis has been limited by low mapping resolution, high labor costs, and large sample size requirements for detecting modest effects. Here, we present a novel approach to map trait loci using artificial selection. First, we generated populations of 10-100 million haploid and diploid segregants by crossing two budding yeast strains of different heat tolerance for up to 12 generations. We then subjected these large segregant pools to heat stress for up to 12 d, enriching for beneficial alleles. Finally, we sequenced total DNA from the pools before and during selection to measure the changes in parental allele frequency. We mapped 21 intervals with significant changes in genetic background in response to selection, which is several times more than found with traditional linkage methods. Nine of these regions contained two or fewer genes, yielding much higher resolution than previous genomic linkage studies. Multiple members of the RAS/cAMP signaling pathway were implicated, along with genes previously not annotated with heat stress response function. Surprisingly, at most selected loci, allele frequencies stopped changing before the end of the selection experiment, but alleles did not become fixed. Furthermore, we were able to detect the same set of trait loci in a population of diploid individuals with similar power and resolution, and observed primarily additive effects, similar to what is seen for complex trait genetics in other diploid organisms such as humans.

Full text

Page 1

10.1101/gr.116731.110Access the most recent version at doi:
2011 21: 1131-1138 originally published online March 21, 2011 Genome Res.

Leopold Parts, Francisco A. Cubillos, Jonas Warringer, et al.
population under selection
Revealing the genetic structure of a trait by sequencing a


Material
Supplemental

http://genome.cshlp.org/content/suppl/2011/03/18/gr.116731.110.DC1.html
References

http://genome.cshlp.org/content/21/7/1131.full.html#ref-list-1
This article cites 40 articles, 10 of which can be accessed free at:
Open Access
Freely available online through the Genome Research Open Access option.
service
Email alerting

click here
top right corner of the article or
Receive free email alerts when new articles cite this article - sign up in the box at the

http://genome.cshlp.org/subscriptions
go to: Genome ResearchTo subscribe to
Copyright © 2011 by Cold Spring Harbor Laboratory Press
Cold Spring Harbor Laboratory Press on September 22, 2011 - Published by genome.cshlp.orgDownloaded from

Page 2

Method
Revealing the genetic structure of a trait by sequencing
a population under selection
Leopold Parts,1,6Francisco A. Cubillos,2Jonas Warringer,3,4Kanika Jain,2
Francisco Salinas,2Suzannah J. Bumpstead,1Mikael Molin,3Amin Zia,5Jared T. Simpson,1
Michael A. Quail,1Alan Moses,5Edward J. Louis,2Richard Durbin,1and Gianni Liti2,6
1The Wellcome Trust Sanger Institute, Hinxton CB10 1SA, United Kingdom;2Centre for Genetics and Genomics, Queen’s Medical
Centre, University of Nottingham, Nottingham NG7 2UH, United Kingdom;3Department of Cell and Molecular Biology, University
of Gothenburg, 41390 Gothenburg, Sweden;4Centre for Integrative Genetics (CIGENE), Norwegian University of Life Sciences
(UMB), 1432 A˚s, Norway;5Department of Cell and Systems Biology, University of Toronto, Toronto, Ontario M5S 2J4, Canada
One approach to understanding the genetic basis of traits is to study their pattern of inheritance among offspring of
phenotypically different parents. Previously, such analysis has been limited by low mapping resolution, high labor costs,
and large sample size requirements for detecting modest effects. Here, we present a novel approach to map trait loci using
artificial selection. First, we generated populations of 10–100 million haploid and diploid segregants by crossing two
budding yeast strains of different heat tolerance for up to 12 generations. We then subjected these large segregant pools to
heat stress for up to 12 d, enriching for beneficial alleles. Finally, we sequenced total DNA from the pools before and
during selection to measure the changes in parental allele frequency. We mapped 21 intervals with significant changes in
genetic background in response to selection, which is several times more than found with traditional linkage methods.
Nine of these regions contained two or fewer genes, yielding much higher resolution than previous genomic linkage
studies. Multiple members of the RAS/cAMP signaling pathway were implicated, along with genes previously not an-
notated with heat stress response function. Surprisingly, at most selected loci, allele frequencies stopped changing before
the end of the selection experiment, but alleles did not become fixed. Furthermore, we were able to detect the same set of
trait loci in a population of diploid individuals with similar power and resolution, and observed primarily additive effects,
similar to what is seen for complex trait genetics in other diploid organisms such as humans.
[Supplemental material is available for this article. The sequence data from this study have been submitted to the NCBI
Sequence Read Archive (SRA) (http:/ /www.ncbi.nlm.nih.gov/Traces/sra/sra.cgi) under study accession no. ERP000500,
with individual sample accession numbers provided in Supplemental Table S3.]
A central challenge of modern genetics is to identify genes and
pathwaysresponsiblefor variation in quantitative traits.Inthelast
decade, efforts of global international collaborations have revealed
numerous loci that influence disease risk in humans by genotyp-
ing and phenotyping very large cohorts of individuals. However,
the effects of single alleles are almost all modest, and explain only
a small portion of the heritable variability (Manolio et al. 2009).
Furthermore, whiletrait loci are found, association peaksgenerally
spanalargeregion,anddonotpointtotheunderlyingmechanism
responsible for the association. Thus, studies in model organisms,
where consequences of genetic variation can be analyzed using
reverse genetic tools, have been important for understanding the
genetics of complex traits (Yvert et al. 2003; Deutschbauer and
Davis 2005; Perlstein et al. 2006; Nogami et al. 2007; Demogines
et al. 2008;Sinhaet al. 2008;Smith andKruglyak2008;Gerke etal.
2009; Liti et al. 2009b; Romano et al. 2010).
Mapping the effect of naturally occurring alleles on traits is
notstraightforwardeveninmodelorganisms(HunterandCrawford
2008). Designed crosses often use substantially manipulated labo-
ratory strains (Brem et al. 2002; Steinmetz et al. 2002) and produce
segregants that have to be laboriously genotyped and phenotyped.
Linkage analysis on the resulting individuals can suffer from low
resolution due toa limited number of crossoverevents (Darvasi and
Soller 1995), but more rounds of crossing alleviate the problem
(Wang et al. 2003). Developing and maintaining sufficiently large
outbredpopulationstoresemblehumancohortsusedinassociation
mapping is costly (Valdar et al. 2006).
Recently, analysis of a very large pool of recombinant yeast
strains has been used to identify quantitative trait loci (QTLs) for
multiple traits without characterizing individual segregants (Segre `
etal.2006;Ehrenreichetal.2010;Wengeretal.2010).Whilemany
QTLsweredetected,theproblemoffindingallresponsiblelociand
localizing the trait genes within the linkage regions, which typi-
cally span many genes, remains. Furthermore, such analyses in
yeast have previously been limited to haploid samples, in which
genetic architecture may differ from that in diploids. Here, we
present a precise and sensitive approach to QTL mapping, extend-
ing the method recently proposed by Ehrenreich et al. (2010) to
sensitively identify trait loci at high resolution, in some cases down
to single genes, in both haploid and diploid populations.
Results and Discussion
Strategy for high resolution QTL mapping
We used a three-step process for QTL mapping (Fig. 1). First, we
generated very large pools of progeny between two phenotypically
6Corresponding authors.
E-mail gianni.liti@nottingham.ac.uk.
E-mail leopold.parts@sanger.ac.uk.
Article published online before print. Article, supplemental material, and pub-
lication date are at http://www.genome.org/cgi/doi/10.1101/gr.116731.110.
Freely available online through the Genome Research Open Access option.
21:1131–1138 ? 2011 by Cold Spring Harbor Laboratory Press; ISSN 1088-9051/11; www.genome.org
Genome Research
www.genome.org
1131
Cold Spring Harbor Laboratory Press on September 22, 2011 - Published by genome.cshlp.orgDownloaded from

Page 3

different yeast strains. We chose YPS128, a heat tolerant North
American (NA) oak tree bark strain, and DVGBP6044, a heat sen-
sitive West African (WA) palm wine strain as parents, and placed
a different selectable marker at the same genomic position in each
(Supplemental Table S1). We then systematically forced the yeast
cells through multiple rounds of random mating and sporulation
(Methods; Supplemental Fig. S1; Supplemental Material SI), cre-
ating advanced intercross lines (AILs) with reduced linkage be-
tween nearby loci. We produced haploid as well as heterozygous
diploid pools of sixth and 12th generation progeny (F6 and F12
AILs), consisting of 10–100 million random segregants each.
We applied selective pressure to half of each pool by growing
it asexually in a restrictive condition (40°C), to enrich for fit in-
dividuals with beneficial alleles. In parallel, we grew the other half
of the pool in a permissive condition (23°C) as a control. Finally,
we sequenced the pools before and at multiple timepoints during
selection to directly assess the changes in parental allele frequen-
cies throughout the genome. The entire procedure was performed
in two biological replicates starting from the same F1 hybrid.
Genetic content of AIL pools before selection
We confirmed that the AIL pools were well-suited for QTL map-
ping. First, we established that nearly all loci were segregating be-
fore selection, with both parental alleles represented. Sequencing
total DNA followed by allele frequency estimation (Methods)
showed that >99% of the mappable genome was segregating in
the F6 pool, and 97% in the F12 pool with minor allele fre-
quency >10% (Supplemental Material SI). Some other parts of the
genome with minor allele frequencies >10% were also selected for
during the intercross rounds without reaching fixation, likely due
to alleles favoring sporulation, mating, or resistance to selection
steps used in the cross (Supplemental Material SI). This allowed us
to map several regions involved in these processes as a byproduct
of our approach, including a previously uncharacterized sporula-
tion QTL in chromosome V (Supplemental Table S2; Supplemental
Material SI).
Usingmanyroundsofcrossesshouldexpandthegeneticmap
due to reduced linkage between nearby loci (Fig. 2A; Darvasi and
Soller 1995). To confirm this, we genotyped 30 markers in 96 in-
dividualsegregantsfromeachofthreegenerations,F1,F6,andF12,
in three regions to assess the change in recombination fraction
between adjacent markers (Data set 1). The genetic distance
(measured as 100 times the average number of recombination
events) between two chromosome XIII loci separated by 204 kb
increased from 88 in F1 to 125 in F6 and 180 in F12 (Fig. 2A). This
increase is less than expected under the assumption of inde-
pendent recombination events and random mating, and is likely
due to the considerable variation in recombination rates along
the genome, with about half of the events concentrated in re-
combination hotspots (discussed in SupplementalMaterial SI). We
further sequenced two segregants from the F6 pool at low coverage
(Supplemental Material SI) and observed 64 and 68 recombination
events (Fig. 2B), an 125% increase compared to 30 events observed
Figure 1.
segregating pool of individuals of various fitness, and growing the pool in a restrictive condition that enriches for beneficial alleles that can be detected via
sequencing total DNA from the pool.
Overall strategy. A three-step QTL mapping strategy by crossing two phenotypically different strains for many generations to create a large
Figure 2.
100 times the average number of recombination events from first to 12th generation (bottom) of a 200-kb chromosome XIII locus genotyped at nine
markers (top). (B) Genetic background of two segregants from a first (F1) and sixth (F6) generation cross shows a sharp increase in recombination events.
Recombination landscape after multiple rounds of intercrosses. (A) Expansion of the genetic map, measured in recombination units (ru) of
1132 Genome Research
www.genome.org
Parts et al.
Cold Spring Harbor Laboratory Press on September 22, 2011 - Published by genome.cshlp.orgDownloaded from

Page 4

on average in a set of 96 F1 segregants genotyped at ;200 evenly
spaced loci (Cubillos et al. 2011).
Changes in population allele frequency in response
to heat selection indicate protective alleles
To identify the alleles underlying variation in high temperature
tolerance, we sequenced DNA from the F12 haploid pool to an
average genome coverage of 253 to 1503 (Supplemental Table S3)
after 0 (T0), 96 (T1), 192 (T2), and 288 (T3, at least 25 genera-
tions, discussed in Supplemental Material SI) hours of growth at
40°C. There were 21 regions where the inferred allele frequency of
the T2 pool was significantly different compared to the control
experiment propagated at 23°C in both of the two biological rep-
licates (Fig. 3A–C; Supplemental Table S4; Methods). We desig-
nated these regions as QTLs, with corresponding false positive rate
Figure 3.
F12 pool before (blue) and after (green) selection. Lines in gene regions in C denote segregating sites (black) and nonsynonymous segregating sites (red).
The sites with intolerable mutations determined by SIFT analysis (Supplemental Material SI; Data set S3) are highlighted with arrows and designated with
the amino acid change. (D) Individual examples of mapped QTLs that show differences in QTL strength, beneficial allele, effect of intercross rounds, and
ploidy. Each window spans 80 kb and is centered on the locus with the largest allele frequency change in F12 T2 across two replicas. Shaded regions
indicate 90% and 95% confidence intervals of the allele frequencies (Supplemental Material SI).
Changes in allele frequencies pinpoint QTLs. (A–C) WA allele frequency of whole genome (A), chromosome II (B), and IRA1 region (C) of the
High resolution QTL mapping
Genome Research
www.genome.org
1133
Cold Spring Harbor Laboratory Press on September 22, 2011 - Published by genome.cshlp.orgDownloaded from

Page 5

<10?3determined from the changes of allele frequencies in the
control (Methods). As we chose this particular cutoff conserva-
tively, there remains a set of lower confidence QTLs with smaller
allele frequency changes that we do not consider here.
The NA allele was selected in two thirds (14/21) of the cases,
consistent with it being the more heat resistant strain (Liti et al.
2009a). However, seven WA alleles were also selected, indicating
antagonistic variants whose effect could only be observed when
decoupled from the rest of the genetic context. The selected alleles
are specific to the heat stress condition, as the same pool propa-
gated at 23°C had no significant changes in allele frequency after
192 h, and selection under oxidative stress (paraquat, 1.5 mM)
yielded a different set of loci with large allele frequency changes
(Supplemental Fig. S2).
Thereisariskthatlongtermculturingunderstressconditions
will select for adaptive mutations rising to high frequencies and
dominating the pool. However, theoretical models suggest this
result is unlikely (Supplemental Material SI). We genotyped 960
segregantsfromtheF12haploidpoolafter240hofselection(T2.5)
at 24 loci (Methods, Data set 2), and observed 787 unique haplo-
types, with no haplotype (meaning here the set of QTL genotypes)
represented more than six times. This is consistent with expecta-
tion under the model of independent selected segregants; if an
adaptive clone had become dominant, its haplotype would have
risen to high frequency (Supplemental Material SI). This suggests
that the selection we observe is against haplotypes that cannot
survive or grow at high temperature, rather than for a haplotype
that performs better than all the others.
In addition to changes in chromosomal allele frequencies, all
the mitochondrial genes were greatly reduced in copy number
upon heat selection (Supplemental Table S5). Interestingly, 960/
960 of the genotyped segregants exhibited a petite, nonrespiration
phenotype when grown using a nonfermentable carbon source
(glycerol and ethanol), indicating the loss of a mitochondrial ge-
nome, most probably as a response to the accumulation of reactive
oxygenspecies(ROS)during heatstress(DavidsonandSchiestl2001;
Rikhvanov et al. 2001). In contrast, none of the 96 F6 or F12 segre-
gants isolated before heat selection exhibited the petite phenotype.
Prolonged artificial selection on AILs improves sensitivity
and resolution
The additional value in using prolonged artificial selection is
twofold. First, it allows alleles with smaller fitness effects to rise in
frequencyandbecomedetectable(Fig.3D).Only10ofthe21QTLs
had significantly changed in allele frequency compared to the
control experiment during the first 96 h under selection (T1), in-
dicating that a longer experiment is required to find all QTLs
(Supplemental Table S6). Second, extended selection can drive al-
lele frequencies to an equilibrium, with no further significant
changestakingplaceapartfromrandomdrift.Indeed,weobserved
that the largest inferred allele frequency change between T2 and
T3 was 4%, and frequencies of 19/21 beneficial alleles changed
<3% (Supplemental Table S6). We estimate by resampling that no
changes >5% take place after T2 even though frequencies of 19
QTL alleles are >13% from fixation (Supplemental Material SI;
Supplemental Table S6). Possible further interpretations of this are
discussed below.
Our method is more sensitive than conventional linkage
mapping. Analysis of 96 genotyped and phenotyped F1 segregants
fromthesamecrosshaspreviouslyfoundonestrongQTLforgrowth
in high temperature at the right end of chromosome XIII (Cubillos
et al. 2011) (log-odds [LOD] score: 12.8, variance explained 66%).
However, none of the other 20 QTL regions had a LOD score
above the 5% FDR cutoff.
The advantage of using AILs with reduced linkage is evident
from narrow mapped intervals, in some cases localizing to single
genes (Fig. 3C,D; Supplemental Fig. S3). We designated mapped
intervals as the segregating sites for which the allele frequency
change is within one standard deviation of the change of the QTL
peak (Supplemental Material SI; Supplemental Table S4). The
resulting regions had a median size of 6.4 kb, and overlapped a
median number of four genes (Supplemental Table S4), in contrast
to regions selected during the intercross, which had a median size
16.3 kb, overlapping a median number of 10.5 genes (Supple-
mental Table S2). For example, for a chromosome II QTL, we could
visually map the selected variant down to a small region of the
IRA1 gene (Fig. 3C), that also harbors the strongest candidate ge-
netic variant from bioinformatic analysis (Supplemental Tables S7,
S8; Supplemental Fig. S4). This resolution is in contrast to that
from previous studies based on crosses between strains, including
Ehrenreich et al. (2010), which typically map to large regions
containing many genes, and can be further improved by using
more advanced statistical methods for estimating the selected in-
terval. The effects of many rounds of crossing and prolonged se-
lection on mapping resolution have been quantitatively addressed
in earlier work (Darvasi and Soller 1995) and are explored in sim-
ulation studies in the Supporting Information.
The RAS/cAMP signaling pathway regulates quantitative
growth at high temperature
The two loci that were fixed in the selected population were IRA1
on chromosome II and a subtelomeric locus of chromosome XIII
(Fig. 3). IRA1 is a GTPase-activating protein that negatively regu-
lates the RAS signaling module (Tanaka et al. 1989). The RAS path-
way is a general hotspot for natural variation (Smith and Kruglyak
2008),aknowntargetofadaptivemutation(KaoandSherlock2008),
and a regulator of protein kinase A (PKA) that in turn controls the
synthesis and activity of stress response genes (Santangelo 2006).
Interestingly, three additional regulators of the same RAS/cAMP
module(IRA2,GPB1,andCYR1),aswellassomeofitstargets(BCY1),
were contained in QTL intervals with an increase in NA allele fre-
quencies in the F12 pool, indicating the involvement of PKA sig-
naling pathway components both upstream and downstream of
cAMP in the heat resistance phenotype (Fig. 4D).
We validated the three strongest mapped QTLs. The QTL at
the right end of chromosome XIII was previously confirmed by
decreased growth rate of the F1 hybrid of parents upon truncation
of the NA subtelomere (Cubillos et al. 2011). However, due to lack
of assembled sequence in the region in the founder strains for this
experiment, we could not identify the responsible gene. We vali-
dated by reciprocal hemizygosity (Steinmetz et al. 2002), deleting
both alleles from a diploid hybrid in turn, that IRA1 and IRA2 al-
leles affect high temperature growth. The effect was evident from
a plating assay, growth curves, and competition experiments (Fig.
4A,B; Supplemental Fig. S5A,B). These genes affect both growth
rate (doubling time) and efficiency (final density) of segregants,
with IRA1 having a stronger effect than IRA2, consistent with the
difference in their final allele frequencies (Supplemental Fig.
S5A,B). Interestingly, the same IRA1 and IRA2 alleles do not have
a strong differential effect on growth in other stress conditions
(Supplemental Fig. S5C), even though RAS activity is involved in
the response to these stresses (Park et al. 2005).
Parts et al.
1134Genome Research
www.genome.org
Cold Spring Harbor Laboratory Press on September 22, 2011 - Published by genome.cshlp.orgDownloaded from

Page 6

We further constructed all four possible double hemizygous
combinations of IRA1 and IRA2 and grew them at 40°C. The
doubling time for the strain with WA alleles was 3.3-fold longer
compared to the multiplicative expectation (Supplemental Mate-
rial SI), indicating a negative epistatic interaction between these
alleles, which is consistent with partially redundant function
(Supplemental Fig. S6). Furthermore, at 40°C we detected a nine-
foldhigherlevelofinternalcAMPinthehybridcarryingWAalleles
of the IRA genes compared to the hybrid carrying NA alleles (P <
0.005, one-sided t-test) (Fig. 4C). This is consistent with RAS hy-
peractivity in the WA strain due to loss of negative regulation from
IRA genes, resulting in a higher level of internal cAMP, contribut-
ing to its heat sensitivity (Fig. 4D). The RAS/cAMP signaling path-
way has previously been implicated in the accumulation of ROS
under stress (Hlavata et al. 2003). Our results are consistent with
this functionality being under selection alongside the selection for
removal of the mitochondria.
Different QTL haplotypes are maintained after selection
It is surprising that for 19 QTLs both alleles remained segregating
in the pool after 192 h under selection (T2), and did not signifi-
cantly change in frequency in a further 96 h in the restrictive
condition (Supplemental Figs. S7, S8; Supplemental Table S6). This
suggests that we have saturated for individual alleles with strong
independenteffects that are present in the founding strains. It also
indicatesthat all the haplotypes remainingin the pool have nearly
equal fitness in this stress condition, or are so rare even by T3 that
change in their frequency does not have a major effect on the
average pool genotype. These observations are consistent with
an abundance of negative epistatic interactions, where particular
combinations of alleles are selected against.
To test this explanation, we isolated and genotyped 960 seg-
regants from the F12 pool after 240 h of selection (T2.5), and
looked for scarcity and abundance of specific allele combinations
at 19 QTLs (Supplemental Table S9, Data set 2, and Supplemental
Material SI). None of the interchromosomal two-locus genotype
combinations was significantly different from the expectation
under independence after correcting for multiple testing (lowest
multiple testing corrected P > 0.1, two-sided Fisher’s exact test,
Supplemental Table S9).
The apparent lack of allele fixation and strong interchro-
mosomal interactions after 12 d under selection could be ex-
plained by various models involving one or more loci. First, it is
possible that individual beneficial alleles are still changing in fre-
quency, albeit slowly, and would fix if the selection was carried out
for a much longer period. Second, there could be two-locus epis-
tasis within a linked region that we are unable to detect from the
genotypingdata.Finally, thedataarealso consistentwithcomplex
control and interactions involving multiple genes (Phillips 2008).
Figure 4.
WA/NAhybridswereindividuallydeletedfortheIRAallelesandusedtoassesstheircontributiontohightemperaturegrowth.Platespottingassayusing10-fold
serial dilution demonstrates better growth of the hybrid when the NA allele is present. (B) Competition experiment on hybrids with IRA1/IRA2 reciprocal
hemizygousdeletions(suchasA)thatresemblestheselectivestepappliedtothepool.HybridscarryingtheNAalleleoutcompeteoneswithWAalleleafter192
h (T2) of growth at 40°C. (C) Internal level of cAMP is reduced at 40°C, but unchanged at 30°C for WA/NA hybrids with WA alleles deleted at both IRA1 and
IRA2 loci, compared to NA alleles deleted. (D) RAS/cAMP signaling contributes to natural variation in heat sensitivity. Defective function of the WA alleles of
IRA1andIRA2athightemperatureresultsinhyperactiveRAS,leadingtohighlevelofcAMPandhighPKAactivityinhibitingtheheattranscriptioninduction.As
a response to heat stress, the majority of the QTLs selected in the pool are from the NA genetic background (red: NA; blue: WA). Dashed arrow indicates
unknown mechanism. Figure adapted from Figure 2 of Santangelo (2006) and reprinted with permission from the American Society for Microbiology.
IRA1 and IRA2 are high temperature growth QTLs. (A) Reciprocal hemizygosity confirms that IRA1 and IRA2 are high temperature growth QTLs.
High resolution QTL mapping
Genome Research
www.genome.org
1135
Cold Spring Harbor Laboratory Press on September 22, 2011 - Published by genome.cshlp.orgDownloaded from

Page 7

QTL mapping in a heterozygous diploid population
Importantly for drawing comparisons with genetic studies of
complex traits in humans and other diploid organisms, 18 of 21
heat resistance QTL alleles significantly changed in frequency in
the pool of heterozygous diploid individuals (Supplemental Table
S6; Data set 3). The process of selection was slower for the diploid
pool, as allele frequencies continued to change between T2 and T3
(SupplementalTableS6;Fig.3D).Examiningthediploidpoolallele
frequency after selection at T3, the NA chrXIII QTL allele is dom-
inant with only the homozygous deleterious genotype (WA/WA)
being removed from the pool, and the NA IRA1 allele is recessive
with the beneficial allele (NA/NA genotype) being fixed. For 10 of
the other 16 detected loci it is surprising that the allele frequencies
after 12 d of selection on the diploid segregants were within 5% of
the haploid pool after selection (Supplemental Table S6; Data set
3), consistent with the selected alleles having additive effects
as observed for most human GWAS hits (Manolio et al. 2009).
However, as we did not observe data from consecutive timepoints
without significant allele frequency changes for the diploid pools,
it is also possible that the allele frequencies had not yet reached
equilibrium.
Extensions and applications of the method
It is straightforward to apply our method to any selectable trait,
including ones that do not affect fitness. For example, cell sorting
(Ehrenreich et al. 2010) to select for cell size or GFP expression on
specific promoters, ageing the population for detecting chrono-
logical life span QTLs (Fabrizio and Longo 2003) or washing the
platetodetectcelladhesiontraits(ReynoldsandFink2001),canall
be used. A similar approach may also be adapted to other model
genetic systems, including Drosophila melanogaster and Caeno-
rhabditis elegans, which are amenable to crossing in bulk and large
experimental population sizes. In these other models, the ratio
of recombination rate to gene density determines the ability to
identify responsible genes.
The heterozygous diploid intercross pool can be used for
other genetic studies, such as dissecting mechanisms that con-
tribute to heterosis (Lippman and Zamir 2007), and extended to
includemoreofthegeneticdiversityinthespecies.Weexpecttobe
able to cross a larger number of parental strains to expand the
range of standing variation segregating in the pool. Therefore, we
have the potential to establish an artificial outbred yeast pop-
ulation that can be used as a model for natural diploid genome-
wide association studies as carried out in humans.
Methods
Intercross
Parental strains YPS128 (MATa, hoTHphMX4, ura3TKanMX4) and
DBVPG6044 (MATa, hoTHphMX4, ura3TKanMX4, lys2TURA3)
were crossed in complete media (YPDA) and grown overnight.
Patches were replica plated in synthetic minimal media (MIN) to
select for diploid F1 hybrids. F1 hybrids were isolated and stored
at?80°C.TwoF1hybridreplicasweregrownovernightandreplica
plated on KAc at 23°C to be sporulated for 10 d until 90% sporu-
lationefficiency.Cellswerecollectedandresuspendedin 0.5mLof
sterile water, treated with an equal amount of ether and vortexed
for10mintokillunsporulatedcells(DawesandHardie1974).Cells
were washed four times in sterile water, resuspended in 900 mL of
sterile water, and treated with 100 mL of Zymolase (10 mg/mL) to
remove ascus. Cell mixtures were vortexed for 5 min to increase
spore dispersion and inter-ascus mating. For the heterozygous
diploid pool, we forced one extra round of mating and selection
for LYS+/URA3+ cells. Full details are presented in Supplemental
Material SI.
Selection experiment
Pools of 10–100 million cells were collected from sporulation
media and treated with ether and zymolase. Spores were plated in
YPDA and incubated until full growth was obtained. Each plate
was incubated for 48 h, and then resuspended in distilled water.
Ten percent of the cells were used for next replating, and the rest
for DNA extraction.
Genotyping
SNP genotypes were obtained by real time PCR coupled to high
resolution melting (HRM) using the Corbet Rotorgene and Quan-
tace PCR HRM mix. Sequenom genotyping of 960 segregants was
performed using the iPLEX Gold Assay (Sequenom Inc.) (Supple-
mental Material SI).
DNA isolation, library preparation, and sequencing
DNA was extracted using the phenol chloroform protocol. Multi-
plexed PCR-free Illumina sequencing libraries were prepared as in
Kozarewa et al. (2009) with modifications (Supplemental Material
SI). Fragments with 200–300-bp inserts were gel-purified and se-
quenced using standard Illumina SBS v4 chemistry for 2 3 76 cy-
cles plus extra seven cycles to determine the tag sequence of each
cluster.
Sequencing data handling
Sequencing reads were mapped to the S288c reference genome
obtainedfrom theSGRPprojectwebsite(http://www.sanger.ac.uk/
research/projects/genomeinformatics/sgrp.html) using BWA (Li
and Durbin 2009), with option ‘‘-n 8.’’ Pileup files comprising the
genotypes of mapped reads were created for segregating sites
inferred from both low-coverage capillary sequencing (Liti et al.
2009a) and the parental strain shotgun sequence mapping to the
S288c assembly, and sites further filtered.
Parental strain analysis
The parental strain sequence was mapped similarly to the selec-
tion experiment. We used the SAMtools (Li et al. 2009) variant
caller with default settings to call differences from the reference
sequence, andused these data to update the list of segregating sites
used in the allele frequency analysis.
Segregant analysis
A site was called to be from one parent, if it was covered by at least
15 sequencing reads with base and mapping qualities at least 30,
and 80% of them had the parental allele. We conservatively
refrainedfrommakingacallatlow-coveragevariants,subtelomeric
regions up to 30 kb, and variants with ambiguous mapping data.
We called a recombination event if a region of at least 2 kb from
one parent was followed by a region of at least 2 kb from the other,
and at least five calls were made in both regions.
Allele frequency inference
Under a simple model, there is an unobserved WA allele frequency
flat each locus l; we want to infer the posterior distribution of
Parts et al.
1136 Genome Research
www.genome.org
Cold Spring Harbor Laboratory Press on September 22, 2011 - Published by genome.cshlp.orgDownloaded from

Page 8

flafter observing the sequence data. We assume all reads to come
from different segregants after filtering segregating sites to be dis-
tant, thus every segregant i has one allele ai observed at some locus
l9 distance diaway from l. We take d to be infinity if the loci are on
different chromosomes. For that segregant, there is an unobserved
allele blat locus l, and the probability that these loci are linked,
with no recombination event occurring during the intercross be-
tween them, is qi = exp(?dir), where r is the recombination rate.
We took r = 30 (1 + [g ? 1]/2), where g is the number of intercross
rounds, as there is on average 30 crossovers per tetrad, and every
intercross after the first one has a 50% chance of introducing
a switch between parental haplotypes. The likelihood of the allele
frequency at locus l is thus
P(D | fl) =Q
=Pðaijbi=‘WA’ÞPðbi=‘WA’jflÞ+Pðaijbi=‘NA’ÞPðbi=‘NA’jflÞ
=qiai=‘WA’ð1?qiÞai=‘NA’fl+qiai=‘NA’ð1?qiÞai=‘WA’ð1?flÞ
»qiflai=‘WA’ð1?flÞai=‘NA’:
Here, we have discarded likelihood terms that require a re-
combinationevent,aswewillfilterqitobelarge.Wecalculatedthe
posterior (beta) distribution of flby applying Bayes rule: P(fl| A) }
P(A | fl) P(fl) =Q
ness). This inference procedure corresponds to a smoothing ap-
proach within a fixed window with the width determined by the
recombination rate (;6 kb for qi > 0.9), and has the effect of dis-
criminatingagainstextremeallelefrequencies.Theposteriormean
and confidence intervals were obtained from the approximated
Beta distribution.
iP(ai | fl), where
PðaijflÞ=Pðai;bi=‘WA’jflÞ+Pðai;bi=‘NA’jflÞ
iP(ai | fl) P(fl), where the beta prior P(fl) is un-
informative, and we filter qi > 0.9 (0.75 for Fig. 3A,B for smooth-
Allele frequency change
We called a QTL if the inferred allele frequency changed in the
same direction by at least 10% in both biological replicas, and the
change was larger than four times the average standard deviation
of the inferred allele frequencies. One QTL was called in any 50-kb
window, corresponding to the variant with the largest combined
allele frequency change over two replicas. We assessed the signif-
icance of the calls using the null distribution of allele frequency
changes from the control experiment, where the initial pool was
propagated in permissive temperature alongside the selected pool
for 144 h (T2). Due to the repetitive nature of subtelomeric regions
resulting in a lack of assemblies and low sequencing coverage,
we did not consider loci within 30 kb of the end of chromosomes.
We fit a normal distribution to the allele frequency changes at the
26,871 loci assessed (Supplemental Fig. S9), and calculated the
probabilityofobservingachangeofatleast10%ineitherdirection
to be <10?7. After Bonferroni-correcting for the 26,871 tests, the
P-value remained <10?3.
F1 segregants
We used standard marker regression for 200 genotyped markers
and heat growth rate phenotype to map QTLs significant at 5%
false discovery rate (FDR) using a standard linear model and 1000
permutations in rQTL (Broman et al. 2003).
Interaction tests
We tested for scarcity and abundance of two-locus genotype
combinations by the two-sided Fisher’s exact test using the fisher.
testfunctioninRonthetwo-locusgenotypecountsforeachpairof
genotyped loci.
Reciprocal hemizygosity
IRA1 and IRA2 were deleted individually or in the four possible
combinations (Supplemental Table S1) in hybrid strain YCC22F
using the standard single-step PCR gene deletion method (Wach
et al. 1994). We performed a temperature growth assay by plating
serial dilutions of cells in YPDA and incubated the plates at 30°C
and 40°C for 48 h.
Competitive growth
We competed the reciprocal hemizygous hybrids (naD/WA vs.
waD/NA) by mixing equal numbers of cells, and growing at either
30°C or 40°C for 96 h. Pyrosequencing was used to assess the allele
frequency in the pools.
Phenotyping
Individual yeast segregants and reciprocal hemizygous were phe-
notyped using high-resolution microcultivation instruments Bio-
screen C (Growth curve Oy, Finland) for quantitative growth as
previously described (Warringer and Blomberg 2003; Liti et al.
2009a).
cAMP determination
IntracellularcAMP was determined using a commercially available
kit (LANCE cAMP 384 kit, Perkin-Elmer). Values presented are av-
erage values determined in three to six replicate cultures per strain
and error bars indicate SEM.
Acknowledgments
We thank all the members of the Sanger Sequencing, Sequenom
Genotyping, and Sample Logistics teams for generating the se-
quence and genotype data, and G. Russo, E. Scovacricchi, and A.
Mott for technical help. We thank C. Nieduszynski for comments
and suggestions, and V. Mustonen for discussions on epistasis.
Research attheWellcomeTrustSangerInstitute(L.P, S.J.B.,M.A.Q.,
J.T.S., and R.D.)is supported by the WellcomeTrust (WT077192/Z/
05/Z). G.L., F.A.C., K.J., F.S., and E.J.L. were supported by the
WellcomeTrust(WT084507MA), theRoyalSociety,andtheBBSRC
(BBF0152161). F.S. was supported by the Mecesup (UCH0604) and
Becas Chile. J.W. was supported by the Royal Swedish Academy of
Sciencesand The Carl Trygger Foundation. M.M. was supported by
Magnus Bergvalls Stiftelse. A.M.M. and A.Z. were supported by
Canada Foundation for Innovation and CIHR grant no. 202372.
References
Brem RB, Yvert G, Clinton R, Kruglyak L. 2002. Genetic dissection of
transcriptional regulation in budding yeast. Science 296: 752–755.
Broman KW, Wu H, Sen S, Churchill GA. 2003. R/qtl: QTL mapping in
experimental crosses. Bioinformatics 19: 889–890.
Cubillos FA, Billi E, Zo ¨rgo ¨ E, Parts L, Fargier P, Omholt S, Blomberg A,
Warringer J, Louis EJ, Liti G.2011. Assessing the complex architecture of
polygenic traits in diverged yeast populations. Mol Ecol 20: 1401–1413.
Darvasi A, Soller M. 1995. Advanced intercross lines, an experimental
population for fine genetic mapping. Genetics 141: 1199–1207.
Davidson JF, Schiestl RH. 2001. Cytotoxic and genotoxic consequences of
heat stress are dependent on the presence of oxygen in Saccharomyces
cerevisiae. J Bacteriol 183: 4580–4587.
Dawes IW, Hardie ID. 1974. Selective killing of vegetative cells in sporulated
yeast cultures by exposure to diethyl ether. Mol Gen Genet 131: 281–289.
Demogines A, Smith E, Kruglyak L, Alani E. 2008. Identification and
dissection of a complex DNA repair sensitivity phenotype in Baker’s
yeast. PLoS Genet 4: e1000123. doi: 10.1371/journal.pgen.1000123.
Deutschbauer AM,DavisRW.2005. Quantitativetrait locimappedtosingle-
nucleotide resolution in yeast. Nat Genet 37: 1333–1340.
High resolution QTL mapping
Genome Research
www.genome.org
1137
Cold Spring Harbor Laboratory Press on September 22, 2011 - Published by genome.cshlp.orgDownloaded from

Page 9

Ehrenreich IM, Torabi N, Jia Y, Kent J, Martis S, Shapiro JA, Gresham D,
Caudy AA, Kruglyak L. 2010. Dissection of genetically complex traits
with extremely large pools of yeast segregants. Nature 464: 1039–1042.
Fabrizio P, Longo VD. 2003. The chronological life span of Saccharomyces
cerevisiae. Aging Cell 2: 73–81.
Gerke J, Lorenz K, Cohen B. 2009. Genetic interactions between
transcription factors cause natural variation in yeast. Science 323: 498–
501.
Hlavata L, Aguilaniu H, Pichova A, Nystrom T. 2003. The oncogenic
RAS2(val19) mutation locks respiration, independently of PKA, in
a mode prone to generate ROS. EMBO J 22: 3337–3345.
Hunter KW, Crawford NP. 2008. The future of mouse QTL mapping to
diagnosediseaseinmiceintheageofwhole-genomeassociationstudies.
Annu Rev Genet 42: 131–141.
Kao KC, Sherlock G. 2008. Molecular characterization of clonal interference
during adaptive evolution in asexual populations of Saccharomyces
cerevisiae. Nat Genet 40: 1499–1504.
Kozarewa I, Ning Z, Quail MA, Sanders MJ, Berriman M, Turner DJ. 2009.
Amplification-free Illumina sequencing-library preparation facilitates
improvedmappingandassemblyof(G+C)-biasedgenomes.NatMethods
6: 291–295.
Li H, Durbin R. 2009. Fast and accurate short read alignment with Burrows-
Wheeler transform. Bioinformatics 25: 1754–1760.
Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, Marth G,
Abecasis G, Durbin R. 2009. The Sequence Alignment/Map format and
SAMtools. Bioinformatics 25: 2078–2079.
Lippman ZB, Zamir D. 2007. Heterosis: Revisiting the magic. Trends Genet
23: 60–66.
Liti G, Carter DM, Moses AM, Warringer J, Parts L, James SA, Davey RP,
Roberts IN, Burt A, Koufopanou V, et al. 2009a. Population genomics of
domestic and wild yeasts. Nature 458: 337–341.
Liti G, Haricharan S, Cubillos FA, Tierney AL, Sharp S, Bertuch AA, Parts L,
Bailes E, Louis EJ. 2009b. Segregating YKU80 and TLC1 alleles
underlying natural variation in telomere properties in wild yeast. PLoS
Genet 5: e1000659. doi: 10.1371/journal.pgen.1000659.
Manolio TA, Collins FS, Cox NJ, Goldstein DB, Hindorff LA, Hunter DJ,
McCarthyMI,RamosEM,CardonLR,ChakravartiA,etal.2009.Finding
the missing heritability of complex diseases. Nature 461: 747–753.
Nogami S, Ohya Y, Yvert G. 2007. Genetic complexity and quantitative trait
loci mapping of yeast morphological traits. PLoS Genet 3: e31. doi:
10.1371/journal.pgen.0030031.
Park JI, Grant CM, Dawes IW. 2005. The high-affinity cAMP
phosphodiesterase of Saccharomyces cerevisiae is the major determinant
of cAMPlevelsin stationary phase: Involvement of different branches of
the Ras-cyclic AMP pathway in stress responses. Biochem Biophys Res
Commun 327: 311–319.
Perlstein EO, Ruderfer DM, Ramachandran G, Haggarty SJ, Kruglyak L,
Schreiber SL. 2006. Revealing complex traits with small molecules and
naturally recombinant yeast strains. Chem Biol 13: 319–327.
Phillips PC. 2008. Epistasis—the essential role of gene interactions in the
structure and evolution of genetic systems. Nat Rev Genet 9: 855–867.
Reynolds TB, Fink GR. 2001. Bakers’ yeast, a model for fungal biofilm
formation. Science 291: 878–881.
Rikhvanov EG, Varakina NN, Rusaleva TM, Rachenko EI, Kiseleva VA,
Voinikov VK. 2001. [Heat shock-induced changes in the respiration of
the yeast Saccharomyces cerevisiae]. [Article in Russian]Mikrobiologiia 70:
531–535.
Romano GH, Gurvich Y, Lavi O, Ulitsky I, Shamir R, Kupiec M. 2010.
Different sets of QTLs influence fitness variation in yeast. Mol Syst Biol 6:
346. doi: 10.1038/msb.2010.1.
Santangelo GM. 2006. Glucose signaling in Saccharomyces cerevisiae.
Microbiol Mol Biol Rev 70: 253–282.
Segre ` AV, Murray AW, Leu JY. 2006. High-resolution mutation mapping
reveals parallel experimental evolution in yeast. PLoS Biol 4: e256. doi:
10.1371/journal.pbio.0040256.
Sinha H, David L, Pascon RC, Clauder-Munster S, Krishnakumar S, Nguyen
M, Shi G, Dean J, Davis RW, Oefner PJ, et al. 2008. Sequential
elimination of major-effect contributors identifies additional
quantitative trait loci conditioning high-temperature growth in yeast.
Genetics 180: 1661–1670.
Smith EN, Kruglyak L. 2008. Gene-environment interaction in yeast gene
expression. PLoS Biol 6: e83. doi: 10.1371/journal.pbio.0060083.
Steinmetz LM, Sinha H, Richards DR, Spiegelman JI, Oefner PJ, McCusker
JH, Davis RW. 2002. Dissecting the architecture of a quantitative trait
locus in yeast. Nature 416: 326–330.
Tanaka K, Matsumoto K, Toh EA. 1989. IRA1, an inhibitory regulator of the
RAS-cyclicAMPpathwayinSaccharomycescerevisiae.MolCellBiol9:757–
768.
Valdar W, Solberg LC, Gauguier D, Burnett S, Klenerman P, Cookson WO,
Taylor MS, Rawlins JN, Mott R, Flint J. 2006. Genome-wide genetic
association of complex traits in heterogeneous stock mice. Nat Genet 38:
879–887.
Wach A, Brachat A, Pohlmann R, Philippsen P. 1994. New heterologous
modules for classical or PCR-based gene disruptions in Saccharomyces
cerevisiae. Yeast 10: 1793–1808.
Wang X, Le Roy I, Nicodeme E, Li R, Wagner R, Petros C, Churchill GA,
Harris S, Darvasi A, Kirilovsky J, et al. 2003. Using advanced intercross
lines for high-resolution mapping of HDL cholesterol quantitative trait
loci. Genome Res 13: 1654–1664.
Warringer J, Blomberg A. 2003. Automated screening in environmental
arrays allows analysis of quantitative phenotypic profiles in
Saccharomyces cerevisiae. Yeast 20: 53–67.
Wenger JW, Schwartz K, Sherlock G. 2010. Bulk segregant analysis by high-
throughput sequencing reveals a novel xylose utilization gene from
Saccharomyces cerevisiae. PLoS Genet 6: e1000942. doi: 10.1371/
journal.pgen.1000942.
Yvert G, Brem RB, Whittle J, Akey JM, Foss E, Smith EN, Mackelprang R,
Kruglyak L. 2003. Trans-acting regulatory variation in Saccharomyces
cerevisiae and the role of transcription factors. Nat Genet 35: 57–64.
Received October 20, 2010; accepted in revised form March 16, 2011.
Parts et al.
1138 Genome Research
www.genome.org
Cold Spring Harbor Laboratory Press on September 22, 2011 - Published by genome.cshlp.orgDownloaded from