TREE 2303 No.

of Pages 12

Opinion
Genomic Quantitative
Genetics to Study Evolution in
the Wild
Phillip Gienapp,1,* Simone Fior,2 Frédéric Guillaume,3
Jesse R. Lasky,4 Victoria L. Sork,5 and Katalin Csilléry3,6
Quantitative genetic theory provides a means of estimating the evolutionary
Trends
potential of natural populations. However, this approach was previously only
Information about genetic (co)var-
feasible in systems where the genetic relatedness between individuals could be iances of traits is essential to under-
inferred from pedigrees or experimental crosses. The genomic revolution stand and predict evolutionary
change. This has gained additional
opened up the possibility of obtaining the realized proportion of genome shared importance in the light of contempor-
among individuals in natural populations of virtually any species, which could ary human-induced environmental
promise (more) accurate estimates of quantitative genetic parameters in virtu- change to which species and popula-
tions need to adapt.
ally any species. Such a ‘genomic’ quantitative genetics approach relies on
fewer assumptions, offers a greater methodological flexibility, and is thus Quantitative genetics studies genetic
(co)variances based on the phenotypic
expected to greatly enhance our understanding of evolution in natural pop-
resemblance of related individuals.
ulations, for example, in the context of adaptation to environmental change, Relatedness among individuals can
eco-evolutionary dynamics, and biodiversity conservation. be known from observed pedigrees
or breeding experiments but such
information can only be gathered for
Understanding Trait Evolution in Natural Populations Using Quantitative a limited range of species.

Genetics The advances in genomics tools now

Reliably predicting responses to selection is essential in animal and plant breeding, for studying allow us to genotype virtually any spe-
evolution in natural populations, and [31_TD$IF]for understanding population and species response to cies for thousands of genomic mar-
rapid anthropogenic environmental change (e.g., [1]). Building on the seminal work of Fisher [2], kers. Relatedness estimates from
such genomic markers have been
who first proposed to decompose the variation in a continuous trait into genetic and environ- used in human genetics as well as
mental components, breeders and evolutionary biologists have long used quantitative genetic animal and plant breeding, but doing
(QG) theory to estimate the heritable proportion of trait variation and covariation with other traits so also in wild populations has the
to predict response to natural and artificial selection [3]. Partitioning the variance in a trait has potential to considerably advance
our understanding of adaptation in
been possible by assuming a simple model of genetic determinism: each trait is affected by the wild.
many loci of individually (infinitesimally) small effects, known as the infinitesimal model (see
Glossary). Despite this strong, and not necessarily realistic assumption underlying QG [4], it has
been very efficient and successful in predicting selection responses in animal and plant
breeding over the [312_TD$IF]past century [5].

1
The key requirement for any QG analyses is information about the relatedness among individ- Department of Animal Ecology,
Netherlands Institute of Ecology
uals. The resemblance between relatives is the basis for separating the heritable proportion of (NIOO-KNAW), Wageningen, The
the trait variance (the additive genetic variance) that determines the response to selection and Netherlands
2
potential for evolution. The generalization of QG theory to include any kind of relatives in a single Plant Ecological Genetics, ETH
Zurich, Switzerland
statistical framework called the ‘animal model’ has become the major tool in animal breeding to 3
Department of Evolutionary Biology
predict the ‘genetic merit’ or breeding value of individuals [6]. The animal model also opened up and Environmental Studies, University
the possibility to study evolution in the wild, because it was no longer necessary to perform of Zurich, Switzerland
4
Department of Biology, Pennsylvania
crosses to obtain many pairs of individuals of the same relationship, instead, any degree of State University, [308_TD$IF]University Park, PA,
relatedness was regarded as informative. Nevertheless, the application of QG in wild USA

Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy http://dx.doi.org/10.1016/j.tree.2017.09.004 1
© 2017 Elsevier Ltd. All rights reserved.
 TREE 2303 No. of Pages 12

populations remained a tool of privilege, limited to long-term studies, where pedigrees [31_TD$IF]could be
5
Department of Ecology [309_TD$IF]and
Evolutionary Biology, University of
established from individual observations, biasing the approach towards a handful of mammal California [310_TD$IF]Los Angeles, Los Angeles,
and bird species [7]. CA, USA
6
Biodiversity and Conservation
Biology, WSL Swiss Federal Research
With the genomic revolution, QG moved away from the spotlight, and the focus shifted to Institute, Birmensdorf, Switzerland
searching for the genetic determinants of phenotypic variation, that is, mapping quantita-
tive trait loci (QTL). Thanks to lower genotyping costs, mapping at the finest possible level
was enabled by genome-wide association studies (GWAS) that aim to find markers, most *Correspondence:
[email protected] (P. Gienapp).
often single nucleotide polymorphisms (SNPs), associated with phenotypic variation (e.
g., [8]). However, GWAS [314_TD$IF]has not yet met the high expectations: in many cases, a large part
of the phenotypic variance remained unexplained by SNPs that reached genome-wide
significance (the so-called missing heritability problem); a problem that is now partly
resolved by increased sample sizes [8]. Furthermore, even the largest GWAS studies in
humans have not yet reached a plateau of the number of ‘risk factors’ (significant SNPs)
associated with trait variation [8]. Not so surprisingly, GWAS conducted in natural plant and
animal populations [315_TD$IF]has limited power and often failed to identify SNPs associated with traits
(reviewed in [9]); yet the lack of any genome-wide significant loci indicates the absence of
major effect loci. Some of these studies instead found that the proportion of variance
explained by a chromosome correlated with its size [10,11], as expected if these traits are
under the influence of many loci of small effects. All these findings support that the
infinitesimal model of classical QG theory is a reasonable approximation for most quantita-
tive traits, including life history, behavior, or morphology [12]. The QG approach also has the
advantage of not relying on unknown underlying molecular or functional details of the
genetic basis of quantitative traits and directly provides estimates of the evolutionary
potential of such traits [13]. In this Opinion paper, we promote a so-called ‘genomic
QG’, which makes use of genomic data almost exclusively for estimating the relatedness
between individuals, thus opening the door to many applications of QG to [316_TD$IF]questions
previously off-limits in many natural systems. Genomic QG has the potential to improve
our ability to mitigate species’ responses to environmental change (e.g., [1]), and contribute
to the field of conservation genetics, such as identifying ‘pre-adapted’ individuals in
programs of assisted migration to ‘help’ species keep up with climate change by
estimating genetic merit of translocated individuals (e.g., [14]).

From Pedigrees to Genome Sharing

Relatedness is traditionally based on pedigrees at the base of which individuals are assumed
to be unrelated (founders) [15]. Given a pedigree, it can be determined whether two alleles at
a particular locus that are identical by state (IBS) are also identical by descent (IBD)
because they were inherited from the same ancestor. Due to Mendelian segregation, there
is inherent variation in IBD across loci. For example, full-sibs carry, on average, 50% of
their alleles IBD, but the exact percentage of genome sharing and hence their realized
relatedness is subject to variation (Figure 1A). The amount of variation around the expec-
tations depends upon the pedigree-relationship class, and is zero only for parents and their
offspring and monozygotic twins. Pedigree-based relatedness has well known shortcom-
ings, such as its arbitrariness with respect to a founder population and its imprecision
regarding IBD allele sharing (e.g., [16]). Nevertheless, it has been standard in the vast majority
of past QG studies.

Even though different relatedness estimators that approximate IBD probabilities have been
suggested in the microsatellite era, they had too large sampling errors (e.g., [17]); thus it was
widely accepted that markers should only be used to check, correct, and complete pedigrees
rather than [317_TD$IF]replace them [18]. With the advent of genomic data (Box 1), it has become possible
to estimate the realized proportion of the genome that two individuals share IBD. The proportion

2 Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy

 TREE 2303 No. of Pages 12

Box 1. Methods to Estimate the Genomic Relationship Matrix (GRM)

Glossary
Approximate Identity-by-Descent (IBD) Probabilities at Single Locus
Assisted migration: many species
This approach includes a whole class of maximum-likelihood (e.g., [61,64]) and method of moments estimators (see [17] have limited dispersal abilities or their
for an overview). dispersal is impaired by habitat
fragmentation, which means that
Cumulative Excess of Recent Coalescence they cannot reach other, isolated
populations or currently unoccupied
A coalescent tree is used to trace the origin of alleles in the current population back to a most recent common ancestor
but suitable habitat. In assisted
(MRCA). Rousset [39], suggested to define kinship coefficients as the degree of excess coalescent time to their MRCA
migration, or assisted gene flow,
at each independently segregating segment of the genome, however, no estimator or software have been proposed
individuals are artificially moved from
yet.
one population to improve another
population’s probability to persist or
Average Allelic Correlation (Also Known as Unified Additive Relationship) to colonize novel but suitable habitat,
This is the most standard class of approaches to estimate the GRM in animal and plant breeding and human genetics for example, in the context of climate
(see Table I for software). In breeding, the method appeared with the uprise of genomic selection (GS), and [29_TD$IF]is calculated change.
as the correlation between genotypes of two individuals across all loci [65]. In human genetics, the approach appeared Evolvability: can be defined in the
in Yang et al. [28], and was explained in Powell et al. [66]. The approach circumvents the problem of the [293_TD$IF]arbitrary broader and narrower sense. In the
reference population by defining it as the current population. As a result, IBD is no longer a probability, but a correlation broader sense it simply is the ability
coefficient that can be used to predict if a gamete carries the same allele as another irrespective of its origin (i.e., IBS). of a trait to evolve (including from
Adjustment for linkage disequilibrium (LD) [294_TD$IF]has been proposed by [16], which can lead to improved estimates, especially novel mutations). In the narrower
with sparse SNP data [67]. When using dense SNP data, LD-adjustment attributes too much weight to the SNPs with sense it is the additive genetic
low minor allele frequencies. variance scaled by the trait mean:
evolvability ¼ trait Vmean
A
2 , or the
Haplotype Sharing additive genetic coefficient of
Haplotype sharing approaches require both marker data and a genetic map (e.g., [68]). A common approach in GS is to variation (CVA):
use [295_TD$IF]a dense SNP chip on a few key individuals to identify haplotypes, and a sparse chip on the rest of the individuals
CV A ¼ traitsmean
A
.
(reviewed in [23]).
Genome-wide association study
Table I. Examples of Software to Estimate the GRM (GWAS): aims to identify loci
Software PLINK (v1.07) GCTA LDAK BEAGLE determining a trait by testing for
associations between molecular
Method Approximate IBD probabilities Average allelic Average allelic Haplotype
markers and trait values. In the
at single locus correlation correlation sharing
common case when SNP chips are
Refs (software) [69]i [70]ii [16]iii [68]iv used, the underlying assumption is
that the markers are in tight linkage
with or within the loci responsible for
trait variation.
Genomic relationship matrix
(GRM): the pedigree-based
relatedness matrix, often denoted A,
of genome shared between all pairs of individuals in a population can be summarized with the contains all pairwise relatedness
genomic relationship matrix (GRM) (see Box 1 for methods). estimates (which are twice the
kinship coefficient u) in the off-
diagonals and the individuals’
The number of markers necessary to accurately estimate the GRM ultimately depends on the inbreeding coefficient in the diagonal.
effective number of independently segregating chromosome segments[318_TD$IF]. This number is influ- The GRM, often also denoted G, is
enced by the effective population size, the recombination rate[319_TD$IF], and the genome size [19] and [368_TD$IF] the same but with the relatedness
also by the [320_TD$IF]mating system, population structure and history, which concertedly determine estimates derived from genomic
markers. Various approaches to
linkage disequilibrium decay. Generally speaking, species with [321_TD$IF]big genomes, where the estimate relatedness from markers
difference between expected and realized relatedness can be high [19,20][32_TD$IF], need more markers exist (see Box 1) and GRMs can be
than species with [32_TD$IF]small genomes. A greater number of markers is also expected to be constructed with any of them.
Heritability: the narrow-sense
necessary in outcrossing populations in comparison to selfing, partially selfing or inbred
heritability (h2) is the ratio of
populations because outcrossing populations generally have large effective population sizes the additive genetic variance
and high recombination rates [20,21]. It is difficult to provide a quantitative guidance for the (VA) over the total phenotypic
variance (VP) of a trait: h ¼ VV A .
2
required number of markers for a specific study. Thus, performing simulations using character- P
The narrow-sense heritability is used
istics of the studied species and populations, and exploring multiple GRM estimation methods
to predict the response to selection
(Box 1) might be the best approach [324_TD$IF](see Outstanding Questions). of a trait when using the breeders’
equation. When only the whole
The ‘Dawn’ of Genomic Quantitative Genetics genetic variance (VG) is known,
which also includes nonadditive
For humans and species of economic importance, genomic data became available much
genetic effects (dominance and
earlier than for other species, thus the first genomic QG tools appeared in these fields. Many of

Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy 3

 TREE 2303 No. of Pages 12

these methods have not reached evolutionary [325_TD$IF]biology yet. First, we review these tools, then epistasis), broad-sense heritability
discuss the initial applications in natural populations. (H2) is estimated.
Identity-by-descent (IBD): a
genomic section is IBD in two
Animal and Plant Breeding individuals if they have inherited it
Animal breeders traditionally estimated breeding values based on phenotypic scores, such as from a common ancestor.
milk yield, of a focal individual and its relatives, using [326_TD$IF]the animal model (Box 2). With high density Identity-by-state (IBS): a genomic
section is IBS in two individuals if it
SNP chips (Table 1) becoming available for many domesticated species, breeders were the first has the same nucleotide sequence in
to suggest the use of the GRM in [326_TD$IF]the animal model to predict genomic breeding values (GEBVs) these two individuals. IBS can arise
[22][327_TD$IF]. They also developed other approaches for genomic selection (GS) that even bypass the by chance (mutation) or because the
individuals inherited it from the same
animal model (Box 2). GS revolutionized animal breeding and the field has experienced a surge
ancestor, in which case it would also
of method development [23]. It was also demonstrated that [326_TD$IF]the animal model coupled with a be identical by descent.
GRM instead of a pedigree leads [328_TD$IF]to more accurate estimates of the variance components than Infinitesimal model: this model
pedigree-based equivalents [23,24]. Further, most pre[329_TD$IF]-genomic breeding applications ignored assumes that the genetic component
of a trait is determined by an infinite
estimating non[30_TD$IF]-additive genetic effects because they generally require complex breeding
number of unlinked genes that each
designs or pedigree structures, and also because it was believed that their effects were has an infinitesimally small, additive
negligible (e.g., [5]). A handful of recent empirical studies from tree breeders demonstrated effect. This will lead to continuous
that, in comparison to classical pedigree based approaches, the use of a GRM in an animal genetic variation in the trait and this
model is hence generally used to
model so-called ‘quantitative’ or
Box 2. Two Key Statistical Models to Estimate QG Parameters Using Genomic Data ‘complex’ traits. It is, however, also
[296_TD$IF]Two main types of statistical models (Table I) dominate the QG literature and both can make use of genomic data. In the applicable to discrete traits, for
animal model, the pedigree can be replaced with the GRM, and in genome regressions, genomic data is a necessary example litter size, and even binary
ingredient. traits if trait expression depends on
one or more thresholds that the
summed allelic effects need to pass.
Animal Model
Linkage disequilibrium: the non-
[297_TD$IF]The animal model is most often used to estimate variance components, such as the additive genetic variance, which random association of alleles at
serves to estimate the heritability of traits or genetic correlations between traits. Breeding values are the estimated different loci. Loci are in linkage
random effects and can be used to characterize individuals[298_TD$IF]; however, several authors warn not to use estimated disequilibrium, or linked, if certain
breeding values (EBVs) in further statistical analysis without incorporating their estimated standard errors (e.g., [51]), for combinations of alleles occur more
example in a hierarchical Bayesian framework. Generally large pedigrees are required but it can also be used with often than expected by chance.
[29_TD$IF]common garden experimental data. Linkage disequilibrium can be
caused by a multitude of factors,
including selection, recombination
Genome Regressions
rate, mating system, population
structure or physical proximity in the
Genome regressions are principally used in a prediction context to help breeders identify superior individuals. [30_TD$IF]These
genome.
models are fitted in a training population to estimate breeding values of individuals of unknown trait value in a candidate
Nonadditive genetic effects:
population (such as [301_TD$IF]milk yield at an early age or [302_TD$IF]in a male cow). GEBVs are generally much more accurate than EBVs. [30_TD$IF]For
dominance, epistasis and similar
a highly heritable trait, we expect particularly high accuracy is expected with a large number of individuals in the training
effects. Dominance means that the
population [304_TD$IF]and when the number of segregating chromosome segments is small (i.e., small genome and small effective
effect of one allele masks (is
population size). GS does not work well across breeds. Instead, the best performance is achieved when individuals in
‘dominant’ over) the effect of the
the candidate population are related to individuals in the training population.
other (recessive) allele. Incomplete
and co-dominance are cases when
Table I. Statistical Models to Estimate QG Parameters Using Genomic Data the dominant allele is not completely
masking the recessive allele and the
Animal model Genome regression
heterozygous phenotype is
Refs [6] [22] somewhat intermediate between the
X  homozygous dominant and
The model y i ¼ m þ ai þ ei yi ¼ m þ SNPi;j  gj þ ei
homozygous recessive phenotype.
where ai is the breeding value of individual i. where SNPi,j is the genotype of
Epistasis, or gene–gene interactions,
Since ai’s are unknown, they are estimated individual i at loci j, and gj is the
occurs when the allelic effect at one
from the covariance between relatives in effect of the SNP on the phenotype
locus depend on the alleles of one or
additive genetic effects given by 2*kinship*VA,
more other loci, the genetic
thus using a pedigree or a GRM
background.
Explanation of yi is a trait value of individual i, m is the population mean, and ei is the residual term Quantitative trait locus (QTL): a
common model terms genomic section, (i.e., a locus) that
contributes to the variation (together
Breeding value Using the best linear unbiased predictor (BLUP) BLUPs assuming equal variances
with other QTL) of a quantitative trait
(BV) estimation in a mixed effects model across SNPs is possible, but Bayesian
(i.e., a trait that shows continuous
methods generally work better
phenotypic variation).

4 Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy

 TREE 2303 No. of Pages 12

model can yield substantially more accurate separation of the additive and non[30_TD$IF]-additive Realized relatedness: the
components of genetic variance [25,26]. relatedness of two individuals
calculated from a pedigree (kinship)
is their expected genetic similarity (or
Human Genetics proportion of genome sharing). There
Human populations cannot be manipulated experimentally, so different ‘non-invasive’ genomic is, however, variation around this
QG methods have been developed in this field. Most notably, it was suggested to exploit expectation (except for parent-
offspring pairs and monozygotic
variation in realized genome sharing within a pedigree relationship class to estimate the twins) because of the segregation of
heritability: Visscher et al. [27] proposed this idea within full-sib families, and Yang et al. independent genome segments
[28] among unrelated individuals. Most importantly, these methods led to more accurate during meiosis (Mendelian noise).
This variation is lowered by more
estimates of the variance components. Nevertheless, population parameters such as linkage
independently segregating segments
disequilibrium between marker and quantitative trait loci can lead to different performance of (e.g., [19,27]). An interesting
this method in different species or populations. To explore this further, Visscher and Goddard consequence of this is that for
[29] investigated the sampling error [31_TD$IF]of heritability estimates as a function of the sampling species with ‘small’ genomes (i.e.,
with fewer independently segregating
scheme, the sample size and the effective population size, and concluded that overall, more
segments), QG estimates based on
individuals are needed to achieve the same precision with random population samples in realized relatedness will be more
comparison to pedigrees, and higher sample size is required when there is little variance in precise than estimates based on
relatedness such with large effective population size or in expanding populations. expected relatedness.
Single nucleotide polymorphism
(SNP): a variation at a specific
The First ‘Wild’ Applications genomic position. For example, at a
Applications of the genomic QG approach in wild populations have been lagging behind specific position in a genome, the
breeding and human genetics for obvious genomic resource reasons, even though the idea base C appears in most individuals,
but in a minority the base A appears.
to use molecular markers to estimate heritability in natural populations had been proposed This specific SNP would have the
back in 1996 by Ritland [30]. Most early applications, based on a few microsatellites, yielded two nucleotide variations – C or A –
estimates too imprecise and often out of range [30,31]. Recently, with dense marker panels as its alleles.
becoming accessible in some wild pedigreed populations, it has been shown that marker-
based relatedness estimates agreed with those from pedigrees (see Figure 1A[32_TD$IF],B) and also
that heritability estimates based on high-density markers and pedigrees agree [3_TD$IF]in Soay sheep
(Ovis aries) [32] [36_TD$IF]and great tits (Parus major) [10][34_TD$IF]. Some empirical studies also investigated
how many markers are required to estimate the heritability by comparing estimates from a
decreasing proportion of the available loci. For example, in the selfing plant species, Medicago
truncatula, heritability estimates with 25 000 SNPs (i.e., one marker every 10 kb, which
approximates the distance of linkage disequilibrium decay) were nearly unbiased in comparison
to those obtained with >5 million SNPs (Figure 1C). Similarly, for a small island population of
a wild ungulate, Soay sheep, heritability with 10 000 SNPs yielded unbiased estimates of
the heritability obtained with 33 000 SNPs across several traits (Figure 1D).

The Genomic QG Toolbox for Ecology and Evolution

Bringing QG to the wild by using the pedigree-based animal model has greatly advanced our
understanding of evolution [7]. Today, the acquisition of high-density marker data is possible in
virtually any species (Table 1), which opens up the possibility to apply the QG research program
in the absence of a pedigree or controlled cross in any population of any species. Genomic QG
also offers a greater methodological flexibility than classical QG and many evolutionary and
global change research questions could benefit from it.

Genomic Estimated Breeding Values

GEBVs could have many potential uses outside animal and plant breeding, such as in the
context of conservation biology[35_TD$IF], to identify suitable individuals for assisted migration or release
programs. The objective of assisted migration is to ‘help’ species migrating to keep up with the
pace of climate change, and it is [367_TD$IF]being used for forest tree management [14]. It has been
suggested that GEBVs can be combined with gene–environment association analysis to
predict phenotypes and identify changes in performance along environmental gradients
[33]; Box 3 provides a proof-of-concept example of this approach in Arabidopsis thaliana.

Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy 5

 TREE 2303 No. of Pages 12

(A) (B)
0.8 0.80

0.6

Correlaon coefficient
Genomic relatedness

0.75
0.4

0.2
0.70

0.0

0.0 0.2 0.4 0.6 0.8 0.025 0.1 0.3 0.5 0.7 0.9

Pedigree relatedness Proporon of SNPs used

(C) (D)
0.8 0.30

0.7

0.25
Heritability

Heritability

0.6

0.20
0.5

0.4
0.15

2k 25k 250k 5M 0.025 0.1 0.3 0.5 0.7 0.9

Number of SNPs used Proporon of SNPs used

Figure 1. Genomic Relatedness and Heritability Estimates. (A) Relationship between marker-based and pedigree relatedness in one of the wild study
populations with the deepest pedigree, the St. Kilda Soay sheep population. Realized or ‘genomic’ relatedness, estimated from ca. 37000 SNP markers, is plotted
against IBD relatedness, estimated from the pedigree. The variation of realized relatedness estimates with the same IBD relatedness class, seen as vertical spread, is
due to Mendelian segregation (and sampling error). The unbroken and broken grey lines indicate the 1:1 relationship and the actual regression, respectively. (B) shows,
for the same population, that the correlation between pedigree and marker-based relatedness estimates increases with the number of SNPs used to estimate the GRM.
(C) and (D) show the effect of number of markers on heritability estimates in Medicago truncatula and Soay sheep, respectively. Boxes and whiskers indicate the
variation in estimates from the repeated samples of the reduced data sets. When few markers are used to estimate the GRM heritability, estimates are downward
biased. Increasing the number of markers reduces this bias, but, at some point, adding more markers has little effect on the estimate anymore, only on its variation. In (B)
and (D) the x-axis shows the proportion of SNPs used in relation to the complete data set and the absolute numbers are 926, 1852, 3704, 11 111, 18 518, 25 926,
respectively. (A), (B), and (D) adapted, with permission, from [32]; (C) adapted, with permission, from [38].

6 Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy

 TREE 2303 No. of Pages 12

Table 1. Overview of High-Throughput Genotyping Approaches and Sequencing Approaches that Are Potentially Suitable To Estimate
Relatedness
Data type Genomic Pros Cons Cost Examples of studies
resources needed applying this approach
to calculate the GRM

SNP chip Reference ‘Ready to use’ - No bioinformatics work Availability mostly limited to model Low, if Animal and plant
genome and breeding systems, and possibly chip breeding [23,24],
close relatives. Ascertainment bias: available human genetics [28],
new populations can have different wild species [32]
polymorphisms [56]

RRL- or RAD- None, but Flexibility of experimental design assisted Bias on relatedness estimates due to Medium [57,58]
sequencing improved [290_TD$IF]with a by software predictions on number of allele dropout and high proportion of
reference markers. Established bioinformatics for missing data [57]
genome data analysis with easy-to-use pipelines

Whole-genome Reference High-confidence genome-wide High costs for large genomes, and Very high [38,59]
sequencing genome information and use of established can yield excessive amount of
bioinformatics tools information [38]

Low-coverage Reference Dense genome representation including Requires bioinformatics knowledge High [60,61]
whole-genome genome causal variants that can be suitable for to use software building on
sequencing parallel association studies. probabilistic framework.

Sequence or Reference Targeted sequencing of selected loci, Requires high-quality genomic High [63]
Exome capture genome of closely which can include causal variants resources for bait design and to
related species or suitable for parallel association studies. minimize allele bias [62]
transcriptome

Genome Reference Minimal sequencing effort to survey high- No studies have yet built a GRM. Low
skimming genome copy regions of the genome, typically Complex evolution of high-copy
organellar and ribosomal DNA nuclear loci can limit applicability
in natural populations.

Since GS is based on a statistical model built on data from a ‘training population’ (Box 2[37_TD$IF]), in
order to obtain reliable predictions of the breeding values, it is necessary to include individuals
across the full environmental range over that predictions are desired. A further potential use of
GEBVs is in breeding programs of endangered species, where genomic data could help to
optimize breeding of remaining individuals to minimize inbreeding or outbreeding depression
[34]. Releasing individuals bred in captivity into the wild is often used to restore extinct
populations or help threatened populations to survive, but programs might fail if individuals
are maladapted to the new environment (e.g., [35]). Identifying potentially well-adapted indi-
viduals based on their GEBVs, as described in Box 2, could be an option, or combined with
allele-specific genetic rescue, such as when disease resistance genes are known (see exam-
ples in [36]). As pointed out above, it remains unclear whether estimating GEBVs will be
generally feasible in natural populations, but in an outbreeding tree species GEBVs [38_TD$IF]were
estimated with reasonable [39_TD$IF]accuracy [37]. The fact that QG estimates based on genomic
markers agreed with ‘standard’ QG estimates in the studies on mammals, birds and plants
[10,32,38] also indicates that it should be possible [340_TD$IF]across many species.

Realized Genome Sharing to Quantify Fitness

The GRM is just the first step in genomic QG, but it can also be used in another innovative way in
evolutionary biology: [182_TD$IF]the average relatedness of an individual with the population, estimated
from a GRM, could be used [341_TD$IF]as a measure of inclusive fitness integrated across the current
population [342_TD$IF]because individuals whose ancestors produced many offspring [34_TD$IF]and have a higher
than average relatedness with the population. Such a fitness measure could also be [34_TD$IF]useful for
behavioral ecology, where estimating relatedness among interacting individuals is necessary to
understand social behaviors. Here, the average genomic relatedness of an individual could

Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy 7

 TREE 2303 No. of Pages 12

better estimate its inclusive fitness. Particularly, relatedness estimates based on coalescence
could be adapted here (see also [39] and Box 1). Estimating reproductive success is a
challenging endeavor in wild populations for several reasons [40] but could potentially be
tackled by estimating GRMs.

Evolutionary Potential in the Wild

Evolutionary potential is a scaled measure of the additive genetic variance (VA); it is the
heritability when it is scaled with the total phenotypic variance and evolvability when it is
scaled by the trait mean [41]. Bringing genomic QG to the wild will allow assessing the
evolutionary potential of a population in situ, which is a considerable advancement because
it estimates the additive genetic variance and covariance of phenotypic traits under the
conditions in which natural selection is acting, which is not the case when using [345_TD$IF]breeding
experiments. Such estimates are not guaranteed to apply under natural conditions as VA might
vary across environments and strongly depends on population size and inbreeding levels
(reviewed in [42]). Estimates of trait heritability in natural populations, however, are also prone to
biases due to plastic responses to environmental variables (see [43] for a recent review).
Consequently, a good understanding of environmental but also individual-level variables, such
age or body condition, that affect the trait is necessary to account for this ‘environmental’
variation by including the relevant variables in the model. Alternatively, if the exact variables
responsible for plastic trait variation are unknown or difficult to estimate but are known to vary
spatially or temporally, appropriately modelling this spatial or temporal variation is a suitable
approach. For example, in the case of an environmental variable that varies in discrete ‘units’,
such as the effect of spring [346_TD$IF]temperatures on timing of flowering, breeding or migration, fitting
the temporal or spatial ‘unit’ as a (fixed or random) factor would be appropriate. If variation is
more gradual, fitting an appropriate autocorrelation in the model can remove this plastic
variation in traits (e.g., [44]). Another approach to remove confounding effects between genetic
and environmental similarities is to estimate QG parameters from variation in relatedness within
families, such as full-sibs [27], which could be possible in clutches of birds, fish or insects or
half-sibs in plants (e.g., [45]). These in situ estimates of the evolutionary potential can be useful
in many fields of ecology, evolution and conservation biology. Climate change [347_TD$IF]has introduced
new selective pressures such as increased drought, heat or seasonal shifts [348_TD$IF]into many eco-
systems but for most species we lack in situ or even any estimates of VA at ecologically key traits

Box 3. Using Genomic Relatedness and [305_TD$IF]Home Environments to Identify Locally Adapted Genotypes
A common challenge in conservation genetics is to identify locally-adapted genotypes for a given site. Here, we
demonstrate that GRMs for wild accessions collected across diverse climates (‘ecotypes’) can be applied to predict
which genotypes are adapted to cold sites and which are adapted to warm sites. Our approach relies on a three-way
correlation between genomic similarity (GRMs), similarity in home environments (winter cold), and similarity in locally
adaptive traits (unobserved and predicted). As such, correlation between population structure and environmental
gradients causing local adaptation is not a problem for our approach, rather this correlation is leveraged to predict
adaptive trait variation. By contrast, population structure-environment correlation is problematic for genome-environ-
ment association studies that seek to identify individual causal loci involved in local adaptation. We reanalyzed published
data on Arabidopsis thaliana [71–73] and global climate [74]. We used >200 000 SNPs [73] to estimate a GRM based
on average allelic correlation (Box 1). We fitted an animal model where an ecotype’s native winter cold environment
(minimum winter temperature) was used as ‘phenotype’ and modeled as a function of the GRM with 10-fold cross
validation to predict native temperatures as GEBVs [75]. Based on the assumption that populations are locally adapted
along temperature gradients, these GEBVs can also be interpreted as predictions of a ‘latent cold-adapted phenotype’,
that is, an unobserved phenotype assumed to confer high fitness in particular temperatures. We next show that these
predictions closely correspond to which genotypes are locally adapted along temperature gradients. Although here we
validated our predictions with experimental data, note that no experiments were used to generate our predictions;
predictions are based solely on the relationship between GRM and native environments. Individual relative fitness of 157
ecotypes was measured in four common gardens by Fournier-Level et al. [72]. In each garden, we regressed relative
fitness against predicted latent cold-adapted phenotypes to identify the expected most fit latent phenotype at each site
(Figure I). The fittest latent cold-adapted phenotype at each site varied closely with differences in winter cold among sites
(central panel, R2[291_TD$IF] = 0.92). This relationship indicates that the GRM association with native environments can be used to

8 Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy

 TREE 2303 No. of Pages 12

predict which genotypes are adapted to cold sites and which are adapted to warm sites, information that might guide
conservation and restoration actions.

Finland UK

relave fitness
relave fitness

0.8
0.8
Observed

Observed

0.4
0.4

0.0
0.0

–10 –5 0 5 –20 –10 0 5

Latent cold phenotype Latent cold phenotype

6
4
2

Predicted latent
cold phenotype
0

with maximum
–2

fitness
–6 –4

R2 = 0.92

–15 –10 –5 0 5

Minimum temperature of coldest month (°C)

Figure I. Relative Fitness versus Predicted Latent Cold-Adapted Phenotypes. Change in latent cold adapted
phenotype (GEBVs) with maximum expected fitness across four common gardens differing in winter cold (lower panel).
Two insets show relative fitness data (scaled to maximum of unity) for different GEBVs at two common garden sites,
curve shows a fitted quadratic regression [72].

including phenology or drought tolerance (e.g., [46,47]). Climate change is also expected to
drive changes in community assemblages due to different dispersal abilities at different species,
as has been shown for alpine plants [48]. In situ evaluation of the evolutionary potential of key
life-history traits is necessary to assess whether these species will be able to adapt to the
changed competition regime.

Estimates of VA could also be particularly useful for models of eco-evolutionary dynamics [49].
For example, Ellner [50] suggested a framework to partition change in a population into
contributions of evolution, non-heritable trait change and environmental change. To date, this

Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy 9

 TREE 2303 No. of Pages 12

theoretical framework has been limited to experimental systems, but with genomic QG it could Outstanding Questions
be applied [349_TD$IF]to populations in natural settings. How many markers are necessary to
reliably estimate relatedness? The
number of markers that is necessary
With the decreasing costs of genotyping, measuring phenotypes of individuals can become to precisely estimate genome sharing
the limiting factor for QG studies in natural populations. Although the variance in (genomic) among individuals as a measure of
breeding values is not VA (e.g., [51]), it could be used as a proxy for it, which would relatedness depends on the physical
size of the genome and extent of link-
allow prediction of evolutionary potential in new populations without the need to
age disequilibrium. Linkage disequilib-
phenotype individuals. However, genotype-by-environment interactions might limit the rium is affected by species or
application of this approach to species where reaction norms are known across environ- population specific variables as mating
mental gradients. system or population structure. Con-
sequently, for the same physical
genome size the number of markers
Predicting Response to Selection necessary to reliably infer relatedness
The evolutionary potential informs about the capacity of a population to respond to selection. may vary. Simulation studies testing
Knowledge of VA for a trait or for fitness directly applies to predictive QG models of eco- the necessary number of markers for
various life-histories would hence be
evolutionary dynamics. This is especially relevant in the context of predicting shifts of species’
useful to address this.
ranges caused by climate change, and many voices have called for the development of eco-
evolutionary forecasting models in that context (e.g., [52]). Quantitative genetics theory Can too high or too little variation in
provides a well formed framework for predicting eco-evolutionary dynamics [350_TD$IF]within single relatedness impede genomic QG?
or among competitively interacting species (e.g., [53]). Current research efforts provide Variation in relatedness is crucial for
any quantitative genetic analysis. In
practical examples of implementation of this framework, together with modeling of species breeding experiments this variation
ecological niche distributions, to predict changes of species’ distribution in Drosophilid flies in can obviously be controlled but in wild
Australia [54] or endemic alpine plants in Austria [55]. However, this approach is data hungry populations, where GRMs are the only
and needs estimates of VA for climate-adaptation traits and the selection gradients acting on way for quantitative genetic analyses, it
will depend on the species’ life-history
them. There is hope that by using genomic QG tools such as environment-related GEBVs in and might impede such analyses. For
natural populations (Box 3) we will be able to alleviate the dearth of estimates of QG example, in species with large popu-
parameters necessary to correctly forecast the fate of natural species facing a warming lations and long dispersal distances
relatedness may on average be too
climate.
low for quantitative genetic analyses,
while it could be too high in selfing
Concluding Remarks plants with limited dispersal. Appropri-
Genomic QG has the potential to advance our understanding of evolutionary processes in ate sampling schemes may alleviate
this problem but it remains to be seen
natural populations in several ways. Its main advantage is that we no longer need pedigrees or
whether marker-based relatedness
breeding experiments for QG analysis, which means that we can study additive genetic (co) estimates are precise enough [352_TD$IF]for
variances, but also selection, in virtually any species in a natural setting. Genomic QG genomic QG to also work in such
parameters estimates can be more precise than pedigree-based estimates, [351_TD$IF]they have fewer cases.

assumptions and offer a greater methodological flexibility. This ability to conduct quantitative
Is the potential usefulness of genomic
genetic analyses in natural populations could allow more accurate predictions of species’
QG limited by available phenotypic
persistence under climate change and enable informed conservation strategies. As we pointed information? Establishing deep pedi-
out, the feasibility of this approach will partly depend on the specific study system. The relative grees to detect distant relatedness
ease with which high-density markers can be obtained make us cautiously optimistic about the among individuals obviously requires
long-term monitoring of populations.
potential of marker-based quantitative genetics to considerably contribute to our understand- Marker-based relatedness estimates
ing of evolutionary processes in natural populations and mitigating climate change (see can detect these relationships within
Outstanding Questions). a single ‘time slice’ of a population,
for example, by sampling all individuals
alive in a year. This, however, means
[35_TD$IF]Acknowledgments
that much fewer phenotypes are
We thank Thomas Mitchell-Olds, Jean-Luc Jannink, Marcel Visser and two anonymous reviewers for constructive recorded and can be analyzed. The
comments on the manuscript. The idea for this manuscript developed at [354_TD$IF]a Monte Verità Conference in 2016 on ‘The question hence is whether the
genomic basis of eco-evolutionary change’ organized by the ETH Zurich’s center for [35_TD$IF]Adaptation to a Changing increased precision of marker-based
Environment (ACE). We would like to acknowledge the organizers for providing such a fruitful, scientific atmosphere. relatedness estimate can ‘compen-
Camillo Bérénos, Josephine Pemberton, John Stanton-Geddes and Peter Tiffin kindly allowed us to re-use their figures sate’ for the reduced number of phe-
notypes in the analysis or whether still
and provided the data to do so. SF was partly supported by the Swiss National Science Foundation (project
multiple years need to be sampled and
31003A_160123). FG is supported by grant PP00P3_144846 from the Swiss National Science Foundation. KC was analyzed.
supported by an ACE Fellowship while working on the manuscript.

10 Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy

 TREE 2303 No. of Pages 12

Resources
[306_TD$IF]http://zzz.bwh.harvard.edu/plink/
i

ii
http://cnsgenomics.com/software/gcta/
iii
http://dougspeed.com/ldak/
iv
http://faculty.washington.edu/browning/beagle/beagle.html

References
1. Gonzalez, A. et al. (2013) Evolutionary rescue: an emerging focus family testing for disentangling additive from nonadditive genetic
at the intersection between ecology and evolution. Philos. Trans. effects. G3 6, 743–753
R. Soc. B 268, 20120404 27. Visscher, P.M. et al. (2006) Assumption-free estimation of herita-
2. Fisher, R.A. (1918) The correlation between relatives on the bility from genome-wide identity-by-descent sharing between full
supposition of Mendelian inheritance. Trans. R. Soc. 52, 399–433 siblings. PLoS Genet. 2, e41
3. Falconer, D.S. and Mackay, T.F. (1996) Introduction to Quantita- 28. Yang, J. et al. (2010) Common SNPs explain a large proportion of
tive Genetics, Longman the heritability for human height. Nat. Genet. 42, 565–569
4. Nelson, R.M. et al. (2013) A century after Fisher: time for a new 29. Visscher, P.M. and Goddard, M.E. (2015) A general unified frame-
paradigm in quantitative genetics. Trends Genet. 29, 669–676 work to assess the sampling variance of heritability estimates
5. Hill, W.G. (2010) Understanding and using quantitative genetic using pedigree or marker-based relationships. Genetics 199,
variation. Philos. Trans. R. Soc. B 365, 73–85 223–232

6. Henderson, C.R. (1975) Best linear unbiased estimation and 30. Garant, D. and Kruuk, L.E.B. (2005) How to use molecular marker
prediction under a selection model. Biometrics 31, 423–447 data to measure evolutionary parameters in wild populations.
Mol. Ecol. 14, 1843–1859
7. Charmantier, A. et al., eds (2014) Quantitative Genetics in the
Wild, Oxford University Press 31. Coltman, D.W. (2005) Testing marker-based estimates of herita-
bility in the wild. Mol. Ecol. 14, 2593–2599
8. Visscher, P.M. et al. (2017) 10 years of GWAS discovery: biology,
function, and translation. Am. J. Hum. Genet. 101, 2–22 32. Bérénos, C. et al. (2014) Estimating quantitative genetic param-
eters in wild populations: a comparison of pedigree and genomic
9. Gienapp, P. et al. (2017) Environment-dependent genotype–phe-
approaches. Mol. Ecol. 23, 3434–3451
notype associations in avian breeding time. Front. Genet. 8, 102
33. Lasky, J.R. et al. (2015) Genome-environment associations in
10. Robinson, M.R. et al. (2013) Partitioning of genetic variation
sorghum landraces predict adaptive traits. Sci. Adv. 1, e1400218
across the genome using multimarker methods in a wild bird
population. Mol. Ecol. 22, 3963–3980 34. Frankham, R. (2008) Genetic adaptation to captivity in species
conservation programs. Mol. Ecol. 17, 325–333
11. Santure, A.W. et al. (2015) Replicated analysis of the genetic
architecture of quantitative traits in two wild great tit populations. 35. Griffiths, R.A. and Pavajeau, L. (2008) Captive breeding, reintro-
Mol. Ecol. 24, 6148–6162 duction, and the conservation of amphibians. Conserv. Biol. 22,
852–861
12. Hill, W.G. (2012) Quantitative genetics in the genomics era. Curr.
Genomics 13, 196–206 36. Harrisson, K.A. et al. (2014) Using genomics to characterize
evolutionary potential for conservation of wild populations. Evol.
13. Travisano, M. and Shaw, R.G. (2013) Lost in the map. Evolution
Appl. 7, 1008–1025
67, 305–314
37. Beaulieu, J. et al. (2014) Accuracy of genomic selection models in
14. Aitken, S.N. and Whitlock, M.C. (2013) Assisted gene flow to
a large population of open-pollinated families in white spruce.
facilitate local adaptation to climate change. Ann. Rev. Ecol. Evol.
Heredity 113, 343–352
Syst. 44, 367–388
38. Stanton-Geddes, J. et al. (2013) Estimating heritability using
15. Lynch, M. and Walsh, B. (1998) Genetics and Analysis of Quan-
genomic data. Methods Ecol. Evol. 4, 1151–1158
titative Traits, Sinauer
39. Rousset, F. (2002) Inbreeding and relatedness coefficients: what
16. Speed, D. et al. (2012) Improved heritability estimation from
do they measure? Heredity 88, 371–380
genome-wide SNPs. Am. J. Hum. Genet. 91, 1011–1021
40. Shaw, R.G. et al. (2008) Unifying life-history analyses for inference
17. Csilléry, K. et al. (2006) Performance of marker-based related-
of fitness and population growth. Am. Nat. 172, E35–E47
ness estimators in natural populations of outbred vertebrates.
Genetics 173, 2091–2101 41. Houle, D. (1992) Comparing evolvability and variability of quanti-
tative traits. Genetics 130, 195–204
18. Pemberton, J.M. (2008) Wild pedigrees: the way forward. Proc.
R. Soc. B Biol. Sci. 275, 613–621 42. Hoffmann, A.A. et al. (2017) Revisiting adaptive potential, popu-
lation size, and conservation. Trends Ecol. Evol. 32, 506–517
19. Hill, W.G. and Weir, B.S. (2011) Variation in actual relationship as
a consequence of mendelian sampling and linkage. Genet. Res. 43. Wood, C.W. and Brodie, E.D., III (2016) Evolutionary response
93, 47–64 when selection and genetic variation covary across environ-
ments. Ecol. Lett. 19, 1189–1200
20. Wang, J. (2016) Pedigrees or markers: which are better in esti-
mating relatedness and inbreeding coefficient? Theor. Popul. 44. Germain, R.R. et al. (2016) Direct and indirect genetic and fine-
Biol. 107, 4–13 scale location effects on breeding date in song sparrows. J. Anim.
Ecol. 85, 1613–1624
21. Gay, L. et al. (2013) Pedigree-free estimates of heritability in the
wild: promising prospects for selfing populations. PLoS One 8, 45. Bontemps, A. et al. (2016) In situ marker-based assessment of
e66983 leaf trait evolutionary potential in a marginal European beech
population. J. Evol. Biol. 29, 514–527
22. Meuwissen, T. et al. (2001) Prediction of total genetic value using
genome-wide dense marker maps. Genetics 157, 1819–1829 46. Alberto, F.J. et al. (2013) Potential for evolutionary responses to
climate change – evidence from tree populations. Glob. Change
23. Meuwissen, T. et al. (2016) Genomic selection: a paradigm shift in
Biol. 19, 1645–1661
animal breeding. Anim. Front. 6, 6–14
47. Gienapp, P. et al. (2014) Why climate change will invariably lead to
24. Jannink, J.-L. et al. (2010) Genomic selection in plant breeding:
selection on phenology. Proc. R. Soc. B Biol. Sci. 281, 20141611
from theory to practice. Brief. Funct. Genomics 9, 166–177
48. Alexander, J.M. et al. (2015) Novel competitors shape species’
25. Muñoz, P. et al. (2014) Unraveling additive from nonadditive
responses to climate change. Nature 525, 515–518
effects using genomic relationship matrices. Genetics 198,
1759–1768 49. Hendry, A.P. (2017) Eco-evolutionary dynamics, Princeton Uni-
versity Press
26. Gamal El-Dien, O. et al. (2016) Implementation of the realized
genomic relationship matrix to open-pollinated white spruce

Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy 11

 TREE 2303 No. of Pages 12

50. Ellner, S.P. et al. (2011) Does rapid evolution matter? Measuring 63. Zhuang, Z. et al. (2012) Detecting identity by descent and homo-
the rate of contemporary evolution and its impacts on ecological zygosity mapping in whole-exome sequencing data. PLoS One 7,
dynamics. Ecol. Lett. 14, 603–614 e47618
51. Hadfield, J.D. et al. (2010) The misuse of BLUP in ecology and 64. Milligan, B.G. (2003) Maximum-likelihood estimation of related-
evolution. Am. Nat. 175, 116–125 ness. Genetics 163, 1153–1167
52. Urban, M.C. et al. (2016) Improving the forecast for biodiversity 65. Goddard, M.E. et al. (2011) Using the genomic relationship matrix
under climate change. Science 353, aad8466 to predict the accuracy of genomic selection. J. Anim. Breed.
53. Duputié, A. et al. (2012) How do genetic correlations affect Genet. 128, 409–421
species range shifts in a changing environment? Ecol. Lett. 15, 66. Powell, J.E. et al. (2010) Reconciling the analysis of IBD and IBS in
251–259 complex traits. Nat. Rev. Genet. 11, 800–805
54. Bush, A. et al. (2016) Incorporating evolutionary adaptation in 67. Lee, S.H. et al. (2013) Estimation of SNP heritability from dense
species distribution modelling reduces projected vulnerability to genotype data. Am. J. Hum. Genet. 93, 1151–1155
climate change. Ecol. Lett. 19, 1468–1478 68. Browning, B.L. and Browning, S.R. (2013) Improving the accu-
55. Cotto, O. et al. (2017) A dynamic eco-evolutionary model predicts racy and efficiency of identity-by-descent detection in population
slow response of alpine plants to climate warming. Nat. Commun. data. Genetics 194, 459–471
8, 15399 69. Purcell, S. et al. (2007) PLINK: a tool set for whole-genome
56. Pérez-Enciso, M. (2014) Genomic relationships computed from association and population-based linkage analyses. Am. J.
either next-generation sequence or array SNP data. J. Anim. Hum. Genet. 81, 559–575
Breed. Genet. 131, 85–96 70. Yang, J. et al. (2011) GCTA: A tool for genome-wide complex trait
57. Dodds, K.G. et al. (2015) Construction of relatedness matrices analysis. Am. J. Hum. Genet. 88, 76–82
using genotyping-by-sequencing data. BMC Genomics 16, 1047 71. Anastasio, A.E. et al. (2011) Source verification of mis-identified
58. Palaiokostas, C. et al. (2016) Genomic prediction of resistance to Arabidopsis thaliana accessions. Plant J. 67, 554–566
pasteurellosis in gilthead sea bream (Sparus aurata) using 2b- 72. Fournier-Level, A. et al. (2011) A map of local adaptation in
RAD sequencing. G3 (Bethesda) 6, 3693–3700 Arabidopsis thaliana. Science 334, 86–89
59. Morrison, A.C. et al. (2013) Whole-genome sequence-based 73. Horton, M.W. et al. (2012) Genome-wide patterns of genetic
analysis of high-density lipoprotein cholesterol. Nat. Genet. 45, variation in worldwide Arabidopsis thaliana accessions from
899–901 the RegMap panel. Nat. Genet. 44, 212–216
60. Snyder-Mackler, N. et al. (2016) Efficient genome-wide sequenc- 74. Hijmans, R.J. et al. (2005) Very high resolution interpolated cli-
ing and low-coverage pedigree analysis from noninvasively col- mate surfaces for global land areas. Int. J. Climatol. 25, 1965–
lected samples. Genetics 203, 699–714 1978
61. Ackerman, M.S. et al. (2017) Estimating seven coefficients of 75. Endelman, J.B. (2011) Ridge regression and other kernels for
pairwise relatedness using population genomic data. Genetics genomic selection with R package rrBLUP. Plant Genome 4,
206, 105–118 250–255
62. Meynert, A.M. et al. (2014) Variant detection sensitivity and biases
in whole genome and exome sequencing. BMC Bioinformatics
15, 247

12 Trends in Ecology & Evolution, Month Year, Vol. xx, No. yy