DETECTING SELECTION IN NATURAL POPULATIONS: MAKING SENSE OF GENOME SCANS AND TOWARDS ALTERNATIVE SOLUTIONS

Abstract

Genomewide scans for natural selection (GWSS) have become increasingly common over the last 15 years due to increased availability of genome-scale genetic data. Here, we report a representative survey of GWSS from 1999 to present and find that (i) between 1999 and 2009, 35 of 49 (71%) GWSS focused on human, while from 2010 to present, only 38 of 83 (46%) of GWSS focused on human, indicating increased focus on nonmodel organisms; (ii) the large majority of GWSS incorporate interpopulation or interspecific comparisons using, for example F_ST, cross-population extended haplotype homozygosity or the ratio of nonsynonymous to synonymous substitutions; (iii) most GWSS focus on detection of directional selection rather than other modes such as balancing selection; and (iv) in human GWSS, there is a clear shift after 2004 from microsatellite markers to dense SNP data. A survey of GWSS meant to identify loci positively selected in response to severe hypoxic conditions support an approach to GWSS in which a list of a priori candidate genes based on potential selective pressures are used to filter the list of significant hits a posteriori. We also discuss four frequently ignored determinants of genomic heterogeneity that complicate GWSS: mutation, recombination, selection and the genetic architecture of adaptive traits. We recommend that GWSS methodology should better incorporate aspects of genomewide heterogeneity using empirical estimates of relevant parameters and/or realistic, whole-chromosome simulations to improve interpretation of GWSS results. Finally, we argue that knowledge of potential selective agents improves interpretation of GWSS results and that new methods focused on correlations between environmental variables and genetic variation can help automate this approach.

Introduction

The genome provides an organic record of evolution that is frequently likened to a palimpsest (Delwiche 2004; Weiss & Kawasaki 2006)—a writing medium that is recycled, continuously written over and reoriented so as to partially or wholly obscure older text (Fig. 1A). By this metaphor, chromosomes are the parchment and DNA sequence the text. Mutation obfuscates older genetic text; recombination and chromosomal rearrangements change the content, sense and/or order of the text; and natural selection may secure permanent erasure and replacement of older text (Fig. 1B). In the latter case, reference to other copies of the genetic text—in closely related species or populations where the text has not been altered by natural selection—may enable inference of the original, ancestral genetic text (Fig. 1C).

Fig. 1.

A palimpsest as a metaphor for the genome and the population genetic processes that change it. (A) The Codex Nitriensis is a palimpsest. The lower, faded text is written in Greek and dates to the sixth century A.D., while the upper, bolder text is written in Syriac Aramaic and dates to several centuries later. (B) A genetic text in which the content and/or sense of the text is changed by mutation, chromosomal rearrangement and recombination. These events obscure or permanently alter the original text. Initially, an individual chromosome text is affected by a C-to-A point mutation, while another is affected by a chromosomal inversion of five nucleotides. Recombination can bring these separate mutations together in individual sequence texts. Mutations are in black, while the original sequence is in grey. Relative sizes of letters indicate their frequency in the population. If the A is advantageous, it may eventually fix in the population. At fixation, the A is most commonly found in combination with the noninverted sequence because that is the sequence it originally arose upon. (C) Comparison of the focal group’s sequence text to that of a closely related outgroup (population or species) can help with inference of the ancestral sequence text. While some methods for detecting natural selection (F_ST outlier, d_N/d_S, McDonald–Kreitman test, etc.) require such comparisons, other methods can potentially identify the targets of natural selection without comparison to outgroup sequences. The majority of studies documented here do incorporate a test that utilizes out-group comparison (see text).

The modern evolutionary biologist attempting to infer past events from the historical but palimpsest-like text of a species’ genome is therefore faced with an exciting though exacting task: identify regions of the genome critical to adaptation despite the muddled historical record encoded in the palimpsest-like genome. Increasingly, the task of identifying targets of natural selection is performed using genomewide, population-level data. Indeed, genomewide scans for natural selection (GWSS), in which anomalous patterns of genetic diversity are linked to selective events, have produced a number of important results. For example, in humans, frequency of a null variant of CYP3A5 is positively correlated with population distance from the equator; given that CYP3A5 functions in salt homoeostasis, it has been suggested that climatic environmental variables act as selective agents at this locus (Thompson et al. 2004). Subsequently, a number of GWSS corroborated this locus as a target of selection in Europeans and Asians (Carlson et al. 2005; Voight et al. 2006; Olesyk et al. 2008). As an interesting parallel, based on a comparison between the genomes of wild and domestic camels, Jirimutu et al. (2012) found that 11 copies of CYP2J (a member of the same cytochrome P450 family to which CYP3A5 belongs) are found in the wild camel; this far exceeds the copy number of this gene in other mammals (e.g. humans have only one copy). The selective pressure for maintenance of this high copy number is likely also related to salt homoeostasis, as camels are able to ingest large quantities of salt without developing hypertension (Jirimutu et al. 2012). As the number of species investigated using methods of GWSS increases, interspecific comparisons such as this that consider targets of selection and putative selective pressures will refine our understanding of a variety of evolutionary processes including convergent evolution.

The use of genomewide data, which, unlike candidate gene approaches, interrogates variation across the genome, is meant to identify selective targets unbiased by a priori expectations (Ellegren 2014). Yet, a number of factors may undermine this bias-free hope for GWSS. Even though the explicit bias of a candidate gene study is eliminated in GWSS, empirical and simulation studies have shown that some selective events are inherently more difficult to identify. For example, selection on standing variation (Hermisson & Pennings 2005; Przeworski et al. 2005) and selection targeting molecular variants with complex mutational properties (Zhang et al. 2012; Haasl & Paysuer 2013) involve population genetic dynamics that often differ from those underlying stereotypical signatures of selection such as extended haplotype homozygosity (Sabeti et al. 2002). Thus, GWSS based on standard summary statistics and methods may fail to identify a range of selective events, including soft sweeps, polygenic selection and selection targeting genetic variants such as microsatellites or copy number variants (Innan & Kim 2004; Pritchard & Di Rienzo 2010; Haasl et al. 2014). At the biological level, another potential bias derives from the fact that different taxa are characterized by a remarkable diversity of demographic and natural histories as well as a wide variety of environmental factors that may act as selective pressures. Frequently, GWSS are performed with the expectation that certain categories of genes are likely to stand out due to what is known of the focal species biology. When studying high-altitude populations, for example, the understandable tendency is to focus on selection targeting genes associated with adaptation to hypoxic conditions despite the fact that whole-genome sequences or dense genotypes are available (e.g. in yak, Qiu et al. 2012; in human, Tibetan and Andean populations, Bigham et al. 2010; Wuren et al. 2014; in pig, Dong et al. 2014). As we will argue, interpretation of GWSS results is improved by consideration of candidate genes determined a priori.

Here, we survey the findings of >100 GWSS to date. Taking a broad view of biodiversity and ecological circumstance, two extreme possibilities might be found among this catalogue of recent GWSS: (i) GWSS identify a disparate array of selective targets with little overlap between studies or (ii) within and among species, the targets and modes of selection identified by GWSS are largely similar. The latter case would signal something profound about evolution, as this would suggest a subset of the genome’s diversity is the primary source of evolutionary change at both micro- and macroscales. Indeed, previous authors have found some interspecific evidence that supports disproportionate targeting of certain DNA regions. For example, Marden (2013) showed that the results of candidate gene studies and GWSS in organisms as diverse as Clamydomonas, Drosophila mojavensis, the Red abalone snail, the Bactrian camel and humans are enriched for metabolic enzymes. Using a GWSS, Vernot et al. (2012) found that the number, although not effect size, of regulatory variants under selection far exceeded the number of selected variants in protein-coding genes. Similarly, a GWSS comparing variation in 2773 protein-coding genes between normal and dwarf forms of the whitefish Coregonus clupeaformis found very few divergence outliers that were protein-coding mutations, suggesting an abundance of regulatory mutations under selection (Hebert et al. 2013).

Yet, it is important to consider the possibility that convergence of natural selection on a subset of molecular targets might result from something other than a true biological bias towards a subset of critical proteins. For example, apparent biological bias may result from inability to detect unusual modes or targets of selection, failure to correct for complications such as variation in recombination rate or focus on a biased set of organisms and/or environments.

The goal of this perspective article was threefold. First, we briefly discuss major genetic factors that complicate GWSS and may lead to nonbiological biases in results. In particular, we discuss how variability in mutation, recombination, natural selection and the genetic architecture of adaptive traits affect the success of GWSS. Second, we survey recent GWSS that include a variety of methods and cover a broad taxonomic range. The nonstandardized nature of GWSS (still in its infancy) precludes us from performing a true, quantitative meta-analysis of this catalogue of GWSS. However, we discuss the most important genetic, evolutionary and methodological trends observed in this representative set of GWSS and discuss whether the data seem to conform to disparate or similar selective targets across studies and species. Furthermore, we perform a more detailed comparison of GWSS focused on the intense selective pressure of hypoxic conditions at high altitude. Finally, based on genetic complications discussed in the first section and early empirical trends identified in our survey of GWSS, we recommend solutions and best practices to improve the efficacy and impact of future GWSS.

Complicating genetic factors in GWSS

Genomewide scans for natural selection convert heterogeneity in patterns of variation across the genome into inferences about natural selection. All factors that cause variation to differ from one locus to the next therefore affect the success of GWSS. Here, we briefly describe challenges and predictions generated by four determinants of genomic heterogeneity: mutation, recombination, selection and the genetic architecture of adaptive traits.

Selection targets variants that arise through a wide spectrum of mutational events, including single-nucleotide substitutions, insertions, deletions, transpositions and inversions (Fig. 1). The mutational class of a variant affects the signature of selection. For example, microsatellites mutate by adding or subtracting repeats to a tandem array. With realistic mutation rates, this process recurrently generates the same adaptive allele on short timescales, violating the common assumption that beneficial alleles have single mutational origins. Additionally, microsatellites often harbour many alleles, leading to complex fitness surfaces (Haasl & Paysuer 2013). Collectively, these characteristics predict little power for standard approaches to find instances of positive selection that involve microsatellites (Haasl et al. 2014). Furthermore, the rates at which the full variety of mutational events occurs span several orders of magnitude. In humans, single-nucleotide mutations happen at a rate of 10⁻⁸–10⁻⁹/site/generation (Nachman & Crowell 2000; Roach et al. 2010), microsatellite mutation rates range from 10⁻² to 10⁻⁶ (Weber & Wong 1993; Sun et al. 2012), and large-scale copy number variants arise at a genomewide rate of 10⁻² (Itsara et al. 2010). There is heterogeneity even among single-nucleotide changes, including an order of magnitude elevation in rate at CpG dinucleotides (Campbell et al. 2012). Beneficial mutations appear at different rates across the genome and signatures of selection vary among classes of mutational variants.

Although it is possible to pinpoint specific mutations targeted by positive selection, most GWSS approaches look for the effects of selection on linked diversity. The length of sequence over which polymorphism is distorted (relative to neutral predictions) is inversely related to the local meiotic recombination rate (Maynard Smith & Haigh 1974; Kaplan et al. 1989). As a result, frequency increases in beneficial variants (‘selective sweeps’) with the same selective intensity will be easier to detect in regions with little recombination. Indeed, a positive correlation between nucleotide diversity and recombination rate across the Drosophila melanogaster genome provided the first general evidence for recurrent selective sweeps (Begun & Aquadro 1992). Genomic variation in the recombination rate assumes two forms. Broadscale rate differences among chromosomes or on megabase scales within chromosomes (Broman et al. 1998; Kong et al. 2002; Jensen-Seaman et al. 2004; Shifman et al. 2006; Backström et al. 2010; Wong et al. 2010) likely reflect meiotic constraints, including crossover interference, suppressed recombination near centromeres and requirements for at least one crossover per chromosome or per chromosome arm (Hassold & Hunt 2001; Pardo-Manuel de Villena & Sapienza 2001; Fledel-Alon et al. 2009). In multiple species, crossovers disproportionately occur at a subset of sites (‘hot spots’) interdigitated by stretches of sequence that rarely experience recombination (‘coldspots’); variation in the location and intensity of hot spots produces dramatic fluctuations in recombination rate on the fine scale (Gerton et al. 2000; Jeffreys et al. 2001; Myers et al. 2005; Coop et al. 2008; Kong et al. 2010; Comeron et al. 2012). The degree of recombination rate heterogeneity varies among species (Smukowski & Noor 2011; Kaur & Rockman 2014), suggesting caution when GWSS are applied to taxa without independent information about the rate of crossing over. Recombination rates also vary among individuals (Brooks & Marks 1986; Broman et al. 1998; Koehler et al. 2002; Kong et al. 2010; Comeron et al. 2012). Theory describing the effects of interindividual differences on signatures of selection is needed (Comeron et al. 2012).

Methods of GWSS usually assume that overall patterns of genomic diversity reflect neutral processes, including nonequilibrium demographic history. However, recurrent selection shapes linked variation. The effects of purifying selection (background selection) and selective sweeps on linked diversity depend on the intensity of selection, the local recombination rate and the mutation rate to selected alleles (Maynard Smith & Haigh 1974; Kaplan et al. 1989; Stephan et al. 1992; Charlesworth et al. 1993). Because these parameters vary along genomes, recurrent selection generates heterogeneous patterns of polymorphism. For example, nucleotide diversity covaries with local recombination rate in a variety of species (Cutter & Payseur 2013). Ignoring recurrent linked selection complicates GWSS in two ways. First, positive selection and purifying selection can be conflated. By reducing diversity, background selection also elevates relative measures of population differentiation (Charlesworth et al. 1997), which provide the basis of several common GWSS methods (such as F_ST-outlier approaches). Second, appropriate thresholds for identifying selective sweeps are unclear. Using patterns of variation at sites affected by linked selection to formulate baseline expectations (as in the commonly employed outlier strategy) violates the basic null model (neutrality) of GWSS and muddles comparisons among genomic windows. Species with large population sizes are especially susceptible to this problem (Leffler et al. 2012; Corbett-Detig et al. 2015). In one notable example, signs of linked selection seem to be pervasive across the Drosophila genome (Begun et al. 2007; Sella et al. 2009; Langley et al. 2012). Ironically, the ability to detect individual instances of selection can decrease as the fraction of the genome affected by linked selection grows.

Finally, genomic regions, genes or variants identified by GWSS are expected to control variation in an organismal trait that in turn affects fitness. How phenotypic selection is projected on to the genome is determined by the genetic architecture of the selected trait. Much of the theory underlying GWSS assumes that selection on individual variants is strong, a situation that arises when adaptive trait differences are conferred by one or a few mutations. Even in this simple scenario, the signature of selection depends on characteristics of adaptive mutations, including dominance (Teshima & Przeworski 2006) and starting allele frequencies (Hermisson & Pennings 2005; Przeworski et al. 2005). When selection targets complex phenotypes—at which variation reflects the action of many mutations—GWSS are less likely to succeed (Pritchard & Di Rienzo 2010). As the number of causative mutations grows, the intensity of selection experienced by each mutation decreases, and the resulting signature of selection is dampened. Selection on a highly polygenic trait generates minimal changes in the frequencies of causative variants; the response to selection mostly comes from covariances among variants (Latta 1998; McKay & Latta 2002; Le Corre & Kremer 2003). In this case, common GWSS approaches fail and alternative strategies are required (Le Corre & Kremer 2012; Berg & Coop 2014; Kemper et al. 2014). When adaptive trait differences are instead generated by a moderate number of substitutions, theory predicts an exponential distribution of phenotypic effects and selection coefficients among substitutions (Orr 1998, 2002). The key point is that the same selection differential applied to phenotypes with contrasting genetic architectures leaves distinct imprints on genomic patterns of variation (Le Corre & Kremer 2012). Because selection affects multiple phenotypes, differences in inheritance provide another source of genomic heterogeneity.

A survey of empirical GWSS

To identify a representative set of GWSS over the last 15 years, we queried the online database Web of Science using a number of different queries, including: ‘selection and genome*wide’, ‘selection and genome and scan’, ‘genome-wide scan’, and ‘genomic scan and selection’. These queries were deemed sufficiently vague to collect the majority of GWSS, while including key terms that would limit the number of query hits. In addition to query results that were clearly not relevant, we rejected a number of GWSS from inclusion in our study for a variety of reasons. We did not include genome scans that used amplified fragment length polymorphisms (AFLPs) as genetic markers. These ecological genomic studies represent an important first look at genomic level data in these nongenetic model organisms. However, AFLPs are usually dominant markers, which limits them to F_ST-outlier approaches (Luikart et al. 2003), and are plagued by fragment-size homoplasy that reduces power to detect natural selection by ~15% (Caballero et al. 2008). We excluded most studies that search for the genetic targets of artificial selection, including GWSS applied to different breeds of domesticated animals. Exceptions to this include cases where GWSS were used to identify selective targets associated with domestication from the wild (Vigouroux et al. 2002; Chapman et al. 2008) or adaptation to natural selective pressures, such as domestic pigs to high altitude (Dong et al. 2014). We included several GWSS with relatively low marker density—for example scans that only use several thousand SNPs or ~100 microsatellites. While these studies provide lower power to detect targets of natural selection, we included them to increase taxonomic diversity in the data set and because these genetic data span the full genome. We also included several instances of genomic scans that analyse exome or transcriptome sequences only and refer to these studies as exomic scans for natural selection (ESS). Finally, we note that the set of GWSS and ESS included here are meant to be representative rather than comprehensive. For example, although we include several studies that report the draft genome sequence of a species and use d_N/d_S to scan the newly obtained genome for positive selection, a complete accounting of such studies is beyond the scope of this review.

Qualitative trends

Table 1 lists details of 132 GWSS and ESS. Additional information, including marker number, focal population(s) and major findings, is included in Table S1 (Supporting information). Not surprisingly, the predominant subject species of GWSS is human. The primary driver of this trend is no doubt the abundance of publicly available SNP data from a diversity of natural human populations; sources include the HapMap project (International HapMap Consortium 2005), Human Genome Diversity Panel (Cann et al. 2002) and 1000 Genomes Project (1000 Genomes Project Consortium 2012). These data make it possible to perform GWSS of importance from the computational laboratory alone. Furthermore, these data also provide reference sets of human population genetic variation for studies in which newly sampled human populations are the focus [e.g. Oceanians (Kimura et al. 2008); Indian ethnic groups (Metspalu et al. 2011); Sardinians (Piras et al. 2012); and pygmy populations from the Philippines and Papua New Guinea (Migliano et al. 2013)]. From 1999 to 2009, 35 of 49 (71%) GWSS focused on human, while from 2010 to present, only 38 of 83 (46%) of GWSS focused on human; the decreasing percentage of human studies indicates that genomewide data are becoming easier to obtain in nonmodel organisms.

Table 1.

Summary of representative genomewide scans for natural selection (GWSS) to date

References	Type	Species	Methodology	Marker type
Jaquiery et al. (2012)	gwss	Acyrthosiphon pisum   (pea aphid species complex)	FST outlier	STR
Schubert et al. (2014)	GWSS	Ancient and extant horses	Comparative scans for selection	WGS
Gagnaire et al. (2012a)	ess	Anguilla rostrata (American eel)	FST outlier; logistic regression	SNP
Gagnaire et al. (2012b)	ess	Anguilla rostrata and   A. anguilla (eels)	Extension of McDonald–Kreitman test	SNP
White et al. (2011)	GWSS	Anopheles gambiae (mosquito)	FST outlier; SFS	SNP
Wallberg et al. (2014)	GWSS	Apis mellifera (honeybee)	FST outlier	SNP
Chavez-Galarza et al.(2013)	gwss	Apis mellifera iberiensis   (honeybee in Iberia)	FST outlier	SNP
Hancock et al. (2011b)	GWSS-GLMM	Arabadopsis thaliana	GLMM	SNP
Huber et al. (2014)	GWSS	Arabadopsis thaliana	SweepFinder and FST outlier	WGS
Lobreaux & Melodelima (2015)	GWSS-GLMM	Arabadopsis thaliana	GLMM (climatic variables)	SNP
Qiu et al. (2012)	GWSS	Bos grunniens and   Bos taurus (yak and cattle)	Comparative genomics	WGS
Edea et al. (2014)	GWSS	Bos taurus (cattle)	LD outlier	SNP
Jirimutu et al. (2012)	GWSS	Camelus bactrianus ferus   (wild Bactrian camel)	dN/dS	WGS
Akey et al. (2010)	gwss	Canis familiaris (dog)	FST outlier	SNP
Quilez et al. (2011)	gwss	Canis familiaris   (dog, breed: Boxer)	Regions of homozygosity	SNP
Pollinger et al. (2005)	gwss	Canis familiaris   (dog, breed: Dauschund)	FST outlier; regions of homozygosity	STR
Hagenblad et al. (2009)	gwss	Canis lupus (Eurasian   wolf in Scandinavia)	Ewens–Watterson; lnRV/H; FST outlier	STR
Hebert et al. (2013)	ESS	Coregonus clupeaformis   (whitefish)	FST outlier	SNP
Tsumura et al. (2014)	gwss	Cryptomeria japonica   (Japanese cedar)	FST outlier	SNP
Pool et al. (2012)	GWSS	Drosophila melanogaster	Modified SweepFinder	WGS
Langley et al. (2012)	GWSS	Drosophila melanogaster	McDonald–Kreitman test; FST outlier	WGS
Reinhardt et al. (2014)	GWSS	Drosophila melanogaster	FST outlier	GWS
Begun et al. (2007)	GWSS	Drosophila simulans	Modified HKA test; SFS	GWS
Shapiro et al. (2007)	gwss	Drosophila spp.	Ka/Ks; SFS; McDonald–Kreitman test	SNP
Gu et al. (2009)	gwss	Equus ferus caballus   (Thoroughbred)	Ewens–Watterson test; FST outlier	STR
Steane et al. (2014)	gwss-glmm	Eucalyptus tricarpa   (Red ironbark eucalyptus)	Bayescan	DArT
Zhan et al. (2013)	GWSS	Falco peregrinus and   Falco cherrug (falcons)	Comparative genomics	WGS
Star et al. (2011)	GWSS	Gadus morhua (Atlantic cod)	Comparative genomics	WGS
Makinen et al. (2008)	gwss	Gasterosteus aculeatus   (three-spined stickleback)	FST outlier; lnRH	STR; indel
Kane & Rieseberg (2007)	gwss	Helianthus annuus (sunflower)	lnRV and lnRH; FST outlier	STR
Chapman et al. (2008)	gwss	Helianthus annuus (sunflower)	lnRV and lnRH	STR
Huttley et al. (1999)	GWSS	Homo sapiens	Extended LD	STR
Akey et al. (2002)	gwss	Homo sapiens	Variety of FST -based methods	SNP
Akey et al. (2002)	gwss	Homo sapiens	FST outlier	SNP
Payseur et al. (2002)	GWSS	Homo sapiens	SFS	STR
Kayser et al. (2003)	gwss	Homo sapiens	lnRV; RST outlier	STR
Storz et al. (2004)	gwss	Homo sapiens	FST outlier; SFS	STR
Shriver et al. (2004)	gwss	Homo sapiens	FST outlier	SNP
Bustamante et al. (2005)	ESS	Homo sapiens	dN/dS	WES
Carlson et al. (2005)	GWSS	Homo sapiens	SFS	SNP
International HapMap Consortium (2005)	GWSS	Homo sapiens	LRH; population differentiation	SNP
Weir et al. (2005)	GWSS	Homo sapiens	FST outlier	SNP
Voight et al. (2006)	GWSS	Homo sapiens	iHS	SNP
Wang et al. (2006)	GWSS	Homo sapiens	LD decay	SNP
Mattiangeli et al. (2006)	gwss	Homo sapiens	Ewens–Watterson test	STR
Kelley et al. (2006)	GWSS	Homo sapiens	SFS	SNP
Zhang et al. (2006)	GWSS	Homo sapiens	WGLRH	SNP
Bubb et al. (2006)	GWSS	Homo sapiens	High SNP density	WGS
Williamson et al. (2007)	GLMM	Homo sapiens	CLRT	SNP
International HapMap Consortium (2007)	GWSS	Homo sapiens	iHS; EHH	SNP
Sabeti et al. (2007)	GWSS	Homo sapiens	iHS; XP-EHH	SNP
Tang et al. (2007)	GWSS	Homo sapiens	Modified EHH; LRH	SNP
Kimura et al. (2007)	GWSS	Homo sapiens	Haplotype homozygosity (Rsb)	SNP
Haygood et al. (2007)	ESS	Homo sapiens	dN/dS	WGS
Hancock et al. (2008)	gwss-glmm	Homo sapiens	GLMM (climatic variables)	SNP
Olesyk et al. (2008)	GWSS	Homo sapiens	FST and heterozygosity outliers	SNP
Johansson & Gyllensten (2008)	GWSS	Homo sapiens	(Haplotype length + FST) outliers	SNP
Kimura et al. (2008)	GWSS	Homo sapiens	LRH; SFS	SNP
O’Reilly et al. (2008)	GWSS	Homo sapiens	Novel recombination rate-based test	SNP
Myles et al. (2008)	GWSS	Homo sapiens	FST outlier	SNP
Amato et al. (2009)	GWSS	Homo sapiens	FST outlier	SNP
Pickrell et al. (2009)	GWSS	Homo sapiens	iHS; XP-EHH; FST outlier	SNP
López Herráez et al. (2009)	GWSS	Homo sapiens	Modified Rsb	SNP
Chen et al. (2009)	GWSS	Homo sapiens	Modified McDonald–Kreitman test	Indel
Andres et al. (2009)	ESS	Homo sapiens	CLRT	WES
Hancock et al. (2010)	GWSS-GLMM	Homo sapiens	GLMM (four ecoregion variables)	SNP
Yi et al. (2010)	ESS	Homo sapiens	PBS	WES
Albrechtsen et al. (2010)	GWSS	Homo sapiens	Excessive Identity by descent	SNP
Bigham et al. (2010)	GWSS	Homo sapiens	lnRH; WGRLH; SFS	SNP; CNV
Simonson et al. (2010)	GWSS	Homo sapiens	iHS; XP-EHH	SNP
Beall et al. (2010)	GWSS	Homo sapiens	Allele frequency differences	SNP
Lappalainen et al. (2010)	GWSS	Homo sapiens	iHS, LRH, EHH; FST outlier	SNP
Chen et al. (2010)	GWSS	Homo sapiens	XP-CLR	SNP
Xu et al. (2010)	GWSS	Homo sapiens	iHS; XP-EHH; XP-CLR; FST outlier	SNP
Metspalu et al. (2011)	GWSS	Homo sapiens	XP-EHH; iHS	SNP
Fumagalli et al. (2011)	GWSS-GLMM	Homo sapiens	GLMM	SNP
Hancock et al. (2011a)	GWSS-GLMM	Homo sapiens	GLMM	SNP
Granka et al. (2012)	GWSS	Homo sapiens	iHS; XP-EHH	SNP
Piras et al. (2012)	GWSS	Homo sapiens	EHH and XP-EHH	SNP
Vernot et al. (2012)	GWSS	Homo sapiens	FST outlier vs. DNase I peak	WGS
Zhang et al. (2012)	GWSS	Homo sapiens	CNV frequency differentiation	SNP; CNV
Jarvis et al. (2012)	GWSS	Homo sapiens	FST outlier; XP-EHH; iHS	SNP
Andersen et al. (2012)	GWSS	Homo sapiens	Composite of multiple methods	SNP; WGS
Scheinfeldt et al. (2012)	GWSS	Homo sapiens	Locus-specific branch length	SNP
Suo et al. (2007)	GWSS	Homo sapiens	iHS; XP-EHH	SNP
Migliano et al. (2013)	GWSS	Homo sapiens	iHS; XP-EHH	SNP
Somel et al. (2013)	ESS	Homo sapiens	dN/dS	SNP
Hider et al. (2013)	GWSS	Homo sapiens	SFS; Rsb; PBS	WGS
Frichot et al. (2013)	GWSS-GLMM	Homo sapiens	Latent factor mixed models	SNP
Raj et al. (2013)	GWSS	Homo sapiens	iHS; FST outlier	SNP
Liu et al. (2013)	gwss	Homo sapiens	Long-range haplotype method	SNP
Bhatia et al. (2014)	GWSS	Homo sapiens	Deviations in local ancestry	SNP
Colonna et al. (2014)	GWSS	Homo sapiens	iHS; XP-EHH; FST outlier	SNP; indel
Eichstaedt et al. (2014)	GWSS	Homo sapiens	iHS; XP-EHH; FST outlier	SNP
Clemente et al. (2014)	GWSS	Homo sapiens	iHS; SFS	SNP
Haasl et al. (2014)	GWSS	Homo sapiens	Novel ksk2 test	WGS
Ali et al. (2014)	GWSS	Homo sapiens	iHS; XP-EHH	SNP
Wuren et al. (2014)	GWSS	Homo sapiens	iHS; XP-EHH	SNP
Fangy et al. (2014)	GWSS	Homo sapiens	iHS and Derived Intraallelic   Nucleotide Diversity test	WGS
Enard et al. (2014)	ESS	Homo sapiens	iHS; XP-EHH	WGS
Sjostrand et al. (2014)	GWSS	Homo sapiens	Novel Maximum Frequency of Private   Haplotypes test	SNP
Leffler et al. (2013)	GWSS	Homo sapiens, Pan troglodytes	Haplotype sharing between species	WGS
Nielsen et al. (2005)	ESS	Homo sapiens, Pan troglodytes	dN/dS likelihood ratio test	WES
Clark et al. (2003)	ESS	Homo sapiens, Pan troglodytes,   Mus musculus	dN/dS in the human lineage	WES
Enard et al. (2010)	GWSS	Four primates	Novel version of HKA test	SNP
Westram et al. (2014)	ESS	Littorina saxatilis   (marine snail)	FST outlier	WES
Rhesus macaque Genome Sequencing and Analysis Consortium (2007)	GWSS	Macaca mulatta	dN/dS likelihood ratio test	WGS
George et al. (2011)	ESS	Numerous primates	dN/dS for each orthologous set of   genes	WES
Branca et al. (2011)	GWSS	Medicago truncatula   (a legume, Barrel clover)	Extreme 100 kb windows for π,   recombination and LD	WGS
Yoder et al. (2014)	GWSS-GLMM	Medicago truncatula   (a legume, Barrel clover)	GLMM (climatic variables)	SNP
Srivastava et al. (2012)	ess	Melospiza melodia   (song sparrow)	Comparative genomics	SNP
Puzey & Vallejo-Marin (2014)	GWSS	Mimulus guttatus   (monkey flower)	SFS	WGS
Ihle et al. (2006)	gwss	Mus musculus (house mouse)	lnRV and lnRH	STR
Teschke et al. (2008)	gwss	Mus musculus domesticus and   Mus musculus musculus   (house mouse)	lnRH	STR
Limborg et al. (2014)	gwss	Oncorhynchus gorbuscha   (pink salmon)	FST outlier	SNP
Lv et al. (2014)	gwss-glmm	Ovis aries (sheep)	GLMM	SNP
Eckert et al. (2010)	ess	Pinus taeda (Loblolly pine)	GLMM (heterozygosity of SNPs)	SNP
Frichot et al. (2013)	ess-glmm	Pinus taeda (loblolly pine)	Novel GLMM approach: latent   factor mixed models	SNP
Ochola et al. (2010)	ESS	Plasmodium falciparum	HKA test; SFS	WGS
Amambua-Nqwa et al. (2012)	GWSS	Plasmodium falciparum	SFS	WES
Park et al. (2012)	GWSS	Plasmodium falciparum	XP-EHH on isolates resistant to >1   of 12 antimalarial drugs	SNP
Nygaard et al. (2010)	GWSS	Plasmodium spp.   (seven species)	Modified McDonald–Kreitman   test; SFS	WGS
Fraser et al. (2015)	GWSS	Poecilia reticulata (guppy)	FST outlier	WGS
Evans et al. (2014)	GWSS	Populus trichocarpa   (black cottonwood)	FST outlier; iHS	SNP
Cai et al. (2013)	GWSS	Pseudopodoces humilis   (Ground tit)	Comparative genomics	SNP
Vincent et al. (2013)	gwss-glmm	Salmo salar (Atlantic salmon)	GLMM (49 environmental variables)	SNP
Zueva et al. (2014)	gwss-glmm	Salmo salar (Atlantic salmon)	FST outlier and LFMM   (Frichot et al. 2013)	SNP
Casa et al. (2005)	gwss	Sorghum bicolor	Ewens–Watterson test; lnRH;   FST outlier	STR
Thomas et al. (2012)	GWSS	Staphylococcus aureus   (bacterium)	SFS (balancing sel.): Tajima’s   D > 2.03; π/K > 0.12	WGS
Dong et al. (2014)	GWSS	Sus scrofa (pig)	FST outlier	SNP
Cavagnagh et al. (2013)	ess	Triticum aestivum (wheat)	FST outlier; pairwise haplotype sharing	SNP
Sun et al. (2013)	ESS	Tursiops truncates   (common bottlenose dolphin)	dN/dS	WES
Vigouroux et al. (2002)	ess	Zea mays (maize)	Ewens–Watterson test	STR

Regarding type of scan: ESS, exonic scan for selection; GLMM, use of generalized linear mixed model methodology; lowercase indicates a relatively small number of markers used. Regarding methodology: CLRT, composite likelihood ratio test; iHS, integrated haplotype statistics; EHH, extended haplotype homozygosity; XP-EHH, cross-population EHH; LRH, long-range haplotype test; WGRLH, whole-genome LRH; SFS, site frequency spectrum statistic(s); PBS, population branch statistic; XP-CLR, cross-population composite likelihood ratio; HKA, Hudson–Kreitman–Aguade test. Regarding marker type: STR, microsatellite (short tandem repeat); CNV, copy number variant; WGS, whole-genome sequence; WES, whole-exome sequence.

The majority of GWSS in Table 1 rely on intraspecific data and methods that compare genetic variation between populations to identify targets of natural selection. Of these methods, the most common are (i) simple F_ST-outlier approaches, in which SNPs with extreme F_ST among pairs of populations are associated with linked selection, and (ii) the cross-population extended haplotype homozygosity test (XP-EHH; Sabeti et al. 2007). Given that it is easier to obtain data from a single population, this trend suggests that biologists prefer to apply analyses that rely on multipopulation comparative data. One reason for this may be that statistics of the site frequency spectrum require a relatively large number of genetic markers to estimate. On the other hand, an F_ST comparison can be made for every marker included regardless of the total number of markers. Importantly, the scope of comparative approaches has expanded with the advent of recently developed methods that use generalized linear mixed models (GLMMs; Hancock et al. 2008, 2010; Frichot et al. 2013), which seek correlations between environmental parameters (potential selective pressures) and genetic variants across multiple populations exposed to different values of these parameters. In these studies, samples are often drawn from individuals spanning wide geographic distances.

Our survey of the GWSS literature also reveals a strong methodological and biological bias towards attempting to detect positive, directional natural selection. Very few of the studies included in our survey address, or attempt to analyse, other forms of natural selection. Exceptions include a small number of scans that intentionally sought signatures of balancing selection. Bubb et al. (2006) identified 16 regions of high SNP density outside of the human leucocyte antigen (HLA) system and loci for ABO blood antigens that provided suggestive evidence for balancing selection within human populations. Intriguingly, Andres et al. (2009) performed an ESS in which signatures of long-term balancing selection in humans were discovered in loci related to cellular structure, including keratins. Leffler et al. (2013) discovered 125 regions in addition to loci of the HLA system in which humans and chimpanzee (Pan troglodytes) shared haplotypes, suggesting long-term balancing selection. Parasites and infectious organisms are relatively overrepresented for scans focused on balancing selection, presumably because loci with greater-than-average genetic variation are critical to the successful infection of host organisms. In various species of Plasmodium, the causative parasite of malaria, two separate scans identified loci subject to balancing selection based on summaries of the site frequency spectrum, including loci involved in host–parasite interaction (Nygaard et al. 2010; Ochola et al. 2010). Thomas et al. (2012) scanned 16 strains of the bacterium Staphylococcus aureus, which can become methicillin resistant (MRSA) and generate serious threats to health care (David & Daum 2010); based on summaries of the site frequency spectrum, the authors discovered 186 windows in 99 genes putatively affected by balancing selection.

Early GWSS focused on the human genome used relatively small numbers of markers (Huttley et al. 1999; Akey et al. 2002; Payseur et al. 2002; Kayser et al. 2003; Shriver et al. 2004; Storz et al. 2004); before 2005, the largest number of markers applied in a human GWSS was 26 530 SNPs (Akey et al. 2002). Preferences for marker number and type changed dramatically in 2005 with the advent of new technologies and large publically available data sets. With one exception (Mattiangeli et al. 2006), all GWSS focused on human with a publication date of 2005 or later used SNPs or whole-genome sequences; in cases where SNPs were used, >50 000 SNPs were genotyped and the majority of studies used ~1 million SNPs. In other species, where comparative data are lacking, it is difficult to assess the strength of this trend, but certainly, other species are now genotyped or sequenced at high coverage with some frequency: 8.3 million SNPs in honeybee, Apis mellifera (Wallberg et al. 2014), and whole-genome sequences for >100 guppies of the species Poecilia reticulata (Fraser et al. 2015).

Interpreting GWSS results: the case of hypoxia as a selective pressure

Humans have adapted to hypoxic conditions in three distinct high-altitude environments: the Tibetan Plateau, the Ethiopian highlands and the Andean Altiplano (Bigham et al. 2010). We compared the results of GWSS in human populations living in these regions as well as several recent studies focused on yak, cattle, pig and ground tit in these same geographic regions (Tables 2 and S2, Supporting information). The GWSS included in Table 1 are too disparate to serve as the basis of a meaningful meta-analysis. However, focusing on this relatively small number of studies in which animals have adapted to the same, unequivocally strong selective pressure revealed valuable insights regarding the interpretation of GWSS results. First, even when a strong selective pressure exists to aid interpretation of results, discrepancies still arise among studies. Second, and more positively, this example shows it is possible to delimit different evolutionary genetic responses to a common selective pressure.

Table 2.

Genes of the hypoxia-inducible factors (HIF) pathway producing signatures of positive selection in various high-altitude adapted populations

References Species Region	Wuren et al. (2014) Human Tibet	Eichstaedt et al. (2014) Human Andes	Scheinfeldt et al. (2012) Human Ethiopia	Simonson et al. (2010) Human Tibet	Beall et al. (2010) Human Tibet	Bigham et al. (2010) Human Tibet	Bigham et al. (2010) Human Andes	Xu et al. (2010) Human Tibet	Yi et al. (2010) Human Tibet	Qiu et al.(2012) Yak Tibet	Edea et al.(2014) Cattle Ethiopia	Dong et al.(2014) Pig Tibet	Cai et al.(2013) Ground Tit Tibet
EGLN1	X			X		X	X	X	X
ELGN3					X
PTEN				X
HIF1A					X					X
EPAS1	X			X	X	X		X	X
PPARA	X			X
EDNRA				X
ANGPTL4				X
VEGFB		X
VEGFC												X
MMP3										X
PRKAA1							X
NOS2A							X

Genes in bold face are members of the HIF pathway. Underlined genes are downstream targets of HIF pathway genes.

See Table S2 (Supporting information) for extended findings of these studies.

Humans from low-altitude regions of the world acclimate to hypoxic conditions via erythropoiesis and thereby increased haemoglobin concentrations (Storz 2010). Key to this acclimation (rather than adaptation) response is a regulatory pathway whose central transcription factors are known as hypoxia-inducible factors (HIFs). However, the quick physiological fix of increased erythropoiesis represents a short-term solution; blood viscosity increases with the greater number of erythrocytes, which ultimately hampers blood flow and limits tissue oxygenation (Villafuerte et al. 2004). Surprisingly, Tibetan natives possess haemoglobin concentrations similar to individuals living at sea level, while Andean natives possess significant increases in haemoglobin concentrations relative to low-altitude populations (Beall et al. 1998). This suggests that the genetic architecture of high-altitude adaptation may be different in Andeans and Tibetans.

Indeed, while EPAS1—which codes for the oxygen-sensitive domain of the transcription factor HIF-2—is a top adaptive hit in all GWSS focused on humans of the Tibetan Plateau (Beall et al. 2010; Bigham et al. 2010; Simonson et al. 2010; Xu et al. 2010; Yi et al. 2010; Wuren et al. 2014), the results of GWSS focused on Andean and Ethiopian high-altitude populations do not identify EPAS1 as a target of positive selection (Bigham et al. 2010; Scheinfeldt et al. 2012; Eichstaedt et al. 2014). EGLN1, which codes for a repressor of EPAS1 production, showed signatures of natural selection in both Andeans and Tibetans, but the adaptive patterns of genetic variation at EGLN1 are clearly distinct between the two populations (Bigham et al. 2010). Note that Eichstaedt et al. (2014) did not uncover EGLN1, which shows that distinct studies using different samples, genetic markers and/or methods can arrive at different results despite the strong selective pressure acting to shape relevant genomic regions. Furthermore, Bigham et al. (2010) identified 14 and 37 1 Mb regions of significance based on multiple, corroborating signatures of selection in Tibetans and Andeans, respectively. None of these regions overlapped with each other. Further evidence of the variable genetic architecture of high-altitude adaptation was provided by a GWSS focused on native populations of the Ethiopian highlands, where no enrichment for HIF pathway genes was discovered (Scheinfeldt et al. 2012)—a result that was distinct from both Andeans and Tibetans. Interestingly, top signatures of selection included genes related to immune function, suggesting that distinct pathogenic exposures at high altitude might represent as great a selective pressure as hypoxia itself (Scheinfeldt et al. 2012).

Several GWSS have also examined the effect of high-altitude environment on the evolution of domesticated animals. A comparison of genetic variation between yaks of the Tibetan Plateau and lowland cattle revealed that HIF1A, a subunit of the HIF-1 transcription factor, was targeted by positive selection in yaks (Qiu et al. 2012); the same gene appears to be targeted by selection in human populations of the Tibetan Plateau (Beall et al. 2010). In addition, downstream targets of HIF pathway regulation such as ARG2 as well as numerous proteins key to the metabolism of polysaccharides, amino acids and fatty acids appear to be targeted by selection in yaks. Similarly, metabolic genes of cattle living in the Ethiopian highlands bear strong signatures of selection (Edea et al. 2014). Pigs living in the Tibetan Plateau also bear signatures of selection for genes involved in angiogenesis, response to hypoxia and nucleic acid metabolism (Dong et al. 2014).

In a telling comparison with these mammalian examples, calculation of d_N/d_S ratios for the ground tit (Pseudopodoces humilis), a bird living in the Tibetan Plateau, in comparison with numerous avian species of low altitude revealed positive selection on genes associated with cardiac function and hormone behaviour (Cai et al. 2013). Putative targets of selection in the ground tit genome were not enriched for (i) metabolic genes, as in high-altitude domesticated mammals, or (ii) genes of the HIF pathway, as in all mammals.

The results from this focused set of studies provide several important insights regarding the interpretation of GWSS. First, several studies mentioned here relied upon an a priori list of candidate genes to filter the list of genes deemed significant in the GWSS (Simonson et al. 2010; Eichstaedt et al. 2014). At face value, this approach may seem strange, as it counters the unbiased nature of GWSS. However, GWSS provide a list of putative regions targeted by selection. Depending on the number of markers, these regions may be quite large and include numerous genes and regulatory regions. Moreover, the list is likely to contain numerous false positives. Then, what approach should we take to filter the list for the most likely targets of selection? One common approach, found frequently in the human GWSS literature, is to find reassurance in the fact that well-established targets of natural selection such as LCT (lactase) are present in the list of significant hits and then suggest that the rest of the list is sure to include numerous true targets of selection. Even if this inference is correct, this approach does little to further our understanding of human biology and the selective forces that have helped shape human adaptation throughout the history of the human lineage. We have trouble connecting selected genes to the causative selective pressure precisely because the unbiased method of GWSS makes no a priori assumptions regarding what classes of genes might be targeted by selection. Indeed, it is in this very situation that the researcher is tempted to cherry-pick the list of significant hits for genes with interesting functions and construct plausible though likely fanciful stories of adaptation (Barrett & Hoekstra 2011; Pavlidis et al. 2012).

Rather than relying upon potentially spurious a posteriori evaluations of a list of selected genes, it therefore seems prudent to list our a priori assumptions of the genes or pathways we expect to find before performing a GWSS. Furthermore, interpretation of the results of GWSS focused on high-altitude adaptation has the advantage of dealing with a clearly defined, strong selective pressure. In this context, when a gene such as EPAS1 is shown to bear the top signature of selection (Beall et al. 2010; Yi et al. 2010), researchers have little reason to doubt the validity of this result. Knowledge of the primary selective pressures acting on a population also facilitates the further exploration of the correlation between phenotype and putatively selected genotype. Beall et al. (2010), for example, showed that the single-nucleotide variants at high frequency in the sequence of EPAS1 in high-altitude Tibetans were associated with low haemoglobin concentration. This finding is congruent with the counterintuitive fact that Tibetans possess low haemoglobin concentrations (Beall et al. 1998), particularly given that Andeans show high haemoglobin concentration (Storz 2010) and GWSS of Andean genomes revealed no selection on EPAS1 (Bigham et al. 2010; Eichstaedt et al. 2014).

Second, intraspecific and interspecific comparisons of adaptive response to hypoxic conditions make clear the radically different genetic architectures that can result from an identical selective pressure. Hypoxic conditions in the Tibetan Plateau, Andean Altiplano and Ethiopian highlands have apparently all elicited highly different genetic adaptations in response to this selective pressure (Bigham et al. 2010; Scheinfeldt et al. 2012; Eichstaedt et al. 2014). Moreover, consideration of domesticated mammals and a wild bird species widens the adaptive response even further. Finally, GWSS focused on humans of the Tibetan Plateau yielded largely similar sets of significant hits—namely genes central to the HIF pathway (Table 2; Beall et al. 2010; Bigham et al. 2010; Simonson et al. 2010; Wuren et al. 2014; Xu et al. 2010; Yi et al. 2010). Reassuringly, this suggests that GWSS reliably uncover adaptive genes of large effect despite varied methodological approaches. However, we again note that the focus of these studies on a key, unambiguous selective pressure makes the adaptive evolution of HIF pathway genes more convincing.

Recommendations and best practices

The genetic complications outlined above make clear that identification of adaptive alleles in species or populations using GWSS is made difficult by genomic heterogeneity in mutation, recombination, selection and the genetic architecture of adaptive phenotypes. By definition, GWSS cover the entire genome. Therefore, GWSS methodology should better incorporate aspects of genomewide heterogeneity.

We provide two recommendations. First, information about key determinants of genomic diversity can be used to adjust genomewide patterns. For example, local estimates of rates of recombination and deleterious mutation could be used to fit a model of background selection to levels of polymorphism across the genome (Reed et al. 2005). The best-fit model could serve as a new null model for identifying instances of positive selection (Comeron 2014). Because measures of genomic heterogeneity are immediately available for genetic model organisms and therefore can be incorporated into analyses, we believe these species are currently the best targets for GWSS. In other organisms, we recommend using surrogates of genomic heterogeneity to improve the interpretation of results. For example, in many species, recombination rates are highly correlated with the distance from the centromere. Even if no recombination rate estimates are available for a species, researchers conducting GWSS could use distance from the centromere as a rough gauge of relative recombination rate. Our most general recommendation is to simply be aware that a set of GWSS results are shaped by genomic heterogeneity. In addition to the use of empirical measurements of genomic heterogeneity in mutation or recombination, it is also important to consider selective targets other than single-nucleotide variants; researchers should be aware of the differing effects selection can have on these genetic variants and, in some cases, methods that have been developed to aid in their detection (Sebat et al. 2004; Feuk et al. 2006; Haasl & Payseur 2013; Haasl et al. 2014).

A second general recommendation is to measure the consequences of heterogeneity in mutation, recombination, selection and the genetic architecture for genomic patterns of diversity using simulations that sample a range of reasonable parameter values. Because genomic heterogeneity forms patterns along chromosomes, whole-chromosome simulations should be especially useful. In empirical cases in which information about genomic heterogeneity is available, these simulations could be built directly into the GWSS inference procedure, using an approach such as approximate Bayesian computation. Simulation results could establish useful guidelines for interpreting GWSS even in the absence of genomic heterogeneity measures. For example, the level of variation in mutation rate that produces high false-positive rates could be determined for a range of selective sweep scenarios the investigator wishes to detect in the GWSS. In this manner, the plausibility of alternatives to selection could be gauged.

Genomewide scans for natural selection can also be improved by considering knowledge of potential selective agents. In our discussion of GWSS focused on adaptation to high-altitude conditions, we suggested that a major advantage of these studies was the presence of an unequivocal selective pressure affecting the subject populations. This advantage came to bear near the end of these studies during interpretation of the results of each GWSS. The known selective agent facilitated the identification of plausible targets of selection from the list of significant genomic regions. Yet, the Tibetan Plateau, Andean Altiplano and Ethiopian highlands represent some of the most extreme terrestrial environments on the Earth. Is it possible to identify unambiguous selective pressures affecting populations in more pedestrian regions of Earth? The short answer is no.

However, a suite of recently developed methods do not require a priori determination of selective pressures. Instead, these approaches, which employ GLMMs, simply require that the researcher identify a set of environmental parameters that may act as selective pressures (Hancock et al. 2008, 2010; Frichot et al. 2013). These approaches search for correlations between values of these environmental parameters (or synthetic combinations of them) and genetic variation in individuals sampled from across a geographic range that includes substantial variation in these environmental parameters. The results of GWSS-GLMMs simultaneously identify the most likely selective pressures and the genomic regions subject to natural selection as a result of these pressures. Again, the main advantage to this type of approach is that it links the otherwise anonymous list of putative selective targets with ecological and biological information. This combination of information makes it less tempting to tell stories about adaptation (Pavlidis et al. 2012) and to scan the genomic palimpsest for signatures of selection that are biologically relevant.