TARGETED CAPTURE IN EVOLUTIONARY AND ECOLOGICAL GENOMICS

Abstract

The rapid expansion of next-generation sequencing has yielded a powerful array of tools to address fundamental biological questions at a scale that was inconceivable just a few years ago. Various genome partitioning strategies to sequence select subsets of the genome have emerged as powerful alternatives to whole genome sequencing in ecological and evolutionary genomic studies. High throughput targeted capture is one such strategy that involves the parallel enrichment of pre-selected genomic regions of interest. The growing use of targeted capture demonstrates its potential power to address a range of research questions, yet these approaches have yet to expand broadly across labs focused on evolutionary and ecological genomics. In part, the use of targeted capture has been hindered by the logistics of capture design and implementation in species without established reference genomes. Here we aim to 1) increase the accessibility of targeted capture to researchers working in non-model taxa by discussing capture methods that circumvent the need of a reference genome, 2) highlight the evolutionary and ecological applications where this approach is emerging as a powerful sequencing strategy, and 3) discuss the future of targeted capture and other genome partitioning approaches in light of the increasing accessibility of whole genome sequencing. Given the practical advantages and increasing feasibility of high-throughput targeted capture, we anticipate an ongoing expansion of capture-based approaches in evolutionary and ecological research, synergistic with an expansion of whole genome sequencing.

Introduction

The ability to address many fundamental evolutionary and ecological questions is no longer constrained simply by the generation of sequence data. Instead, as next generation sequencing (NGS) has become more accessible, a major challenge has become choosing which sequencing strategies to pursue (Ekblom & Galindo 2011; Davey et al. 2011; McCormack et al. 2013b; Ellegren 2014). The power of a given NGS experiment to address a central research question would ideally drive such decisions. However, often these choices come down to more practical considerations, such as cost, ease of use, or researcher expertise level (Ekblom & Galindo 2011). Reference genomes remain integral to most NGS analytical frameworks, yet de novo whole genome sequencing (WGS) and assembly remains prohibitively costly, time-consuming, and computationally difficult for widespread adoption by individual labs. Thus, the challenges of NGS data can be particularly acute for biologists interested in species without established reference genomes (hereafter non-reference species).

Fortunately, diverse genome partitioning approaches have also been developed that enable the collection of genome-wide data at substantially reduced effort and cost compared to WGS (Davey et al. 2011). Two approaches, restriction-site-associated DNA sequencing (RAD-seq and related approaches; Miller et al. 2007; Baird et al. 2008; Elshire et al. 2011; Peterson et al. 2012; Wang et al. 2012) and whole transcriptome shotgun sequencing (RNA-seq; Wang et al. 2009), have quickly become the predominant genome partitioning methods used in evolutionary studies. Both RAD-seq and RNA-seq are relatively simple to implement and can be applied to an array of evolutionary questions within and between species (Davey & Blaxter 2010; Ekblom & Galindo 2011). As a simple NGS derivative of more traditional marker based approaches (Miller et al. 2007), RAD-seq in particular has emerged as the gateway genomic approach for most non-reference species.

Although these partitioning approaches are providing a wealth of insights in non-reference species, they can also be strongly limiting for some research questions (Ku et al. 2012; Rubin et al. 2012; Arnold et al. 2013; Henning et al. 2014) or even more effective when used in concert with other NGS strategies. High-throughput targeted capture is a general class of methods that achieves genome partitioning through selective enrichment of specific subsets of the genome prior to NGS. Targeted capture approaches were developed as more cost-effective and high-throughput alternatives to WGS and multiplex PCR, respectively, to obtain large datasets of orthologous loci across many individuals (Olson 2007). The first proof-of-principle high throughput capture studies targeted large subsets of the human genome using arrays (6726 exons, ~5 Mb, Albert et al. 2007; 204,490 exons, 42.7 Mb, Hodges et al. 2007; ~10,000 exons, 6.7 Mb, Porreca et al. 2007; 304 kb of the X chromosome, Okou et al. 2007), demonstrating the massive scaling potential of this approach. The subsequent development of in-solution targeted capture (Gnirke et al. 2009) provided numerous technical improvements over array-based platforms (Gnirke et al. 2009; Tewhey et al. 2009a; Mamanova et al. 2010) and has emerged as the industry standard. In addition to advantages in scalability and cost-effectiveness, targeted capture generally provides enhanced data quality relative to alternative genome partitioning approaches, including lower variance in target coverage, more accurate SNP calling, higher reproducibility, and longer assembled contigs (Gnirke et al. 2009; Tewhey et al. 2009a; Ku et al. 2012; Harvey et al. 2013).

The benefits of high throughput targeted capture were immediately apparent in biomedical fields (Olson 2007; Hodges et al. 2007). By focusing NGS efforts on “high-value genomic regions” (Hodges et al. 2007), such as exons or structural variants, targeted capture has yielded tremendous power to identify genetic variants associated with simple and complex human diseases (Choi et al. 2009; Ng. et al. 2009; O’Roak et al. 2011; Worthey et al. 2011; Calvo et al. 2012; Rivière et al. 2012; Zaidi et al. 2013; Gee et al. 2014; Iossifov et al. 2014; Guo et al. 2015). In parallel with its medical applications, targeted capture began to emerge as a powerful approach to address evolutionary questions in humans (Briggs et al. 2009; Burbano et al. 2010; Krause et al. 2010; Li et al. 2010; Yi et al. 2010). Although initially restricted to species with sequenced genomes (e.g., chimpanzee, Perry et al. 2010; Drosophila melanogaster, Wang et al. 2010; maize, Fu et al. 2010), more recent extensions to non-reference species have demonstrated the potential to effectively use targeted capture across diverse taxa (Cosart et al. 2011; Mason et al. 2011; Vallender 2011; Bi et al., 2012, 2013; Good et al. 2015). Nonetheless, targeted capture remains relatively uncommon in evolutionary biology studies.

A major impediment to the widespread use of targeted capture in non-reference species is the challenge of designing a capture probe set (Elshire et al. 2011), which by definition requires a priori knowledge of target sequences. Fulfilling this simple requirement can be daunting. The goal of this review is to discuss how some of these practical challenges can be overcome in non-reference species and to draw attention to the utility of targeted capture in addressing central questions in evolutionary and ecological genomics. We assume a basic understanding of hybridization-based capture methods, the details of which have been reviewed elsewhere (e.g., Mamanova et al. 2010). Finally, given the ongoing trend of decreasing NGS costs, we consider the extent to which targeted capture and other genome partitioning approaches are transitory genomic tools on the path to routine whole genome sequencing.

Targeted capture without a reference genome

Targeted capture approaches rely on prior sequence knowledge. Thus, the first major implementation challenge is identifying the genomic sequences to be used for capture design. Several studies have focused on technical aspects of capture design and implementation, including probe design (Tewhey et al. 2009a; Ávila-Arcos et al. 2011; Clark et al. 2011), library preparation protocols (Mamanova et al. 2010; Meyer & Kircher 2010; Harakalova et al. 2011; Kircher et al. 2012; Rohland & Reich 2012), and bioinformatics (Sulonen et al. 2011; Asmann et al. 2012; Cosart et al. 2014; Jiang et al. 2015). Few have discussed complexities that arise with identifying genomic regions of interest in non-reference species (but see Bi et al. 2012). Several solutions that circumvent the need of a reference genome for focal species have now emerged, enabling an expansion of targeted capture to a wide array of taxa. Broadly, these solutions fall into PCR capture, de novo assembly capture, and divergent reference capture (Fig. 1, Table 1).

Figure 1.

The steps to synthesize a targeted capture probe set in non-reference taxa using four alternative approaches. In PCR capture, target sequences are identified and amplified with PCR. Amplicons are then biotinylated to create long single-stranded probes to capture these sequences across a diverse range of species. For de novo assembly capture, initial transcriptomic, RAD-seq, or WGS experiments are used to generate de novo assemblies from which a targeted capture probe set is designed. In transcriptome-based approaches, probes are designed to target regions within a single exon corresponding to a transcript or set of transcripts of a gene. For RAD-seq or WGS capture, probes may target population-informative markers from the assembly. Finally, using a reference genome assembly of a divergent species, genomic sequences of interest and their locations (e.g., often the chromosome and the start and stop position of the interval) are identified and a probe set is designed to tile across the targeted regions.

Table 1.

The scalability, cost, taxonomic range of application (the level of species divergence from probes), and required genomic knowledge associated with three approaches of targeted capture in non-reference species.

	Non-reference targeted capture methods
	PCR	de novo assembly	Divergent genome
Feasible range of targeted loci	up to ~100	up to hundreds of thousands	up to hundreds of thousands
Total probe synthesis cost (maximum number of targets)	low	high	intermediate
Probe synthesis cost per probe	high	intermediate	low
Taxonomic range of application	Depends on targets, but relatively high	Transcriptome: high RAD-seq: moderate WGS: high to moderate depending on target	Depends on targets (high for exons or UCEs, low for quickly evolving loci)
Genomic knowledge required	PCR primer sequences	De novo transcriptome, RAD, or whole genome assembly	Annotated genome assembly or multisequence alignment

Under the PCR-based approach, PCR products can be used as capture probes for high throughput sequencing of a relatively small number of genomic regions (Maricic et al. 2010; Mariac et al. 2014; Peñalba et al. 2014; Tsangaras et al. 2014). Here PCR is performed on a set of target loci, those PCR products are biotinylated or affixed to an array and used to capture homologous sequences for subsequent NGS (Fig. 1). This approach leverages the extensive availability of primer sequences for many non-reference species (Portik et al. 2012; Peñalba et al. 2014). In addition, probes generated from long-range PCR may increase capture efficiency of divergent sequences (e.g., up to 27% mitochondrial divergence; Peñalba et al. 2014), thus increasing the taxonomic range of this approach (Table 1). PCR-based capture is perhaps the simplest and least expensive application of targeted capture (assuming capture of a modest numbers of loci) because it avoids the most cost-intensive aspect of target capture: the synthesis of large probe sets. However, this approach is relatively low throughput and most appropriate for questions that require only a handful of loci (e.g., up to 100 loci, Table 1), such as phylogenetic or phylogeographic studies (Mariac et al. 2014; Peñalba et al. 2014).

Several capture approaches for non-model taxa leverage other sequencing strategies to develop de novo sequence assemblies from which to design a capture (Fig. 1). These approaches are appealing because they can theoretically be applied to any species where a de novo sequence assembly has been or could be generated. For instance, de novo RNA-seq transcriptomes or expressed sequence tag (EST) data may be used to design probes corresponding to exonic regions (Bi et al. 2012; Salmon et al. 2012; Bi et al. 2013; Hebert et al. 2013; Neves et al. 2013; Müller et al. 2014; Good et al. 2015). A minor technical challenge lies in identifying exon boundaries, because capture probes that span exon boundaries will result in low coverage of these regions and lower overall capture performance (Bi et al. 2012; Neves et al. 2013). Exon boundaries tend to be conserved and can usually be identified through comparison to annotation from fairly divergent reference genomes (Bi et al. 2012). Similarly, a low coverage WGS experiment could be used to generate a de novo assembly with probes designed to target anonymous or informative genomic regions (Linnen et al. 2013). Low coverage WGS data may also be aligned to a closely related reference genome to identify conserved exonic regions for capture. Shotgun WGS may be less tractable for species with large, low complexity genomes (see Box 2), though repetitive regions can be masked to avoid non-specific capture. As the costs of sequencing continue to drop, WGS will likely become an increasingly attractive alternative for capture design.

Box 2. Genome and locus-specific characteristics.

Intrinsic qualities of study organisms, such as genome size and complexity, should weigh heavily in decisions on sequencing strategy. Small, compact genomes present relatively few challenges to WGS (Davey et al. 2011; Riberio et al. 2012), while large genomes characterized by a high proportion repetitive elements (e.g., many eukaryotic genomes) may be more amenable to reduced representation sequencing approaches (Hirsch et al. 2014; Sims et al. 2014). However, targeted capture of repetitive sequences can result in substantial off-target capture and interfere with efficient on-target capture (Mercer et al. 2014; Saintenac et al. 2014). Approaches have been developed to de-enrichment for repetitive sequences, which may help to alleviate these problems (e.g., Fu et al. 2010); regardless, for many studies it is preferable to avoid these regions in favor of functional elements within unique genomic sequences. To this end, targeted capture has become synonymous with exome capture and has been used extensively to characterize protein-coding variation in plants species with large polyploid genomes and high transposable element abundance (Saintenac et al. 2011; Winfield et al. 2012; Zhou & Holliday 2012; Uitdewilligen et al. 2013; Neves et al. 2014). These studies demonstrate the utility of targeted capture to generate genome-wide datasets in species that have previously been intractable for WGS efforts.

Although targeted capture has primarily been applied to protein-coding exons (Hodges et al. 2007), regulatory regions such as UTR, miRNA, promoter regions, and other non-coding elements can also be targeted in a highly parallel fashion (Bainbridge et al. 2011; Hebert et al. 2013; Linnen et al. 2013; Carneiro et al. 2014a; Carneiro et al. 2014b; Schiessl et al. 2014). In addition, non-coding regions flanking exons and other baited regions are often indirectly targeted (Samuels et al. 2013). For instance, often regions up to 250 bp from baited sequences are recovered in targeted capture studies, albeit generally at lower coverage (Yi et al. 2010; Chevalier et al. 2014). The extent of flanking coverage is simply a function of genomic library insert size. The performance of targeted capture on non-coding regions has not been explored in detail, although some studies indicate reduced capture efficiency (Bainbridge et al. 2011; Vallender 2011). Bainbridge et al. (2011) found that both conserved UTR and other regulatory regions tended to have substantially lower coverage depth compared to coding DNA sequence (CDS), which may be correlated with GC-content (see inset Figure following Bainbridge et al. 2011). Across different capture studies, loci with intermediate GC-content are often most highly represented (45%, Tewhey et al. 2009a; 53%, Saintenac et al. 2011; 55%, Ávila-Arcos et al. 2011) because these loci amplify well and hybridize most efficiently to probes (Sims et al. 2014). Introns and intergenic regions often contain a higher frequency of repetitive elements (Tewhey et al. 2009a), however non-coding capture probes can avoid annotated repetitive DNA (Carneiro et al. 2014a). An increasing number of studies are including non-coding regions in their capture designs, however the potential costs associated with capturing these regions should be considered when designing a capture.

Finally, the uniformity in coverage across targeted loci may be leveraged to infer copy number variation (CNV) based on relative coverage (Saintenac et al. 2011; Schiessl et al. 2014; Jiang et al. 2015). The utility of targeted capture in this respect is a tremendous advantage because structural variation (e.g., whole genome or gene duplications and chromosomal rearrangements) can play an important role in adaptation and speciation (Lowry & Willis 2010; Kondrashov 2012), yet identifying and characterizing structural variation is often difficult. However, stochastic variation in coverage due to biases introduced during library preparation or hybridization can lead to incorrect CNV inferences (Sims et al. 2014), and care should be taken to correct for these issues.

Figure.

The performance of targeted capture across various genomic regions in humans. Depicted is the percent coverage relative to coding DNA sequence (CDS) as a function of deviation from GC-content of CDS (55.9%) (adapted from Bainbridge et al. 2011).

A capture could also be developed to target a panel of informative RAD markers (RAD-tags) identified from an initial RAD-seq experiment. This two-step approach, sometimes referred to as Rapture (M. Miller, pers. comm.), combines the power and ease of RAD-seq SNP discovery with the more robust technical performance of capture (i.e., higher repeatability and lower variance among samples) thereby enabling efficient and cost-effective generation of large population datasets. It is worth noting that several alternative technologies have also been developed for high-throughput genotyping of pre-defined genomic regions (e.g., see Fluidigm Access Array and other microfluidic approaches; Tewhey et al. 2009b). The details of these alternatives are beyond the scope of this review, but the choice to use capture-based resequencing versus an alternative genotyping approach usually depends on inherent cost tradeoffs between sample throughput and the total number of markers that can be efficiently assayed.

Capture approaches based on de novo sequence assemblies require performing two separate NGS experiments, decreasing cost-effectiveness (see Table 1). However, these initial experiments may be performed on a limited set of samples. For instance, a reference transcriptome can be constructed from a set of tissues from a single individual and then used to capture orthologous sequences from species across moderate evolutionary scales (Bi et al. 2012). Sequencing pools of individuals may also be preferable to identify the most informative or interesting genomic regions to sequence at a fraction of the cost of sequencing individual libraries (Schlötterer et al. 2014). However, capture approaches that first require de novo SNP discovery in a small subsample may suffer from ascertainment biases, including dropout of rare SNPs, which may affect downstream population genetic inferences (e.g., estimation of site frequency spectrum and related statistics; Clark et al. 2005; Lachance & Tishkoff 2013; McTavish & Hillis 2015). Steps may be taken to mitigate ascertainment biases by including as many samples as possible in the initial SNP discovery phase, correcting for SNP detection probabilities, or incorporating SNP ascertainment bias in population genetic analyses (Clark et al. 2005; Lachange & Tishkoff 2013).

A final approach is to design target probes based on a divergent reference genome (Cosart et al. 2011; George et al. 2011; Saintenac et al. 2011; Jin et al. 2012; Nadeau et al. 2012; Good et al. 2013; Hancock-Hanser et al. 2013; Hedtke et al. 2013; Li et al. 2013; Ilves & Lopez-Fernandez 2014) or a multisequence alignment (e.g., ultraconserved element capture; Faircloth et al. 2012; Lemmon et al. 2012; McCormack et al. 2012; Faircloth et al. 2013; McCormack et al. 2013a; Crawford et al. 2015; Leaché et al. 2015). The principle behind this strategy is straightforward: orthologous regions of the genome between divergent species can be captured if their sequences are sufficiently conserved. An appealing aspect of this approach is that annotated genomes can be exploited to interpret capture data from non-reference species without relying upon the biased representation inherent to transcriptomes. Furthermore, this approach is highly scalable and allows for capture of a handful of genome regions or even whole genomes, such as in metagenomic applications (Bos et al. 2011, Table 1). Unlike other non-reference capture approaches, divergent reference capture incurs no cost until the actual synthesis of probes because target sequence identification relies on pre-established genomic resources. Nonetheless, quantifying capture performance relative to reference divergence is critical for understanding the utility of divergent reference-based capture.

Several studies have examined the ability of probes designed from a single reference genome to capture DNA from a range of species (Vallender 2011; Jin et al. 2012; Lemmon et al. 2012; Good et al. 2013; Hancock-Hanser et al. 2013; Hedtke et al. 2013). For instance, commercially available human exome kits perform equivalently in humans and chimpanzees (~90–91% capture sensitivity) yet decline in sensitivity and depth of coverage when applied to species with >5% sequence divergence, such as macaques (Vallender 2011; Jin et al. 2012). A capture based on de novo transcriptomes from the alpine chipmunk (Tamias alpinus) was used to capture T. alpinus and three other species of chipmunks ranging up to 1.5% sequence divergence (Bi et al. 2012; 2013; Good et al. 2015). In these experiments, capture specificity and capture sensitivity were similar across all species. This same experiment included probes targeting anonymous regions from the genome of a more divergent species of squirrel (Ictidomys tridecemlineatus; ~9% sequence divergence). These divergent targets showed declines in normalized coverage in chipmunk species, consistent with the above primate studies, but high levels of overall target recovery nonetheless (~90% capture sensitivity). Hedtke et al. (2013) designed a targeted exon capture from the western clawed frog (Xenopus tropicalis) reference genome and applied it to 16 frog species spanning ~250 million years of divergence (up to ~10% sequence divergence) from the reference. Fold enrichment and capture sensitivity were strongly associated with species divergence time and loci with higher sequence divergence performed most poorly (Hedtke et al. 2013). Capture sensitivity for species >200 million years diverged was also highly variable (range 0% to ~100%), but generally less than 60%. These studies demonstrate that capture success does not seem to be compromised between closely related species but it declines abruptly at moderate levels of divergence (5–9%).

When should targeted capture be used?

No single NGS approach is ideal for all research applications. Rather, sequencing strategies should be tailored to specific research aims (Davey et al. 2011; Good 2011; Ekblom & Galindo 2011). Thus, the benefits and limits of sequencing approaches are best discussed in the context of a specific research aim. Below we highlight several active areas of evolutionary and ecological research that may benefit from targeted capture approaches, often when used in tandem with other genome partitioning approaches.

Genetic mapping of phenotypic traits

Genetic mapping remains one of the primary approaches to identify the genetic mechanisms underlying phenotypic evolution (Lynch & Walsh 1998; Stinchcombe & Hoekstra 2008). The development of NGS-based genotyping has led to tremendous advances in mapping quantitative trait loci (QTL; Baird et al. 2008; Andolfatto et al. 2011; Baxter et al. 2011; Martin et al. 2012; Weber et al. 2013; Henning et al. 2014). QTL mapping through crossing experiments or association/admixture studies requires a genetic map of the locations and relative positions of markers in the genome. To this end, RAD-seq and related approaches provide simple and inexpensive means to generate a dense map of anonymous markers (i.e., RAD-tags; Baird et al. 2008; Narum et al. 2013). However, mapping studies rarely yield single gene or marker resolution. In species with annotated reference genomes, RAD-tags can be aligned to the genome sequence to facilitate identification of candidate functional elements associated with traits (Fig. 2; Andolfatto et al. 2011). In contrast, it is very difficult to assess the proximity of QTL-associated RAD-tags to candidate genes in non-reference species (Fig. 2; Mascher & Stein 2014). In these instances targeted capture can be used to anchor anonymous genetic maps to known genes (or other functional elements), thereby informing the detection of causative loci (Fig. 2).

Figure 2.

The process of identifying causal loci underlying a phenotypic trait using different genetic mapping and sequencing strategies. With an available reference genome, RAD-seq is used to create a high-density linkage map that can be aligned to the genome to anchor RAD-tags (black ticks) to a physical map (genes are shown as red ticks and the causal locus is shown as a blue tick). To precisely localize candidate loci, initial genetic mapping can be followed by fine-scale RAD-seq mapping. Candidate loci found within the fine-mapped region can be assayed to verify their phenotypic function. Genetic mapping with RAD-seq is less powerful without a reference genome because of the difficulty identifying candidate loci. In lieu of a reference genome, high-resolution mapping with RAD-tags can be combined with low-resolution anchoring of known genes to identify candidate genes associated with a QTL and permit follow-up functional assays. High-resolution mapping with highly multiplexed exome capture genotyping could also directly reveal candidate genes to functionally test.

Several previous studies have demonstrated the power of genetic mapping with a combination of anonymous markers and a priori candidate genes (e.g., Brown et al. 2003; Steiner et al. 2007; Martin et al. 2012). Targeted capture can be used to quickly anchor RAD-tag maps to hundreds of candidate genes, providing a powerful surrogate for a whole genome reference (Baxter et al. 2011; Mascher et al. 2014; Neves et al. 2014; Fig. 2). Anchoring a RAD-based genetic map could be achieved though genotyping a relatively small subset of a larger QTL mapping panel with exon capture (i.e., low-resolution anchoring, see Fig. 2). This would still require performing exon capture on a large number of individuals, which is cost-prohibitive if only one or a few samples are captured at a time. However, standard capture platforms have the capacity for extensive custom multiplexing of individuals (Kircher et al. 2012; Rohland & Reich 2012), greatly reducing the cost of anchoring a map. Indeed, constructing high-density anchored linkage maps with only exome genotyping would also be a feasible, albeit more expensive, option that could provide many of the benefits of RAD-tags (i.e., high marker density and rapid genotyping capability; Fig. 2).

Exome data can also simplify additional fine-scale mapping to identify actual causal variants associated with phenotypic traits (Linnen et al. 2013; Chevalier et al. 2014). In contrast to anonymous RAD-tags that may lie in repetitive regions far from genes, causal variants underlying QTL may often be linked to gene coding sequences or their proximal regulatory sequences (Chevalier et al. 2014). Thus, exome genotyping used in isolation or in combination with genotyping of anonymous regions can provide a powerful and efficient means of determining which genomic regions are associated with a trait of interest (Linnen et al. 2013). To date, few studies have used targeted capture for genetic mapping of phenotypic traits outside of human disease (although see del Viso et al. 2012, Linnen et al. 2013, Tennessen et al. 2013, and Chevalier et al. 2014). Given the success of targeted capture in mapping genetic variants associated with human disease (Bamshad et al. 2011), we anticipate the capture-assisted QTL mapping of ecologically important phenotypic traits will emerge as a parallel utility.

Detecting selection in the genome

The relative importance of natural selection as a force governing evolutionary change has been a continuous source of debate amongst biologists (Fisher 1930; Wright 1932; Ford 1964; Lynch 2007). Genome technologies have finally begun to allow us to understand the frequency, mode, and distribution of selection across the genomes of diverse organisms (Begun et al. 2007; Drosophila 12 Genomes Consortium 2007; Jensen et al. 2008; Kosiol et al. 2008; Hohenlohe et al. 2010; Rubin et al. 2010; Hernandez et al. 2011; Lohmueller et al. 2011), providing important insights into this debate. However, the massive amounts of data required to assess broad-scale patterns of selection has limited the above studies to taxa with well-developed reference genomes. As a result, we still know very little about how selection shapes genomic variation across that majority of life. To advance our understanding of the evolutionary processes shaping genetic diversity it is imperative to extend these investigations to diverse natural populations. Targeted capture provides several benefits over alternative sequencing strategies to achieve this goal.

Quantifying molecular evolution using genome-wide patterns of protein divergence between species (d_N/d_S) is one powerful approach to detect selection that has been mostly restricted to species with reference genomes (e.g., Bazykin et al. 2004; Nielsen et al. 2005; Drosophila 12 Genomes Consortium 2007; Kosiol et al. 2008). Tests for selection based on protein divergence are appealing because they rely upon the well-established foundations of molecular evolution, are relatively robust to confounding demographic effects (Nielsen 2001), and enable functional inferences regarding the targets of selection. Targeted exome capture offers a highly cost-effective approach to extend these tests beyond classic model systems to a broader range of taxa (Burbano et al. 2010; Aagard et al. 2013; Vilstrup et al. 2013; George et al. 2011; Good et al. 2013). For instance, George et al. (2011) performed an exome capture of several Old and New World monkeys and revealed novel targets of positive selection on genes associated with keratinization or the conversion of epithelial cells to keratin. Moreover, comparative exome data can be used to move beyond the post hoc inferences provided by standard genome-wide scans for selection towards a priori hypothesis testing of specific genetic pathways or classes of genes (e.g., Smadja et al. 2012; Nadeau et al. 2012; Aagaard et al. 2013; Good et al. 2013). Thus, in addition to elucidating the frequency and mode of selection, comparative molecular evolutionary studies of targeted exome data can provide essential information on the ecological or life history drivers of molecular evolution.

Tests of selection using species divergence do not rely on population sampling within species; however, population genetic approaches to infer selection based on patterns of polymorphism within species (e.g., F_ST, site-frequency spectrum, linkage disequilibrium) or polymorphism compared to divergence between species (e.g., HKA test, McDonald-Kreitman test) require sampling many individuals to accurately estimate allele frequencies (Li et al. 2010; Hvilsom et al. 2012; Tennessen et al. 2012; Fu et al. 2013). While this level of population sampling is often not feasible with WGS or RNA-seq, exome capture provides a powerful platform to collect population-level genomic data. Targeted capture has frequently been used to estimate the efficacy of selection in populations (Hvilson et al. 2012; Veeramah et al. 2014; Bataillon et al. 2015) and to infer local adaptation by identifying highly differentiated loci between populations (Yi et al. 2010; Nadeau et al. 2012; Smadja et al. 2012; Hebert et al. 2013; Müller et al. 2014) or signatures of selective sweeps (Aagaard et al. 2013; Linnen et al. 2013; Tsangaras et al. 2014). The potential power of capture-based population genomic studies is exemplified by Yi et al. (2010) who used whole exome capture to investigate the genetic basis of high altitude adaptation in Tibetans. They found strong evidence that a transcription factor, EPAS1, involved in red blood cell production was involved in local adaptation to hypoxia based on patterns of differentiation between low elevation Han Chinese and high elevation Tibetan populations.

Finally, targeted capture is easily scalable from hundreds (Smadja et al. 2012) to hundreds of thousands of genomic regions (Hodges et al. 2007; Yi et al. 2010) and can include both coding and linked non-coding regions (Bainbridge et al. 2011; Hebert et al. 2013; Carneiro et al. 2014a; Carneiro et al. 2014b; Schiessl et al. 2014). Non-coding capture data sets (e.g., introns, intergenic regions) can serve as appropriate null models of genomic variation for inferring selective sweeps within the genome (e.g., Carneiro et al. 2014b) or regions of elevated differentiation between species (e.g., Carneiro et al. 2014a). Alternatively, UTRs and conserved non-coding regions often show strong signatures of adaptive evolution (Halligan et al. 2013; Carneiro et al. 2014b) and may be targeted to investigate the history of selection within these regions. Sequencing long contiguous regions of exons, flanking introns, up- or downstream regulatory sequence or even large tracts on non-coding intergenic regions generally provides more power to detect selective sweeps (Thornton & Jensen 2007) and to estimate the overall strength and genomic rate of selective sweeps (Jensen et al. 2008) when compared to sequencing shorter regions. Furthermore, the lower coverage variance and general high quality of targeted capture data offers several advantages over alternative approaches. For example, RAD-seq data can suffer from numerous data quality issues, such as allelic bias or dropout, which can impact population genetic inferences (Gautier et al. 2013; Henning et al. 2014). Here the increased data quality and the dynamic flexibility of targeted capture to scale both number and size of targets can provide a decided advantage over RAD-seq and other genome partitioning approaches.

Phylogenetics

Until recently, NGS-based genomics has arguably had a less profound impact on phylogenetics compared to other fields of biology (McCormack et al. 2013b; Jarvis et al. 2014; Mirarab et al. 2014; Misof et al. 2014). In addition to the computational challenges of inferring phylogenies from millions of DNA base pairs, certain phylogenetic problems are not easily resolved with the addition of more sequence data (Philippe et al. 2011). One way to ameliorate these issues is to sequence a tailored subset of informative orthologous loci to maximize phylogenetic signal while minimizing non-phylogenetic signal and computational effort (Philippe et al. 2005; Philippe et al. 2011). Here, the flexibility of targeted capture offers a tremendous advantage. Customized capture designs can target hundreds or thousands of loci (Faircloth et al. 2012; Mandel et al. 2014), slowly or quickly evolving loci (Mason et al. 2011; Lemmon & Lemmon 2013; McCormack et al. 2013a), and nuclear or organelle loci (Hedtke et al. 2013; Vilstrup et al. 2013; Ilves & Lopez-Fernandez 2014) to resolve an array of phylogenetic questions. As such, phylogeneticists are increasingly turning to targeted capture approaches to resolve their questions.

To maximize phylogenetic information, it is important to select loci with evolutionary rates appropriate to resolve a given relationship (Philippe et al. 2011). For instance, deep phylogenetic nodes are resolved with slowly evolving sequences that retain signals of orthology across distant taxa (McCormack et al. 2012). Genomic data from restriction enzyme-based approaches are beginning to be used for phylogenetic inference mostly below the genus level (e.g., Emerson et al. 2010; Eaton & Ree 2013). However, these data seem to be unreliable beyond moderate phylogenetic depth because restriction cut sites are often not conserved across taxa, leading to high levels of missing data and incorrect topologies (Rubin et al. 2012; Wagner et al. 2013; Leaché et al. 2015). High throughput targeted capture of slowly evolving ultraconserved element (UCE) markers has emerged as a very promising method to resolve deep nodes in the vertebrate phylogeny (Faircloth et al. 2012; Lemmon et al. 2012; McCormack et al. 2012; Faircloth et al. 2013; McCormack et al. 2013a; Crawford et al. 2015; Leaché et al. 2015). Capture of UCEs is appealing because one core set of loci can be identified from highly divergent reference genomes and then broadly applied to diverse taxa, enhancing comparisons among datasets while obviating the need to continually re-design custom probes. This method also should work across more shallow evolutionary time scales because UCEs flank more variable regions (Faircloth et al. 2012) that can provide resolution of more recent splitting events (Smith et al. 2014).

One looming issue with UCE markers is the extent to which selection might impact their application to various evolutionary questions. By design, UCEs are expected to experience among the strongest levels of purifying selection in the genome. While most of the phylogenetic signal from UCE regions comes from less-constrained sites linked to UCEs, these linked regions should nevertheless be strongly influenced by background selection (Charlesworth et al. 1993; 1995; Hahn 2008; Charlesworth 2012; Halligan et al. 2013). Thus, UCE markers are expected to have small effective population sizes and highly skewed allele frequencies relative to other genomic regions. It is important to note that patterns of genetic variation near protein-coding regions are also likely to be strongly influenced by background selection (Halligan et al. 2013), though the context dependence is better understood. These issues might not matter for some phylogenetic questions, but background selection does have profound impacts on most inferences that depend on patterns of population-level variation (Hammer et al. 2010; Hahn 2008). Thus, the validity of using UCE-linked markers to address population genetic questions should be considered carefully (e.g., coalescent history, effective population size, gene flow, phylogeography, etc.).

Protein-coding sequence can also be used for phylogenetic reconstruction at moderate to deep evolutionary scales due to their increased conservation relative to most non-coding regions (e.g., ~20% of mammalian protein-coding bases are evolutionarily constrained compared to ~4% genome-wide; Lindblad-Toh et al. 2011). Protein-coding sequences may be more appropriate than UCEs for organisms such as plants with large, complex, and repetitive genomes where UCE identification may be difficult (Mandel et al. 2014). Phylogenetic studies in frogs (Hedtke et al. 2013), cichlids (Ilves & López-Fernández 2014), and flowering plants (Mandel et al. 2014) have used protein-coding regions to show the power of these datasets for resolving complex phylogenetic histories.

Ancient DNA and metagenomic applications

Ancient or historic DNA (hereafter aDNA) extracted from paleontological remains, sediments, and museum specimens can provide a detailed understanding of the evolutionary history of lineages (Krings et al. 1997; Willerslev & Cooper 2005; Briggs et al. 2009; Green et al. 2010; Stoneking & Krause 2011; Orlando et al. 2013). Ancient DNA studies were traditionally hindered by limitations of PCR and DNA sequencing (Fig. 3). For instance, the short and highly damaged fragments that characterize aDNA often impair traditional PCR amplification (Knapp & Hofreiter 2010; Rowe et al. 2011; Tin et al. 2014). However, aDNA samples are highly amenable to many NGS approaches where library preparation requires shearing genomic DNA to the size range generally observed in aDNA (~40–500 bp, Pääbo 1989; Green et al. 2008; Sawyer et al. 2012; Fig. 3). As such, the analysis of aDNA has been greatly enabled by NGS (Green et al. 2010; Krause et al. 2010; Rasmussen et al. 2010; Meyer et al. 2012).

Figure 3.

The relative suitability of different NGS approaches for nucleic acid sources of varying levels of degradation (ranging from fresh samples to ancient samples) and non-target contamination (ranging from no contamination to extremely low levels of target sequence). Darker colored parts of the bars indicate a higher suitability of the approach whereas light colors indicate a low suitability.

An enduring challenge of working with aDNA samples from paleontological samples is to isolate sequences of interest from a complex blend of bacterial DNA and target DNA (Green et al. 2006; Àvila-Arcos et al. 2011; Carpenter et al. 2013). For example, remains of ~55,000 year old bone fragments used to generate the Neandertal reference genome were comprised of ~99% contaminant environmental DNA (Green et al. 2010). Whole genome shotgun sequencing of these types of aDNA is highly inefficient (Fig. 3). To overcome this issue, ancient hominin researchers have used capture to sequence targets ranging from complete mtDNA genomes (Briggs et al. 2009; Krause et al. 2010) to thousands of exons (Burbano et al. 2010; Castellano et al. 2014). By designing probes that capture specific target sequences, the problem of sequencing large amount of contaminant DNA is greatly diminished. These studies were some of the first to employ targeted capture outside of biomedical research and targeted capture has continued to expand across aDNA research (e.g., Àvila-Arcos et al. 2011; Bos et al. 2011; Schuenemann et al. 2011; Carpenter et al. 2013; Castellano et al. 2014; Enk et al. 2014).

Natural history museum specimens provide a vast repository of biological data that suffer from largely the same problems that afflict aDNA from paleontological remains (e.g., DNA degradation and, to a lesser extent, contamination). However, the utility of natural history collections for targeted capture investigations is just beginning to be realized (Mason et al. 2011; Bi et al. 2013; Hedtke et al. 2013; Vilstrup et al. 2013; Tsangaras et al. 2014). Bi et al. (2013) provided an exciting advance in genomic analysis of museum samples by capturing ~12,000 exons from forty alpine chipmunk (Tamias alpinus) specimens, including 20 contemporary samples and 20 collected in 1915. Exome data from the historic samples were compared to modern samples to infer recent shifts in genetic diversity and population structure coincident with range restrictions associated with climate change. This study provides a glimpse into the potential role of targeted capture in expanding natural history museum collections for genomics research (Nachman 2013). Given the vast number of specimens currently archived in museums, efficient incorporation of museum collections into genomics represents one of the most promising, but as of yet underutilized, applications of targeted capture.

A technically related issue is the isolation and sequencing of DNA from an environmental sample (metagenomics) to study pathogen evolution or microbial community composition (e.g., microbiome research). In particular, targeted capture has greatly enabled the recovery of pathogen sequences from host tissues (Bos et al. 2011; Kent et al. 2011; Schuenemann et al. 2011; Geniez et al. 2012; Bent et al. 2013; Wagner et al. 2014; Bos et al. 2015). For instance, Bos et al. (2011) used probes from modern strains of the pathogenic bacterium Yersinia pestis to reconstruct whole genome sequences from ancient Y. pestis strains found within the teeth of Black Death victims from 1348–1350. They found that the ancient Y. pestis strain responsible for the Black Death gave rise to all modern Y. pestis lineages. Later, this approach was used to demonstrate that the Plague of Justinian (541–543 AD) was likely caused by an independent emergence of Y. pestis in humans from rodent vectors (Wagner et al. 2014).

A central goal of community genomics is to characterize species diversity within a community. While shotgun-sequencing approaches have provided detailed profiles of microbial genetic diversity in environmental samples (Breitbart et al. 2002; Tyson et al. 2004), WGS is still difficult and expensive when compared to traditional methods of 16S rRNA sequencing (Hugenholz et al. 1998). The use of one or a few DNA markers for species identification is limited in many ways but remains a common approach in metagenomics (Valentini et al. 2009; Portik et al. 2012). Targeted capture can provide a high-throughput approach to analyze community composition using species barcodes and to explore functional variation in a community (e.g., by targeting genes underlying specific ecological functions; Denonfoux et al. 2013). Gut microbiomes constitute one type of community that has gained substantial interest recently for its potential role governing health and contributing to broader ecological and evolutionary patterns (Muegge et al. 2011; The Human Microbiome Project Consortium 2012). Sufficient sequence data are accumulating for the human microbiome to facilitate targeted capture of large portions of the diversity that comprise this community. These general principles can be extended to any type of species detection, including non-invasive sampling (Perry et al. 2010; Kidd et al. 2014). While relatively few studies have delved into the use of targeted capture for species detection purposes, this also represents a promising avenue of future research. Indeed, genetic surveys of environmental DNA (eDNA) are emerging as a powerful tool for species detection and biodiversity monitoring (Schnell et al. 2012; Wilcox et al. 2013; Bohmann et al. 2014). Thus far eDNA studies have predominantly relied upon PCR-based approaches (Bohmann et al. 2014). However, further development of targeted capture for eDNA applications seems inevitable given the strong parallels in technical challenges presented by eDNA and aDNA.

Is target capture a fleeting method?

Targeted capture holds tremendous potential to advance evolutionary and ecological research (Fig. 4). Genome partitioning approaches predominate many evolutionary studies simply because WGS remains prohibitively expensive or difficult for problems that require large sample sizes and for species with large genomes (although see Nystedt et al. 2013 and Neale et al. 2014). But it would be naïve to assume that many of the current limitations of WGS will persist into the future. NGS costs continue to plummet, while sequencing technologies and assembly methods continue to evolve. The difficulties associated whole genome assembly are quickly being overcome with technologies that generate long reads spanning repetitive regions (Treangen & Salzberg 2012; Huddleston et al. 2014; Goodwin et al. 2015) and with powerful new assembly approaches (Putnam et al. 2015). Given this, it is reasonable to speculate that targeted capture and other genome partitioning approaches will soon become obsolete (Ku et al. 2012). What roles might targeted capture play, if any, in genomic studies once WGS becomes universally accessible? Are there intrinsic advantages to targeted capture over WGS?

Figure 4.

The relative suitability of targeted capture (purple), RAD-seq (blue), and transcriptomic approaches (green) for genetic mapping of phenotypic traits, population genetics (includes inferring population history and detecting population level signatures of selection), molecular evolution (e.g., rates of protein evolution), phylogenetics, ancient DNA, and metagenomics. Specific height of each bar is arbitrary but increasing height corresponds to increasing suitability.

Many evolutionary and ecological questions simply do not require whole genome sequences (Davey et al. 2011). Even as WGS becomes economically feasible, genome partitioning approaches will remain preferred if they are cheaper for broader sampling or the data are easier to analyze. The inherent tradeoff between sample size and genome coverage means that, regardless of sequencing costs, exome capture will always enable sequencing larger samples of individuals than WGS given the same sequencing effort. The latter strategy may be favored or required for addressing detailed population genetic questions, identifying rare variants associated with phenotypes (Tennessen et al. 2012), or identifying epistatic interactions among genes (Wei et al. 2014). Eventually, however, the costs of WGS and targeted capture may drop to the point where the large differences in throughput between the approaches outweigh minor differences in cost. Regardless, targeted capture approaches are likely to remain well suited for certain questions. In phylogenetic contexts targeted sequencing of loci with desired rates of evolution may often lead to better resolved topologies than whole genome data. Targeted capture also offers the ability to isolate specific sequences of interest from a blend of DNA of many organisms, which holds tremendous utility for aDNA research, metagenomics, and host-parasite or pathogen research. While we are careful to not disregard the importance of non-coding DNA (e.g., transposable elements) in evolution, for many questions we are specifically interested in coding variation and their regulatory regions and thus not the vast majority of the genome.

Ultimately, the increasing feasibility of WGS will enable rather than nullify targeted capture approaches (Teer & Mullikin 2010). We expect targeted capture and other genome partitioning approaches to remain a vital tool in evolutionary and ecological research in the near future. Our genomics tool set continues to diversify and we no longer have to rely on a single approach for all research questions. Rather, we are afforded the opportunity to customize methods to suit our specific research. Targeted capture is only one method to choose from, but given the flexibility of this approach we believe its broad implementation would benefit many evolutionary and ecological disciplines.

Box 1. Definitions.

aDNA

ancient DNA, DNA isolated from ancient or historic sources including paleontological remains, fossil remains, sediment or ice cores, and museum specimens.

Array capture

targeted capture using probes fixed on a glass slide.

Bait/probe

Single-stranded oligonucleotides with a complementary sequence to a portion of a targeted genomic interval.

Capture sensitivity

the percentage of targets covered by at least one mapped read.

Capture specificity

the percentage of unique reads mapping to target sequences.

eDNA

environmental DNA, DNA isolated from an environmental sample (e.g., water, sediment, gut) containing a blend of DNA sequences from many organisms.

Exome

the portion of the genome coding for exons.

Exome capture

a targeted capture of the exome or component of the exome (e.g., often protein-coding exons).

Fold enrichment

the fold-increase in the number of reads covering targeted regions over the expected number from a shotgun whole genome approach.

Hybridization

the step of targeted capture where single-stranded genomic sequences bind to complementary probe sequences for enrichment.

In-solution capture

targeted capture using biotinylated probes that allow hybridization to occur in a liquid solution. Biotin molecules binds with high affinity to streptavidin beads, which allows target sequences bound to probes to be isolated from non-target sequences.

Non-reference species

a species lacking a reference genome.

UCE

ultraconserved element, a region in the genome that shows extraordinary sequence conservation across deeply divergent taxa, likely as a result of strong purifying selection.

Box 3. Genomic samples.

The quality and quantity of available tissue or nucleic acid samples can be a major limiting factor in any genetic study. Low quality nucleic acid sources (e.g., aDNA and eDNA) often preclude transcriptome approaches because of the rapid rates of mRNA degradation (Sharova et al. 2009; Fig. 3) and changes in transcript profiles as a function of the time lag between tissue collection and preservation (Sanoudou et al. 2004; Birdsill et al. 2010). Post-mortem enzymatic, hydrolytic, and oxidative reactions degrade and damage genomic DNA over time (Pääbo et al. 2004; Briggs et al. 2007), but at a much slower rate compared to RNA. Whole genome sequencing, targeted capture, and restriction-enzyme digest approaches have all been applied to degraded genomic samples (Green et al. 2010; Meyer et al. 2012; Bi et al. 2013; Tin et al. 2014), although the extent of DNA degradation may dictate which sequencing approaches are most appropriate. For instance, additional DNA digestion with restriction enzymes may produce fragments that are too short for analysis and lead to extensive data loss (Tin et al. 2014, Burrell et al. 2015; Fig. 3). Shotgun sequencing and targeted capture can be applied to degraded DNA without additional fragmentation, although the choice of which of these strategies to pursue with degraded DNA depends largely on the study organism, the amount of sample, and the study question (Burrell et al. 2015).