The use of museum samples for large-scale sequence capture: a study of congeneric horseshoe bats (family Rhinolophidae)

Introduction

Museum collections have long served as repositories for physical biological data in the form of specimens and tissue samples. Some museums have amassed material from across the globe with the intention of documenting the existence of all unique forms of life. As a consequence, collections frequently house many poorly known, rare, endangered, and extinct taxa, some of which may not have been seen again subsequent to their initial collection (Sinha, 1973; Novotny & Basset, 2000; Thessen, Patterson & Murray, 2012).

Aside from the importance of museum collections for research and education, they may also represent immensely rich genetic resources. For many years, researchers have incorporated sequences from museum specimens into studies of systematics, evolution, and conservation (Carstens & Knowles, 2007; Paijmans, Gilbert & Hofreiter, 2013; Wiley et al., 2013). More recently, several larger-scale campaigns have been launched to genetically index specimens held in museums (Schindel et al., 2011; Lim, 2012; Brugler et al., 2014). However, many of the traditional collection, storage and curation methods found in museums are incompatible with contemporary genetic research. In particular, the practice of fixing material in formaldehyde or ethanol may damage or degrade nucleic acids, and thereby reduce any potential viability in genetic studies, which use traditional Sanger sequencing. Thus, the recovery of a single gene, such as the mitochondrial cytochrome oxidase 1 (COI), in museum specimens has commonly been performed (Almeida et al., 2014; Brace et al., 2014; Li & Liu, 2014). For example, the frozen tissue collections of mammals and birds held at, respectively, the Royal Ontario Museum and the National Museum of Natural History of the Smithsonian Institution, have been DNA barcoded using COI sequences (Schindel et al., 2011; Lim, 2012).

Furthermore, even when material is suitable, obtaining sufficient tissue from specimens for genetic analyses is often viewed as unacceptably destructive, with some rarer material placed ‘off-limits’ to genetic investigation. These conflicts of interest are unfortunate, not least because the systematics of many taxa, including even relatively well-described groups such as vertebrates, is frequently unclear (Guschanski et al., 2013; Tsagkogeorga et al., 2013). In such cases, the best hope of settling taxonomic disputes and standardizing nomenclature is to match curated specimens to genetic sequences.

The advent of high-throughput sequencing technologies (‘Next Generation Sequencing’) has provided new and exciting opportunities for utilizing collection-derived material (Bi et al., 2013; Nachman, 2013). In particular, the short fragment sizes, that often characterize degraded DNA isolated from museum specimens, pose less of a problem for high-throughput sequencing technologies than for traditional Sanger sequencing. This is because current sequence read lengths obtained from common platforms, such as Illumina, are relatively short (approximately 100–250 bp), whereas traditional Sanger sequencing read lengths are much longer and so are more suited to longer stretches of DNA. The second main advantage of high-throughput sequencing is the sheer volume of data generated, allowing millions of short reads to be sequenced in parallel and assembled into potentially longer ‘contigs’. Indeed, the ability to obtain very large volumes of sequence data from a small amount of starting material means that, at least in theory, high-throughput sequencing can also reduce the need to resample valuable specimens on multiple occasions for different projects. It follows that new sequencing methods allow researchers to sequence a wide range of taxa and loci, and to a much greater depth, for a fraction of the cost per base when compared to Sanger sequencing (Blencowe, Ahmad & Lee, 2009; Zhou et al., 2010; Rowe et al., 2011; Treangen & Salzberg, 2012).

Previous efforts towards high-throughput sequencing of museum specimens have been limited to either relatively few taxa (Bi et al., 2012, 2013) or to small numbers of loci (Schindel et al., 2011; Guschanski et al., 2013). However, some recent studies have applied shotgun sequencing to assemble the genomes of collection-based specimens of mammals, insects, fungi and plant species, although coverage has tended to be low (Rowe et al., 2011; Staats et al., 2013).

One application of high-throughput sequencing that appears to be especially promising for working with museum collections is so-called ‘target sequence capture’. Here, a set of focal genes are ‘pulled-down’ from a larger pool of DNA fragments through the use of sequence-specific baits. Because baits manufactured from a single taxon have been shown to recover loci across related taxa (Mason et al., 2011; Enk et al., 2014), this method is potentially very useful for phylogenomic studies (Mason et al., 2011; Enk et al., 2014). Moreover, the bait sequences themselves are short, and so this approach is also well-suited to fragmented DNA (Mason et al., 2011; Tin, Economo & Mikheyev, 2014) of the sort that is often isolated from specimens in collections.

Although a relatively new method, several studies have applied target sequence capture to study samples held in collections. Templeton et al. (2013) used baits to recover the mitogenomes from 2500-year-old human archaeological remains, and other studies have also reported successfully capturing and sequencing mitochondrial genes of other species held in collections (Hebert et al., 2003; Mason et al., 2011; Sawyer et al., 2012; Paijmans et al., 2013). Compared to nuclear DNA, mitochondrial DNA is less prone to deterioration during preservation and, as a result of its high copy number, is more readily recovered from limited material (Mason et al., 2011; Guschanski et al., 2013). Nonetheless, mitochondrial DNA also displays several characteristics that may limit its use for evolutionary inference, such as its maternal mode of inheritance, lack of recombination, and the frequent phylogenetic discordance seen between mitochondrial and nuclear gene trees (Desalle & Giddings, 1986; Cronin, 1991; Guschanski et al., 2013; Morgan, Creevey & O'Connell, 2014). Recently, sequence capture has also been used to pull-down nuclear DNA sequences of museum specimens of rodents (Bi et al., 2013), endangered primates (Guschanski et al., 2013), and extinct species of proboscideans (Enk et al., 2014).

The success of recent advances in the detection and recovery of DNA from so-called ‘ancient’ tissue (Bos et al., 2011; Fu et al., 2013; Devault et al., 2014; Meyer et al., 2014) has allowed access to the wealth of preserved specimens and tissues in museums across the world. Many similarities can be drawn between ancient tissues and those from museums and most techniques are transferable and may prove to be more effective.

In the present study, we applied a target sequence capture method to samples obtained from museum specimens of horseshoe bats (Rhinolophidae: Chiroptera) with the aim of recovering several thousand candidate genes, including both nuclear and mitochondrial genes. Horseshoe bats number approximately 80 recognized species (Csorba, Ujhelyi & Thomas, 2003), all of which belong to the single genus Rhinolophus. Despite their relative lack of overall morphological divergence, horseshoe bats can often be differentiated from each other on the basis of body size, the shape and size of their noseleaf (Csorba et al., 2003), and their peak echolocation call frequency (Kingston & Rossiter, 2004; Stoffberg, 2007; Sedlock & Weyandt, 2009; Stoffberg et al., 2010; Stoffberg, Jacobs & Matthee, 2011; Taylor et al., 2012). Horseshoe bats have long been recognized as important model organisms in auditory research (Long & Schnitzler, 1975; Fenton & Bell, 1981; Jones & Ransome, 1993; Li et al., 2007), and have also been subject of many molecular studies ranging from population genetics (Rossiter et al., 2000; Chen et al., 2009) and phylogeography (Flanders et al. 2009, 2011; Mao et al. 2010, 2013; Dool et al. 2013) to social structure (Rossiter et al., 2000; Puechmaille et al., 2014; Ward et al., 2014).

Molecular dating analyses calibrated with fossils suggest horseshoe bats arose approximately 15 Mya (Teeling et al., 2005; Foley et al., 2015) when they diverged from their sister taxon, the Hipposideridae. They subsequently radiated into a number of clades, currently recognized as approximately 15 species groups, with rapid diversification being seen over the past few million years (Csorba et al., 2003; Guillén-Servent, Francis & Ricklefs, 2003). Findings have revealed numerous new cryptic taxa present within the genus, suggesting that current numbers of species are underestimated (Kingston & Rossiter, 2004; Sun et al., 2008; Monadjem et al., 2010; Taylor et al., 2012). To date, most phylogenetic studies of horseshoe bats have focused on one or only a few loci (Guillén-Servent et al., 2003; Li et al., 2006; Stoffberg et al., 2010; Taylor et al., 2012). Recently, Foley et al. (2015) examined approximately 10 kb in 40 species, including ten Rhinolophus spp., and produced a well-resolved phylogeny for the taxa included.

To examine the utility of target sequence capture across congeneric taxa, we designed baits against coding sequences in two species of horseshoe bat, and used these to obtain new data from across the group. As a result of the scarcity of many of these species, we used material held in museum collections that encompass a range of ages and preservation methods. In particular, we were interested in determing to what extent the success of capture varied with phylogenetic relatedness between ‘bait’ and target species.

Material and methods

Tissue collection

For target sequence capture, we obtained tissue samples from ten Rhinolophus species (Table 1): Rhinolophus subrufus, Rhinolophus inops, Rhinolophus shameli, Rhinolophus bocharicus, Rhinolophus clivosus, Rhinolophus virgo, Rhinolophus malayanus, Rhinolophus macrotis, Rhinolophus imaizumi, and Rhinolophus formosae. These species cover six of the 15 recognized species groups based on Guillén-Servent et al. (2003) and were selected to cover a range of phylogenetic distances from the two sister taxa whose transciptomes were used for the bait design (Fig. 1). Briefly, the bait model species belong to the pearsoni-group (R. pearsoni and R. yunanensis), which is sister to the euryotis-group, represented here by R. subrufus, R. inops, and R. shameli. For the purposes of relating capture to phylogenetic distance, we defined these two sister groups (pearsoni and euryotis) (Fig. 1) as having a most recent common ancestor (MRCA) 11 Mya. The [pearsoni, euryotis] clade is then sister to the [megaphyllus, [pusillus, philippinensis]] clade containing three species groups, all of which share a common ancestor with the [pearsoni, euryotis] bait model clade 12 Mya. The most phylogenetically distant species groups from the bait taxa are the ferrumequinum species group (MRCA = 14 Mya) and, finally, the trifoliatus-group (MRCA = 15 Mya).

Table 1. Sample collection, preservation information and summary of sequence data

ID	Species	Species group	Most recent common ancestor	Range	Museum	Collection date	Storage media	Tissue type	DNA extraction concentration (ng μL−1)	DNA integrity number (DIN)	Number of contigs assembled	Percentage unique contigs matching genes	Number of genes with sequence coverage ≥ 90%	Number of genes recovered
190759	subrufus	euryotis	11	Philippines	FMNH	2006	DMSO	Muscle	2.71	8.5	24361	77.41	1204	3151
206207	inops	euryotis	11	Philippines	FMNH	2008	DMSO	Muscle	6.36	8.5	31212	59.72	1296	3163
23674	shameli	euryotis	11	Mainland SE Asia	HNHM	1996	96% ETOH	Muscle	59	7.3	34123	56.41	1562	3177
16417	bocharicus	ferrumequinum	14	Middle East	HNHM	1921	70% ETOH	Internal organ	2.41	NA	52357	86.55	835	3024
150066	clivosus	ferrumequinum	14	Africa	FMNH	1992	Flash frozen	Heart	16.3	8.9	32545	66.36	1344	3186
168915	virgo	megaphyllus	12	Philippines	FMNH	2000	DMSO	Muscle	3.54	8.1	35926	58.66	1411	3165
22726	malayanus	megaphyllus	12	Mainland SE Asia	HNHM	2007	96% ETOH	Muscle	46.3	8.7	85249	16.61	662	3093
202721	macrotis	philippinensis	12	Philppines	FMNH	2008	DMSO	Muscle	2.13	NA	84271	20.10	956	3141
16382	imaizumi	pusillus	12	Japan	HNHM	1996	70% ETOH	Muscle	0.4	1.9	30167	75.74	1446	3168
21237	formosae	trifoliatus	15	Taiwan	HNHM	2005	96% ETOH	Wing punch	36.9	8.6	30351	69.06	1352	3144

All specimens sampled were held at the Field Museum of Natural History, Chicago, USA (FMNH) or the Hungarian Natural History Museum, Budapest, Hungary (HNHM). The tissues selected encompass a range of conditions expected to impact on nucleic acid integrity, including collection dates that ranged from 1921 to 2007, and different storage conditions that included 70% and 96% ethanol, flash frozen, and dimethyl sulphoxide.

Quantification of bait capture success

For each species, the raw reads from both MiSeq runs were assembled into contigs using the software TRINITY r20140717 (Grabherr et al., 2011). TRINITY will only assemble contigs with a reasonable level of overlap between k-mers. All contigs were then blasted against the original bait model sequences using the same parameters as described above (Altschul et al., 1997). To quantify the number of recoverable genes from sequencing, we recorded the percentage of unique contigs constructed by TRINITY, which were successfully matched to a gene from the bait design (Table 1). Because each locus was recovered as a series of short sequences, to estimate the success of recovering whole genes, we recorded the number of total genes that exhibited ≥ 90% coverage across their entire sequence (Fig. 2).

Results

Discussion

To assess the viability of target sequence capture on museum specimens across a diverse clade, we performed target sequence capture of 3520 genes using a bait designed from RNA-Seq data from two species of horseshoe bat. In total, we were able to recover most (89.23%) of the target genes from multiple, previously unsequenced taxa seperated by up to 15 Myr (Guillén-Servent et al., 2003). We detected no clear pattern of variation in the number of genes recovered in each of the target species in relation to their phylogenetic distance to the ‘bait’ species (Fig. 2). Indeed, we were able to recover most of the genes with at least partial sequence coverage in 10 distinct species, which together span a wide range of the phylogenetic diversity in this group (Guillén-Servent et al., 2003; Stoffberg et al., 2010).

Although the DNA concentration of samples was standardized following extraction, the amount of DNA extracted from museum tissue is often indicative of its degree of fragmentation, and can thus be a proxy for the quality of the DNA. This was confirmed when running all samples on a TapeStation (Agilent). Samples that were low in concentration at extraction appeared to be fragmented into much shorter sequences compared to samples with a high DIN. The DIN is used as a measure of fragmentation, which is used to infer the quality of the DNA (Gassman & McHoull, 2014). It is perhaps important to note that the overall DNA concentration from a museum specimen is not a good correlate of capture success, again as a result of the detrimental affects of preservation (Enk, Rouillard & Poinar, 2013). Simply by omitting the shearing step for the most degraded samples, we were able to reduce any further damage and retain the samples for successful target sequence capture. Most samples used in the present study were recent collections (10–20 years); however, one sample (R. bocharicus) was collected in 1921. Although the date of collection did appear to have some bearing on the number of genes recovered based on results of R. bocharicus (Fig. 2), a wider spread of collection date would be needed to verify this trend. Because R. bocharicus was also stored in a low quality preservation medium, this may also be the reason for its partially lower recovery of genes.

Although some tissues preserved in low quality storage media did recover numbers of genes comparable to high quality storage methods, low quality media appears to be highly variable in its ability. To corroborate this, further testing of high quality storage media would be desirable as a result of the low sample number in the present study. Sample storage media for the tissues used in the present study are also conflated with the date of specimen collection, with older samples tending to be stored in poorer quality media. Future work would ideally look at capture success across a larger sample of older specimens and more recent high quality media preserved specimens.

We used a range of tissues in the present study; however, DNA can be recovered from internal organs (heart, liver, etc.) with favourable results, minimizing the external destructive sampling of specimens and making use of tissues often destroyed in the preservation process (e.g. skin preparation). Even very old tissues (R. bocharicus) and tissues with low quantity and quality (fragmented) DNA (R. imaizumi) were able to be sequenced with comparable success to tissues collected within the past 10 years (R. malayanus) and those flash frozen in the field (R. clivosus), with this latter field preservation method being the most favourable for recovery of nucleic acids.

By contrast to the mean number of genes recovered with high coverage (1207 genes), there was some variation among species and individuals. As a consequence, the number of genes recovered in all 10 species was relatively low (449 genes with ≥ 90% sequence coverage). Apart from the factors considered (e.g. storage media and phylogenetic distance), there are others that may affect the capture success inherent to the short fragment length of Illumina sequencing. Multiple filtering stages and overly short sequence length for contig recovery and BLAST results can negatively affect gene sequence recovery. Minor adjustments to blast parameters, such as sequence length or expected value, may mean important portions of genes are not ‘recovered’ for the final analysis (Altschul et al., 1997; Moreno-Hagelsieb & Latimer, 2008).

The overall success of the capture bait method reinforces its support as a robust technique for use with museum specimens from related species, in this case separated by up to 15 Myr. As is demonstrated by the general lack of distinct patterns of correlation between all measurements taken, there appears to be little to support any single factor affecting the success of bait capture in museum specimens. We did not find any significant correlation with the final number of target genes recovered by the baits. It should be noted, however, that, in the present study, we were unable to account for many external factors, such as the time between collection and preservation, number of freezing/thawing cycles, etc., and, as a result of small sample sizes, we were unable to draw explicit conclusion regarding age of specimen or variety of storage media.

One important concern with utilizing museum-based material is that of contamination from outside sources. DNA from humans, bacteria, fungi, and other specimens (Malmstrom et al., 2005; Miller et al., 2009) can all drastically affect the output of whole genome sequencing because multiple taxa may be sequenced simultaneously. The problems of exogenous DNA may also not be resolved during the analysis stage, especially where short sequences are hard to tell apart, or where data from nontarget taxa have completely swamped the signal of the target species. Although we did not focus on contamination per se, we found no evidence of this based on preliminary analyses of approximately 100 genes. In particular, when our captured gene sequences were aligned with published mammal orthologues and used to build phylogenetic trees, we found that the former almost always formed monophyletic clades to the exclusion of human and M. lucifugus sequences. Here, we found no direct link between the initial DNA concentration at extraction and the number of contigs successfully assembled by TRINITY or between the number of contigs assembled and the number of genes recovered (see Supporting information, Figs S3 and S4).

Our results provide further support for the application of target sequence capture within the field of systematics. Target sequence capture could have particular importance for museum collections by opening up possibilities for incorporating genetic material from type specimens. These specimens underpin the morphological descriptions of almost all recognized species and, as such, serve as benchmarks for field identification and taxonomy (Kingdon, 1997; Csorba et al., 2003), as well as for conservation action (DeSalle & Amato, 2004). Yet, despite this, only a relatively small proportion of type specimens has been sequenced for inclusion in studies of phylogenetics and phylogenomics (Schindel et al., 2011; Staats et al., 2013; Brugler et al., 2014). Notable exceptions include the current efforts by the American Museum of Natural History to obtain COI barcodes from their mammal type specimens (Brugler et al., 2014). The rarity of such initiatives is unfortunate, not least because surprisingly large numbers of field researchers are unable to assign a taxonomic name to their study organisms with complete confidence (Wilson, 2000; Kim & Byrne, 2006; Bennett & Balick, 2014). This is especially true of researchers working on taxa that are cryptic or otherwise morphologically difficult to tell apart (Jones & Vanparijs, 1993; Kingston et al., 2001; Soisook et al., 2008). Sequencing of type material would benefit large numbers of taxonomists, including collectors and museum staff, who are engaged in laborious comparisons involving costly international loans and visits. If more type specimens were sequenced, then verifying the status of new species would be much simpler (Schindel et al., 2011; Brugler et al., 2014) and, effectively, digital through the use of online databases (Rhead et al., 2010; Kasprzyk, 2011; Goodstein et al., 2012). Additionally, it would help settle long-standing taxonomic debates, including the verification of the taxonomic distinctiveness of numerous specimens that were described in the past but are now viewed as questionable (Zhang et al., 2009; Helgen et al., 2013).

To conclude, the present study demonstrates the value of sequence capture for obtaining large volumes of target sequence data from museum specimens including some that show high levels of degradation. Moreover, the method was seen to work well across congeneric species that diverged by as much as 15 Mya. The comparative ease by which this method can recover large numbers of orthologous loci unlocks new opportunities for incorporating museum specimens, including types, into phylogenomic studies (Bi et al., 2013; Nachman, 2013; Tin et al., 2014). In particular, the removal of tiny amounts of material from museum specimens may potentially add enormous additional value to the content of collections.