Prospects for using DNA barcoding to identify spiders in species-rich genera

While previous research has indicated the utility of DNA barcoding in identifying spider species sampled from a localized region, the eff ectiveness of this method over a broader geographic scale and with denser taxon sampling has not yet been extensively considered. Using both new and published data from 1801 individuals belonging to 361 morphospecies, this study examined intraand interspecifi c divergences for 19 genera that were each represented by at least 10 morphospecies. We particularly focused on increasing species-level sampling in order to better characterize levels of interspecifi c divergence within species-rich genera and to examine the prevalence of a “barcode gap” (discontinuity between intraand interspecifi c divergences). Overall, the mean intraspecifi c divergence value was found to be 2.15%, the average maximum intraspecifi c divergence was 3.16%, while the mean divergence between nearest interspecifi c neighbours was 6.77%, demonstrating the typical presence of a barcode gap. Of the 66% of morphospecies that formed monophyletic sequence clusters, the majority (92.5%) possessed a barcode gap. We also examine possible biological explanations for the large proportion of paraphyletic and polyphyletic clusters and discuss the need for further taxonomic investigations. Th e overlap between intraand interspecifi c divergences was not unexpected for some ‘species’, such as Pardosa groenlandica, since prior morphological studies have suggested that it is an example of a species complex. However, other cases of high intraspecifi c divergences may refl ect cryptic species diversity, indicating the need for a taxonomic approach that combines both morphological and molecular methods. Th e list of the species, COI sequences, and source references used in the analysis is published as a dataset under doi: 10.3897/zookeys.16.239.app.A.ds. Th e list of analyzed species, mean and maximum intraspecifi c divergences, distances to the nearest neighbouring species in its genus, general localities, and lifestyle characteristics is published as a dataset under doi: 10.3897/zookeys.16.239.app.B.ds.


Introduction
DNA barcoding -the use of short, standardized gene regions for species identifi cation and discovery (Hebert et al. 2003) -has proven to be very eff ective in identifying specimens to species level for many diff erent groups of organisms, from butterfl ies (Lukhtanov et al. 2009) to birds (Kerr et al. 2007) to plants (Kress et al. 2005).For groups in which identifi cation can be very diffi cult, the potential utility of DNA barcoding is immense.Spiders are one such group, with 40,700 spider species and subspecies belonging to 109 families recognised worldwide (Platnick 2009).
Morphology-based identifi cations of spiders are time consuming and problematic for several reasons.With some groups, it is diffi cult or impossible to identify juveniles, which are the most abundant individuals at certain times of the year.Another challenge for identifi cation is striking sexual dimorphism, especially in some weavers (Nephila and Micrathena), or the lack of information on diagnostic characters for one sex.In fact, 46% of spider descriptions consider just one sex, and 1.5% are based upon juveniles only (Platnick 2009), subsequently making it diffi cult to match sexes and life stages.Even identifi cation of known adults is time consuming, because the majority of species require detailed examination and even dissection of the sexual organs for authoritative identifi cation (Locket and Millidge 1951).Th ese factors represent major obstacles for large ecological studies and biotic surveys of spiders but are ones that could be overcome by DNA barcoding.DNA barcoding could also assist in both distinguishing species with similar morphologies (sister species), a common issue with spiders due to their overall long-term evolutionary morphological stasis (Bond et al. 2001), as well as in resolving the status of subspecies.
In addition to assisting with the identifi cation of known species and helping to solve taxonomic problems, DNA barcoding is likely to be useful for species discovery in spiders, where many species await discovery (Fig. 1).Although the majority of undescribed species are expected to inhabit the tropics, certain temperate regions of the world are clearly underexplored.Interestingly, last year twice as many new taxa were described from China as from North America (Platnick 2009).Th erefore, DNA barcoding could play an important role in an iterative taxonomic program, fl agging cases of potential new species or cryptic species (Hebert et al. 2004) among both juveniles and adults, thus focusing further collecting or taxonomic attention where it is needed most.Many authors now emphasize that the combination of morphological, molecular, and other types of data is the best approach for identifying and describing new species (Dayrat 2005;DeSalle et al. 2005;Gibbs 2009).In this respect, strong collaboration between molecular systematists and traditional taxonomists is necessary to increase the accuracy, speed, and accessibility of identifi cations (Hebert and Barrett 2005).Th e present paper represents an example of such a collaboration.
To date, only one large-scale study (Barrett and Hebert 2005) has examined the eff e ctiveness of DNA barcoding for species-level identifi cation across the spiders.Th eir study showed that barcoding of spiders held considerable promise, with mean interspecifi c divergence (16.4%) far exceeding mean intraspecifi c divergence (1.4%), allowing for a 100% success rate in matching duplicate specimens to their species among the taxa included in their study.However, two important constraints limit the generality of their conclusions: only a small fraction of known diversity was included, less than 0.005% of spider species (Prendini 2005), and taxon sampling was sparse, with only 5 genera being represented by more than 4 species.While the collections used in their study were augmented by museum material, most of the specimens were collected from southern Ontario (Barrett R., pers. comm.), leading some to question the utility of DNA barcoding for spiders that are more broadly sampled (Prendini 2005).One reason for this is that increased geographical coverage may lead to greater intraspecifi c variability, reducing or potentially swamping the "barcode gap", which is a typical discontinuity between levels of intraspecifi c and interspecifi c divergence that allows for successful identifi cation.Secondly, since allopatric speciation is thought to be the dominant mode of speciation (Coyne and Orr 2004), many more closely related species pairs are also expected to be encountered as geographic coverage expands.While locality information could be incorporated into identifi cation algorithms to address this issue, it would be simplest if barcoding were to work globally across specimens from any site from any lineage, even species-rich ones, especially since a desired application is to be able to detect invading species.
Some progress has been made to address these issues in spiders, with recent studies using COI to examine divergences within and between species collected regionally or globally (Table 1).While one study that considered 61 species in the family Pholcidae  found a maximum intraspecifi c divergence value of 10.9%, which was greater than the minimum interspecifi c divergence value of 8.7%, overall there was little overlap between the two categories of divergence (Astrin et al. 2006).Binford et al. (2008) also found slight overlap in intraspecifi c and interspecifi c divergence values in 40 species from 4 genera as well as one anomalously high intraspecifi c divergence value (16.3%) in Loxosceles speluncarum, a taxon which they suggested may represent a species complex.Garb and Gillespie (2006), however, did fi nd evidence of a barcode gap in their 64 widely collected specimens representing 33 species, even with a maximum intraspecifi c divergence value of 8.9% (Table 1).
Currently, global-scale taxonomic campaigns are underway to test barcoding success in geographically widespread and species-rich clades.In the continent-wide and large-scale regional campaigns that have been conducted thus far, high identifi cation success and low intraspecifi c variability have been found in birds of North America and Argentina (Kerr et al. 2007(Kerr et al. , 2009)).Increased intraspecifi c divergence did, however, arise in central Asian butterfl ies that were sampled across a broad range, but this problem was addressed through a clustering method rather than a divergence threshold concept (Lukhtanov et al. 2009).DNA barcoding has therefore been shown to be successful in the face of high species richness sampled allopatrically, but the need for broader sampling and testing of more taxonomic groups remains.
In this study, we build upon previous studies examining COI sequence diversity in spiders (Table 1) to explore patterns of genetic variability in this group more broadly.We specifi cally aim to extend taxon coverage within species-rich genera to better estimate minimum interspecifi c and maximum intraspecifi c divergences.We use this information to draw inferences about whether or not closely related species sampled allopatrically can be distinguished using either a threshold or clustering approach.Based on original and publicly available data, we investigate genetic patterns within and among species belonging to 19 species-rich genera to examine the evidence for a barcode gap in broadly distributed species.

Collection procedure
Spiders were collected by various methods (pitfall traps, catching by hand, sweeping, and beating trees) from multiple sites in Canada each year from 2005 to 2008.Th ese include: Guelph, Grey County, Point Pelee, Kawartha, and Bruce Peninsula in Ontario; Lundar Beach Provincial Park, Riding Mountain National Park, and Churchill in Manitoba; Grasslands National Park in Saskatchewan; and Banff National Park and Waterton, Alberta.Spiders from the genera Agelenopsis, Cicurina, Dolomedes, Misumenops, Pardosa, Tetragnatha, Th eridion, and Xysticus were used in the analysis.

Protocol for barcoding
Following photography of the whole specimen, a section of one leg was removed and placed in a well of a 96-well plate containing ethanol.One well was left empty in each plate to serve as a negative control.A standard Glass-Fibre protocol, (Ivanova et al. 2006) was used for DNA extraction.First, 50 μL of a mixture of 5 mL of insect lysis buff er and 500 μL of Proteinase K (20mg/L) was added to each well of the plate.Th e plate was then incubated overnight at 50°C before being centrifuged.A liquidhandling robot carried out the remaining series steps that included the addition of binding mix into each well followed by vacuuming and washing using protein wash buff er (Ivanova et al. 2006).
For the polymerase chain reaction (PCR), the standard PCR cocktail concentrations of the Canadian Centre for DNA barcoding were employed (Ivanova and Grainger 2007).To amplify the target gene, cytochrome c oxidase subunit I (COI), either the Folmer primer pair (tailed with M13), a universal metazoan primer (Folmer et al. 1994), or Lep primers (Hebert et al. 2004) were used.Th e PCR thermal regime was as follows: 94°C for a minute; 5 replicates of 94°C for a minute, 45°C for 40 seconds, and 72°C for one minute; 35 cycles of a minute at 94°C, 40 seconds at 51°C, and 72°C for a minute; and concluding with fi ve minutes at 72°C.
All PCR products were sequenced bi-directionally on an ABI3730XL using the primer pair of M13F and M13R or Lep-F or Lep-R.Th e forward and reverse sequences were used to generate a single consensus sequence using CodonCode Aligner v. 3.0.2(CodonCode Corporation).

Accessible data
In order to include as many genera in the analysis as possible, publicly available data in BOLD (Ratnasingham and Hebert 2007) were combined with the newly obtained sequences.Most of the data from BOLD were originally obtained from GenBank and include specimens collected around the world by various methods and researchers.All species names, BOLD or GenBank accession codes, and general localities are provided in Appendix A. For original data, additional locality details as well as the sequence data are available in the following project in BOLD: "Spiders in species rich genera".

Analysis of genetic divergence patterns
Genetic divergences were examined in the 19 genera (Agelenopsis, Anoteropsis, Cicurina, Cyclosa, Dolomedes, Dysdera, Geolycosa, Hypochilus, Latrodectus, Misumenops, Neriene, Orsonwelles, Pardosa, Pholcus, Pimoa, Schizocosa, Tetragnatha, Th eridion, and Xysticus) that were represented by 10 or more morphospecies, at least two of which were represented by more than two sequences, when the unpublished and published data were combined.Only specimens with species-level identifi cations and sequence lengths greater than 420 base pairs were included in this analysis, resulting in a dataset consisting of 1801 individuals and 361 species.
With the sequence analysis tools available in BOLD, distance summary analysis as well as nearest-neighbour (NN) analysis were performed for each genus separately using the Kimura-2-Parameter (K2P) genetic distance model (Kimura 1980) with pairwise deletion of missing sites.For each species, mean intraspecifi c divergence, maximum intraspecifi c divergence, and mean NN distance (average distance to the most closely related species) were computed.Th e averages and ranges of these values were tallied and compared among genera.
Using the 200 morphospecies represented by two or more individuals, we checked for the presence of a "barcode gap", a disjunction between levels of intraspecifi c and interspecifi c variability, by plotting maximum intraspecifi c divergences against mean NN distances.Additionally, we explored whether this pattern varied among morphospecies displaying diff erent topological patterns in the Neighbour-joining (NJ) (Saitou and Nei 1987) phenogram for each genus.Each morphospecies was assigned to one of four topological categories depending upon its sequence clustering pattern: monophyletic (and reciprocally monophyletic from the nearest neighbour), nested (monophyletic cluster nested within a paraphyletic species), paraphyletic, or intermingled (polyphyletic).Four anomalous specimens each failed to group with its own morphospecies and additionally had an average distance of more than 10% from its morphospecies; these specimens were excluded from further analysis under the assumption of misidentifi cation.Morphospecies that formed monophyletic and paraphyletic clusters were retained regardless of the genetic distances.Th e relationship between topology category and presence/absence of a barcode gap was plotted for all data together and for just those specimens examined in the present study (for the three genera that were represented by more than 5 species once GenBank data were excluded: Pardosa, Tetragnatha, and Xysticus).

Life histories and geographic distributions of species
To examine the relationship between ecology and genetic divergence patterns, each species was placed into one of three categories based on lifestyle: web-builders, sit-andwait predators that do not build webs, or active predators.Agelenopsis spp., Cicurina spp., Cyclosa spp., Hypochilus spp., Latrodectus spp., Neriene spp., Orsonwelles spp., Pimoa spp., Pholcus spp., Tetragnatha spp., and Th eridion spp.were classifi ed as webbuilders; Dolomedes spp., Misumenops spp., and Xysticus spp. as sit-and-wait predators; and Anoteropsis spp., Dysdera spp., Geolycosa spp., Pardosa spp., and Schizocosa spp. as active predators (Comstock 1965;Dondale andRedner 1978, 1990;Dondale et al. 2003).Each species was also assigned to one of three approximate geographical distribution categories, based on the locality information available for the specimens of that species: regional (local to regional: collected from a single Canadian province or US state or single other country), continental (regional to continental: collected from at least two provinces or states or countries), or intercontinental (collected from at least two continents).If increasing geographical scope results in greater intraspecifi c and lower interspecifi c divergences, then the ratio between NN distance and intraspecifi c distance is expected to decrease as geographical sampling increases, resulting in lower barcode identifi cation success.One-way ANOVAs were used to test for diff erences in genetic patterns among lifestyles of regionally sampled species as well as among the geographical distribution categories, while box-and-whiskers plots were used to explore the data graphically.

Range of divergence values
When all available data were considered, divergence values were fairly variable across genera.Mean intraspecifi c divergences within morphospecies ranged from a low of 0.61% in Cyclosa to a high of 10.86% in Hypochilus.Th e latter genus also included the highest observed intraspecifi c divergence value -17.74% (Table 2; Appendix B).Although some of these divergence values are very high, they were still, in most cases, less than the NN distances.Average NN distances ranged from 2.38% in Geolycosa to 14.10% in Pimoa.However, 0% divergences were found in 5 species pairs belonging to 3 genera: Cicurina caliga / C. hoodensis, Cicurina vespera / C. madla, Latrodectus variegates / L. mirabilis, Latrodectus corallinus / L. diaguit, and Schizocosa ocreata / S. rovneri (Table 2).
While a greater diff erence between intraspecifi c divergences and NN distances was expected when published data (which originated from many sources) were excluded, some high intraspecifi c divergence values remained in our new dataset.In Xysticus, the average of mean intraspecifi c divergences was 2.12%, with a maximum intraspecifi c divergence of 9.21% in Xysticus durus (Table 2).However, even though mean NN distances were reduced in both Pardosa and Xysticus when only new data were considered, no cases remained of species with NN distances of 0%.
Box-and-whisker plots of average intraspecifi c divergences showed that 5 of the genera had extreme outliers, values that were more than 3 times the interquartile range from the third quartile (Fig. 2).Th ese outliers could represent cryptic species complexes and should be more closely examined.Separating the regionally sampled species into the three lifestyle categories did not yield any signifi cant diff erences in mean or maximum intraspecifi c divergence (F=2.01 and F=0.68 respectively with df=2, p=0.14 and p=0.51).In the comparisons of mean and maximum intraspecifi c divergences between the three geographical distribution categories, no signifi cant diff erence was found (F=0.15 and F=0.54 with df=2, p=0.86 and p=0.58).Contrary to predictions, the ratios of NN distance to both mean and maximum intraspecifi c divergence did not decrease with increased geographic coverage, with no signifi cant diff erence being found between the three categories (F=0.87 and F=0.87 with df=2, p=0.42 and p=0.42).

Types of sequence topologies within morphospecies and presence of a barcode gap
Among the newly collected data, a greater proportion of the morphospecies formed monophyletic sequence clusters.Of the 24 morphospecies, 22 (92%) were monophyletic, as opposed to 66.5% of the morphospecies examined when all of the data were combined.Th e remaining two morphospecies examined in the present study possessed a paraphyletic topology.When GenBank data were included, 14% of morphospecies were paraphyletic, 11% were nested, and 8% were intermingled.Th e proportion of monophyletic morphospecies in these two datasets ultimately aff ected how frequently a barcode gap was present, with a gap observed in 83.3% of morphospecies when only new data were considered and 71.5% when all data were used.Among those morphospecies within the monophyletic category, however, there was very little diff erence between the two datasets in the proportion that lacked a barcode gap.When GenBank data were excluded, a barcode gap was absent for 13.6% of monophyletic morphospecies, while 7.5% of monophyletic morphospecies lacked a gap when the data were combined (Fig. 4).A barcode gap was more frequently absent in the combined dataset for species belonging to the three other

Divergences of broadly sampled species
To address the issue of broader geographic sampling within species potentially diminishing the barcode gap, intraspecifi c divergence values and NN distances were considered for 3 morphospecies represented by specimens collected from at least 2 geographically distant sites, more than 1000 km apart.A barcode gap was clearly present in 2 of these cases.Xysticus emertoni, collected from 2 sites in Southern Ontario as well as in Saskatchewan, showed a maximum intraspecifi c divergence value of 2.2% while the NN distance was 3.46%.A maximum intraspecifi c divergence value of 2.4% and a NN distance of 7.26% were found in Tetragnatha versicolor, a species collected from Manitoba and 3 sites in Southern Ontario.However, Pardosa groenlandica, which was collected from Michigan, Alberta, and Churchill, Manitoba, had a maximum intraspecifi c divergence value of 3.87% and a nearest neighbour distance of only 1.26%.Because locality information was available from the source papers for some of the GenBank records, it was possible to examine genetic variability in species collected globally.While maximum intraspecifi c divergence (5.11%) did exceed the NN distance of 0.17% in Dysdera crocata, a species collected from the Canary Islands, Canada, and Spain, a barcode gap was present for three other broadly sampled species.Latrodectus geometricus, sampled from Argentina, Hawaii, and Florida, showed a maximum intraspecifi c divergence value of 2.39% and an NN distance of 12.39%.While the maximum intraspecifi c divergence value was high in Pholcus phalangioides (7.13%), a species collected from Brazil, Spain, and Germany, it remained below the NN distance of 17.72%.Represented by individuals from Canada and French Polynesia, a barcode gap was also detected in Tetragnatha laboriosa, with a maximum intraspecifi c divergence value of 2.67% and an NN distance of 13.88%.

Discussion
Our study sought to determine the suitability of DNA barcoding for identifying spiders by examining genetic variability in 19 species-rich genera.Overall, a barcode gap -a discontinuity in levels of intraspecifi c compared to interspecifi c genetic divergences -was detected for a large proportion of the species, even though some very high values of intraspecifi c divergence and very low NN distances were observed.We explore possible biological and artefactual explanations for these extreme values.Th is study furthers our understanding of genetic divergence patterns, the degree of concordance between taxonomy and genetic results, and geographical distributions of lineages.Further study on genetic patterns and processes of evolution will enhance our ability to create a barcode-based species identifi cation system in spiders as well as advance our knowledge of spider biology.

Presence of a barcode gap
While expanding to a global scale led to overlap in intraspecifi c and interspecifi c divergence values in a greater proportion of species than reported in an earlier study (Barrett and Hebert 2005), a barcode gap was still present for 89.7% of monophyletic and nested morphospecies.Some of the morphospecies that lacked a barcode gap in our study possessed very high intraspecifi c divergence values (up to 17.74% in Hypochilus pococki), more than doubling the highest observed by Barrett and Hebert (2005).Th eir overall mean intraspecifi c divergence value of 1.4% was also considerably lower than the 2.15% observed in this study.However, even the value obtained from their local study was higher than intraspecifi c divergences reported in other taxonomic groups.For example, mean intraspecifi c divergence was 0.6% in 81 of 87 species of bats examined in Guyana (Clare et al. 2007) and just 0.23% in North American birds (Kerr et al. 2007).In a study of 300 aphid species, Foottit et al. (2008) found an overall average intraspecifi c divergence of just 0.2%.When 39 of these species that were represented by replicate individuals from widely separated geographical regions were considered, divergences were still modest, with 30 of the species showing <0.7% divergence and the remaining 9 possessing divergences ranging from 0.7 to 3.1%.
High intraspecifi c divergence values were not the only cause of more spider morphospecies lacking a barcode gap in our study; interspecifi c divergence values (or NN distances) were also low in comparison to previous fi ndings.Th e average minimum interspecifi c divergence within genera was 9.2% in Barrett and Hebert (2005) but was only 2.61% in our study.Although Paquin and Hedin (2004) did observe some similarly low values for interspecifi c divergence, most other previous studies on spiders have found interspecifi c divergence values closer to those of Barrett and Hebert (2005) (Table 1).Th e denser taxon sampling in the present study certainly reduced inter-species distances, but this low average value was also infl uenced by the fi ve species pairs with a NN distance of 0%.In light of the broad range and diverse sources of data included in the present study, the high rate of barcode gap detection is perhaps more remarkable than its absence in a small fraction of cases.

Potential causes of anomalous values in published data
While the causes of the high intraspecifi c divergence values (especially in Hypochilus, Dysdera, and Pimoa, where the average mean exceeded 4%), the low interspecifi c divergence values (which averaged below 3% in Geolycosa, Anoteropsis, and Agelenopsis), and the large number of paraphyletic and interspersed morphospecies cannot accurately be determined in the scope of this study, previous research has provided multiple possible explanations.
Cases of extreme intraspecifi c divergence have been previously revealed in some spider morphospecies.Within single populations in California, Bond et al. (2001) found haplotypes of the trapdoor spider Aptostichus simus with divergences ranging from 6-12%.Th eir results suggested that a morphological species concept could greatly underestimate true evolutionary diversity since spider genitalia may not evolve as rapidly and divergently as previously believed (Bond et al. 2001;Hedin 1997).Morphospecies demonstrating high genetic divergence without comparable morphological divergence may therefore represent species complexes (Bond et al. 2001).Pardosa groenlandica, one of the species that was widely sampled in this study, and with high intraspecifi c divergence value (max =3.87%), may also represent a species complex but may not be completely cryptic.Dondale (1999) reached the same conclusion after observing signifi cant geographical variety in the epigynal ratio (length: width ratio of the sclerotized region covering the internal female genitalia) within P. groenlandica, suggesting the need for further investigation of this morphospecies.
Th e previous studies from which the 5 species pairs having 0% NN distances originated were unable to establish the exact cause of the lack of interspecifi c divergence.However, for the species pairs of Cicurina caliga / C. hoodensis and Cicurina vespera / C. madla, Paquin and Hedin (2004) suggested that the explanation could either be introgression or the species being synonymous.Synonymy was also used to explain the lack of genetic divergence between Latrodectus variegates / L. mirabilis (Garb et al. 2004), while incomplete lineage sorting and hybridization were proposed for Schizocosa ocreata / S. rovneri and Latrodectus corallinus / L. diaguita (Hebets and Vink 2007;Garb et al. 2004).
Although 90.9% of paraphyletic and interspersed 'species' lacked a barcode gap, 22% of morphospecies in this study possessed these sequence patterns.It would be benefi cial to consider why this pattern occurred and whether these morphospecies are in fact valid taxa.Firstly, it is important to consider that paraphyly can occur naturally and is an expectation of many speciation processes (Kuntner and Agnarsson 2006).Potential causes for species displaying an interspersed pattern include hybridization, molecular introgression, and rapid morphological divergence due to selection on certain traits.Another possible explanation is convergent evolution, which is thought to rarely occur, but that has been recently suggested to explain colour polymorphisms in the family Th eridiidae (Oxford 2009), behaviors in Hawaiian Tetragnatha (Blackledge and Gillespie 2004), and secondary sex traits in Padilla (Andriamalala 2007).However, such types of characters are seldom used as the basis for species discrimination in spiders.Since a far smaller proportion of species in the new dataset were paraphyletic or interspersed (8%), misidentifi cations may also explain some cases in the published data.

Prospect of a DNA barcode identifi cation system for spiders
For large-scale barcode-based identifi cation of spiders to be successfully implemented in monitoring or ecological studies, future research should consider using a clustering method rather than a threshold method, circumventing complexities introduced by the absence of a barcode gap in 10.3% of monophyletic and nested spider morphospecies.In Lepidoptera, the proportion of species forming discrete clusters only decreased by 2.7% when an additional population was included (Lukhtanov et al. 2009).Increased geographical coverage is therefore not an obstacle to the implementation of barcoding as an identifi cation method even across large spatial scales.Additionally, the high success in this study did not seem to be compromised by the geographic distribution of the species, with the ratio of NN distance to intraspecifi c divergence not decreasing with increasing geographic sampling.In order to further validate this conclusion, it would be benefi cial to collect data on a greater number of widely distributed species.Furthermore, to determine whether or not the high intraspecifi c divergences observed in this study do indicate cryptic species, ecological or breeding studies could be conducted, along with genitalic investigations.
Despite the clear need for further systematic studies and for harmonization of molecular and taxonomic data, the results of this study are promising.Rapid identifi cation of spiders using DNA barcoding methods is expected to become increasingly accurate, aff ordable, and attainable.Due to their diversity and important position in terrestrial foodwebs, biosurveillance protocols that include spiders are highly desirable.We propose that collaboration towards building a global community resource -consisting of expert-identifi ed specimens in permanent collections linked to online specimen and sequence records -will be the most productive step towards understanding and enabling research on global spider biodiversity.

Figure 1 .
Figure 1.Cumulative number of spider species described over time, including only species that are currently valid (description years for all valid species follow Platnick 2009).

Figure 2 .
Figure 2. Box-and-whisker plots of average intraspecifi c divergence for 16 genera represented by more than 3 species (Neriene, Pimoa, and Th eridion were excluded).Th e life history of each genus is also indicated.

Figure 3 .
Figure 3. Bar graph with standard errors showing the diff erences in mean (in black) and maximum (in grey) intraspecifi c divergence as well as nearest-neighbor distance (in white) between the three geographical distribution categories.

Figure 4 .
Figure 4. Maximum intraspecifi c divergence compared with nearest-neighbour distance of monophyletic morphospecies for all data and using only new data, which have been identifi ed by a single spider taxonomist.Most species (92.5%) fall above the 1:1 line, indicating the presence of a "barcode gap".

Figure 5 .
Figure 5. Maximum intraspecifi c divergence compared with nearest-neighbor distance using all data for the four categories of topology: A monophyletic (133 cases), B nested (23 cases), C paraphyletic (28 cases), and D intermingled (16 case).See Methods for defi nitions.89.7% of monophyletic and nested species fall above the 1:1 line, indicating the presence of a barcode gap, while 90.9% of paraphyletic and intermingled species fall below this line.

Table 1 .
Summary of the scope and results of several previous studies that used COI sequences to study spider diversity.For studies that did not provide ranges of divergence values but did have all of their sequences deposited in GenBank, ranges were recalculated based on the studies' original data (indicated by *).

Table 2 .
Summary of COI divergence patterns within and among species belonging to 19 diverse spider genera.Divergence values for 3 of the genera are also provided using new data only (indicated in bold).