News

Growing barcode libraries challenge understanding. There are already about 200,000 mtCOI barcodes from about 25,000 species in BOLD, the Barcode of Life Data Systems Database. The burgeoning data sets hint at insights into biological diversity, revealed by looking at many species at once. A map of counties of the United States shows both large and small scale patterns, shaped by history, geography, and politics.  Viewed through the lens of mitochondrial variation, what would a map of species show? Are differences among and within species similar in birds and butterflies? Do species boundaries differ in marine vs terrestrial species, or in tropical vs. temperate zones? 

 

   

Here I offer one possible way of visualizing differences in barcode data sets using as an example the BOLD  “Hesperiidae of the ACG 1” Project containing 2,185 COI sequences from 355 species of skipper butterfly in ACG conservation area in Costa Rica (Hajibabei et al Proc Natl Acad Sci USA 2006 103:968). The BOLD-generated neighbor-joining tree of ACG Hesperiid COI sequences shown at left offers a traditional way of comparing sequences and is an essential step in looking at individual species and their close genetic neighbors. However the NJ tree contains only about 100 sequences from 20 species per page, and so runs to 22 pages. In the future it will likely be desirable to compare much larger data sets from, say, all 3700 known species of world skipper butterflies. 

For DNA barcoding, the essential information is differences among and within species. The higher-level groupings of species which are inevitably generated by a tree are of less interest. (In the following analysis distances are used simply to examine patterns of variation, NOT to determine whether they are sufficient for diagnosing species.) 

One useful approach is to generate histograms of differences within and between species. BOLD has a “Nearest Neighbor” analytic function which generates a table of mean and maximum variation within each species, “nearest neighbor” distance to the next closest species, and histogram summaries of the results.  

The histograms quickly show distances within most species are small and minimum distances between species are generally larger. Histograms are summaries with unlimited capacity. However, one might want to know more about individual species. For example, do species with higher intraspecific distances also show greater interspecific distances? One also wonders about the variation below 1% in both panels. In Beautiful Evidence, Edward Tufte points out histograms display relatively small amounts of data, usually 1 value per column. How to generate something with more information, more like the US Counties map, but not 22 pages long?

The graph at left uses the same 2 essential parameters: distance within each species and distance to nearest genetic neighbor. Because the usual distances within and between species are very different, plotting on a logarithmic scale allows one to inspect the variation in each set simultaneously. The results with 278 of the ACG skipper species (all those for which more than one individual was sampled, thereby generating a mean intraspecific distance) are shown. For each species, there is a black dot showing intraspecific distance and a red dot directly above or below showing distance to nearest neighbor. Sorting by intra- and interspecific distance allows the relative distances for each species to be seen. This graph highlights the relatively few species with nearest neighbor distances less than the mean intraspecific distance for that species. A line drawn at 1% appears to separate most of the intraspecific from interspecific values.

This graph is remarkably compressible, as shown by the small inset in the US county map above and in the figure at right. Here this is used to compare variation in Costa Rican skippers (278 species in 1 Family) to that in Australian fish (172 species in 1 Class) (Ward et al 2006 Phil Trans Royal Soc B 360:1471). The distribution of intraspecific variation seems quite similar while the nearest neighbor distances in fish are higher, presumably reflecting less dense sampling of a larger taxon. In the Fish paper, the red dots at bottom were thought to be ID errors, so perhaps some of the those in the skipper data set showing zero distance between species are taxonomic errors as well. This graphic approach could be useful in comparing patterning of intra- and inter-specific variation in marine vs terrestrial, tropical vs temperate, and allopatric vs sympatric species.

 

Two recent articles suggest how and how not to learn about applying mtCOI sequences to identifying species.  In Zoologica Scripta 2006 35:441 researchers from Koenig Zoological Research Museum, Bonn, analyze 113 specimens of 61 morphologically-defined species of pholcid (daddy long-legs) spiders. Important for this analysis and for future study, collection locations are given and voucher numbers are provided for each specimen and DNA extract.

(Some pholcid spiders vibrate in their webs when disturbed, moving so rapidly they become invisible; here is a wonderful video)

16s and COI sequences were successfully amplified using a single primer pair for each gene from 79% and 80% of specimens, respectively. It is striking that strong clustering within species was observed despite using short segments of mtDNA (COI, 312 bp; 16s 287 bp), which are less than half as long as the standard 648 bp COI barcode. In NJ trees with either mtDNA sequence, all morphologic conspecifics grouped together and were reciprocally monophyletic (ie no overlaps between species). Likely splits based on large intraspecific distances and differing geographic distributions were observed in 6 (25%) of 24 multiply-sampled morphospecies.

The authors go on to propose graphic and statistical metrics to calibrate how well simple distances can define species limits. They find that mtDNA distances will often diagnose species: “tree-based taxon clustering and statistical taxon analysis indicate that molecular evidence does coincide with morphological hypotheses” and “we disagree with [Meyer and Paulay’s] point that independently of the group of organisms studied, a “barcoding gap” between interspecific and intraspecific distance values would likely disappear in studies featuring both dense within-species sampling and closely related species”, ie distance-based clustering often corresponds to species limits. 

This study uses vouchered specimens from known locations and accurate modern sequencing technology, focuses on a relatively small clade (959 known species pholcid spiders), and analyzes in a positive way how distance measures might be used to define species, helping us learn about DNA barcoding as a tool for species identification.   

In another recent study Syst Biol 55:715 2006 researchers from National University of Singapore examine COI sequences deposited in GenBank from Diptera (flies, mosquitos, and gnats). They found 449 of the 150,000 known species of dipterans represented, with multiple sequences from 127 species, and analyze these to “test two key claims of molecular taxonomy”. The scientists found that there were often large differences in COI within species and also frequent overlaps between species such that some sequences were more closely related to those of another species than to conspecifics. The litany of failures is quite long, including “even when two COI sequences are identical, there is a 6% chance they belong to different species”.

I do not understand why the authors put so much effort into analyzing such a heterogeneous set of data, except that they are worried about molecular taxonomy in general and DNA barcoding in particular. To my reading this study suggests that many GenBank records contain errors, either because current morphologic taxonomy is incorrect (for example, study cited above suggests probable splits in 25% of pholcid spiders), specimens used for GenBank records are incorrectly identified, or because DNA sequences in GenBank contain errors due to human factors or older sequencing technology. There must be some limitations to COI barcode identification of dipteran species, mostly presumably closely-related young species, but this study has not shown where such problems might lie. 

I hope that future studies will use more of the “best practices” demonstrated in Astrin et al’s study of pholcid spiders and so help us learn more about how to apply COI sequences to species identification.    

Thrips are tiny (.5 to 2 mm) plant feeding insects; approximately 4500 species are known, and some are serious agricultural pests. Kladothrips is an Australian genus of at least 35 species which form galls on Acacia trees. In Biol J Linn Soc 2006 88:555 researchers from Flinders University, Australia, apply mtDNA analysis to show that two gall morpho-types of Kladothrips rugosus represent different species. 

Originally described in 1907, K. rugosus is widely distributed across south and western Australia. Two gall types were noted, but no morphologic differences could be found in the thrips themselves. McLeish, Chapman, and Mound found pairwise uncorrected mtCOI p-distances were 0.0-0.6% within gall morphotypes, and 7.4-7.8% between, similar to distances within and among other gall thrips species. The authors aver the usual taxonomic distaste for distance measures (“Distance values are not intended as a means of identifying different species here, which is a problematic approach for species depiction, but as useful descriptors of genetic variation”). I translate this as distance measures can be used help discover new species, but are verboten in official species descriptions. 

The only morphologic differences are that “abdominal segments I-III are as brown as IV-VII, the metathorax is scarcely paler than the brown mesothorax and prothorax, and the sculptured reticles on the posterior half of of tergites II-III are all small and equiangular.” Phew! Not many persons could decipher such abstruse morphologic terminology, whereas DNA-based identification promises more democratic access to species identification. The main limiting factors are technological and likely solvable: establishing reference libraries and developing inexpensive DNA analytic methods. 

The authors found a third genetic cluster in K. rugosus, but were unable to discover any morphologic characters, so did not describe this as a new species. This seems scientifically inconsistent, and the authors seem to agree: “This lack of morphologic divergence has evident problems for traditional taxonomy..we suggest that “morpho-taxonomy” is little more than an historical artifact in the methodology of species recognition, despite commonly providing the most practical methods”

I hope the large data sets emerging from the barcode initiative and other genetic surveys will enable taxonomists to develop consistent methods of species delimitation, whether in thrips or thresher sharks, and the sequences themselves or their diagnostic nucleotide characters will be routinely incorporated into species descriptions.