Blog

In 2003, Paul Hebert and colleagues proposed a universal identification system employing short DNA sequences as identifiers for animal and plant species. Inspired by the Universal Product Code labels that stores use to track merchandise, he named these short sequences “DNA barcodes.” My colleagues and I set down thoughts inspired by this new name:

“Commercial barcodes and the barcode of life

Jesse Ausubel, Mark Stoeckle, Paul Waggoner

September 2004 

Although new methods of sequencing and visualization have displaced the one that produced autoradiographs that show blurry gray stripes of a gel indicating presence or absence of particular bits of DNA, the analogy between the commercial barcode and the barcode of life may be traced to it. However, the power of the analogy comes from other similarities: large capacities to differentiate mind-boggling diversity, ability of digits to distinguish unambiguously, rapidity and economy of identification, ability of parts of the code to distinguish categories, and avoidance of a Tower of Babel by uniformity. We elaborate briefly.

Without the final digit that checks accuracy, the quartets of bars and spaces in the Universal Product Code (UPC) have 10 alternatives at 11 locations, creating an ample 1011 capacity to identify manufacturers and their products. Instead of operating in quartets, sequences of CATG operate in trios that specify synthesis of an amino acid. Each trio of the four alternative CATG has 43 or 64 alternatives. A 600-unit sequence of DNA comprising 200 trios with 64200 alternatives opens ample capacity to identify millions of species. Such large capacities are needed to differentiate the diversity of an economy or a forest.

Because one product number in a UPC differs from another by discrete, digital steps rather than by the shades of verbal descriptions, the numbers identify the product–unambiguously. A barcode of life written as a sequence of CATG along a uniform locality of genomes differs from another by four discrete, unambiguous steps rather than by gradations of words, shapes, and colors. Barcodes gain power because digital beats analogue at making unambiguous distinctions.

Speed and economy also propel use of barcodes. Behind the beep of a UPC scanner lies orchestration that began with the initial conception of bars for numbers a half-century ago. Users and inventors orchestrated optics, electronics, and software to develop miniature, robust equipment that made the barcode an affordable master key to supermarket inventories and suppliers (Swartz 1999). Now that the price of DNA identification of a species has fallen to about $10 (Randhawa 2004), the orchestration can begin to provide a barcode of life. Uniformity fosters frequent use and thus learning and economy.

Product codes can identify products with increasing resolution. At the first level of resolution, the first bars of a UPC on a carton resolve the manufacturer. At the second level, the last bars resolve the product line. Opening the box and reading the serial number would resolve the individual. In analogous manner, extending a DNA barcode through more and more sequences would resolve from kingdoms to species, subspecies, and finally individuals. For our goal, Ockham’s razor prescribes as short a barcode of life as suffices to distinguish species.

Uniformity bestows the universality implied by the U in UPC. Scanners in hardware, grocery, and convenience stores must all call the same light bulb by the same 12 digits. Recently agreement between America and Europe added a thirteenth digit, made uniformity more widespread, and brought universality closer to realization (NY Times 12 July 2004, page C1). The power of standardization, whether in railroad gauges or typewriter keyboards, is one of the strongest lessons of the history of technology.

Finally, the success of a short DNA sequence distinguishing species will rest on reasoning, testing, and agreement, not just an appealing analogy. Reasoning will select a uniform locality on genomes that varies enough but not too much among species, testing will establish whether barcodes of that uniform locality correspond to established binomial names across several species, and then agreement will foster an expanding compilation of matching barcodes and binomial names.”

Robert Koch (1842-1910), father of medical microbiology, isolated agents of mankind’s major plagues: Vibrio cholera, Bacillus anthracis (anthrax and bubonic plague), and Mycobacterium tuberculosis. He laid down four conditions, “Koch’s postulates“, for establishing that an organism is the agent of disease, and subsequent generations of researchers applied these principles to determine the etiology of a multitude of infectious diseases. One legacy of Koch’s postulates was that isolation of organisms in pure culture became the backbone of diagnostic and research microbiology.

A century later, genetics has replaced culture as the essential framework for exploring microbial life. Metagenomic analysis of environmental samples, including from anatomic sites, has identified an unsuspected plethora of organisms, most of which are unculturable, at least under standard laboratory conditions. Even for organisms that can be grown in the laboratory, genetic detection is often the preferred diagnostic method, including for example detection of HIV, Neisseria gonorrhea, and Chlamydia sp.  Following Carl Woese’s early lead (PNAS 1977, 74:5088), microbiologists have generally included a standard locus, 16s rRNA, in genetic work, enabling phylogenetic trees spanning the diversity of life, and allowing each new isolate to be analyzed in conjunction with the work of others (as of 24 oct 2008, 75,257 16S rRNA sequences in GenBank).

Are genetic methods equally necessary for eukaryotes? In October 2008 Mol Ecol researchers from Cardiff University analyze mitochondrial COI differences among nine species of British lumbricid earthworms which were first described between 1758 and 1843, over 150 years ago. Partial COI sequences (a 582 bp segment which overlaps 648 bp DNA barcode region) from 71 individuals showed 2-5 deeply divergent clusters (average 13-15% sequence difference) in 4 of the 8 multiply-sampled species, and small divergences within each cluster, “indicative of the presence of multiple previously undescribed species”.  COI sequences from 270 individuals of one species, Allolobophora chlorotica, collected at 24 British and 5 mainland European sites showed 5 divergent clusters and surprisingly no clear geographic distribution pattern; over half the sites had 2 or more lineages, and one site had 4 lineages. As expected the same clusters were found by comparing another mitochondrial gene, 16s rRNA. Two of the lineages were found only in green color morphs; prior work indicated this form has distinct ecological preferences compared to pink morph Allo. chlorotica and that F1 hybrids are sterile, suggesting species status. As an aside, if earthworm specialists find morphological and ecological differences and mating incompatibility, why not designate as distinct species? As another example, two forms of European corn borer Ostrinia nubilalis are sympatric, genetically distinct, develop on different host plants, have different mating pheromones, and exhibit >95% reproductive isolation, yet are described as “host races” rather than separate species (Science 2005, 308:258). It sometimes seems there is an arbitrary aspect of how species status is awarded, or perhaps the process is slow.  

To see if mtDNA clusters were also reflected in nuclear genome, King et al performed AFLP (amplified fragment length polymorphism) mapping on 4-12 individuals from each of the 5 lineages. The nuclear results corresponded exactly to COI clusters except that the 2 green morph forms could not be distinguished, suggesting these are either a single interbreeding species (despite 14% mtCOI sequence difference!) or are young species which have not yet accumulated differences in nuclear DNA. It is hard to see how a 14% sequence difference could accumulate in mtDNA without accompanying nuclear changes, so I wonder if one of the genetic forms might reflect a relatively recent introgression from another earthworm species which has not yet been sequenced. It will be interesting to see whether the two green morph lineages, which were often found together at the same site, show assortative mating or restricted fertility. The authors conclude “extraordinary species-level genetic diversity was revealed among the British earthworms”….”four of nine ecologically generalist earthworms are probably complexes of multiple cryptic species”. And finally “further earthworm research in areas such as ecology and ecotoxicology, should be conducted in the knowledge that there are multiple cryptic species within many earthworm species”.

I conclude that genetics is equally essential for eukaryotic taxonomy as for microbiology. I believe there is no getting around the need to genetically reexamine most or all of the species named in the past 200 years to see if what we recognize as single and distinct species are really so. If there can be cryptic species in large visible animals such as birds, and males and females can be given different species names in fish, then there must be many more such oversights among the less easily observed. A standardized approach (ie DNA barcoding) is the most expeditious way forward and will leave a permanent marked trail that can easily be followed by non-experts who wish to identify their specimens. As in bacteria, standardizing on a single locus (ie barcode region COI for animals) enables new work to be seamlessly combined with old, leveraging its value (497,851 barcode records from 48,459 species in BOLD so far). Regarding higher-level evolutionary relationships, I find routine dismissal based on mathematical modeling of mtDNA single-locus trees, but not much effort to see what the potential is. Perhaps translated amino acid sequences and/or GC content can be informative for deeper branches, and nucleotide sequences for family- and generic-level relationships. At the very least, mtDNA trees serve to generate hypotheses, which can be  corroborated or disproved by more extensive genetic, morphologic, ecologic, behavioral, or fossil record data.

As of October 11, 2008 researchers have deposited 14,594 DNA barcodes in BOLD representing 2,586 avian species, 26% of world’s 9,933 birds. You can browse taxonomic coverage to date at All Birds Barcoding Initiative (ABBI) and BOLD taxonomy browser sites. Coverage includes representatives of all 27 orders and 159 families of world birds and nearly half of avian genera [1,014/2,101 (48%)]. To uncover possible hidden diversity, most researchers are sampling species across their geographic ranges rather than focusing on named subspecies, many or most of which appear to represent clinal variation (see for example Zink 2004, Phillimore and Owens 2006).

How far along are researchers toward mapping COI barcode resolution of avian species? Birds are of particular interest because species limits are generally well-defined, supported by a wealth of morphologic, ecological, behavioral, and other genetic data. Looked at regionally, there is good coverage in northern North America, parts of Central and South America, western Europe, Korea, and New Zealand, so it should be possible to see how well COI barcodes distinguish among local species in these areas. Published studies so far show >95% resolution of named species and have identified genetically divergent clusters which may represent unrecognized cryptic or “hidden” species (Vilaca et al 2006, Yoo et al 2006, Nyari 2007, Kerr et al 2007). As an aside, “cryptic” is an awkward term for genetically divergent populations of birds since most of these have diagnostic differences in morphology or behavior; “hidden” is more accurate to my ear. On a separate note, COI surveys have regularly revealed misidentified voucher specimens of birds, suggesting routine application of DNA barcode analysis could enhance quality of avian collections. 

Looked at globally, there is 100% coverage of 104 polytypic genera (having 2 or more species) representing 324 birds, so this should include the sister species and/or “nearest neighbors” for these, plus there are 853 monotypic genera (having only 1 species) in world birds, which are likely or known to be genetically divergent from birds in other genera. In addition, there are likely many other sister species or “nearest neighbors” within the remaining 1,982 birds with DNA barcodes so far (for example, BOLD includes 28 of 29 Dendroica sp wood warblers). It would be interesting to look at the nearest neighbor differences within the global data set. To my knowledge, comparisons among regions with COI barcode data have been not been published. My impression based on other avian genetic work is that named taxa in different biogeographic regions are genetically distinct, plus there are many unrecognized genetic divisions within species that range across biogeographic regions. I look forward to trans-regional and global comparisons!

In 29 july 2008 Fish Biology scientists from Macquarie University, Sydney describe successful recovery of mitochondrial DNA from contemporary and historical shark teeth and jaws. After developing the method on 11 recently collected teeth from Gray nurse shark Carcharias taurus and Ornate wobbegong (excellent name!) Orectolobus halei, Ahonen and Stowe applied it to 20-40 year old museum specimens, including 5 jaws from 3 species and 19 individual teeth from 2 species. They collected approximately 0.02-0.06 g of “tooth powder” by drilling several small holes into a tooth or jaw; DNA was extracted using a standard silica-based method or Qiagen DNAeasy tissue kit.

The authors are interested in historical population sizes for sharks; following the theory that genetic variation within species is an indicator of population size, they picked the hypervariable control region as their target. As an aside, results so far with mitochondrial surveys including DNA barcoding generally show very low variation within most animal species and no relationship between intraspecific variation and census population size. In any case, a 700 bp fragment of mtDNA control region was amplified with a single pair of primers. The two extraction methods gave similar results.  DNA was amplified and sequenced from 100% of the contemporary samples and 15/34 (44%) historical samples. 700 bp is a relatively long sequence to amplify from historical samples, suggesting it may be possible to obtain standard COI barcodes (648 bp) from museum skeletons of sharks and bony fish, which would be particularly useful for those species which are rare or otherwise difficult to collect. A standard set of fish primers (see for example Hubert et al June 2008 PLoS ONE) amplifies COI barcode region from most fish (more than 5,000 species so far, including including representatives of all major divisions of Chondrichthyes (cartilaginous fish) and Osteichthyes (bony fish), both marine and freshwater).

To date most fish specimens are preserved in formaldehyde, which makes routine DNA recovery difficult or impossible. If DNA can be recovered from skeletons, there are many museum specimens that might be used. For example, the American Museum of Natural History Icthyology Department collection includes over 35,000 fish skeletons as compared to about 2,500 tissue samples so far.

In August 2008 Sys Biol (open access!) researchers from East Carolina University apply mtDNA analysis as the necessary first step in defining three new species of trapdoor spider, previously subsumed as a single species Aptostichus atomarius. According to Bond and Stockman, “the genus Aptostichus is species rich, consisting of 30+ species (most undescribed) found predominantly throughout southern California.” I note that online World Spider Catalog version 9.0, 2008, lists four Aptostichus species, all described between 1891 and 1919, so apparently there is a lot of more work to be done, including updating the reference lists. One of the new names is A. stephencolberti, which led to what must be the first appearance of a spider taxonomist on national television (link to TV episode).

The authors describe the challenge for delimiting species in these California trapdoor spiders: “Highly structured, genetically-divergent, yet morphologically homogeneous species (eg nonvagile cryptic species[my note: nonvagile refers to organisms with limited dispersal]), although often ignored or overlooked, provide one of the greatest challenges to delimiting species. Populations, or very small groups of populations constitute diverent genetic lineages but present somewhat of a contradiction because they lack the “requisite” characteristics” often used when delimiting species. Morphological approaches to species delimitation in many of these groups grossly oversimplify and underestimate diversity; in short these traditional applications fail if our interests extend beyond what can simply be diagnosed with a visual and/or anthropormorphic-based assessment.”

So on the one hand, these spiders comprise multiple genetically distinct lineages (up to 24% sequence difference in 12S/16S mtDNA) with geographically restricted ranges; on the other hand, they all look more or less alike. How to decide which are species? The authors apply “cohesion species concept” by asking if the lineages are “genetically and/or ecologically interchangeable.” The authors provide helpfully provide explicit details of their decision making process. The short version is that genetically distinct, geographically disjunct lineages are counted as separate species, and parapatric or sympatric lineages are counted as different species only if they are NOT “ecologically interchangeable (EI).” EI is calculated from a defined set of ecological and climatic parameters.

Under some criteria, the authors note these spiders could be split into “more than 20” [or even] “~60” groups, which they describe as “an unreasonable number of species-level lineages.” This conjecture may be true; I hope that more scientists apply similarly explicit criteria for species delimitation as described here so we can learn more about how finely divided biodiversity is, in addition to our judgment about what is a “reasonable” number of species. Genetics is a powerful window into biology, of course. In birds the frequency of extra-pair matings (up to 96% pairs and 75% offspring in fairy wrens, for example (Double and Cockburn 2000)) was unsuspected until genetic testing was applied to parents and offspring. 

The genetic framework in this study is based on 1300 bp of 12S/16S mtDNA (167 individuals, 75 locations), plus 905 bp nuclear ITS sequence in a subset of 22 individuals. Looking ahead, I hope that in their next study of spider phylogeography the authors include COI as an mtDNA locus (full-length sequence is 1500 bp, so that would likely have given the same phylogenetic signal as 12S/16S); this would enable the authors and others to combine their data with the reference COI DNA barcode databases.

I close with an observation about spider genetic data. To my eye, there are surprisingly few genetic data on spiders so far. A search in GenBank for Order Araneae (spiders) shows 9,445 sequences (representing any gene) from 1,852 species (4.6% world total of 40,432 species (World Spider Catalog)). Looking at mitochondrial genes, there are 2,629 COI sequences from 1,071 species (2.6% world) and 2,268 12S/16S sequences from 1,041 species (2.6% world). Thus it appears that only about 1/40th of world’s spiders have a uniform gene locus deposited in GenBank, and on average, only 2 individuals per species have been sequenced. The Spider Tree of Life project plans to sequence 50 loci (including COI and 12S/16S) from about 500 species, so that will help. I hope that arachnologists will follow the approach in this paper and include a standard genetic locus (most usefully COI) as part of species descriptions and analyze multiple individuals per species. Among other applications, this might help identify currently unidentifiable juvenile forms, like the wind-blown “little aeronaut[s]” that arrived on silk threads in vast numbers on the Beagle when it was sixty miles distant from land, November 1, 1832 (Voyage of the Beagle).

In October 2008 Wired reporter Gary Wolf profiles birth and rapid growth of standardized DNA-based species identification (ie DNA barcoding). His article centers around time spent in Costa Rica with Dan Janzen, Winnie Hallwachs, and their band of parataxonomists in Area de Conservacion Guanacaste; additional legwork includes visits to worried taxonomists at University of California Berkeley (“Honestly, I never thought it would get this far,” says Kipling Will), and University of Guelph, Ontario. He concludes with an evocative analogy: “barcodes are not just devices to put names on animals; they are also clever traps to catch all the people in the world whose curiosity impels them toward data as if toward light.”

An article in October 2008 Scientific American, with Sci Am’s trademark excellent illustrations, (web version; pdf) examines hows and whys of DNA-based future of species identification (I am co-author with Paul Hebert). After discussing the many practical applications for identifying known species, we  conclude with our own analogy: “Just as the speed and economy of aerial photography caused it to supplant ground surveys as the first line of land analysis, DNA barcoding can be a rapid, relatively inexpensive first step in species discovery.”