[Taxacom] inapplicability of mtDNA barcoding to insects

Karl Magnacca kmagnacca at alumni.wesleyan.edu
Sun Sep 30 00:17:56 CDT 2007

On Fri, September 28, 2007 10:42 am, Doug Yanega wrote:
>>Why imply through nay-saying that barcoding should be
>>dismissed outright because it is not 100% accurate 100% of the time?
> Because if we rely exclusively on COI to make species IDs, and are
> sinking huge amounts of funding into it (whether or not it is to the
> exclusion of morphological taxonomy), then ~70% accuracy is certainly
> not much of a selling point for that MASSIVE investment.

That depends on how the 70% is distributed.  If it works on 70% of
families, including some where identification is currently difficult, then
that's useful, since you can define where it will work and where it won't.
 If it works on an average of 70% of the species in any given family or
genus, then that's not very useful.

> The failure of COI barcoding *is* a problem if COI is being treated
> as the only viable source of taxonomic data, and - if what Karl says
> is true - there are apparently a number of barcoders who are NOT
> using COI as an adjunct to traditional taxonomy, but as a replacement
> for it (or treating taxonomy as an adjunct to COI).

My point was more that there are people who have jumped on the bandwagon
and gotten excited about finding things that *might* be cryptic species,
without really doing the grunt work biology to determine if they really
*are* - in the process, overlooking alternative explanations as to why the
morphology and genes conflict.

There are also places where having a greater knowledge of other methods
would vastly improve the usefulness of even the current barcoding methods.
 For example, there were several talks at the conference where people were
only able to obtain short (100-200 bp) sequences from degraded specimens. 
Unsurprisingly, these often failed to reliably separate species using the
standard neighbor-joining method.  Yet I don't recall any of them trying
parsimony or other phylogenetic algorithms, which would give much better
results when fewer variable sites are available or sequences only
partially overlap.  As with using COI as opposed to other sites (see
below), NJ is useful because it's fast and the phylogeny is unimportant,
but that doesn't mean you stick with it when you run up against its

> At least several
> MAJOR papers on the topic I have read concluded that the
> discrepancies between what the taxonomists called species versus what
> COI grouped as species meant that the taxonomy was wrong (i.e., there
> were cryptic species the morphological taxonomists had completely
> overlooked).

Again, this is part of what I see as the problem of overpublicity causing
problems down the road.  Finding populations with widely divergent
sequences indicates places where there might be cryptic species, but you
need to investigate further to see what else correlates with it before you
can make the claim formally (AFAIK, no one has actually published new
species based solely on barcodes).  Incidentally, the existence of cryptic
species illustrates an area of failure for traditional taxonomy (often in
the assumption that minute morphological differences represent
intraspecies variation), but certainly nobody's giving up on that yet.

> It leaves us with a dilemma as far as any debate here: if everyone
> here claims to be a "good" barcoder, for whom COI is just one of an
> array of character sets used to determine species, but if those folks
> acknowledge that there are "bad" barcoders that believe in using (or
> trusting) COI alone, then we may actually be condemning a common
> enemy, and should be acting as allies rather than adversaries.
> So, let's sit back one second, take a breather, and ask two questions:
> (1) Can we agree upon a definition of "barcoding"? As far as I knew,
> it meant using COI, and ONLY COI in order to make species assignments.

It means doing that *once you have an established database of sequences
where it's known to work*.  So far that exists for a relatively small
number of taxa.  If it doesn't work, then other data sets need to be
investigated, including other genes.  But (despite my concerns in the
previous message) I think it is a good idea to have a consistent default
data set that can be used as widely as any one data set can.

> (2) Can we agree that there ARE researchers who use COI only, or give
> COI results *greater weight* than taxonomy?

Yes, but those people seems to be mainly making bold pronouncements about
the existence of thousands of cryptic species rather than actually
describing species based on COI.  Perhaps I'm being too trusting in the
scientific process, but I think that as it evolves and we understand more
about how well and where it works and fails, those people will be weeded
out.  As Rich Pyle pointed out, even some of the people who made those
early bold pronouncements have realized that it's only a piece in the

> If so, can we then treat the debate here as a matter of whether THAT
> approach is worthy of support, or should be put aside in favor of
> genuinely *integrated* studies?

This seems to be a somewhat simplistic way of framing it.  I think it's
worth investigating whether COI works, but that absolutely doesn't mean we
should close ourselves off to other alternatives.  And part of that is
integrating morphology and multiple DNA data sets to make sure we've
delineated species boundaries correctly, at least to the extent that it's
possible (that of course get into another whole issue).

Karl Magnacca, UC-Berkeley
ESPM Dept., 137 Mulford Hall #3114

More information about the Taxacom mailing list