Made in the USA - Freedom Breeder
News & Events:
Posted by WW on December 12, 2002 at 12:33:59:
In Reply to: Re: mtDNA - pros and cons posted by troy h on December 12, 2002 at 11:41:45:
::You certainly can, but only with some provisos. The problem is that different markers, *especially* different DNA sequences, may well have different histories. There is a large literature on the problems of combining different sources of evidence in one analysis.
:as a species may well have different histories, i would think that this would be an advantage in revealing the true history of a species evolutionary history and relationships.
Revealing that a species' genome has more than one history is very interesting. However, simply lining up all the sequences end-to-end and runing PAUP will not give you anything particularly useful.
: using a gene that has only one history, therefore, would seem to me more misleading, providing a researcher with the illusion of confidence that his tree represents a "true" history of that species or group of species.
Correct - the difference between gene tree and species tree.
::: seems to me that sequence data for several nuclear genes would only give you more characters that, because they may have been inherited differently, would give you a much more complete history for that population or species.
::Actually, they will give you a mish-mash, unless you go for gene tree parsimony approaches. For instance, if you wanted to trace the phylogeny of a set of species using a nuclear and a mitochondrialmarker, and one of the nuclear marker had been "injected" into the gene pool by a hybridisation event, then it would have a totally different history. Combining its sequence with that of another gene which represented organismal history and simply analysing everything in a total evidence approach would not make sense.
:i can see where this might be a problem, but one that would be overcome by adding more evidence (more sequences). furthermore, this also points to problems mentioned above - if a gene enters due to a hybridization event, then don't we want to know that? doesn't that evidence give us a more exact representation of actual history of that species or group than a character like mtDNA which you state gives a better "organismal history"?
Considering the phylogenies of the different genes separately will certainly give us a better insight. Mroeover, there are approaches which allow you to reconstruct the species tree from multiple gene, different trees derived from multiple genes (gene tree parsimony). However, simply lining up the sequences end to end will obscure the differences bewteen the genes, and the final tree will simply show which particular set of gene histories was best represented.
For example, if you used a 250 bp section of cytb and a 1000 bp section of a nuclear gene (just to give an example), and the genes had differnet histories, then the final tree would be most heavily influenced by the nuclear gene (assuming similar levels of divergence and %ages of inforamtive characters), and would represen the history of that gene, with cytb just adding noise. On the other hand, if you used 1000 bp of cytb and 250 bp of the nuclear gene, than the final "total evidence" tree would represent the mitochondrial gene phylogeny, with the nuclear gene just adding noise. A total evidence approach is inappropriate where different genes have different histories.
:also, this points to another problem with sequence data in general - are these changes in gene sequences really independent characters? if a sequence is inherited intact, and each sequence has its own history - is a sequence really a bunch of independent characters or a single character?
As far as independence is concerned, it acts as s single character, with a very high number of different character states. Of course, the information content of a sequence is huge, when compared with a simple presence/absence character!
::Thelatter is certainly true. However, the point is that mixing multiple markers with potentially different histories will not give you the correct organismal tree, it will give you nonsense.
:i don't see how this is the case ... papers i read (again i'll cite reeder & weins) suggested that "whole data trees" gave the best phylogenies. others that i've seen (some of JACampbell's stuff) also concur.
I have not seen the Reeder/Wiens paper. If morphological or similar data come into it, then we are dealing with a whole new ball game, since morphological characters don't have a built in set of phylgoenetic infgormation each, unlike a sequence. The point is, one should not uncritically combine conflicting gene sequences without testing whether they ahve different histories.
:one proviso i'd like to see in any single-gene paper is "based on the single gene i sequenced" the phylogeny is. of course, the next guy sequences a different gene and gets a different phylogeny because his gene has a different history. then we get to decide who is closer to true?
Depends - if the sequences are mtiochondrial, then we can of course simply combine them, since they are a single-linkage unit.
: that's why we have parsimony . . . combine data, get the closest answer based on all gene histories and all evidence - not just evidence of a single gene.
That's not parsimony, that is total evidence. Yes, of course we should look for evidence from as wide a set of markers as possible. However, where different genes do have different histories, then fusing them together will give you a tree that represents the history of neither - not all that useful.