>> Thorsten Burmester writes:
>> > Hi there,
> > I have a question on the construction of phylogenetic trees with sequence
> > data using parsimony:
> > Is there theoretical basis why sequences that evolve faster than the
> > others tend to branch off "earlier" in a tree instead of joining their
> > actual "relatives"?
>> Yes. The theoritical reason was given by Joe Felsenstein in the 70's
> and 80's and called "Long branch attraction" (LBA). A discussion took
> place after Felsenstein's 1978 paper. Felsenstein argued that
> parsimony was inconsistant in case of unequal evolutionary rates among
> lineages unless rates are small enough, and that a maximum likelihood
> approach is better. Several cladists including Farris and Sober
I just thought that I'd mention that I doubt that the term "cladist"
would validly apply to Elliot Sober-- especially since in his
book "Reconstructing the past" he uses a maximum likelihood argument to
the parsimony method! I think his claim is that when rates of evolution
are low, the maximum parsimony solution is probably the solution of
I can see how one might view him as a cladist of sorts since he has
defended the use of parsimony (on the above principle).
His main argument with Joe (please correct me if I am wrong) was
that he believed that the following statement was not shown
to be true by Joe in 1978:
A- the parsimony method will work if and only if rates of evolution
It is true only if the other assumptions of his evolutionary
model were true.
What Sober believes (I think) is that
B- low rates of evolution are a sufficient condition to justify
the use of parsimony.
However it are not a necessary condition as implied by A.
> advocated for the use of parsimony anyway, considering Felsenstein's
> result as non-conclusive. This somewhat vehement debate undoubtedly
> raised our knowledge about how tree-building methods work. A few
> references are given below (maybe Joe has some more...).
>> Let me try to explain the LBA effect again. Suppose the actual tree is :
>> _________ sp1
> |_ sp2
> | |_ sp3
> __| |
> | |_ sp4
> |_________ O
>> This example includes a multifurcation, but you can imagine any resolution
> provided that the lengths of newly resolved branches are very short.
>> sp2, sp3 and sp4 evolve slowly: they are quite similar to their ancestor,
> and similar to each other. Therefore most characters will suit the following
>> O: ?
> sp1: ?
> sp2: A
> sp3: A
> sp4: A
>> where A denotes any character state.
>> Now, look at what kind of parsimony-informative characters can occur, assuming
> this scheme. A single one is allowed :
>> O: B
> sp1: B
> sp2: A
> sp3: A
> sp4: A
>> supporting a tree where sp1 branches off near O, namely as an outgroup to sp2,
> sp3 and sp4. This (homoplasic) character type can outnumber synapomorphies
> and make parsimony inconsistant if rates are rally different.
>> The Maximum Likelihood method is less (not) sensitive to this problem.
>> The long branch attraction effect may apply whatever the location of the root.
> Usually, the outgroup branch is a long one, so that long branch attraction becomes
> attraction toward the root.
Actually, I have a question on this point. The problem is usually cast
as unequal RATES of evolution will lead to long branch attraction. But
could it also be attributed, in some cases, to the combination of
unequal rates of evolution AND long elapsed time periods?
In your example above there is only one elevated RATE of evolution--
that on the lineage leading to species 1. It is being attracted to a
(leading to 0) which is undergoing the same rate of change as species
2-4. So in this case two long branches are being attracted but only one
is the result of an increased rate of evolution. The other is just long
and unbroken by virtue of it being a deep divergence and there being
a bias in the tree symmetry.
Am I correct in my inferences?
Andrew J. Roger