How To Demonstrate Long Branch Attraction

Question

[Arjen Ten Have]

In short: I am running a phylogeny of a class of peptidases, find some nice new insights but my tree is troubled by two what are most likely long branch attraction clades. Is there a way to systematically show whether a clade is subject to or the result of long branch attraction?

A bit more detailed: Running trees (rather than running a tree) involves many trees in order to show your tree is stable (I would not say correct). Stable in the sense that adding or removing sequences does not change the topology of your tree. This is also indicated by high bootstraps. Hence I have two clades that jump around and have relatively low bootstraps, in addition they force low bootstraps on neighbouring clades. The two clades show relatively long distances (to the root but also internally). Both clades have at least one sequence that is strange. All the hallmarks are there, I can model them but nevertheless, they are at least to say suboptimal for activity. All this makes me believe these clades are subject to Long Branch Attraction but I would like to really show this is the case. Is there like a generally accepted method for this?

Answer 1

[DG]

Not really Arjen, at least not in the sense that there is one perfect answer. What you have already done is a very good indicator of LBA occurring but there are a few other things you can/should do.

1) Demonstrate that the sequences for the taxa in question have elevated rates of evolution/are significantly more divergent

2) Systematically remove fast evolving sites from the alignment and show the changes in bootstrap support for alternative topologies

3) Perform a topology test (AU Test) using CONSEL to see if the suspected LBA tree is significantly different from a constrained "correct" topology with the misbehaving taxa constrained to be in a more reasonable clade/clades. If you haven't run this program before one thing to keep in mind is that you need to provide it lots of alternative topologies to get robust p-values. Generally it is best to provide the ML tree, alternative topologies you want to test, and lots of bootstrap trees.

[Arjen Ten Have]

Thanks, I already thought on determining entropy levels if subclades as an indication for clade variation. The CONSEL is new to me and makes sense. Good idea.

[DG]

CONSEL can be a little tricky to use if you aren't familiar with it, but you just need to start with site likelihoods for all your proposed trees. I am assuming you are using a maximum-likelihood program for this analysis in the first place. Can I ask which you are using?