I have a multiple sequence alignment that I'm using to construct a phylogenetic tree. Usually, phylogenetic trees are constructed under the assumption that the input sequences are all from the present day -- thus, it forces them all to be leaf nodes.

However, my problem is a bit different. I have genome sequences collected at various dates from about 1960 to the present. Because of the extremely rapid rate of evolution (since I'm dealing with viruses), its likely that many of the older sequences are actually ancestors of the more recent ones. Is there any kind of phylogenetic structure that can represent this kind of relationship? Or can I deduce this from a standard phylogenetic tree?


In addition to what JTT wrote.

Your standard phylogenetic tree is unrooted and, more importantly, usually does not make the molecular clock assumption. That is fine, because most likely your old sequences are not literally the ancestral sequences, they are just very close to the ancestral sequences in question.

In other words, you are just fine with building a phylogenetic tree, as long as you don't force it to be a metric tree. This is a metric tree:

   |  |--------------

And this is not:

   |  |------------------

In the first one, it is assumed that the same amount of evolution happend between each leaf and the root of the tree. In the second, such assumption is not made. Maybe some branches were evolving slower; or maybe the sequences were collected at different times. In any case, you want the second one.

As JTT wrote, many examples of such analyses are very well known in literature. There is an especially nice example of such trees in this paper by Taubenberger et al., which analysed the phylogeny of the influenza virus -- and included even sequences from the 1918 outbreak (isolated from bodies of people who died of influenza in 1918 in the Brevig Mission).

  • $\begingroup$ Thanks for the detailed answer. You're right, even though we know that a certain strain was an ancestor of a pandemic strain (I'm dealing with influenza), and we have the sequence of one sample of that strain, that doesn't mean that exact specimen gave rise to the pandemic strain. I'm going to accept your answer, because I'm not sure whether to accept JTT's very good answer or yours, that builds on it. Please correct me if I'm wrong. $\endgroup$ – user1871183 Sep 30 '13 at 19:15
  • 1
    $\begingroup$ +1 for good points. It frankly did not even cross my mind to make a difference between the metric and the non-metric trees, since I was all the time thinking of not making the assumption of a molecular clock. $\endgroup$ – JTT Oct 1 '13 at 6:52

There are a few published examples of a similar problem where the aim has been, e.g, to infer a tree from the HIV transmission events. Furthermore, there are a few experimentally generated phylogenies, where the ancestral sequences might have been sampled, also. There's a list of these publication with the data at http://phylonetworks.blogspot.fi/p/datasets.html. All of them use the standard phylogenetic methology, and they draw the conclusions from the tree, where the potentially ancestral and more recent samples are included in the tree as leaves.

It is not simple to offer very good advice here, since e.g. the rate and type of population change in addition to all sampling related sources of variation will affect the shape of the resulting phylogenetic tree. In addition, the potential recombination event will affect the tree and the conclusion you will be able to draw from it. If you know that the virus you are doing research on is often involved in recombination, it might be a good idea to check the sequences for possible recombination events first.

Typically the ancestral sequences are clustered with its decendants in the same branch of the tree, which might be more easily intrepreted from a star-shaped or “unrooted” tree. If there are very many sequences in the tree, a “fisheye” visualization might help during the interpretation. It is also quite common for the sequences to form a ladder, if you put the first sample on the bottom of the tree (or use it as an “outgroup”). But as I said, the tree could assume very different shapes depending on your population(s) structure and sampling design.

For example, Pagan and Prank software allow insertion of sequences into an alignment using a certain position on the phylogenetic tree, which is quite close to what you're aiming at, but I do not now recall whether you could insert the ancestral states to the tree, or are they always inferred by the software. This type of solutions have really just started to appear during the recent years.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.