phylogenetic trees built from DNA/protein sequences use sequence differences between these biological sequences as proxies for relatedness.

More specifically, maximum-likelihood phylogenetic trees (which are arguably the most commonly used/most accurate way to build phylogenetic trees) assume an evolutionary model (i.e. a way in which sequences diverge) and try to learn parameters that best explain the sequence divergence we see in our DNA/protein sequences.

In my mind, everything is clear and straight forward if you assume neutral evolution and constant mutation rates. Under this model, sequences will simply randomly mutate at a constant rate, and the more different two sequences are, the more distantly related they will be (on average, at least).

Now, what happens if we assume non-neutral, i.e. positive or negative selection? Can maximum-likelihood trees detect this? How do they handle this? Can they? Should we even build phylogenetic trees on sequences that evolve non-neutrally? I cannot wrap my head around

  • $\begingroup$ Before neutral or non-neutral evolution you should care of site specific evolutionnary rate disccussed in the paper linked by Roger Vadim. $\endgroup$
    – reuns
    Nov 18 '21 at 14:13

Constructing a phylogeny requires a distance/similarity metrics for comparing sequences (in fact, constructing phylogeny can be viewed as hierarchical clustering). One however is not obliged to use evolutionary matrix for this - one can construct a phylogenetic tree using, e.g., Hamming distance and simple neighbor-joining algorithm - and this tree will be already valuable. See, e.g., Bilogical sequence analysis for introduction to the basic techniques.

Using sequence substitution matrices to define inter-sequence distances to construct phylogenies is groudned in the molecular clock hypothesis, according to which the substitution rate remains constant over time. The classical text here is Inferring phylogenies. The difference with other methods of constructing phylogenices is that the length of the tree branches can be interpreted as time, e.g., since the two sequences diverged. This is a strong assumption, but in many acses justified and producing useful results - e.g., this is how one could trace the early HIV evolution. There have been also attempts to improve the branch lengths using actual metadata for the sequence collection, as e.g., in this paper.

Substitution matrices are however rather flexible, in terms of the number of parameters they contain, particularly chen dealing with proteins (see Substitution model and PAM. Moreover, the substitution matrices can be site-specific, although this may result in overparametrization, known as infinitely many parameters problem, when teh results become unreliable. It is however fully reasonable to assume different substitution matrices for, e.g., synonimous and non-synonimous mutations. Here is a recent paper exploring this problem - I suppose the references in it could be a reasonable departing point for further study.

  • $\begingroup$ Thanks for your detailed answer Roger. Could I ask you to tell me your thoughts on the following: Assume you have an initial genetically uniform population of bacteria. You then separate them into two populations A and B. Population A you let evolve without constrains (all genes will accrue genetic variation over time), Population B you put under extreme positive selection, such that over time population B acquires less genetic variation (only some sites are allowed to diverge). How do you think the tree will look like? Won't the branch to A be too short? $\endgroup$
    – PejoPhylo
    Nov 18 '21 at 14:42
  • $\begingroup$ @PejoPhylo under strong selection the tme may flow "with different speed", taht is the substitution rates will be different. There is a so-called $d_N/d_S$ approach actually uses this to detect selection - the synonimous sites are assumed to undergo neutral evolution, while non-synonimous sites may be under a selection pressure - in this case the number of non-synonimous substitutions will be disproportionally big, which is taken as the evidence of selection. $\endgroup$ Nov 18 '21 at 16:04
  • $\begingroup$ Hi Roger, thanks for your further answers, so doesn't that mean that indeed the branch leading to A will be (mistakenly) shorter then the branch leading to B? $\endgroup$
    – PejoPhylo
    Nov 19 '21 at 6:43
  • $\begingroup$ @PejoPhylo A short comment: with only two sequences, one can only measure the distance between them. One needs more sequences to speak of "shorter" branches (usually one uses an outgroup to root and calibrate the tree). In your bacteria example the selection will likely act only on a few sites (e.g., those responsible for the response to antibiotics), the rest will evolve neutrally. But yes, you either need that the majority of the sites evolve neutrally or you only use neutral sites (certain genes or synonimous sites) to reconstruct the phylogeny. Otherwise the estimates may be quite off. $\endgroup$ Nov 19 '21 at 8:17

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.