June 15, 2015 Leave a comment
Figure 1 from Ebel et al. “Adelpha wing pattern and species diversity. (a) Examples of the nine Adelpha mimicry types. The number above each image indicates the number of species and subspecies with the pattern. From top left: A. iphiclus iphiclus, A. naxia naxia, A. thesprotia, A. cocala cocala, A. salmoneus colada, A. boreas boreas, A justina justina, A. zina zina, A. levona, A. rothschildi, A. epione agilla, A. lycorias wallisii, A. ethelda ethelda, A. leuceria juanna, A. gelania gelania, A. seriphia barcanti, A. mesentina mesentina, A. melona deborah. (b) Five species have a unique wing pattern. From left: A. seriphia egregia, A. demialba demialba, A. justina inesae, A. zina pyrczi, A. lycorias lara. (c) Adelpha species richness across the Neotropical region (modified with permission from Mullen et al. 2011).”
This was a funny choice from Will as it seems much more up my street than his. Indeed, my colleague James Nicholls and co. are developing similar phylogenomic methodologies to look at rapid diversification within Inga. James uses a targeted sequence approach that seems to also have worked pretty well. But I am too lazy to make this a post about the pros and cons of different genomic techniques.
In fact I’m not sure what to make this post about. I’m not sure what I think about this paper. On the one hand, it clearly represents an amazing amount of work – processing the samples, doing the bioinformatics and the bazillion versions of phylogenetic reconstruction and the assessment of missing data effects. I could not find any details in the main text, but they also appear to have dated the tree (and apparently better than previous attempts). And then they find neat relationships between toxicity of a common host plant family and convergent mimicry patterns across a variety of species. It’s a really nice story.
On the other hand, I am still not 100% convinced by the robustness of the various available methods of character state reconstruction (not discussed in the text) or of diversification rate shift detection (discussed a little bit). No doubt about it, a better phylogenetic hypothesis helps make these tests more meaningful, but they still rely heavily on accurate dating (at least relatively) and on some degree of rate conservatism across the tree(s) – or else you might infer different rates on every branch.
Grumble grumble. I am consistently amazed that these methods, that seem so dodgy, often recover relationships that make ecological sense (as here). So I should probably stop complaining and concede that they might be recovering real patterns (at least now that the phylogenetic hypotheses are more robust and the effects of missing data or rubbish dating priors are better characterized).
So where to next? Should the focus be on improving these methods so it is easier to detect real patterns, should it be on collecting more data for interesting clades to fill in missing data (species, traits) or should it be on collating multiple such datasets to look for concordant patterns across clades? For instance, here, what are the plants doing? To answer that we need to consider the interaction of multiple plant families with a single butterfly genus. How do we do that?
And what are the limits of these macro-methods? This butterfly clade appears to be a recent and rapid radiation. How do we look at character evolution and predictors of rate shifts when species might still be hybridizing? Is there real scope to link population and phylogenomics? How quickly will technology and bioinformatics pipelines progress in order to use complete genomes (and tons of them) to answer such questions. My gut feeling is actually not that fast and that the next ten years or so are going to see a flurry of methods to continue asking these questions with dodgy, patchy data and then, in 10 years, we will have to start over when we are confronted with a whole different kind of dataset requiring a whole different set of techniques.
As ever, exciting times.
Sorry this post is so late; entirely my fault, not Lynsey’s. This feels like an excellent paper to get back on the horse with, because (as all phylogenetics papers do these days), it makes me feel very old. I feel as if I just popped out for a packet of cigarettes and suddenly the whole world changed – pyRAD? Is that like… PAUP*? What year is it?
And yet, somehow, the problems are the same. We have thousands of loci, but we have to concatenate them. We have a wonderfully resolved tree, but we still have to use the same old ancestral state reconstructions. I’m not criticising this paper, which is excellent, and I’m not even sure I’m criticising the methods, which are the best we can do at present. Yet, somehow, I still feel a bit worried whenever I see any sort of reconstruction – even (especially?) when I’m doing it myself.
Which is why it’s so nice to see methods being applied with care, and in such great diversity, to a system with strong a priori hypotheses. Above Lynsey asks whether we need new models or more data. It’s easy to look at the grey branches (“we don’t know”) on these phylogenies and think that we need more data, but in reality I think some of that would be gap-filling. The data we have has already cost a lot of blood and sweat, and is beyond a phylogeneticist’s wildest dreams a decade ago – is that not enough? What we need are explicit models that tell us what an adaptive radiation looks like. That does mean a lot more sweat, and it could mean even more data collection than these authors have already done, but without it, we’ll have no broad framework within which to place data-rich case studies such as these.
*in this case, the asterisk stands for “absolutely not”.