Thursday, March 12, 2015

Horizontal gene transfer into humans? I am not convinced. Full text of my comments to reporters here

Some news stories about a new paper claiming evidence for horizontal gene transfer into humans and other chordates. I got asked by many reporters about it and some used some of my email comments in their articles.

See for example

 Here is the full text of my responses:

"got asked by another reporter to comment on this

so - have seen the paper 
it is interesting .. but I am not overwhelmed by what they present in the paper itself. For example, the HAS story seems really incomplete as presented (e.g., the Figure showing the tree does not have all the HAS1, HAS2, HAS3 genes even though they imply they studied that). "

I have been looking through the supplemental information. I find it impossible to judge the quality of this paper without being able to see the alignments they used for each phylogenetic tree. I cannot find alignments for the trees even after going to their Figshare site with the trees. I therefore think there is not much to say about the paper until being able to see those. 
Without seeing the alignments I offer multiple alternative hypothesis for their findings
  1. They have identified genes for which they are unable to produce reasonable alingnments. Alignments are central to phylogenetic analysis and if their alignments are poor quality then the trees will show all sorts of anomalies that have nothing to do with phylogenetic history. By scanning through 1000s of genes and flagging those with unusual patterns they may be selectively identifying genes for which producing good alignments between species is tough. I note - clustalw is a bit notorious for not producing idea alignments in some cases.
  2. I do not buy their arguments for why gene loss is not a possible explanation. They need to present more detail on how many gene losses would be required for each gene family under consideration and then present some evidence for why that # of gene losses is less likely than HGT.
  3. They have not even considered as far as I can tell, the possibility of divergent evolution (as opposed to gene loss) in many taxa which could lead to them being unable to identify homologs in some species
  4. I am not convinced by the arguments against long branch attraction as an explanation for some of the tree patterns.
  5. Related to alignments they need to show which regions of alignments they excluded from phylo
  6. Convergent evolution could also explain some of the patterns they observe.
  7. I could go on. I am NOT saying that HGT into chordates is impossible. It seems plausible. But it is up to them to exclude other MORE plausible alternatives and I just do not think they have done that.

Reporter: asking if it was OK to quote me

Yes it is OK to quote from me. I would like to reiterate - I am not saying they are wrong. Just that I would like to see (1) all the data (e.g., alignments) that unreels their conclusions and (2) them do more to exclude other possibilities.

Reporter asking what other analyses could they do
So - I don't want to be difficult, but it is their job to figure out how to do such tests before claiming they have strong evidence for HGT. 
In general, this is pretty typical of claims of HGT. Many researchers show evidence that is consistent with the occurence of HGT (which they did here) but few actually explicitly test alternative hypotheses such as gene loss, bad alignments, convergence, divergence, contamination, random noise, and more. I think their work is certainly interesting, but they just have not tested all of these alternatives. And I personally have grown a bit tired of pointing out how people can do better controls for their papers.

Reporter asking about initial impressions:
I see little here that is particularly convincing evidence for HGT ...

My follow up email
Note - I am not saying that this is a bad paper -- just that I am not overwhelmed by their evidence and especially by what they put in the paper. 
For example, the HAS1 gene story seems incomplete.  Figure 3 seems to show just HAS1 but in the text the say they show the same thing for all HAS genes.  And the tree they show shows a tiny subset of all the available sequences (e.g., HAS1 HAS2 HAS3 and fungal and bacterial homologs).  They claim that they now have proof that HAS1 was transferred near the base of chordates but I just don't see how they tested alternative hypotheses ...

Some related links:
Also here are some presentations from many years ago with some discussion of HGT


  1. Thanks for adding your comments to the paper making headlines this week. This question is certainly far from answered which leads me to question the authors' choice or wording, "Expression of multiple horizontally acquired genes is a hallmark of both vertebrate and invertebrate genomes." They seem to portray the topic as conclusively settled and is misleading.

    I have only quickly scanned the article so far, but I wonder if they address or disregard the transfer of mitochondrial genes within these species. Upon first glance, I see no mention of it.

  2. So it's then your stance that the data is incomplete?

    Casey Luskin over at the Discovery Institute seems to think otherwise: