Rieux: Improved calibration of the human mitochondrial clock using ancient genomes

[GailT]

Dienekes blogged (http://dienekes.blogspot.com/2014/08/new-estimates-of-human-mtdna-node-dates.html) on this new, open access mtDNA study: Improved calibration of the human mitochondrial clock using ancient genomes (http://mbe.oxfordjournals.org/content/early/2014/08/03/molbev.msu222.abstract)

They conclude that the mtDNA mutation rate and dates of major branches can be more accurately estimated using ancient mtDNA compared to assumptions about the dates of major demographic events or the human-chimp common ancestor.

The Rieux age estimates are similar to the Fu et al study, e.g., they estimate U and H to have ages of 53 and 34 thousand years, respectively. I'm still skeptical of the date for H as it seems incompatible with the diversity we see in modern day H. In contrast to Fu et al., Rieux et al. have included detailed info on the samples they used in their analysis so it should be possible to determine how they arrived at this date for H. Rieux et al. also included 3 ancient H samples. I'll post again after I've had time to analyze these samples.

I don't think Dienekes comments on M, N and L3 makes much sense. This paper reports a phylogeny based on only 350 mtDNA sequences (out of more than 20,000 available) so any diagram they present of the mtDNA phylotree will not represent the full diversity of the known phylotree. There are a large number of L3 subclades not represented in their diagram, so a phylotree based on this very limited dataset does not tell you anything at all about the relationship of M and N to L3, or the origin of L3.

[GailT]

Here are a few more details on this paper - Rieux have a very limited sample size of only 320 contemporary mtDNA samples. They have 24 H samples, mostly from an Achilli 2004 paper and some newly sequenced HGDP samples that have not yet appeared on GenBank. They have 18 U5 samples, but 8 of 18 are closely related U5b3a1a samples from Sardinia. So they have obviously not even attempted to select samples that can be used to date subclades of M and N. So this seems to be primarily an effort to date the ancient branching in the mtDNA phylotree, and the Behar et al 2012 study appears to be much more reliable for dating of younger haplogroups or subclades.

I'll take another look when they publish the actual data they used in their analysis.

[GailT]

Here are the mutation rate for the full sequence reported for each study:

Soares 1.665 x 10-8
Fu 2.67 x 10-8
Rieux 2.14 x 10-8

Behar used the Soares mutation rate.

Rieux et al. estimate younger ages for major branch points as a result of their faster mutation rate:

Using tip-calibration, we also estimated the coalescence dates of various nodes of interests in the tree. We obtained a value of 4.1 Mya [2.9-5.4 95% HPD] for the divergence between Hominins and chimpanzee.

Rieux offers a couple of explanations why this young date might be correct:

This estimation may appear too young when compared with the dates that are generally derived from the fossil record. The Toumaï fossil (Sahelanthropus tchadensis), dated to 6-7Myr (Brunet et al. 2002), is usually interpreted as being on the Hominin line, and setting a minimum date for the divergence (Vignaud et al. 2002). The apparent conflict between our tip-based estimation and the fossil record could be explained if Toumaï were somewhat younger than previously reported (Brunet et al. 2005) or if we assumed a more complex speciation scenarios where an initial split was followed by an extended period of gene flow before the final separation.

I looked at the 15 of the 24 H samples they used that are already published in GenBank, and the 15 have an average of 5.06 extra mutations. Using the Soares age calculator (that is based on a slowe mutation rate and thus would predict older ages) I get an age estimate of H of 13,500 years.

Rieux estimates 143,000 years for the TMRCA of all modern human mtDNA. Behar found on average about 52 mutations since the MRCA of all modern humans. If haplogroup H has an average of around 5 extra mutations, it seems very implausible that H could be 24% as old at the MRCA.

[GailT]

I also looked at the 3 ancient H samples that they used in their analysis. Given that the Rieux contemporary H samples seem consistent with Behar's much younger estimate of about 13,000 years for haplogroup H, perhaps there is something in the tip analysis of the three ancient H samples that gives them a much older date for H?

The 3 samples are from Bollongino et al, BLA7, BLA10 and BLA13 (H5, H1c3, and H5) and each was identified in Neolithic communities dated to about 3500 BC. Counting their extra mutations in addition to those that define H, BLA7 has 5, BLA10 has 4 and BLA13 has 2, for an average of 3.66. Using the Soares age calculator, the age of H using these 3 samples would be 9700 years, or adding the 5500 year age of the samples, about 15,200 year before present. The faster mutation rate of Rieux et al. would provide a younger age estimate than Soares. So it seems clear that their is nothing in the tip analysis of the ancient H samples that indicates a much older age for H. I'll have to wait for the HGDP samples, but so far there is nothing that would indicate an age for H that is older than Behar's estimate.