Francois Balloux has some scathing criticism on mitochondrial phylogeography as it is currently practiced (doi: 10.1038/hdy.2009.122). I recommend reading the whole thing. The beginning:
Let us assume I gave a seminar. I would tell the audience about my latest results on the population history of the pigmy shrew. My findings would be based on a stretch of DNA comprising several metabolic genes, showing no signs of genetic recombination. Armed with sequences from a large number of individuals sampled over a broad geographical area, I would make some inference on the colonization routes and times. To make life easier, I would restrict my analysis to the mutations I liked best, with nice names having been given to related sequences, rather than relying on dull mathematical quantities. As I reach one of the key conclusions of the lecture, which would go as follows: 'It is obvious from the distribution of haplotypes Amanda, Eugenie* and Hector_2 that the Outer Hebrides were colonised about 50,000 years ago, this was followed by considerable population fluctuations, a bottleneck during the last Ice Age, a swift recovery and a dramatic recent expansion over the last 200 years and...'. Imagine that, at that climactic stage I was interrupted by someone in the audience. The impertinent would say, 'Sir, can I just ask you whether this confidence in your conclusions may not be misplaced; your analysis is based on a single genetic marker, which comprises genes with a central role in metabolism and is thus likely to have been affected by natural selection'. An awkward silence may ensue, as I would find it difficult to dismiss this criticism easily.
and the end:
Despite mitochondrial sequence variation covarying with climate in humans (Balloux et al., 2009), there are better ways to measure temperature. And, I would argue there are also better genetic markers than mtDNA to infer past population history. I fully appreciate that mtDNA has given us some of the most fundamental results on human evolution at a time when using mtDNA was the only realistic option at hand. I do not question the value of mtDNA in forensics and pedigree reconstruction. It is also likely to remain a valuable tool for inference at a localized geographical scale, particularly when testing specific hypotheses rather than making quantitative inferences on the age or size of the populations studied. It is convenient to type and analyse, and its use in humans raises no serious ethical or societal issue. But all these qualities do not counterbalance the fact that a single locus likely to be under selection is inappropriate for population inference at large geographical scales (or over long periods of time in the context of ancient DNA analysis). We have reached an era in which publicly available data sets of large numbers of complete human genomes are a tangible prospect, and I believe it is now time to move on from the excessive reliance on uniparental markers. Exploiting these new resources of autosomal variation will present significant challenges, but it will not help overcoming them if a large fraction of the community of human population biologists persists in sticking to mtDNA as the marker of choice.
The utility of mtDNA for studying modern populations is indeed limited now that we can study hundreds of thousands of markers per individual. However, it is still a very useful marker for ancient DNA, both because it is often the only game in town because of the relative ease with which it can be typed due to its large copy count, and also because it has proven itself to be capable of generating interesting results, as in the recently discovered discontinuity between Paleolithic and Neolithic Central Europeans, studying the mtDNA diversity of Neandertals compared to humans, or detecting sex-biased gene flow in relatively recently admixed populations.
See some of my previous criticisms on facile correlations between mtDNA time depth and archaeological-historical correlations:
John Hawks also comments at length on the paper. An excerpt:
So what can we do? Fortunately we have lots of options. We can test the proposed demographic hypotheses against the historical record. When we make observations that show that people 1000 years ago had very different frequencies of common haplotypes, well, we know it was selection. There hasn't been any genetically significant bottleneck in the last 1000 years! When we see small Neolithic population samples dominated by haplotypes that are very rare today, again, no historically possible bottleneck could have caused that.I am fundamentally in agreement that bottlenecks, so often invoked in the mtDNA literature, are really a non-issue. Consider why this is the case: every mtDNA paper normally takes a random sample of a few tens or hundreds of people from a population that usually numbers in the thousands or millions. The assumption is that such a small random sample generally preserves -within confidence limits- the haplogroup frequencies in the population. But a bottleneck is exactly such a random sample. You can't, at the same time, use a sample of 100 people to infer haplogroup frequencies, and claim that a bottleneck that reduces the population to a 100 people will radically shift haplogroup frequencies. And, of course, there is absolutely no evidence that any major post-Neolithic human population, save for the Andaman Islanders, the Samaritans, or various such populations ever underwent a bottleneck anywhere near that severity.
However, I am in disagreement that a change of haplotype frequencies across 1,000 years is evidence of selection. A different explanation is that of migration, the introduction of a new population element.
Sometimes, migration is easy to infer. For example, we can be fairly certain that modern Europeans are different from Paleolithic Europeans because of Neolithic and post-Neolithic migration into Europe, because there is an introduction of new haplotypes that were absent in the Paleolithic population. One possible explanation is that instead of "absent" we should say "possibly present at very low frequencies". But, once we see that these haplotypes were present on the early Neolithic migrants, it doesn't take much to put 2+2 together and infer that migration is a likelier explanation.
The same process of migration could be inferred for the Neolithic populations of the Lake Baikal district, where a postulated hiatus in occupation, followed by recolonization by immigrants, proposed on archaeological grounds, coincides with the discovery of a sharp difference between pre- and post-hiatus populations in mtDNA haplotype frequencies. Similarly, the absence of Mongoloid mtDNA before the 7th c. BC in Central Asian samples, followed by its introduction after it, can be parsimoniously explained by admixture, since that admixture is evident also in anthropological and autosomal studies.
In other cases, selection may be a more plausible possibility. For example, the reduction in the frequency of haplogroup I in Denmark since the Viking and Iron Age, or changes of frequency in haplogroups in England since the 11th c. AD, such as the reduction of U5a1 and the increase in H may in fact be due to selection. H was present -although not very frequent- in Neolithic farmers from Central Europe, Corded Ware people from Eulau, and its very high present-day frequency in Europeans (roughly 50%) as there is no plausible source or mechanism that would have brought large numbers of it in Europe.
In conclusion, both migration and selection may help explain shifts in haplotype frequencies over time. As we plug in the holes in our knowledge of the mtDNA distribution across space and time, we will be able to decide between the two.