Despite much recent research, the archaeology of North Africa remains enigmatic, with questions of population continuity versus discontinuity taking centre-stage. Key issues concern the identity of the bearers of the Middle Palaeolithic (or Middle Stone Age) Aterian industry (the age of which has recently risen dramatically from ~40-20,000 years ago (ka) to ≥115-40 ka [1, 2]), and whether or not there was continuity between these and the later Upper Palaeolithic populations of the Maghreb. This question has become more urgent with the discovery that the Aterian is associated in Northwest Africa with a very early appearance of evidence for behavioural modernity, such as perforated Nassarius shell beads, use of ochre and bone tools, and long-distance exchange networks - preceding those of southern Africa and making it likely that the Aterian was carried by anatomically modern (rather than archaic) humans . The fate of the populations using this industry, and their possible connection with others in Africa and with the group who dispersed out of Africa ~60 ka to populate the rest of the world, has naturally become a question of great interest.
Further debates have focused on the question of population replacement in the late Pleistocene and Holocene. The earliest Upper Palaeolithic industry in North Africa is the Dabban, limited to Cyrenaïca (a likely glacial refuge area ) and most likely dating from ~36-50 ka to ~20 ka . Its similarities to Near Eastern Early Upper Palaeolithic industries have suggested an origin in the Levant , rather than locally in the Aterian, and similar blade industries have been found further south by 30 ka . The Dabban was replaced by the Eastern Oranian (or Eastern Iberomaurusian) in Cyrenaïca possibly as early as ~18 ka , by which time the Upper Palaeolithic had extended eastwards for the first time along the coastal belt of the Maghreb, with the Iberomaurusian (or Oranian), the origins of which are similarly controversial, especially as it appears earliest in the Northwest. It began at the Last Glacial Maximum (LGM), ~22-20 ka, with a period of intensification in the Late Glacial ~15-13 ka, although recent work suggests a possible underlying industry as early as ~26 ka [1, 6]. Similar debates concerning continuity versus replacement surround discussion of the widespread Epi-Palaeolithic Capsian industries, which saw expansion southwards from eastern Algeria into an increasingly humid Sahara in the early Holocene and sometimes associated with more gracile Mediterranean skeletal remains, and the subsequent emergence of the Neolithic [4, 7–10].
The phylogeographic analysis of human mitochondrial DNA (mtDNA) has the potential to address questions such as these . In particular, the mtDNA haplogroup U6 has a unique and highly distinctive distribution amongst human mitochondrial DNA lineages. It is found primarily in North Africa and the Canary Islands (albeit with secondary dispersals into Iberia and East Africa), with its highest frequency amongst Algerian Berbers (28%), and it has therefore been proposed to be linked to the ancestors of the indigenous Berber-speaking populations of North Africa [12–15]. Macaulay et al.  described U6 and its sister clade U5 as having evolved from a common ancestor in the Near East, approximately 50 ka; while U5 spread along the northern Mediterranean coast with the European Early Upper Palaeolithic, U6 dispersed along the southern coast, as far as Cyrenaïca, alongside the Dabban industry, ~40-50 ka, with a further expansion into Northwest Africa with the Iberomaurusian culture, ~22 ka. On this view, U6 evolved en route or within North Africa (as U5 evolved within Europe ), the presence of occasional derived U6 lineages in the Near East would signal more recent gene flow from North Africa .
These early studies were based only on information from hypervariable region I (HV-I) in the mtDNA control region and a small number of diagnostic RFLPs in the coding region, assayed in samples from several populations of North Africa and Iberia. In most of these regions, U6 frequencies are ~10% or less , and even appear absent from some Berber communities in Tunisia [18, 19]. The major sub-haplogroup U6a (characterized by control-region transitions from the ancestor of haplogroup U at nucleotide positions 16172, 16219 and 16278) is highly dispersed, occurring throughout North Africa (and at low levels in the Near East and Iberia), and a further nested subclade, U6a1 (characterized by an additional transition at position 16189), follows a similar distribution.
By contrast, U6b (characterized by variants at 16172-16219-16311) has a more limited range to the northwest of North Africa, the north of the Iberian Peninsula and, as a nested derivative (U6b1, characterized by a transition at position 16163), in the Canary Islands . In particular, the U6b1 lineage in the Canary Islands has been considered a founder lineage for the colonization of this archipelago by the Guanches (culturally very similar to Northwest Africans), ~2-3 ka , a conclusion supported by studies of ancient DNA . Hence its arrival suggests itself as a potentially useful calibration point for the mtDNA molecular clock, although the archaeological evidence for the colonisation time is rather insubstantial . In fact, the absence of U6b1 lineages anywhere outside of the Canary Islands (the few exceptions detected in Spain and in Americas being most readily explained as recent migrants from there), and the failure to detect immediate ancestors in North Africa, seem to point to the emergence of this clade within the Canary Islands - although probably soon after their colonization, as it is observed across several islands.
Maca-Meyer et al.  performed the first study of complete U6 mtDNA sequences (with 14 samples), defining a new U6 sub-haplogroup, U6c (characterized by the HV-I transition motif 16169-16172-16189), which was even more geographically restricted than U6b - limited to the west of North Africa and, as a derivative (U6c1, with an additional 16129 substitution), in the Canary archipelago. Using coding-region age estimates as maximum limits for radiation times, they proposed that the proto-U6 spread from the Near East to North Africa ~30 ka, alongside the Iberomaurusian industry, with U6a reflecting an African re-expansion from the Maghreb eastwards in Palaeolithic times, and U6a1 a further reverse movement from East Africa back to the Maghreb, possibly coinciding with the probable Afroasiatic linguistic expansion. The clades U6b and U6c, restricted to West Africa, had more localized expansions; they argued that U6b reached Iberia at the time of the diffusion of the Capsian culture in North Africa.
However, a larger study by Olivieri et al.  was closer to the earlier interpretation of Macaulay et al. . They confirmed the origin of U6, or at least that of its immediate ancestor, in southwest Asia, with an ancient introduction (alongside haplogroup M1, and the Dabban industry) to North Africa via the Levant, possibly during the Greenland Interstadial 12, from ~44-48 ka. They reaffirmed that the various U6 sub-groups originated in the southern Mediterranean area, dispersing subsequently to East Africa.
Coalescence time estimates for U6 and it subclades have varied considerably amongst these studies. Yet these are critical for studies of prehistoric dispersals, since reliable estimates can bracket the timing of demographically significant events. For example, a regionally-specific clade may have arisen from a migration event on the edge in the tree leading to that clade, and if the diversification has then arisen in situ rather than prior to the presumed founder event, the estimate of the time to the most recent common ancestor (TMRCA) can provide a minimum bound on the age of the migration event (motivating the "founder analysis" ). However, success rests on a number of requirements, principally that the phylogeny can be well-estimated and the molecular clock that converts genetic differences into time depth is well-calibrated . Considerable progress has recently been made on both of these fronts by more sophisticated analyses of the richer data source provided by mtDNA complete sequences.
With respect to the molecular clock, there are many factors leading to uncertainty. There is the wide variation in positional mutation rates and violations of the independence of mutations at different positions. Obvious regions are the paired stems of rRNAs and tRNAs , which some authors remove from the analysis , but there are other locations in the mtDNA molecule which can also present a secondary structure related with a functional role . There is the problem of multiple hits and saturation, leading to the curious observation that the total proportion of control-region polymorphisms in the African branches of the tree is lower than in the non-African ones. Selection is also an important issue, with a higher frequency of replacement substitutions in the younger branches of the human mtDNA phylogeny compared to the more internal branches [29–31]. Kivisild et al.  advocated the use of only synonymous diversity for estimation of the TMRCA, which is problematic for age estimations in young lineages, while Soares et al.  implemented a correction for the purifying selection effect on the mutation rate estimated for the entire molecule.
The choice of calibration points is also an important issue. Traditionally, an outgroup is used, where the split time with the human lineage can be assigned in some way. For humans, the closest one is that corresponding to the human-chimpanzee split, for which the fossil evidence is controversial and which is in a time frame very distant from the TMRCA of the mtDNA of Homo sapiens, rendering the application of a strict clock problematic . One recent analysis additionally used the chronometric ages of the available Neanderthal sequences as calibration points . A strategy of multiple calibration points in conjunction with relaxed-clock methods, where the rate is allowed to vary among branches in the tree  is appealing, but this has been hard to implement in the human tree because of unavailability of secure multiple calibration points. Bandelt et al.  advocate that calibrated radiocarbon dates in favourable pioneer-settlement situations with a well-defined founder mtDNA scenario and a rich archaeological record could be used for calibration purposes, but consensus for both radiocarbon dates and founder mtDNA lineages are far from being achieved in most known settlement situations. Endicott and Ho  applied an internal calibration to the human mtDNA tree by specifying priors on the ages of three nodes in the tree associated with demographic signals: the entry into Australia and New Guinea by establishing a minimum of 40 ka for haplogroup P; and the post-Last Glacial Maximum expansion of haplogroups H1 and H3 (unfortunately suggested as 18 ka). This internal calibration was performed using a Bayesian approach with the software BEAST , and resulted in a substitution rate 1.4 times higher than that resulting from the human-chimpanzee calibration. BEAST also, however, allows a reconstruction of effective population size through time, by using the Bayesian skyline plot (BSP ), based on a coalescent model analysed by Markov Chain Monte Carlo sampling. BSPs do not require a pre-specified parametric growth model as do other methods and, although designed for use with population data, they have also been used to attempt parameter estimation from haplogroup data with some apparent success .
Here, we analyse an additional 39 complete U6 genomes, from the full range of the U6 geographic distribution, including the Near East, Iberia and the Canary Islands. This has enabled us to construct a phylogenetic tree including 89 U6 genomes in total, and to re-evaluate the demographic history of the haplogroup, and its role in North African prehistory, in the light of the recent developments in the calibration of mutation rates. A comparison tree was inferred for 141 U5 sequences from the literature, allowing us to test the use of four alternative internal calibration points (the settlement of the Canary Islands, the settlement of Sardinia and its internal population re-expansion, and the split between haplogroups U5 and U6 around the time of the first modern human settlement of the Near East) against the recently developed complete genome clock with a correction for purifying selection.