The Caucasus, which is located on the southeastern boundary between Europe and Asia, is one of the areas through which pre-modern humans and anatomically modern Homo sapiens may have entered Europe from the Near East and Africa. Neanderthals invaded the region at an unknown point in time^{7, 8} and may have occupied the region alongside modern humans from 40,000 years before the present (B.P.). During the excavation of the Mezmaiskaya Cave², which is located in the northern Caucasus, a fragmentary skeleton of an infant was found that contained a set of morphological characteristics which indicated clear affinities to the Neanderthals of western and central Europe². Mitochondrial DNA analysis was undertaken using one of this Neanderthal's ribs.

The preservation of collagen-type debris was used as an indicator of macromolecule preservation in the bone. The amount of collagen-type debris extracted⁹ from 130 mg of the Mezmaiskaya Neanderthal rib fragment was 22% of the average level extracted from modern bones, and the extracted collagen contained 41.6% carbon and 14.7% nitrogen. This is within the values recovered from prehistoric samples displaying good preservation¹⁰. These data suggested that there were low levels of diagenetic modification. The high collagen yield made it possible to date the Neanderthal infant to 29,195 +/- 965 (Ua-14512) years B.P. by using a radiocarbon accelerator. This date does not agree with the previously published dates of more than 45,000 (Le-3841) and 40,660 +/- 1,600 (Le-3599) years for the Mousterian layers in the Mezmaiskaya Cave² with which the skeleton was associated. The most likely reason for this discrepancy is the incorrect identification of the poorly defined layers in this area of the cave. The value obtained from the bone itself rather than from associated material gives the most reliable date for this individual.

Two sections of one rib (90 mg and 123 mg) were used for DNA extraction in two independent laboratories. In the Glasgow laboratory, a total of 345 base pairs (bp) of HVRI was determined from two overlapping polymerase chain reaction (PCR) fragments with lengths of 232 and 256 bp. Forty PCR amplification cycles produced sufficient product to enable direct sequencing. Products from independent PCR amplifications were also cloned into a TA vector and sequenced.

The authenticity of the DNA sequence obtained in the Glasgow laboratory is supported by a number of factors. First, a section of the mtDNA was isolated and sequenced with congruent results in the Stockholm laboratory. Second, the PCR products were generated using Neanderthal-specific primer pairs that, under the amplification conditions, failed to amplify any fragments using modern DNA controls from individuals of different ethnic origins. Third, the retrieval of the sequence was not dependent on the primers used. Fourth, the low level of diagenetic modification indicated that the sample could theoretically contain amplifiable DNA. Last, and most convincingly, the sequence is similar to, and after phylogenetic analysis clusters strongly with, the previously analysed Neanderthal sequence⁴.

Comparison of the 345-bp fragment of HVRI with the Anderson reference sequence¹¹ and the Neanderthal from Feldhofer Cave⁴ revealed 22 differences (17 transitions, 4 transversions and 1 insertion) and 12 differences (11 transitions and 1 transversion), respectively. The Feldhofer Neanderthal HVRI contained 27 differences to Anderson reference sequence¹¹ (over the equivalent 345 bp⁴). The two Neanderthals share 19 substitutions relative to the reference sequence. The cloned PCR products contained all the substitutions that were detected by direct sequencing; six other non-reproducible substitutions occurred in seven different clones. No modern sequences were found in the Glasgow laboratory either by direct sequencing or by sequencing cloned PCR products. The Stockholm laboratory experienced problems with contamination: most of the cloned PCR products that they analysed contained sequences that are found in the modern human mtDNA pool, with two haplotypes predominant. However, three clones contained DNA that was the same as the sequence determined in Glasgow (two of these contained non-reproducible substitutions).

The preservation of 256-bp DNA fragments in bone that is 29,000 years old, that has not been preserved in permafrost and that contained sufficient DNA to enable direct DNA sequencing after amplification is unprecedented and may be attributed to specific features of the microenvironment of the limestone cavern². The retrieval of mtDNA showed a positive correlation to the preservation of collagen content and the skeletal morphology.

Phylogenetic analysis using both distance and parsimony optimizations places the two Neanderthal sequences together, in a distinct clade, basal to modern humans. Neighbour-joining analysis supports this separation ( Fig. 3a). Parsimony analysis, which makes minimal assumptions about the model of evolution and optimizes the fit between the tree and data, produced similar results (Fig. 3b).

The level of pairwise difference found between the two Neanderthals was higher than the average values found in random samples of 300 Caucasoids ( 5.28 +/- 2.24) and Mongoloids (6.27 +/- 2.29)—less than 1% of Caucasoid and Mongoloid pairs differ at 12 or more positions—but comparable to a random sample of 300 Africans (8.36 +/- 3.2), where 37% of pairs differed at 12 or more positions. When analysing ancient DNA there is the possibility of misincorporating nucleotides in the early stages of PCR, especially when the target DNA is possibly damaged and present in low copy number¹². As both Neanderthals were analysed in replicate and the results were consistent, however, errors of this type can be discounted.

The Feldhofer and Mezmaiskaya Neanderthals were separated geographically by over 2,500 km. Given that these two individuals contained closely related mtDNA, which is phylogenetically distinct from modern humans, and displays only a moderate level of sequence diversity compared with some primates¹³, these data provide further support for the hypothesis of a very low gene flow between the Neanderthals and modern humans. In particular, these data reduce the likelihood that Neanderthals contained enough mtDNA sequence diversity to encompass modern human diversity.

The 'out-of-Africa' hypothesis for the origin of modern humans predicts equal distances between the Neanderthal sequences and all modern sequences. We observed this in our analysis—the average pairwise differences between the Neanderthals and 300 randomly selected Africans, Mongoloids and Caucasoids were calculated to be 23.09 +/- 2.86, 23.27 +/- 4.06 and 25.45 +/- 3.27, respectively.

We estimated the age of the most recent common ancestor (MRCA) of the mtDNA of the eastern and western Neanderthals to be 151,000–352,000 years. This coincides with the time of emergence of the Neanderthal lineage in the palaeontological records¹⁴. The divergence of modern human and Neanderthal mtDNA was estimated to be between 365,000 and 853,000 years. Using the same model, we estimated the age of the earliest modern human divergences in mtDNA to be between 106,000 and 246,000 B.P.

The results obtained from this specimen suggest that some other Neanderthal samples may be amenable to molecular analysis. To obtain a more complete picture of the relationship of Neanderthals to modern humans, additional Neanderthals and early modern humans must be analysed, especially from the regions where they may have co-existed. The excellent preservation of this specimen leads to the potential of analysing the entire Neanderthal mitochondrial genome.

Methods
DNA extraction, PCR, cloning and sequencing The DNA extraction methods used in Glasgow4 and Stockholm¹⁵ have been described. We took precautions to prevent contamination from modern DNA⁴. The Neanderthal-specific primer pairs, NL16,055 and NH16,262, and NL16,209 and NH16,400 (5'-TGATTTCAC GGAGGATGGTGA-3') were used in Glasgow and the primers L16,212 (5'-ATGCTTAC AAGCAAGCACA-3') and H16,332 (5'-TTGACTGTAATGTGCTATG-3') were used in Stockholm. The annealing temperatures for the primer pairs NL16,055–NH16,262, NL16,209–NH16,400 and L16,212–H16,332 were 50 °C, 60 °C and 50 °C respectively; 40 cycles were used for the first two pairs, 55 cycles for the third. AmpliTaq Gold (Perkin Elmer Cetus) was used in all PCRs. PCR products were purified using the QIAquick Gel Extraction kit (Qiagen) before direct sequencing using the Dye Terminator sequencing kit (Perkin Elmer) or cloning into the TA vector (Invitrogen) before sequencing with the same kit using the M13 and T7 primers.

Sequence analyses The neighbour-joining and the maximum parsimony branch and bound trees were both constructed using PAUP* 4.0 (ref. 16). For the neighbour-joining analysis, the Tamura-Nei DNA substitution model¹⁷ was used with a gamma distribution of 0.4 (ref. 18), for all other parameters the defaults provided by PAUP* 4.0 were used. The MRCA was calculated using the described methods and assumptions⁵. PAUP* 4.0 was used to calculate pairwise differences between sequences: the data sets used for this were constructed by randomly selecting appropriate samples from a published data set¹⁹.