Our study includes 23 specimens of giraffe, representing all the nine subspecies of G. camelopardalis defined in Dagg and Foster [2], except the Thornicroft's giraffe (thornicrofti) from Zambia and the Nubian giraffe (camelopardalis) from eastern Sudan and western Eritrea (Table 1). The outgroup contains three species belonging to three different families of the suborder Ruminantia: Bos grunniens (Bovidae), Muntiacus muntjak (Cervidae), and Okapia johnstoni (Giraffidae), the closest living relative of G. camelopardalis.

Skin and blood samples were digested in CTAB (hexade Cyl trimethylammonium bromide) using the protocol detailed in Winnepenninckx et al. [7]. DNA was extracted from faecal samples using the method described in Porteous et al. [8]. DNA was purified in chloroform isoamyl alcohol and then precipitated with ethanol.

The mitochondrial DNA fragment selected for this study includes the 3′ end of the Glu-tRNA gene, the complete cytochrome b (Cyb) gene, the complete genes for Thr- and Pro-tRNAs, and the 5′ part of the control region, also named D-loop in mammals. The sequences were obtained using several overlapping PCR amplifications. The exact matching between the overlapping portions of two different PCR fragments has been checked as a proof of authenticity of sequences. Most primers come from previous publications on the Cyb gene [9,10], but two new primers were specifically designed to amplify the D-loop region: 5′-CAT CGG ACA ACT AGC ATC TAT-3′ (direct, position 15222 in the sequence of Muntiacus muntjak, accession number NC_004563) and 5′-CCA GAT GTC TGA TAA AGT TCA-3′ (reverse, position 15882). Amplifications were done in 50 μl, using the following PCR standard conditions: buffer 10X with MgCl₂: 5 μl, dNTP: 5 μl (6.6 mM), Taq DNA polymerase (QBiogen, Illkirch, France): 0.3 μl (2.5 U), and primers: 2.5 μl at 10 μM. The standard PCR conditions used were: 94 °C for 4 min; 94 °C for 1 min, 50–58 °C for 1 min, 72 °C for 1 min (30 cycles). PCR products were sequenced by Genoscreen (Lille, France) with the BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Foster City, CA).

The DNA sequences were aligned by eye using SeAl v2.0a11 [11]. One ambiguous region, i.e., involving ambiguity for the position of the gaps, was excluded from the analyses to avoid erroneous hypotheses of primary homology: positions 15531–15603 in the sequence of Muntiacus muntjak (accession number NC_004563). The reduced alignment of mt sequences consists of 1689 nt. It is available upon request to AH.

Phylogenetic analyses were performed using Maximum Parsimony (MP), Maximum Likelihood (ML), Neighbour-Joining (NJ), and Bayesian methods. The MP analyses were conducted on PAUP 3.1.1 [12] with differential weighting of the character-state transformations using the product CI_ex S (CI_ex: consistency index excluding uninformative characters, S: slope of saturation) [9,13]. Bootstrap percentages (BPMP) were computed after 1000 replicates using the closest stepwise addition option. ML and NJ analyses were carried out under PAUP 4.0b10 [14], and Bootstrap proportions (BPML and BPNJ) were obtained after 1000 replicates. MrModeltest 2.2 [15] was used for choosing the model of DNA substitution that best fits the data. The selected likelihood model was the Hasegawa–Kishino–Yano model [16] with among-site substitution rate heterogeneity described by a gamma distribution (HKY + G). Bayesian posterior probabilities (PP) were calculated on MrBayes 3.1.2 [17] using five independent Markov chains run for 2 000 000 Metropolis-coupled MCMC generations, with tree sampling every 100 generations and a burn-in of 2000 trees.

A total of 12 mitochondrial haplotypes (mitotypes) was identified among the 23 samples. Sequences of these haplotypes have been deposited in the EMBL/GenBank/DDBJ database (accession numbers EF442263-EF442274). In Fig. 2, haplotypes are coded by geographical locality or zoological park with individual identification numbers. In general, giraffes of the same zoological park share the same haplotype (Basel, Sigean, and Vincennes). However, two mitotypes were found in the Lisbon zoo (Nos. 7382 and 6751), which differ by only one nucleotide (Fig. 2; transition C-T in the D-loop, position 1566 of the alignment).

Two haplotypes were found for the subspecies antiquorum: the first one diagnoses two wild giraffes of Zakouma (Chad, ZAK 1 and 2), and the second one characterizes the maternal lineage of the Antwerp zoo (specimens named Sarah and Valère; the latter is a hybrid produced from crossing a female antiquorum from the Antwerp zoo with a male peralta from the Vincennes zoo).

Three haplotypes were found for the subspecies peralta: the first one diagnoses all the wild giraffes of Niger (PER1-3, and DOSSO); the second one defines the three wild giraffes of northern Cameroon (BOUBA, WAZA1, and 2); the third one characterizes the giraffes of the Vincennes zoo (Uma and Rafiki).

The 12 haplotypes can be classified into three distinct lineages, i.e. northern giraffe, Angolan giraffe, and the southeastern group, composed of the subspecies giraffa and tippelskirchi, as their sequence divergences range from 3.1 to 4.4% for the total mitochondrial fragment (1765 nt), from 2.5 to 3.8% for Cyb (1140 nt), and from 3.9 to 6.9% for the 5′ region of the D-loop (464 nt). The northern group includes the populations previously included in the subspecies peralta, antiquorum, rothschildi, and reticulata.

Out of 1689 total unambiguous characters kept for the phylogenetic analyses, 1239 were constant, 207 were variable but parsimony-uninformative and 243 were informative. When the three distant outgroup species were excluded from the analyses, only 119 characters were found variable between populations of G. camelopardalis, including 87 informative characters (see details in Fig. 2).

The four methods of tree reconstruction (MP, ML, NJ, and the Bayesian inference) show that G. camelopardalis is a monophyletic species (Fig. 3; BPMP/ML/NJ=100, PP=1). The three groups of giraffe haplotypes are supported by high values of robustness and several molecular signatures diagnose each group. The two haplotypes of Angolan giraffe are grouped together (BPMP/ML/NJ=100/89/100; PP=0.98), and share six exclusive synapomorphies, including four transitions A→G (pos. 4, 1514, 1623, and 1742), and two transitions T→C (pos. 1606 and 1677). The subspecies giraffa and tippelskirchi are grouped together (BPMP/ML/NJ=98/62/100; PP=0.78), and have in common three exclusive transitions (pos. 3: C→T, pos. 836: A→G, and pos. 1763: G→A). All giraffes of the northern group fall into a well-supported clade (BPMP/ML/NJ=99/88/100; PP=0.98) and share seven exclusive synapomorphies, including three transitions A→G (pos. 209, 356, and 716), three transitions C→T (pos. 592, 785, and 1190), and one transition T→C (pos. 1559). In this group, the subspecies antiquorum and peralta are not found to be monophyletic. The subspecies antiquorum is paraphyletic, as the giraffe of the Antwerp zoo is not directly related to the giraffe of Zakouma, but rather is allied to the giraffes peralta of northern Cameroon (Boubandjida and Waza National Parks) and to those held captive in the Vincennes zoo (BPMP/ML/NJ=66/61/78; PP=0.87). In addition, the subspecies peralta is polyphyletic, as the giraffe peralta from Niger is not grouped with other peralta from Cameroon and Vincennes, but rather is linked to the eastern giraffes of the subspecies rothschildi and reticulata (BPMP/ML/NJ=55/64/55; PP=0.97). Within the northern group, only one clade appears strongly supported: the one joining the wild giraffes of Cameroon and Chad with the captive giraffes held in the Vincennes and Antwerp zoos (BPMP/ML/NJ=97/96/100; PP=1). This clade is diagnosed by two exclusive synapomorphies, including one transition C→T (pos. 272) and one transversion C→A (pos. 1320).

The MP and NJ trees are identical to those reconstructed with ML and Bayesian methods, except for the position of angolensis. In the former trees, Angolan giraffes are grouped with other southern giraffes, i.e. the subspecies giraffa and tippelskirchi (BPMP/NJ=66/74), but in the latter trees, they are allied to northern giraffes, i.e. the clade including the subspecies antiquorum, peralta, reticulata, and rothschildi (data not shown, PP=0.55, BPML=55).

The status of the northern giraffes has always been controversial. In Lydekker [18], the reticulated giraffe, from northeastern Kenya, Somalia and southern Ethiopia, was treated as a separate species, G. reticulata, whereas other northern populations were classified into five subspecies of G. camelopardalis: antiquorum (Kordofan, central Sudan), cottoni (northern Uganda), peralta (Nigeria in the neighbourhood of Lokoja, where the Niger and Benue rivers meet), typica (formerly Nubia and Abyssinia, i.e. south of Egypt, northern and eastern regions of Sudan, Eritrea and north of Ethiopia) and rothschildi (western Kenya and eastern Uganda).

Most of these subspecies were adopted by Dagg and Foster [2], but the reticulated giraffe was defined as a subspecies of G. camelopardalis (reticulata), the subspecies typica was regarded as a synonym of camelopardalis and the subspecies cottoni was included in rothschildi. In addition, the former distribution of peralta was supposed to cover most countries of West and central Africa, including Senegal, southern Mauritania, northern Guinea, southern Mali, Burkina Faso, the North of Ghana, Togo and Benin, southern Niger, Nigeria, Cameroon, southern Chad, and the western part of the Central African Republic (Fig. 1B).

The view proposed by Kingdon [1,4] differed in three major points: (i) the subspecies antiquorum was discarded, as all populations inhabiting the Sahelian zone, from Senegal to Sudan, were included in the subspecies peralta; (ii) the woodland populations from Cameroon to Uganda were put in the subspecies congoensis; and (iii) possible hybrid populations were described in Cameroon between congoensis and peralta, and in Kenya between congoensis and reticulata. As a consequence, rothschildi was not regarded as a valid subspecies.

As pointed out by East [5], considerable uncertainty surrounds the validity and geographical limits of most described subspecies of giraffe. For instance, there is no major geographical barrier separating the distributions of supposed subspecies of northern giraffe, such as peralta and antiquorum in central Africa, camelopardalis and rothschildi in southern Sudan, or reticulata and rothschildi in central Kenya. As a consequence, East [5] recognized only three arbitrary groups of northern giraffe in the IUCN classification: the western giraffe, including peralta, antiquorum and congoensis, the Nubian/Rothschild's giraffe, including camelopardalis and rothschildi, and the reticulated giraffe (subspecies reticulata).

Our molecular data show that all previous classifications are invalid. Within the northern clade, the giraffes of the Vincennes zoo, which were referred to as peralta, are closely related to the peralta of Boubandjida and Waza National Parks (Cameroon), but they do not group with the peralta of Niger. However, they are included in a biogeographically-coherent clade, which contains giraffes living in three adjacent countries – Cameroon, Chad, and Sudan – as the giraffes of the Antwerp zoo were acquired from the Game Preservation Department of Sudan in 1948 and 1949 in the neighbourhood of Khartoum (B. Van Puijenbroeck, personal communication). Interestingly, this grouping agrees with their geographical origin, as the captive population of Vincennes grew from four wild animals originating from Cameroon (Zizi I, male), Chad (Lamy and Diamala II, males) and north of the Central African Republic (Negry, female). Since the mitochondrial genome is inherited by the maternal lineage, the mitotype of Vincennes characterizes one population of the Central African Republic. Consequently, our analyses of mitochondrial sequences suggest that two subspecies of giraffes can be distinguished in West and central Africa: the subspecies peralta, which consists of a very small population, representing less than 200 individuals, in the southwestern region of Niger, and the subspecies antiquorum, which comprises all populations found in northern Cameroon, southern Chad, north of the Central African Republic and southwestern and central Sudan.

The different subspecies of giraffe exhibit moderate nucleotide divergence of the mitochondrial sequences. By comparing the 5′ end of the Cyb gene (417 nt), we found a maximum distance of 4.56% between northern and southern subspecies (angolensis and peralta) and 1.44% within the northern clade (antiquorum and reticulata). These values are higher than that calculated for various breeds of domestic cattle (0.48%, Bos taurus taurus, 75 sequences extracted from complete mitochondrial genomes, data not shown), which have diversified after their domestication some 11 000 years ago [19], and lower than that found between southern and northern subspecies of hartebeest (5.76% between Alcelaphus buselaphus swaynei and A. b. caama), which are supposed to have diverged some 500 000 years ago [20]. Although estimates of divergence time based on the molecular clock are questionable, because the nucleotide substitution rates may vary extensively among mammals, these data suggest that the diversification of extant subspecies of giraffes took place during the Late Pleistocene. This result agrees with the fossil record, as the most ancient remains described for the species G. camelopardalis were dated at around 1 Myr [21].

The climate of the Pleistocene was characterized by cold, arid glacial periods oscillating with warmer, wetter interglacial periods, with associated expansion and contraction of savannah habitats [22]. Recurring isolations of populations into one or more favourable refugia in West, East and South Africa during wetter periods with range expansion events during drier periods have been postulated to explain the phylogeographical patterns in other ruminant species occupying savannah habitats, such as the greater kudu (Tragelaphus strepsiceros), hartebeest (Alcelaphus buselaphus), impala (Aepyceros melampus), roan antelope (Hippotragus equinus), topi (Damaliscus lunatus), and wildebeest (Connochaetes taurinus) [20,23–25]. Our phylogenetic analyses show that northern giraffes are divergent from two southern groups corresponding to the southwestern subspecies angolensis and to the southeastern clade uniting giraffa and tippelskirchi. The important nucleotide divergence between these three groups suggests that previously existing ecological or topographic barriers effectively prevented gene flow over extensive periods of time, leading to reproductive isolation of these lineages. As the giraffes are found predominantly in open or broken savannah habitats, we suggest that the existence of dense tropical forests, and also of vast deserts, may have acted as strong ecological barriers. Rivers, lakes and mountain ranges may have also served as effective barriers to giraffe dispersal. For this reason, the Great Rift Valley and lakes of East Africa may explain the important differentiation between northern and southern populations of giraffes. In southern Africa, there is a significant division between the subspecies angolensis and eastern populations. Since a similar phylogeographical pattern was previously obtained for bovids [23,24], we suggest that the Mega Kalahari sand sea [26] may have acted as a strong ecological barrier during the Late Pleistocene.

The phylogeny of northern giraffes shows that the Nigerian giraffe (subspecies peralta) is more closely related to the giraffes of East Africa (subspecies rothschildi and reticulata) than to those of central Africa (subspecies antiquorum). This result suggests that the current distributions of giraffes in Western and central Africa result from two independent biogeographical events and that a physical barrier prevents gene flow between peralta and antiquorum, given that both share the same latitudinal distribution in the Sahelian zone. The first biogeographical event may have led to the separation of giraffes of Central and East Africa, corresponding respectively to the subspecies antiquorum, and the common ancestor of the subspecies reticulata, rothschildi, and peralta. The Nile River and the Great Lakes of East Africa may have stopped gene flows between these two groups during most parts of the Pleistocene and Holocene epochs. A second, more recent biogeographical event was the dispersal of the ancestor of peralta from East to West Africa. Due to the earlier presence of antiquorum in central Africa, it may be inferred that the ancestor of peralta dispersed at first from East to North Africa and thereafter to West Africa. This hypothesis is confirmed by the many prehistoric rock paintings and engravings indicating that the giraffe was found all over the region now covered by the Sahara desert during the Holocene [27]. Around 6000 years ago, the development of hyperarid conditions resulted in the formation and extension of the Sahara desert [28,29], which may have forced the giraffes of West Africa to migrate in the southwards and to occupy their current Sahelian distribution. Several authors have proposed the existence of a Holocene giant lake in the central part of North Africa. Known as Lake Mega-Chad, it included the present-day Lake Chad, but extended over the Tibesti and Ennedi mountains in the North and East of Chad, respectively [30,31]. Since Lake Mega-Chad was at its maximum level between 7700 and 5500 years BP, and was still well developed between 3700 and 3000 years BP, we can assume that it acted as a natural physical barrier preventing communication between peralta and antiquorum populations. More recently, three biogeographical barriers may have limited the distribution of the Nigerian giraffe in West Africa: in the north, the southern limit of the Sahara desert, in the south, the Niger and Benue Rivers, together with the Upper Guinea rainforests extending from Guinea into Sierra Leone and eastward through Liberia, Ivory Coast, and Ghana into western Togo, and in the East, the forests and mountains (Mandara and Alantika) on the border between Nigeria and Cameroon [32].

Our genetic study on the mtDNA sequences is an important first step that can have a direct and important impact on conservation management. We show, firstly, that the giraffes of western and central Africa belong to two different subspecies, peralta and antiquorum, and secondly, that the giraffes in the Vincennes zoo, which were referred to as peralta, in fact belong to the subspecies antiquorum. This means that the subspecies peralta is today only represented in Niger, with a small population of around 160–180 giraffes. As no giraffe peralta exists in European zoos, no re-introduction programme could be conducted in the future to deal with the conservation of this endangered population. We recommend therefore that urgent and major measures be taken to protect the last giraffes of West Africa.