First insights into past biodiversity of giraffes based on mitochondrial sequences from museum specimens

1,2,7 Institut de Systématique, Évolution, Biodiversité (ISYEB), Sorbonne Université, MNHN, CNRS, EPHE, UA, Muséum national d’histoire naturelle, 55 rue Buffon CP 51 75005 Paris, France. 3 5 chemin du bas d’Anville 17750 Etaules, France. 4 92210 Saint Cloud, France. 5 Deceased [18 May 2020]. Former address: Direction Générale Déléguée aux Musées, Jardins Botaniques et Zoologiques, Muséum national d’histoire naturelle, 57 rue Cuvier 75005 Paris, France. 6 Centre d’Ecologie et des Sciences de la Conservation, UMR 7204 MNHN CNRS-UPMC, Muséum national d’histoire naturelle 75005 Paris, France.

A mitochondrial fragment, covering the cytochrome b gene (Cytb), tRNA genes for Threonine and Proline, and the 5′ region of the control region (CR) (length = 1764 base pairs (bp)), was sequenced for giraff es of seven of the nine recognized subspecies (n = 23; G. c. camelopardalis and G. c. thornicrofti missing) in the molecular study of Hassanin et al. (2007). Their phylogenetic analyses have indicated that northern giraff es constitute a monophyletic group, distinct from that of southern giraff es. Brown et al. (2007) analysed two mitochondrial fragments (cytochrome b, tRNA genes for Threonine and Proline, and 5′ region of the CR; length = 1705 bp) and 14 nuclear microsatellite loci for many wild  Dagg 1962). B. At present (after Muller et al. 2018). The subspecies are distinguished by diff erent colours on both maps, whereby the assignment of colours for the nine currently recognized subspecies (B) was modifi ed from https://giraff econservation.org/giraff e-species/. The type locality for each subspecies is indicated by a triangle in map A and detailed in Table 1

Cape of Good Hope, South Africa
Camelopardalis capensis Lesson, 1842 Camelopardalis australis Swainson, 1835 Camelopardalis maculata Weinland, 1863 Giraff a camelopardalis wardi Lydekker, 1904 antiquorum Jardine 1835 SMF-498: unspec. type and SMF-497: paratype South of Darfour, Sudan Giraff a camelopardalis senaariensis Trouessart, 1898 Giraff a camelopardalis congoensis Lydekker, 1903 peralta  Knottnerus-Meyer, 1910 Giraff a reticulata nigrescens Lydekker, 1911 Giraff a camelopardalis australis Rhoads, 1896 rothschildi Lydekker 1903NHMUK-1903.1 Guasin-gisha Plateau east of Mount Elgon, Kenya Giraff a camelopardalis cottoni Lydekker, 1904 angolensis Lydekker 1903NHMUK-1939 Cunene River, Angola Giraff a camelopardalis infumata Noack, 1908 thornicrofti Lydekker 1911NHMUK-1910 Petauke district, Zambia - Table 1. Currently accepted giraff e subspecies (Muller et al. 2018) with their synonyms (modifi ed after Shorrocks 2016). PETZOLD A. et al., Past biodiversity of giraff es individuals (n = 266 and 381, respectively) from six subspecies (G. c. camelopardalis, G. c. antiquorum and G. c. thornicrofti missing). They suggested the existence of at least six species, corresponding to Giraff a angolensis, G. giraff a, G. peralta, G. reticulata, G. rothschildi and G. tippelskirchi. Bock et al. (2014) analysed a similar mitochondrial fragment of 102 giraff es (cytochrome b + 5′ region of the CR; length = 1555 bp) and proposed to synonymize the two subspecies G. c. thornicrofti and G. c. tippelskirchi, because they have very similar mitochondrial haplotypes. Fennessy et al. (2016) and Winter et al. (2018b) recognized four giraff e species based on the analyses of a mitochondrial fragment (cytochrome b + 5′ region of the CR; length = 1556 bp) and several nuclear introns (7 and 21, respectively): G. camelopardalis (northern giraff e), G. giraff a (southern giraff e), G. reticulata (reticulated giraff e) and G. tippelskirchi (Masai giraff e). More recently, Petzold & Hassanin (2020) re-examined all available molecular data using diff erent methods of species delimitation that rely on diff erent species concepts, i.e., multispecies coalescent (Genealogical species concept, Baum & Shaw 1995), population genetic (Genetic species concept, Baker & Bradley 2006) and phylogenetic (Phylogenetic species concept, Donoghue 1985) methods. Their results provided strong support for the distinction of three species only, corresponding to G. camelopardalis (including reticulata), G. giraff a and G. tippelskirchi. This species classifi cation is the one adopted herein. For convenience, we will use the following subspecies names: G. camelopardalis camelopardalis, G. c. antiquorum, G. c. congoensis, G. c. cottoni, G. c. peralta, G. c. reticulata, G. c. rothschildi, G. giraff a giraff a, G. g. angolensis, G. g. capensis, G. g. wardi, G. tippelskirchi tippelskirchi and G. t. thornicrofti. The present study provides further insights on the genetic diversity among giraff e subspecies. It is important to note that most subspecies previously identifi ed using phenotypic features were found monophyletic with mitochondrial data, i.e., G. c. antiquorum, G. c. peralta, G. c. rothschildi, G. g. giraff a, G. g. angolensis, and G. t. thornicrofti. This does not mean that subspecies are fully isolated taxa, as gene fl ow between subspecies can be maintained through dispersing males, at least occasionally. However, the philopatry of females is apparently at the origin of some morphological characteristics. In this perspective, we report herein the fi rst mitochondrial sequences for several key museum specimens. These include the famous giraff e Zarafa from Sennar, which is assumed to belong to the type population of G. c. camelopardalis, two giraff es from Abyssinia, several historical specimens collected by Levaillant and Delalande, which were described as G. g. capensis by Lesson (1842), as well as topotypes of the subspecies G. c. congoensis and G. c. cottoni, and two giraff es collected in Senegal during the 19 th century (see Material and methods for more details). Our three main objectives were: (1) to compare for the fi rst time past giraff e populations with current biodiversity; (2) to include in the phylogenetic analyses some subspecies described in former classifi cations in order to re-evaluate their taxonomic status; and (3) to estimate divergence times among giraff es in order to provide a more comprehensive phylogeographic scenario.

Museum specimens
Bone, tooth, faeces or skin samples were obtained from 41 giraff es, comprising samples from 19 museum specimens (Table 2, illustrated in Fig. 2, sample localities shown in Fig. 3). Among the most important samples are three specimens of the subspecies G. g. capensis (Lesson, 1842) collected by Levaillant (1797) and Delalande (1822), the fi rst southern taxon investigated by the scientifi c community, with populations formerly found north of the Orange River (Levaillant 1797; see Fig. 3), a topotype of the subspecies G. c. cottoni Lydekker, 1904 collected by Powell-Cotton in the Lado enclave (left bank of the White Nile) in northwest Uganda and South Sudan, fi ve specimens from the Haut-Uele Province in Democratic Republic of the Congo, which may be representatives of the type population of the subspecies G. c. congoensis described from the type locality Dungu (Lydekker 1903), two giraff es collected in Bakel (Senegal) by Girardin in 1830, which according to Dagg (1962) may represent the most western distribution point of G. c. peralta (Fig. 1A), and three specimens from the type locality of the nominate subspecies G. c. camelopardalis, i.e., the famous Zarafa from Sennar and two giraff es from Abyssinia (Ethiopia). The ʻGiraff e of Levaillantʼ, anonymous painting made in the late 18 th century and early 19 th century, exhibited in ʻhôtel de Magnyʼ, Jardin des Plantes in Paris (France). B. The ʻGiraff e from Sennaarʼ, representing a lithography of Zarafa (MNHN-1845-211) and the skull of a giraff e from the Cape region (Geoff roy Saint-Hilaire 1827). C. Drawing of the holotype of Giraff a camelopardalis congoensis Lydekker, 1903 (RMCA-452), housed in the Royal Museum of Central Africa, Tervuren (Belgium) (Lydekker 1904). D. Head drawings of the holotypes of G. c. cottoni Lydekker, 1904(NHMUK-1904 and G. g. wardi Lydekker, 1904(NHMUK-1903 (Lydekker 1914  A mitochondrial DNA (mtDNA) fragment, covering the complete cytochrome b gene, two tRNA genes (Threonine and Proline) and the 5′ part of the control region (D-Loop), was amplifi ed by polymerase chain reaction (PCR) using 15 primer sets, which generate overlapping fragments (detailed in Supplementary fi le 2). Amplifi cations were performed on a total volume of 19 μl, comprising 10 μl of SYBR® Green Supermix (BioRAD), 6.8 μl H 2 O, 0.6 μl of each primer (10 μM) and 1 μl of giraff e DNA per sample, using a CFX Connect Real-Time PCR Detection System thermocycler under the following conditions: 4 min at 94°C, followed by 94°C for 30 s, then 1 min at 45-55°C followed by 1 min at 72°C, ending with a single extension of 72°C for 7 min (40 cycles). The PCR products were sent to Eurofi ns Genomics (Ebersberg, Germany) for forward and reverse direction Sanger sequencing. The electropherograms were edited and assembled to a reference sequence using Sequencher ver. 5.1.
(http://www.genecodes.com). All mitochondrial sequences generated for this study were deposited in GenBank (see accession numbers in Table 2 and Supplementary fi le 1).

Haplotype network
The 41 newly generated sequences were aligned in Geneious R10 (Biomatters, Auckland, New Zealand) with published mitochondrial sequences available in NCBI. The fi nal alignment of 548 giraff e samples (length = 1742 bp) was used to construct a median-joining network (Bandelt et al. 1999) with PopArt 1.7 (Leigh & Bryant 2015). Since PopArt does not consider sites with missing data, the two regions corresponding to tRNA-Thr and tRNA-Pro genes, and the last 35 bp of the D-loop were removed from our alignment because they were not sequenced in most previous studies (Brown et al. 2007;Fennessy et al. 2013Fennessy et al. , 2016Bock et al. 2014;Winter et al. 2018aWinter et al. , 2018b.

Phylogenetic analyses based on mitochondrial and nuclear datasets
The nuclear DNA (

Multispecies coalescent analyses of the nuclear dataset
The phased alleles of the 21 introns of the nuDNA-78T dataset (see above) were analysed under BEAST ver. 2.4.8 (Bouckaert et al. 2014) to infer a multispecies coalescent (MSC) tree phylogeny. As in Petzold & Hassanin (2020), the two alleles of the 21 introns were assigned at the level of individuals,

Estimation of mitochondrial and nuclear divergence times
Divergence times among giraff e subspecies were estimated with the nuDNA-78T and mtDNA-91T datasets using the Bayesian approach implemented in BEAST ver. 1.8.4 (Drummond et al. 2012) under the GTR + G + I model and assuming a relaxed clock following a lognormal distribution. The Birth-Death speciation process was specifi ed as tree prior and the most recent common ancestor (MRCA) of Giraff a was used as a calibration point, set at 1.1 ± 0.1 Mya in agreement with the molecular study of Hassanin et al. (2012) and the fossil record (Harris 1991). Analyses were run with random starting seeds for 10 8 generations, sampling trees every 10 000 steps. Tracer ver. 1.7 (Rambaut et al. 2018) was used to visualise the posterior distribution to assess the convergence of model parameters (eff ective sample size (ESS) > 200). The chronograms were summarised in TreeAnnotator ver. 1.10 (Rambaut & Drummond 2007) discarding 25% as burn-in and subsequently displayed with Figtree ver. 1.4.4 (http://tree.bio.ed.ac.uk/software/).

Mitochondrial and nuclear pairwise distances
The 88 giraff e sequences of the mtDNA-91T dataset were analysed in PAUP* ver. 4.0b10 (Swoff ord 2003) to calculate nucleotide pairwise distances (p-distances) within and among haplogroups revealed in the haplotype network of Fig. 3. Nucleotide pairwise distances were also calculated for the 75 giraff e sequences of the nuDNA-78T dataset.

Haplotype network
The haplotype network built from the mtDNA sequences of 548 giraff es is shown in Fig. 3. The network reveals three divergent geographic haplogroups separated by more than 30 mutations, named northern (N), southeastern (SE) and southwestern (SW). The northern and southeastern haplogroups can be further divided into subgroups that are separated by at least 10 mutations.

Fig. 3.
Median-joining network of mitochondrial haplotypes. The network was constructed in PopART 1.7 (Leigh & Bryant 2015) based on the mitochondrial sequences of 548 giraff es. The number of mutations between haplotypes is indicated by perpendicular lines on the branches and is specifi ed if greater than 10. The size of the circles is proportional to the number of individuals sharing a certain haplotype with colours assigned by subspecies. The sample locations are indicated by triangles in the map and highlighted in bold capital letters for museum specimens. The subspecies marked with an asterisk represent formerly recognized subspecies, which were synonymized in recent classifi cations (e.g., Shorrocks 2016). Historical key specimens are highlighted by the respective abbreviation of the museum and the catalogue number. The southwestern (SW) haplogroup contains only giraff es from southwestern Africa, i.e., the 21 haplotypes of 95 individuals assigned to the subspecies G. g. angolensis and the three specimens of the former subspecies G. g. capensis collected by Levaillant (1797) and Delalande (1822) from the Cape Region: one represents a formerly unknown haplotype (MNHN-A7977), and the two others (MNHN-1896-45, MNHN-A10749) are identical with haplotypes of giraff es currently found in the Central Kalahari Game Reserve (Botswana).

Comparison between mtDNA and nuDNA phylogenetic trees
The Bayesian trees, ML trees and BEAST chronograms reconstructed from the mtDNA-91T and nuDNA-78T datasets are available in the Supplementary fi les 4, 5, 6, 7 and 8. Note that the northern haplogroups ʻReticulated IIʼ and ʻMasai IIʼ (Fig. 3) were excluded from the analyses because nuclear introns are not available. In Fig. 4, the chronogram inferred from the mtDNA-91T dataset (Fig. 4A) is compared to the MSC species-tree reconstructed from the nuDNA-78T dataset (Fig. 4B). The divergence times estimated with the nuDNA-78T dataset under BEAST were reported for the main nodes of the MSC species-tree.
The In the MSC species-tree (Fig. 4B), the subspecies G. c. rothschildi can be considered as monophyletic (PP *BEAST = 1) if three giraff es from western Ethiopia (ETH1-3), geographically assigned to G. c. camelopardalis in a previous study, are included into this taxon. Similarly, the subspecies G. c. antiquorum can be considered as monophyletic (PP *BEAST = 0.92) if two giraff es from Shambe NP -South Sudan (SNR1 and SNR2), previously assigned to G. c. camelopardalis, are included into this taxon. The MRCA of southern giraff es was dated at 645± 315 kya. This group can be further divided into two taxa corresponding to the species G. giraff a (PP Bayes / BEAST / *BEAST = 1; BP = 100; 14 ES) and G. tippelskirchi (PP Bayes / BEAST / *BEAST = 1; BP = 100; seven ES). The species G. giraff a includes individuals from the mitochondrial haplogroups ʻSWʼ and ʻSoutheast Africaʼ. The MRCA of G. giraff a was dated at 205 ± 125 kya. The species G. tippelskirchi contains individuals of the mitochondrial haplogroups ʻMasai Iʼ and ʻMasai IIʼ, which are found to be reciprocally monophyletic in the BEAST trees, but with low support (PP BEAST / *BEAST = 0.45 / 0.53 and PP BEAST / *BEAST = 0.74 / 0.53, respectively). The MRCA of G. tippelskirchi was dated at 120 ± 90 kya.

Mitochondrial and nuclear pairwise distances
The mtDNA-91T and nuDNA-78T datasets were used to calculate nucleotide pairwise distances (Table 3 and Table 4, respectively).
The mean mitochondrial distances between the three main haplogroups N, SE and SW range from 2.80% (SE vs SW) to 4.51% (SE vs N). Within haplogroup N, mean distances between the seven subgroups are comprised between 0.96% (ʻKordofan Iʼ vs ʻNubiaʼ) and 1.99% (ʻKordofan IIʼ vs ʻSenegalʼ). The genetic distances between the West African subgroups ʻNigerʼ and ʻSenegalʼ are comprised between 1.74% (WA708 vs MNHN-A10617) and 1.93% (WA036 vs MNHN-A10617). The distances between Taxa I.
X.  Table 3. Minimum and maximum pairwise distances (in %), as well as mean distance (between brackets), calculated using the mtDNA-91T dataset both within and between haplogroups (Fig. 3). Boldface = maximal intrapopulational variation.

XI
Taxa I.

Diversifi cation of giraff es during the Pleistocene epoch
The fi rst fossil remains related to extant giraff es date back to 1 Mya and were found in West Turkana (East Africa) (Harris 1991), which is in agreement with molecular dating estimates of the MRCA of Giraff a based on complete mitochondrial genomes (Hassanin et al. 2012: 1.1 Mya). This corresponds to one of the most arid periods of the Pleistocene epoch (de Menocal 2004), suggesting that the associated expansion of the savannah may have promoted the split between northern and southern giraff es.
Among southern giraff es, the subsequent divergence between the species G. giraff a and G. tippelskirchi took place at 645 ± 315 kya with nuclear data or 780 ± 260 kya with mtDNA data. Around 800 kya, the landscape was shaped by a strongly humid interglacial period, which caused in southwestern Africa the development of a complex hydrographical network nourished by the Okavango and the Zambezi Basin, culminating in the evolution of the Paleo-lake Makgadikgadi (Goudie 2005;Moore et al. 2012). These newly formed physical barriers may have impeded gene fl ow between giraff es separated by the Zambezi River, with G. tippelskirchi in the North and G. giraff a in the South.
The rapid diversifi cation of the northern species G. camelopardalis into several subspecies was dated at 445 ± 155 kya with the mtDNA dataset and 445 ± 245 kya with the nuDNA dataset. These estimates fall into an unusually long interglacial period, which might have been associated with milder savannah conditions (Yin & Guo 2007), favouring the radiation of northern subspecies. The humid conditions led further to the development of a hydrographical network in Niger, with the rise of water levels of Mega Lake Chad, but also in East Africa, with the extension of the White Nile to the proportion of a Paleolake (Williams et al. 2003) and the origin of Lake Victoria at around 400 kya (Johnson et al. 2000). The establishment of this drainage system in East Africa may have isolated (at least temporarily) several giraff e populations, favouring their subsequent diversifi cation at the subspecies level.

The giraff es of the Nile: G. c. camelopardalis, G. c. antiquorum and G. c. rothschildi
Linnaeus (1758) (1553), who saw several living giraff es in the "castle of Cairo", Ray (1693), who refers to Belon du Mans, and Hasselquist (1757), who examined a skin and wrote that giraff es live "in the forests" of Sennar and Ethiopia (indicated as type localities by Linnaeus). Around 1750, the Sultanate of Sennar was located in the current Sudan and covered the regions surrounding the cities of Khartoum and Sennar in the northern parts of the White and Blue Niles. At that time, Ethiopia was smaller than today: bordering southeastern Sennar, it included the regions surrounding the city of Gondar and southern parts of the Blue Nile and its source, the Lake Tana. All fi rst reports on giraff es were based on animals captured by Egyptians, including the giraff es off ered to Frederick II (1245), Lorenzo de Medici (1487), and Charles X (1827), as well as the giraff es held captive in Cairo described by Belon du Mans (1553) and the skins studied by Hasselquist (1757). All these giraff es have been collected by exploring the southern regions close to the borders of the Blue and White Niles. Today, the countries crossed by the two main tributaries of the Nile are, from north to south, Sudan, Ethiopia, South Sudan, northeastern DRC and Uganda. In this area, Lydekker (1914) recognized the fi ve following subspecies: G. c. camelopardalis, G. c. antiquorum, G. c. congoensis, G. c. cottoni and G. c. rothschildi (Fig. 1A). Several of these subspecies were, however, synonymized in subsequent classifi cations: G. c. rothschildi was treated as a synonym of G. c. camelopardalis (East 1998); G. c. congoensis was considered as the synonym of G. c. camelopardalis (Ansell 1968;Dagg 1971) or G. c. antiquorum (Ciofolo & Le Pendu 2013;Dagg 2014;Shorrocks 2016;; and G. c. cottoni was ranged into G. c. rothschildi (Ansell 1968). To better understand the taxonomy of giraff es, we included in this study several museum specimens of the 18 th and 19 th centuries collected within the type localities of the subspecies G. c. camelopardalis, G. c. congoensis and G. c. cottoni.
Fennessy et al. (2016) and Winter et al. (2018b) analysed seven giraff e samples supposed to belong to the nominate subspecies G. c. camelopardalis (Linnaeus 1758), because they were collected in South Sudan and Ethiopia, two countries where the subspecies is still possibly present according to the IUCN . These samples include SNR1 and SNR2, which were collected on the left bank of the White Nile in South Sudan (Shambe NP), BaNP3 and BaNP4, which were collected on the right bank of the White Nile in South Sudan (Bandingilo NP), and ETH1, ETH2 and ETH3, which were collected in western Ethiopia (Gambella NP), far from both White and Blue Nile rivers (see Figs 3, 5). Our network and phylogenetic analyses based on the mitochondrial sequences of giraff es (Figs 3, 4A) show that SNR giraff es belong to the haplogroup ʻKordofan Iʼ, whereas BaNP and ETH giraff es belong to the haplogroup ʻRothschildʼ. In agreement with mtDNA results, our analyses of nuclear introns confi rm that SNR giraff es belong to the subspecies G. c. antiquorum, whereas ETH giraff es belong to the subspecies G. c. rothschildi. Based on similar phylogenetic results, Fennessy et al. (2016) have, however, concluded that the subspecies G. c. rothschildi should be synonymized with G. c. camelopardalis, as previously suggested by some authors (Kingdon 1997;East 1998). Our analyses of museum specimens, however, support another taxonomic interpretation. Indeed, the two mtDNA haplotypes sequenced from three specimens of the 19 th century assigned to G. c. camelopardalis (haplogroup Nubia, Figs 3, 4) form a well-supported monophyletic group (PP = 1; BP = 99), distinct from all other haplogroups previously identifi ed (nucleotide distance > 0.75% from Kordofan I; 1.22% from Kordofan II; > 1.15% from Rothschild): the fi rst haplotype was obtained from the famous Zarafa (MNHN 1845-211), which was captured in 1824 near the mountains located south of the city of Sennar (Salze 1827), probably close to the boundary between Sudan and Ethiopia; the second haplotype was sequenced in two giraff es collected in ʻAbyssiniaʼ (Ethiopian Empire), which both arrived in France in 1843 (MNHN-A8012 and MNHT-1996.121.2). Zarafa was collected near the Blue Nile, but we do not know if it was captured on the left bank or on the right bank of the river. During the 19 th century, several explorers have mentioned the presence of many giraff es in the region between the Blue Nile and Tekezé / Atbara rivers in southeastern Sudan (eastern Sennar) and northern Ethiopia (Abyssinia) (Combes & Tamisier 1838;Baker 1880). Since the haplotype from Sennar is highly similar to the haplotype from Abyssinia (1 mutation), we suggest that the subspecies G. c. camelopardalis was endemic to this region. However, we cannot completely exclude that G. c. camelopardalis was also present in the region of western Sennar between the White Nile and Blue Nile, delimited in the south by the Sobat River in the west, followed in the east by the Baro River and Ethiopian Highlands (Fig. 5).
The taxonomic status of other subspecies related to the Nile was problematic. For instance, some authors have treated G. c. rothschildi as a synonym of G. c. camelopardalis (Ciofolo & Le Pendu 2013;Fennessy et al. 2016). However, our phylogeographic analyses support that G. c. camelopardalis and G. c. rothschildi are two distinct subspecies, and that populations of G. c. rothschildi are found in eastern South Sudan, western Ethiopia, northeastern Uganda and northwestern Kenya, a biogeographical region delimited by the White Nile river in the west, by the Sobat River (or alternatively the Blue Nile) in the north, Ethiopian Highlands in the east and a discontinuous barrier in the south, including Lake Victoria (Fig. 5). Our results therefore corroborate previous hypotheses that giraff es avoid to cross Fig. 5. Giraff e subspecies of the Nile region. The map (extracted from Google Earth; https://www.google.com/intl/de/earth/) shows the geographical barriers (rivers and mountains) that may have isolated (at least temporarily) the subspecies Giraff a camelopardalis camelopardalis (Linnaeus, 1758) (red), G. c. antiquorum (Jardine, 1835) (yellow), G. c. rothschildi Lydekker, 1903 (green) and G. c. reticulata de Winton, 1899 (magenta). The question mark refers to the uncertain geographic origin of Zarafa (left) and the two specimens from Abyssinia (right) (see Discussion for more details).
high mountains (Happold 1978;Dagg 2014) and large rivers (MacClintock 1973;Henderson & Naish 2010). This might concern in particular females living in nursery herds with their calves (Bercovitch & Berry 2010), which tend to avoid the potential risk of overcoming biogeographic barriers (Petzold & Hassanin 2020). In eastern central Kenya, populations of G. c. rothschildi and G. c. reticulata can be found in sympatry (Kingdon 1997) and fi eld observations of intermediate phenotypes have suggested occasional gene fl ow between these two subspecies (Stott & Selsor 1981;Kingdon 1997). In the mtDNA tree, three individuals of G. c. reticulata (RET5, RETRot3, LWC01) cluster within the haplogroup ʻRothschildʼ (Fig. 4A), whereas they are related to other individuals of G. c. reticulata in the nuDNA tree (Fig. 4B). This mito-nuclear discordance suggests that a mitochondrial introgression occurred from G. c. rothschildi to G. c. reticulata at 96 ± 55 kya (Fig. 4B).
Our phylogeographic analyses based on mtDNA sequences suggest that all populations located on the left bank of the White Nile belong to the subspecies G. and a more vertically upright direction of the ossicones" compared to other northern and southern races (Thomas 1898). The type locality is indicated as "southeast of the junction of the Benue and Niger" (Thomas 1898). However, we agree with Lydekker (1904) that Lokoja (Nigeria), which is located in the V-shaped area to the north of the confl uence of the Niger and Benue rivers, is a more probable type locality, because the Benue and Niger rivers may have acted as barriers against giraff e dispersal towards the south of Nigeria (Happold 1969).
During the 20 th century, giraff es were then recorded in other countries of West Africa, such as Senegal, Gambia, Mali, Ghana, and Niger (Sidney 1965;Happold 1969Happold , 1978Nežerková et al. 2004). Consequently, the distribution of G. c. peralta was supposed to cover a large area from Senegal in the west towards the western part of the Central African Republic in the east (Fig. 1A;Dagg 1962;Happold 1969). However, populations have severely decreased over the last decades, because of poaching, habitat loss, increasing aridity (Dagg & Foster 1976) and outbreaks of rinderpest (Roeder et al. 2013), and the West African giraff e is now considered regionally extinct in Burkina Faso, Guinea, Mali, Mauritania and Senegal . The mitochondrial study of Hassanin et al. (2007) suggested that giraff es found in Cameroon, Chad and the Central African Republic rather belong to the subspecies G. c. antiquorum. At present, G. c. peralta is listed as endangered, only represented by 607 individuals in Niger (Zabeirou 2017).
Our results show that the past diversity of West African giraff es was greater than previously assumed, as the mtDNA haplotype sequenced from two museum specimens (MNHN-A10617 and MNHN-A10753) collected in Senegal (Bakel) by Girardin in 1830 (Malte-Brun & Malte-Brun 1839) is distinct from other subspecies of the haplogroup N by at least 1.38% and can be diagnosed by fi ve ES (Supplementary fi le 9). We assume that the Sahara Desert in the north, rainforests in the south and the Niger River in the east may have acted as physical barriers limiting gene fl ow between giraff es from Senegal and those from Niger and Nigeria (G. c. peralta). Both diff er considerably in the colouration and pattern of the coat. The Niger giraff e can be recognized by its pale almond coat covered with numerous light brown spots diff ering in shape and getting smaller on the hindquarters, whereas the beige coat of the Senegal giraff e (Fig. 6) shows large dark brown patches with a clear contour and almost uniform in size on neck, trunk and hindquarters. All these elements indicate therefore that giraff es from Senegal should be placed into their own subspecies that will be formally described below. Unfortunately, the subspecies is now extinct, as the last giraff es of Senegal were killed in the 1970's (Vincke et al. 2005).
Taxonomic status of the giraff es collected in the Cape of Good Hope during the 18 th and 19 th centuries Boddaert (1784) was the fi rst to describe a southern giraff e from the Cape of Good Hope under the name ʻCamelopardalis giraff aʼ, based on specimens with no precise locality housed in the Prince of Orange Museum in The Hague and Naturalis museum in Leiden (Netherlands). At that time, the Cape of Good Hope, also known as the Cape colony, was a Dutch colony covering most of the presentday South Africa, excepting the northeastern regions (see map of Levaillant & de La Borde 1790). Levaillant (1797) and Delalande (1822) caught several giraff es in the vicinity of the Orange River (Fig. 3), which were assigned to the species Giraff a capensis by Lesson (1842) with the type locality "Cape of Good Hope". Some decades later, Lydekker (1904) recognized three subspecies of southern giraff es: capensis for giraff es from the ʻCape colony and adjacent districtsʼ; wardi for giraff es from Northern Transvaal (present-day Limpopo and Pretoria, northeastern South Africa); and angolensis for giraff es from Angola. This taxonomic view was adopted by most zoologists over the next decades, until Ansell (1968) proposed a new classifi cation with only two subspecies: angolensis Lydekker, 1903 for giraff es from Angola and giraff a (Boddaert, 1784) (Fig. 1B) for giraff es from South Africa, the latter including capensis Lesson, 1842 andwardi Lydekker, 1904 as synonyms.
Our mtDNA analyses of three ʻCape giraff esʼ collected by Levaillant (1797) and Delalande (1822) (MNHN-A7977, MNHN-1896-45, MNHN-A10749) show that the two haplotypes fall within the haplogroup SW, which also contains all haplotypes detected for the subspecies G. g. angolensis, including the southwestern giraff es from Angola and central Botswana, as well as an isolated population from Zimbabwe (Winter et al. 2018a). The haplogroup SW is genetically distinct by 3% from individuals of the haplogroup ʻSoutheast Africaʼ, which includes giraff es from northeastern South Africa, southern Mozambique, northern Botswana, southern Zambia and southern Zimbabwe. The results indicate therefore that the subspecies G. g. angolensis and G. g. capensis should be synonymized with G. g. giraff a (giraff es of the Cape of Good Hope), whereas the subspecies G. g. wardi should be rehabilitated for southeastern giraff es.

Taxonomic treatment
The molecular investigation of key specimens of European museum collections provided the unique opportunity to reveal past giraff e biodiversity through the inclusion of (possibly) extinct populations.
Our results have strong taxonomic implications for the classifi cation of giraff es at the subspecies level. We propose the following revision of the classifi cation of the genus Giraff a based on molecular (Petzold & Hassanin 2020; this study: see Supplementary fi les 9 and 10) and morphological diagnosis criteria (Lydekker 1914 Other specimens ETHIOPIA • 1 specimen (skull and skeleton parts); Abyssinia; MNHN-A8012. SUDAN • 1 specimen (skull), "Zarafa"; Sennar; MNHN-1845-211.

Diagnosis
Front of the face sparsely and sides fully spotted, similar pattern to reticulata but with chestnut or sandy patches.

Diagnosis
Spots on the upper part of the fore-limbs and the thighs broken up in a number of very small and irregular ones.

Type material
Holotype SUDAN • 1 ♂ (skull and tanned skin of a male giraff e); South of Darfur; SMF-498.

Distribution
Cameroon, Chad, Central African Republic, Democratic Republic of Congo, South Sudan (holotype).

Remarks
The holotype designation was based on the information provided by Rüppel (1826), who collected two specimens in North Africa. Both specimens can be meanwhile found in the collection of the Senckenberg Museum Frankfurt (Germany) listed as SMF-498 (skull and tanned skin, male): unspecifi ed type and SMF-497 (skull and tanned skin, female): paratype. Thomas, 1898 Diagnosis Elongated skull, large spatulate nasal opening, vertically upright direction of the ossicones, fawncoloured patch below the ears, white sparsely spotted occipital region; two ES in the Cytb gene: 219 C => T, 1080 C => T, and one ES in the CR: 23 A => G.

Distribution
Niger.

Remarks
The type locality has originally been assigned to the junction between the Niger and Benue rivers in Nigeria, but it was corrected in this study to Lokoja (Nigeria) north of the confl uence of the Niger and Benue rivers in accordance with Happold (1969).

Diagnosis
Deep liver-red colour with a coarse network of narrow white lines, one ES in the Cytb gene: 795 C => T; one ES in the CR: 94 C => T.

Distribution
Southern Ethiopia, Kenya (holotype), Somalia. Lydekker, 1903 Diagnosis Lower parts of the legs pure white and unspotted, spots show a tendency to split up into stars, occipital pair of ossicones, one ES in the CR: 129 T => C.

Diagnosis
Beige ground colour covered with dark brown spots following a reticulated pattern separated by narrow lines, skull features detailed in Blainville (1864), one ES in the Cytb gene: 732 A=> G; four ES in the CR: 92 dC, 95 A => G, 359 C => T, 463 A => G.

Remarks
Matschie (1898) mentions two diff erent specimens as syntypes, but the second cannot be found in the collection catalogue and might be considered as lost. Matschie, 1898 Giraff a schillingsi Matschie, 1898: 79.

Diagnosis
Stellate formed spots, shanks olive-coloured and spotted down to the hoofs, anterior horn less developed, one ES in the Cytb gene: 1033 C => T.

Distribution
Southern Kenya, Tanzania. Lydekker, 1911 Diagnosis Low and conical anterior horn, grey colour and scattered spotting of the sides of the face, fawn shanks, three ES in the CR: 39 A => G, 272 T => C, 336 T => A; two ES in the IGF2B1 intron: 60 C => T, 304 G => A.

Remarks
No concrete holotype specimen assigned, as specimens of the Prince of Orange Museum in The Hague and the giraff e of Vosmaer (1787) from the Naturalis museum in Leiden (Netherlands) are ʻwhereabouts unknownʼ. Lydekker, 1904 Giraff a infumata Noack, 1903: 356.

Diagnosis
Irregular spots, spots on the side of the face restricted to the region below and behind the eyes, two ES in the Cytb gene: 634 C => T, 705 A => G.