Genetic Evidence for Two Species of Elephant in Africa

See allHide authors and affiliations

Science  24 Aug 2001:
Vol. 293, Issue 5534, pp. 1473-1477
DOI: 10.1126/science.1059936


Elephants from the tropical forests of Africa are morphologically distinct from savannah or bush elephants. Dart-biopsy samples from 195 free-ranging African elephants in 21 populations were examined for DNA sequence variation in four nuclear genes (1732 base pairs). Phylogenetic distinctions between African forest elephant and savannah elephant populations corresponded to 58% of the difference in the same genes between elephant genera Loxodonta (African) and Elephas (Asian). Large genetic distance, multiple genetically fixed nucleotide site differences, morphological and habitat distinctions, and extremely limited hybridization of gene flow between forest and savannah elephants support the recognition and conservation management of two African species: Loxodonta africana and Loxodonta cyclotis.

Conservation strategies for African elephants have consistently been based on the consensus that all belong to the single species Loxodonta africana(1–3). Yet relative to African savannah elephants, the elephants in Africa's tropical forests are smaller, with straighter and thinner tusks, rounded ears, and distinct skull morphology (2–11). Although forest elephants are sometimes assigned subspecific status and designated L. a. cyclotis, their degree of distinctiveness and of hybridization with savannah elephants has been controversial and often ignored (2–12). Recently, a comprehensive morphological comparison of metric skull measurement from 295 elephants (10, 11) and a provocative molecular report limited to a single individual (13) noted appreciable distinctions between forest and savannah specimens.

Here we report the patterns and extent of sequence divergence for 1732 nucleotides from four nuclear genes (14) among 195 African elephants collected across their range in Africa and from seven Asian elephants (Elephas maximus). African elephants were sampled, with biopsy darts (15, 16), throughout the continent, including individuals from 21 populations in 11 of 37 African elephant range nations (Fig. 1). Based on morphology (2–11) and habitat (17,18), three populations were categorized as African forest elephants, whereas 15 populations in southern, eastern, and north-central Africa were categorized as savannah elephants (Fig. 1). DNA sequences from four nuclear genes, including short exon segments (used to establish homology to mammalian genes) and longer introns (which would evolve rapidly enough to be phylogenetically informative), were determined for all elephants (19). The genes includeBGN [646 base pairs (bp)], CHRNA1 (655 bp),GBA (100 bp), and VIM (331 bp), with sequence from all four genes obtained for 119 individuals. An alignment of variable sites and the composite genotypes are presented in supplemental information (20). Among 1732 bp, 73 sites were variable and 52 were phylogenetically informative. These nucleotide variants defined nine unique savannah genotypes among 58 individuals and 24 unique forest genotypes among 24 individuals. We observed nine genetically fixed nucleotide site differences between Asian and African elephants (BGN 121, 155, 219, and 513 and CHRNA1011, 079, 274, 301, and 548) and one that approaches fixation (BGN 505). There were five fixed site differences between African savannah and forest elephants (BGN 304, 485, 508, 514, and 569) and two that were nearly fixed (CHRNA1 251 andGBA 20) (20).

Figure 1

Locations of sampled African elephant populations. Circles indicate sampling locations and population abbreviations. Green circles are forest populations (the number of elephants sampled is given here in parentheses after the location): DS, Dzanga Sangha (17); LO, Lope (16); and OD, Odzala (3). Red circles are savannah populations: AM, Amboseli (6); BE, Benoue (8); CH, Chobe (5); HW, Hwange (5); KE, Central Kenya (9); KR, Kruger (10); MA, Mashatu (7); NA, Namibia (14); NG, Ngorongoro (10); SA, Savuti (6); SE, Serengeti (7); SW, Sengwa (6); TA, Tarangire (7); WA, Waza (14); and ZZ, Zambezi (7). Black circles are three populations that were not classified a priori in either category: AB, Aberdares (17); GR, Garamba (18); and MK, Mount Kenya (3). Garamba is located in the Guinea-Congolian/Sudanian transition zone of vegetation in Congo, which includes a mixture of forest and secondary grasslands (17) suitable for both African elephant groups. Savannah, forest, and morphologically intermediate elephants have been reported in Garamba (11, 33). The forests of Mount Kenya and Aberdares are currently isolated by surrounding bush (18), and both have elephants that more closely resemble the savannah morphological phenotype. However, these forests may have recently been contiguous with other forest habitat (17, 18) and retained relict forest elephants. Orange indicates current African elephant range (1); historic range is in bright yellow (10). The dotted pattern indicates the extent of tropical forest (hatched) and forest/savanna transitional vegetation zones (17). Pie charts indicate the combined population frequencies of GBA haplotypes: I, [C.C]; II, [C.T]; and III, [T.C] for nucleotide sites 20 and 79, respectively, in Asian, forest, Garamba, and three savannah regional populations.N = number of elephant chromosomes.

Three methods of phylogenetic analysis (minimum evolution, maximum parsimony, and maximum likelihood) (21–23) revealed a concordant deep genetic division between the forest and savannah populations of African elephants (Fig. 2). The forest elephants of Dzanga-Sangha, Lope, and Odzala grouped together, separate from 15 savannah populations, which formed a distinct phylogenetic clade or lineage. An estimated 94% of the observed genetic variation (F ST = 0.94, P < 10−5) (24, 25) was due to differences between forest and savannah elephants and 6% to intragroup differences. Mantel tests (26) revealed only marginal association of genetic versus geographic distance (r = 0.19, P = 0.03), and that association was attributed completely to forest versus savannah population differences (P > 0.05 for forest or savannah populations tested separately).

Figure 2

Phylogenetic relationships for Asian, African forest, and African savannah elephants inferred from combined analyses (21–23) of 1732 bp (BGN, CHRNA1, GBA, andVIM); the two-letter codes for African elephant populations are given in Fig. 1. Asian elephant individuals are coded “Ema.” The minimum evolution (NJ) tree is shown. Concordant trees were obtained by MP (tree length was 248 steps; CI = 0.927, RI = 0.934) and ML (-ln L. = 2774.53539) analyses, which produced the same topology in defining the three groups. Bootstrap resampling support (100 iterations) is listed on branches for NJ (top), MP (middle), and ML (bottom) analyses for nodes supported by all three methods.

Although forest and savannah elephants formed two genetically distinct groups, sequences from populations within the two categories could not be distinguished hierarchical analysis of molecular variance (AMOVA) (24, 25). For example, we could not genetically differentiate the forest elephants in Dzanga-Sangha from those of Lope (F ST P > 0.05). Despite the extensive geographic distances separating them, the savannah populations in southern, eastern, and north-central Africa were genetically indistinguishable (F ST P > 0.05). Forest elephants are genetically more diverse than savannah elephants (Fig. 2). The average number of within-group pairwise differences among 24 forest elephants was 1.74 as compared with a value of 0.06 among 58 savannah elephants (24, 25, 27). Each forest elephant had a unique composite genotype, whereas the 58 savannah elephants defined only nine distinct genotypes (20). Forest elephants displayed larger numbers of heterozygous nucleotide sites than did savannah elephants (an average of 3.54 heterozygous autosomal sites per individual in forest elephants versus 0.39 for savannah elephants) (20). These observations suggest a recent founder event in the history of the savannah metapopulation. A potential time venue for the bottleneck is indicated by fossil evidence, which suggests that the savannah elephant's range greatly expanded at the end of the Pleistocene, after Elephas iolensis, the predominant African species, became extinct (3,12).

The genetic and phylogenetic distinctiveness was evident without exception between 36 sampled forest elephants from three populations and 121 savannah elephants collected in 15 populations throughout sub-Saharan Africa. Each savannah population was genetically closer to every other savannah population than to any of the forest populations, even in cases where the forest population was geographically closer. Individuals from two “indeterminate” populations [Mount Kenya and Aberdares (Fig. 1)] contained exclusively savannah elephant genotypes (see Fig. 2, F ST = 0.88, P < 10−5 in comparing both populations to three forest populations). Genotypes found in the third “indeterminate” population, Garamba, were diverse and predominantly nested within the forest elephant clade in the phylogenetic analyses. The forest populations (including Garamba) were genetically closer to each other than to any savannah populations, including several that were geographically close. A single exceptional Garamba individual, GR0021, contained five signature sequence sites that were diagnostic for savannah elephants (BGN 304T, 485T, 508G, 514G, and 569C), as well as a single site (GBA 79T) that was diagnostic for the forest elephants.

The high level of genetic distinction is demonstrated by calculation of F ST values among savannah, forest, and Asian elephant populations as well as by the computation of genetic distances (average pairwise differences) among them (24, 25, 27). Highly significant differentiation is evident between savannah, forest, and Asian elephants (F ST P< 10−5) but not between Garamba and forest elephants (P > 0.05). The genetic distance (average pairwise difference) between forest and savannah elephants is 9.0, which is 58% of the distance between Asian and African elephant genera (average = 15.5) (24, 25, 27). Tests for molecular evolutionary rate differentials did not reveal significant differences (P > 0.05) for the two African groups (24,28, 29). Considering the estimation from fossil evidence for the divergence time between the two genera as 5 million years ago (12), the results suggest that forest and savannah elephants diverged approximately 2.63 (±0.94) million years ago (24, 27, 29), which is comparable to species-level distinction in other mammalian taxa, including elephants (12, 30, 31). This estimate should be considered as a maximum age, however, because the more recent genetic homogenization of the savannah elephants would inflate genetic distance as a consequence of a recent founder event.

Genetic distinctiveness between forest and savannah elephants is also apparent when individual gene variation is examined. ForGBA, two variable sites in African elephants define three haplotypes ([C.C], [C.T], or [T.C] for nucleotide sites 20 and 79, respectively) that have large forest versus savannah frequency differences (Fig. 1, exact test P < 10−5 for forest versus savannah). The predominant haplotype in savannah elephants is [T.C] (frequency = 0.96), whereas alternative [C.C] and [C.T] haplotypes comprise 100% of the forest elephants, suggesting that reproductive isolation exists between the two groups (Fig. 1). For VIM and CHRNA1, complete and exact haplotypes could not be determined for individuals heterozygous at two or more nucleotide sites, because gametic phase cannot be assessed (for example, for a two-locus genotype, does a double heterozygote G/C,T/A individual contain GT + CA or GA + CT haplotypes?). However, among forest and Garamba elephants, polymorphisms occurred at six nucleotide sites in VIM that were genetically monomorphic in savannah elephants (20). Similar differences in the occurrence of polymorphic nucleotide sites were apparent withinCHRNA1: All sites that were variable among forest and Garamba elephants were fixed in savannah populations, whereas the two sites that were variable in savannah elephants were fixed in forest and Garamba elephants (20). Likewise, both CHRNA1 andVIM had an insertion/deletion variant limited to forest and Garamba elephants (20). The presence of these deletion variants in Dzanga-Sangha, Lope, and Garamba also is consistent with the recent occurrence of gene flow among these forest elephant populations across the Congolian forest.

The X linkage of BGN seen in other mammals (14) was affirmed in elephants by the presence of heterozygous nucleotide sites among females but not among the hemizygous males. Nineteen variable sites in BGN were used to identify 169 haplotypes from 55 males and 57 females. A minimum spanning phylogenetic network of the nine unique BGNhaplotypes observed (Fig. 3) showed clear differentiation of a single distinct Asian haplotype (n= 13 chromosomes), two African savannah haplotypes (n = 103 chromosomes; including Aberdares and Mount Kenya), and six African forest haplotypes (n = 53 chromosomes; including one Garamba individual). For BGN, the number of nucleotide changes separating forest from savannah elephant haplotypes (six steps) was nearly as large as that separating either from the Asian elephant haplotype (seven steps). The BGN haplotypes present in the forest elephant populations were not found among savannah elephants, whereas haplotypes seen in the savannah elephants were not present in the forest populations (P < 10−5, exact test of the forest versus savannah haplotype frequencies) (24, 25, 32), suggesting a high degree of reproductive isolation between the forest and savannah populations. Taken together, the distinction affirmed by independent unlinked nuclear genes (Figs. 1 and 3) (20) offers strong support for the concept of appreciable genetic divergence between the African savannah and forest elephant populations.

Figure 3

Minimum spanning network depicting relationships among nine haplotypes observed for the X-linkedBGN gene for Asian, African forest, and African savannah elephants. Hatch marks indicate the number of nucleotide differences separating each haplotype. Haplotypes were determined using 7 Asian (EMA), 74 African savannah (LAF) and 31 African forest elephants (LCY) for which the sex was known (a total of 55 males and 57 females). Haplotypes unique to each of the three taxa are identified by differences in shading; the number of chromosomes is indicated for each haplotype.

There was no molecular genetic evidence of hybridization among 3 forest and 17 savannah elephant populations [defined a priori, plus Aberdares and Mount Kenya (Fig. 1)]. In Garamba, however, three individuals (GR0021, GR0035, and GR0037) showed genotypes with a combination of forest and savannah taxon-specific alleles, suggesting a history of limited hybridization in the ancestors of this population (20), as has been suggested by some (33), but not all (10, 11). GR0021 grouped with savannah elephants in the phylogenetic analysis, whereas animals GR0035 and GR0037 had largely forest genotypes (Fig. 2) except for the GBA[T.C] haplotype, which is absent in forest elephants but predominant in savannah elephants (Fig. 1). The paucity of gene introgression between forest and savannah populations even near regions of potential physical contact [that is, in north-central Africa or near Garamba (Fig. 1)] suggests that hybridization in nature is rare and perhaps minimized by behavioral or physiological reinforcement. In this regard, no elephant from any population, including Garamba, displayed a predicted F1 hybrid genotype (that is, heterozygous at the genetically fixed sites between savannah and forest elephants), affirming the lack of gene flow or hybridization among the sampled elephants.

The molecular results of a pan-African phylogeographic elephant survey reported here offer support for the idea that a long period of adaptive evolution (estimated at 2.63 ± 0.94 million years) separated the savannah and forest elephant lineages. As such, the results strongly support recognition of species-level distinctions between African elephant taxa (5–11). Although reproductive isolation is the principal criterion for species recognition according to the Biological Species Concept (34), local hybridization or even the presence of a “hybrid zone,” as may have occurred in Garamba, would not preclude species recognition, because the genetic integrity of the parent species remains intact (34, 35). Hybrid zones that fail to spread or homogenize the genetic distinctiveness of contact species have been observed with scores of other species (35, 36). These considerations, along with the combined morphological, ecological, and molecular data, are cogent indicators that there should be species-level recognition forLoxodonta africana (Blumenbach 1797), the African savannah elephant (37) and Loxodonta cyclotis(Matschie, 1900) (4), the African forest elephant. Given the rapid depletion of both forest and savannah elephant numbers in the past century and the ongoing destruction of their habitats, the conservation implications of recognition and species-level management of these distinct taxa are considerable (1, 10,38).

  • * To whom correspondence should be addressed. E-mail: obrien{at}


View Abstract

Stay Connected to Science

Navigate This Article