This study is based on the carnivoran constituent of the mammal faunas of Africa between 7 and 1.2 Ma (henceforth just called “faunas”). It focuses on three time slices: 7–5 Ma, 4–3 Ma, and 2.5–1.4 Ma. These time slices are of special significance. The first, 7–5 Ma, encompasses the time of origin of the first bipedal hominids, and therefore, so current thinking goes, the first members of our lineage, the hominins. The second time slice, 4–3 Ma, is the time of greatest species richness among African carnivorans [44]. The third time slice, 2.5–1.2 Ma, includes the origin of Homo at or near 2.5 Ma, the origin of the modern human body plan with H. ergaster/erectus, at or near 1.8 Ma, and the first time our lineage left Africa, some time after 1.8 Ma. All three time slices include faunas from North, East, and South Africa (except the 4–3 Ma time slice, for which no suitable North African fauna exists). The study uses genera rather than species as its analytical level. This is not desirable, but is unfortunately necessary, as many carnivoran remains are too fragmentary to identify to the species level, and species lists therefore often include large numbers of “Genus sp.,” which would not be helpful in the present case, unless it could be demonstrated conclusively that “Genus sp.” from one site is the same or different from “Genus sp.” from another site, something that is only occasionally possible. Faunas with fewer than five genera have been excluded. This is a small number in itself, but has to be weighed against the value of having more faunas included in the analyses.

Two main statistical methods are used. One is paired-group (UPGMA) clustering [37] with the Raup–Crick similarity measure, which has been found to be especially useful for paleontological data [31]. It weights data on the basis of frequency, so that widespread taxa do not have a disproportionate effect on the results, and also does not disproportionately favor unique taxa, as many similarity indices do. Only one cluster analysis, of all faunas together, was carried out.

As a complement to the cluster analysis, principal coordinate analysis (PCO) was used [46] for both all faunas together and for faunas of each time slice separately. PCO is an ordination technique capable of handling presence/absence data, producing a series of orthogonal coordinate axes based on a selected similarity measure. Here, I have used two different distance measures. Most distance measures tend to weight unique occurrences (in this case taxa unique to a locality) heavily. In a situation like the present one, where the sample sizes are very uneven, this will tend to affect the analyses unduly, such that in the 4–3 Ma time slice, for example, Laetoli, with its many taxa will tend to be placed far apart from the other localities simply because a large fauna will necessarily include many unique taxa (Table 1). To address this problem, I have in the PCO analyses of individual time slices used the Dice index of similarity as the distance measure. The Dice index upweights shared taxon occurrences at the expense of unique ones, and has been shown in simulation studies [2] and [25] to produce the best results among a number of similar indices. The Raup–Crick index, used in the cluster analysis, does not produce useful results when applied in the PCO. In the analysis of all data I have used the Euclidean distance as the distance measure of choice, for two reasons. First, there are fewer unique taxa if all localities are considered, not just those within time slices. Second, over longer time spans, rare or unique taxa can be important to distinguish between stratigraphically discrete units as is, in fact, the case here. Thus, in this situation, downweighting rare or unique taxa is counterproductive. A minimum spanning tree (MST) connecting all localities in the space of the first two coordinates in each analysis of the individual time slices was calculated. The branch lengths of the MST were used to compare with geographic distances between the localities by means of a rank correlation test to investigate to which extent physical proximity dominates the similarity patterns.

Statistical analyses were carried out using PAST, version 1.82 [20], available for download free from: http://folk.uio.no/ohammer/past/.

With only a few exceptions, the analyses of all faunas together align well with expectations (Fig. 1 and Fig. 2). Faunas from the same time slice group together with little overlap, with faunas from more distant localities being somewhat more distant in the analyses as well. The faunas from the 7–5 Ma time slice are separated into two groups at a high degree of significance. Since Toros-Menalla and Sahabi form one group and the East and South African faunas another, this result is in concordance with the proposed separation between a Chado–Libyan bioprovince and the remainder of Africa [24]. The split between Adu-Asa/Sagantole + Langebaanweg and the remaining 7–5 Ma time slice faunas may be stratigraphic, as the two former are the youngest of the localities in this time slice. Thus, they share, e.g., Mellivora, a taxon otherwise only found in younger faunas, whereas several of the other 7–5 Ma faunas (e.g., Upper Nawata, Toros-Menalla) instead have mellivorines that are more primitive than Mellivora.

That Shungura B and Usno group with the 2.5–1.2 Ma faunas while Shungura D–F groups with the 4–3 Ma faunas is almost certainly due to the presence at the former two localities of Panthera and Acinonyx (apart from Laetoli, the only two pre-2.5 Ma localities to have these taxa present) and their absence from Shungura D–F. In all likelihood this is a stochastic phenomenon, as these genera are always rare and could easily be missed if sample sizes are small. The grouping of Okote with Shungura G and Usno has more complex causes, probably related to the absence of taxa such as Canis from these faunas, but this is harder to verify.

The faunas placed in an intermediate position in Fig. 1 are a mixed group. Laetoli and Ahl al Oughlam are among the best sampled faunas and because of this share some taxa that might have been present in the less well sampled faunas, but have not been found due to sampling issues. The same reasoning may explain why Laetoli lies near the two Pliocene South African localities Makapansgat and Kromdraai. The later two, however, are distinct from similarly aged East African localities due to some distinct differences between East and South African faunas. Thus, e.g., Vulpes has been found in these two faunas but not at any East African Pio–Pleistocene site, though Prototocyon/Otocyon has [44]. Vulpes is also present at Ahl al Oughlam. The same situation pertains in the modern fauna, though Vulpes pallida maintains a marginal presence in northern Ethiopia and Eritrea [36]. A phylogeographic study of this animal might indicate whether its current presence in eastern Africa is ancient or due to a recent range extension. The disjunct distribution of Vulpes today may thus be one of long standing. In addition, Kromdraai shares with Laetoli and Ahl al Oughlam Nyctereutes and Viverra (though if the record at Kromdraai really is Viverra is debatable). These genera are now extinct in Africa. Fig. 2 shows that these presences place Makapansgat and Kromdraai in the direction of Laetoli and Ahl al Oughlam along Coordinate 2, but that they are still closer to other faunas of their time slice in the PCO analysis.

In the PCO analysis of the 7–5 Ma time slice, Langebaanweg falls almost at the center of the diagram (Fig. 3A), though closest to Adu-Asa/Sagantole for reasons discussed above. This position of Langebaanweg is probably because it is a large fauna without any unique taxa, thus sharing “something with everybody.” This interpretation is strengthened by the removal of Langebaanweg (Fig. 3B), which has almost no effect on the positions of the remaining faunas. In Fig. 4, placing the MST on a map, the Chado–Libyan bioprovince effect is distinct in the linking of Toros-Menalla with Sahabi rather than directly with East African faunas. Without Langebaanweg the MST is correlated with geographic distance, though adding Langebaanweg removes this correlation, showing that East and South Africa form part of the same bioprovince (at this level of analysis).

The results of the PCO analysis of the 4–3 Ma time slice (Fig. 5 and Fig. 6) are more complex. Coordinate 2 of Fig. 5 mostly serves to separate South Turkwell from the remaining faunas and can be ignored here except to note that this position seems mainly due to the absence from South Turkwell of a number of taxa that unite the other faunas (e.g., Dinofelis). As in the analysis of all time slices (Fig. 1), Laetoli and Makapansgat fall close together. Overall, however, Coordinate 1 separates those faunas, that in the overall analysis (Fig. 1) fall with the 2.5–1.1 Ma time slice fauna from the Okote (Usno and Shungura B from the northern Turkana Basin), from a combined group of southern Turkana Basin and Afar region faunas. Since these groupings do not follow geography, it makes sense that the MST does not either (Fig. 6), with the shortest distance in the MST (Laetoli–Makapansgat) being the longest distance on the map. It therefore does not appear that the area covered by the faunas of this time slice can be conveniently divided into provinces, although some taxa (such as Vulpes) are only found in South Africa. It is, however, of great interest that the faunas of the southern and northern Turkana Basin are not close to each other. This mirrors results from other studies [34], though a hypothesis to explain the pattern has not yet been formulated.

The faunal positions in the PCO of the 2.5–1.2 Ma time slice are relatively closely correlated with geographic distance, though not significantly so (Fig. 6 and Fig. 7). The main exception is the placement of Okote close to the northern Turkana Basin faunas of the Shungura Fm., rather than with the other southern Turkana Basin faunas of the Upper Burgi and KBS Mbs. of the Koobi Fora Fm. This result is mainly due to the presence at Shungura D–F, Shungura G, and Okote of Megantereon. Since Megantereon is a rare taxon at the best of times, this may simply be a sampling issue, though Upper Burgi and KBS are faunas that are fairly well sampled compared to most in this time slice. Another possibility is that it represents a difference in habitat, as Megantereon is considered the large predator that would be most tied to closed habitats. However, studies of faunal and environmental change in the Omo mammal community of this time [5] and [6] do not support this notion. On the other hand, the analysis does not show any signal of the taphonomic break in Shungura Mb. G that is clearly in evidence in the bovids [5]. Nevertheless, even if the position of Okote is ignored as due to sampling, the (lack of) correlation between MST segment length and geographic distance does not change, suggesting that there are no discernible bioprovinces in this time slice either.

It is difficult to find comparisons with this study. Very few studies have compared the carnivoran faunas of South and East Africa [38] and [39]. These studies found no major differences between the two regions in their large carnivorans in the Plio–Pleistocene. At the generic level the results of the present study agree with Turner's, though at the species level, which was Turner's focal level, I find some differences not recorded by Turner, mainly due to expanded faunal lists in the past decade [44]. The smaller carnivorans are, of course, a different matter, but this study does not find any clear pattern to regional differences that would preclude South and East Africa being part of a broad carnivoran bioprovince.

No previous studies have made comparisons between the carnivorans of North and East Africa, except on a locality-specific basis. However, this study agrees with studies on other groups, such as anthracotheres and hippos [24], that in the Late Miocene there is a distinct separation between a Chado–Libyan and an African bioprovince. No such distinction can be seen in the time of Ahl al Oughlam, which though it has a number of unique taxa nevertheless is typically African in faunal composition with only limited influences from Eurasia [16] and [17]. Thus, viewed at the million-year time scale and from the perspective of a carnivoran, post-Miocene Africa formed a single bioprovince, in distinction to the Late Miocene.

At shorter time scales the situation is different. A number of phylogeographic studies of medium-sized and large African mammals have been carried out in the past decade, including giraffe, Giraffa camelopardalis [7] and [21], hartebeest, Alcelaphus bucelaphus [13] and warthog, Phacochoerus africanus [27]. The majority of these studies have found a primary subdivision between a southern clade and a “northern” (mostly eastern, in some cases western + eastern) clade. Thus far, four carnivorans have been studied. In two smaller carnivores, the common genet, Genetta genetta [15] and white-tailed mongoose, Ichneumia albicauda [10] a north–south subdivision was found (in the case of Genetta preceded by a subdivision between a North African population and the sub-Saharan ones). In two larger carnivores, the wild dog, Lycaon pictus [18] and the lion, Panthera leo [11], on the other hand, there were no clear geographic subdivisions among the studied samples.

These phylogeographic results, which broadly speaking address processes occurring over the past 0.5–1 Ma, have generally been interpreted in terms of Pleistocene climatic fluctuations and concomitant changes to the environment such as expanding or contracting forest regions, grasslands, and deserts, suggest several things. The first is that in larger carnivorans gene flow between populations may be rapid enough to disrupt genetic isolating processes occurring over time frames of 10⁵ years or less, although more species need to be studied to determine whether this is a general pattern or specific to wild dog and lion, two of the most mobile of extant carnivorans. Studies of medium-sized taxa such as jackals, Canis sp.; honey badger, Mellivora capensis; caracal, Caracal caracal, or serval, Leptailurus serval would indicate the limits of such a process. Second, smaller carnivorans such as genets and mongooses seem to retain the effects of Pleistocene climatic disruption in their genetic makeup, though again more species need to be examined.

Although the analysis above dealt with the movement of species within genera and the phylogeographic studies with the movement of subspecific units, useful insights can be gained from applying the information from the latter to the former. It must just be kept in mind that the time scale for the phylogeographies is on the order of 10⁵ years while that of the paleontological data is on the order of 10⁶ years. Combining the results suggests the following: •

similarities between East and South African faunas among the larger carnivorans at the generic or specific level are at present irrelevant to considerations of possible migration barriers between African subregions because any differences between regions in this faunal component is quickly erased by their dispersal abilities at time scales below current detection levels in the fossil record;

Such patterns have the potential of being observable in the fossil record. In fact, smaller carnivorans may be among the most informative of mammals in this regard, as they often have large species and genus ranges despite relatively limited dispersal abilities and many of them occupy a wide range of habitats. Indeed, some species, e.g., bat-eared fox, Otocyon megalotis; black-backed jackal, Canis mesomelas and aardwolf, Proteles cristatus have disjunct ranges in the present day. Species with large ranges that would be of considerable interest to study include sidestriped jackal, Canis adustus; slender mongoose, Galerella sanguinea; African civet, Civettictis civetta, and several species of Genetta. On the paleontological side, the phylogeographic results suggest that much more effort should be made to recover and study small carnivorans as among the groups that are the best interregional biogeographic indicators. That the potential exists to recover many more small carnivorans than at present is shown by faunas such as those of Laetoli and Olduvai which have large small carnivoran faunas [29], [30] and [42].

It is of considerable interest that many of the large herbivores studied by phylogeographic means seem to generate phylogeographic signals on time scales similar to those of small carnivores. This surely reflects a fundamental difference between herbivores and large carnivores. The former are foodstuff specialists, but once the preferred food is present they have the ready means of consuming it. The latter are not specialists on the particular foodstuff (meat is, after all, meat), but on the particular vehicle (animal) that the foodstuff comes in. Thus, unlike herbivores, it is obtaining the food that is the problem for a large carnivore, not the presence of the foodstuff per se. The latter perspective is evidently a larger-grained one and less of an impediment to migration than the former.

Hominins take something of an intermediate position in this respect. Given their broad dietary tolerances, at least after the appearance of the genus Homo, we can expect that their temporal scale matches that of the larger carnivorans, which means that the detail of hominin migrations would at the present level of resolution of the fossil record be invisible. Fortunately, resolution of the African Plio–Pleistocene fossil record is continually improving. When the scale of resolution goes significantly below 100,000 years, larger carnivorans may play a significant role in understanding biogeographic patterns at the regional scale and the details of hominin interregional migration patterns may become more amenable to study.