As for many tropical regions, the evolutionary and demographic status of antelope populations in the Udzungwa Mountains, Tanzania, are poorly resolved. We employed genetic information from 618 faecal samples to assess the status of forest antelope species in terms of their distribution, intraspecific diversity and population subdivision within the Udzungwa landscape. Most species were detected in the majority of forest fragments, except for Philantomba monticola. Phylogenetic analyses were consistent with traditional taxonomy with the exception of Cephalophus harveyi which was paraphyletic with respect to C. natalensis. There was strong support for three C. harveyi mtDNA clades within the Udzungwa Mountains although nuclear genetic variation did not partition strongly with these maternal lineages. Significant partitioning of genetic variation between sampling areas was detected for all species except the endangered C. spadix. Overall, our results demonstrate the value of non-invasive genetic sampling in studying the distribution and evolution of rarely observed species.
INTRODUCTION
Antelope constitute a major trophic guild in African forest and woodland ecosystems and play important roles as seed dispersal agents (Gautier-Hion et al., 1980; Beaune et al., 2013), understory browsers (Lunt, 2011) and as a food base for top predators (Boshoff et al., 1994; Hayward et al., 2006) and also human communities through the bushmeat trade (Nasi et al., 2011). Identifying the products of evolutionary and demographic processes in forest antelope is therefore important for delineating appropriate conservation units, accurately documenting their role in ecosystems and predicting the ecological consequences of anthropogenic change. And yet forest antelope diversity is poorly defined at all levels from populations through to species and higher taxonomic ranks, particularly in comparison to savannah antelope (Du Toit & Cumming, 1999; Lorenzen et al., 2012). This may seem surprising for relatively large mammal species but it could be argued that the diversity of forest-associated antelope was overlooked, if not obscured, by taxonomic revisions in the mid-twentieth century, such as Ellerman et al. (1953), with historical consequences which have only recently begun to be addressed by modern analyses (Colyn et al., 2010; Groves & Grubb, 2011). At a local scale, forest antelope distributions and population dynamics are also poorly known primarily due to the difficulty of studying cryptic species in dense habitats leading to stark differences in how studies estimate demographic rates (van Vliet & Nasi, 2008).
Non-invasive genetic sampling from low-quantity DNA sources such as shed hair or faeces is potentially a highly informative tool for characterising populations in terms of taxonomy or conservation management (Kohn & Wayne, 1997; Beja-Pereira et al., 2009). While ecological or behavioural data are often also required for robust conclusions regarding evolutionary or demographic status (Crandall et al., 2000; Fraser & Bernatchez, 2001), non-invasive genetics provides information that would otherwise be very difficult or expensive to obtain.
This study investigates genetic variation within forest antelope by non-invasive faecal DNA sampling across an Afrotropical mountain range, the Udzungwa Mountains in Tanzania. The Udzungwa Mountains (‘the Udzungwas’) are a high priority conservation landscape of global significance due to their remarkable levels of species diversity and endemism (Burgess et al., 2007). This biodiversity, and associated ecosystem services, is under increasing threat from habitat fragmentation and extraction of natural resources, including hunting for bushmeat (Nielsen, 2006; Rovero et al., 2010; Rovero et al., 2012). Despite recent efforts to document the mammalian fauna (Rovero & De Luca, 2007; Rovero et al., 2009), patterns of diversity within forest antelope species across the Udzungwas remain essentially unknown. We present the first multispecies genetic assessment and test predictions at three levels of antelope diversity.
The latest taxonomic treatment of hoofed mammals recognises 204 antelope species within the bovid subfamilies Bovinae and Antilopinae (Groves & Grubb, 2011), a substantial increase on the 92 species recognised by the World Conservation Union (IUCN, 2015). We test the prediction that forest antelope species are distributed uniformly throughout the major Udzungwa forested areas using genetic identification of faecal samples (Bowkett et al., 2009). Regarding our genetic data we predict that mitochondrial haplotypes will be monophyletic with respect to other related species named by Groves & Grubb (2011) within those species groups (i.e. the traditionally recognised polytypic species).
Recognition of evolutionary diversity below the species level in antelopes has most often been attempted through the designation of subspecies. In some, but not all, cases this has led to separate conservation strategies for different subspecies, e.g. Pitra et al. (2006). Phylogeographic study of savannah antelopes has revealed regionally distinct evolutionary histories that in many cases are consistent with named subspecies (Lorenzen et al., 2012). In contrast, very little is known about variation within forest antelopes and the apparent recent Pleistocene radiation of duiker species (Johnston & Anthony, 2012) brings into question at what point divergent lineages should be considered as species. We assess the genetic evidence for intraspecific diversity within the Udzungwas in the context of identifying Evolutionary Significant Units for conservation (Ryder, 1986; Fraser & Bernatchez, 2001). We predict that mitochondrial haplotypes will be monophyletic with respect to available conspecific sequences from other regions and we test representatives of strongly supported mitochondrial clades for significant genetic differentiation at nuclear loci sensu Moritz (1994).
Genetic data can be critical to identifying demographically independent populations within species, e.g. fish stocks (Moritz, 1994; Palsboll et al., 2007). Management of such units may seek to conserve overall evolutionary potential by maintaining local adaptations but also has practical implications for achieving sustainable harvesting of populations — a situation highly relevant to forest antelope, many of which are hunted for bushmeat (van Vliet & Nasi, 2008). Conversely, managers may wish to identify populations that are not demographically or genetically viable without intervention in order to prioritise units for recovery. This may be the case where habitat fragmentation has resulted in population subdivision and reduced gene flow, e.g. Goossens et al. (2005). Relatively few studies have explored the genetic effects of fragmentation in African forests (Radespiel & Bruford, 2014), although examples include anthropoid apes (Bergl & Vigilant, 2006) and cloud forest birds (Callens et al., 2011). Ruiz-Lopez et al. (2016) recently demonstrated that human habitat modification influences the genetic structure of colobus monkeys in the Udzungwas. We predict significant partitioning of forest antelope genetic variation across the Udzungwas and divergence in allele frequencies between discrete forest fragments.
The intention of this study is to demonstrate the value of non-invasive genetic data by providing information that either conflicts with or supports our specific predictions concerning local populations and thereby identifies areas for further research or improves the evidence upon which management decisions are currently based. We do not seek to definitively resolve the taxonomy or phylogeography of those antelope species found in the Udzungwas as this would require range-wide information constituting multiple lines of evidence pertinent to the broader evolution of these taxa. Similarly, robust conclusions concerning meta-population dynamics would require greater sampling effort over a longer time period than feasible within this study. Nevertheless, the genetic patterns revealed by our non-invasive sampling, interpreted with the appropriate caveats, have important implications for forest antelope conservation in the Udzungwa Mountains.
MATERIAL AND METHODS
Study site and subjects
The Udzungwa Mountains are the southernmost and largest block of the Eastern Arc Mountains, a chain of ancient mountain ranges stretching from southern Kenya to southcentral Tanzania. The mountains rise steeply from the Kilombero Valley on their eastern side (200–300 m) to the highest peaks (>2500 m). The Udzungwas encompass a total area of approximately 10 000 km2 including 1600 km2 of forest (Marshall et al., 2010) representing the largest area of rainforest remaining in the Eastern Arc Mountains (Burgess et al., 2007). Many of the main forests are within the Udzungwa Mountains National Park, together with the contiguous Kilombero Nature Reserve, although substantial forest patches outside of these areas are protected only nominally as Forest Reserves (figure1). The major forest fragments are described by Dinesen et al. (2001) and Marshall et al. (2010).
While it is possible that the Udzungwas were once covered by continuous forest this may have been disrupted by early human habitation (Rodgers, 1993) and has certainly been significantly impacted by humans for the last 100 years or more (Struhsaker et al., 2004). Forest cover in the area had been reduced to 26 forest patches by the mid-1990s (Newmark, 1998). A habitat matrix of village agriculture, savannah woodland and extensive areas of fire-maintained grassland separates these forest fragments. The sampling sites included in this study include almost all the major forests in the central Udzungwas, an area of approximately 6500 km2 excluding Image to the north (ca. 106 km2) and fragmented forests on the Mufindi plateau to the west.
Rovero & De Luca (2007) list 12 species of antelope in the Udzungwas based on the taxonomy of Wilson & Reeder (2005). Their list includes three species only found in forest habitats: Abbott's duiker Cephalophus spadix “True, 1890″ (figure 2), blue duiker Philantomba monticola (Thunberg, 1789) and suni Neotragus moschatus (Von Dueben, 1846) and two species that also occur in dry woodland: Harvey's duiker Cephalophus harveyi (Thomas, 1893) (figure 3) and bushbuck Tragelaphus scriptus (Pallas, 1776). The remaining species, including the grey or bush duiker Sylvicapra grimmia (Linnaeus, 1758), are almost entirely restricted to grassland and savannah woodland and are not considered further here.
The more recent taxonomy by Groves & Grubb (2011), and adopted by Wilson & Mittermeier (2011), splits three of the five traditionally recognised species found in the Udzungwas: P. monticola (into 10 species), N. moschatus (3 species) and T. scriptus (8 species). In all cases the new taxa were assigned names previously treated as subspecies or synonyms.
Philantomba monticola within the Udzungwas are assigned to Tanzanian highland blue duiker, P. lugens (Thomas, 1898), together with those in the rest of the Eastern Arcs. The Southern Highlands and Kigoma district. Neotragus moschatus has been split into two species north of the Zambezi River, the nominate moschatus and kirchenpaueri (Pagenstecher, 1885), with a third species, livingstonianus (Kirk, 1865), to the south. N. kirchenpaueri is “not positively recorded” from the Udzungwas according to Groves & Grubb (2011), although given that kirchenpaueri is described from highland areas to the north and south we follow Wilson & Mittermeier (2011) in provisionally assigning Udzungwa populations to this species rather than moschatus. T. scriptus from central Tanzania, including the Udzungwas, are assigned to T. sylvaticus (Sparrman, 1780). As these species designations are provisional, we employ inverted commas when referring specifically to populations in the Udzungwas.
The two remaining species, C. spadix and C. harveyi, retain their previous mono-specific status under Groves & Grubb (2011). In contrast to other reference works, Groves & Grubb (2011) describe the southern limit of harveyi as north of the Udzungwas, although the area is included in the range map of Wilson & Mittermeier (2011) along with highlands in Malawi.
Sample collection and processing
We undertook transect-based surveys in all forest fragments in the central Udzungwas over 10 km2, except for Iyondo, during the dry seasons between late 2006 and 2009 (figure 4 and 5). Total numbers of faecal samples collected from each location were as follows: Mwanihana (212); Luhomero (71); Ruipa (70); Uzungwa Scarp (56); New Dabaga-Ulang'ambi, hereinafter New Dabaga, (50); Lumemo (49); Iwonde (40); Nyanganje (34); Ndundulu (27); Matundu West, hereinafter Mat. west, (16); Nyumbanitu (11); and Kising'a-Rugaro (3). In addition, thirteen tissue samples were opportunistically recovered from skins or dead animals found in poaching camps: Mwanihana (6), Luhomero (1), Uzungwa Scarp (2) and New Dabaga (4). Transects were walked in a triangular configuration (typically 1 km per side) and antelope samples were stored in RNAlater (Ambion Ltd, Huntington, UK). DNA was extracted using the QIAamp DNA Stool Mini Kit (QIAGEN, Crawley, UK) following the manufacturer's instructions but including an extended proteinase K digestion step of 20 to 30 minutes. For tissue samples we used the QIAGEN DNeasy Blood and Tissue Kit with overnight digestion for dried skin samples. Blank extractions were undertaken to control for contamination.
Genetic analysis
Antelope genetic diversity was characterised in the first instance using the mitochondrial control region (or D-loop), a highly variable non-coding marker that has been recommended for species identification in forest antelope (Ntie et al., 2010a; Johnston et al., 2011) and is often used in ungulate phylogeography (Lorenzen et al., 2012). Primers and PCR conditions followed Ntie et al. (2010a). Sequences were edited in AutoAssembler (Applied Biosystems, Foster City, CA, USA) and aligned by MUSCLE (Edgar, 2004) in SEAVIEW (Gouy et al., 2010). All unique haplotype sequences were submitted to GenBank/European Nucleotide Archive ( http://www.ebi.ac.uk/ena/data/view/LT220339-LT220454). Sequence entries can be identified by the sample names used in this study (listed under ‘isolate’).
For Cephalophus species we optimized a panel of nuclear microsatellite markers from those developed for central African species by Ntie et al. (2010b). Loci that amplified consistently for our target species were combined in three pre-PCR multiplexes: MPLX1 = INRA40 (Beja-Pereira et al., 2004), BM1225, Bm2113 and BRRIBO (Bishop et al., 1994), MPLX2A = BM121 (Bishop et al., 1994), and SR12 (Ntie et al., 2010b; modified from Kogi et al., 1995) and MPLX2B = BM143 (Bishop et al., 1994) and INRA05 (Beja-Pereira et al., 2004). MPLX2A and B were then combined for automated sequencer analysis. The majority of these loci failed to amplify consistently in the other antelope species tested. Therefore, microsatellite analysis for this study was restricted to C. harveyi, with the results for C. spadix published in Bowkett et al. (2014). Primers sequences and PCR conditions are given in appendix 1. Fragment analysis was undertaken on a Beckman Coulter capillary sequencer and allele sizes scored using CEQ 8000 software (Beckman Coulter, Fullerton, CA, USA).
Data quality
To check for consistency in sequencing from degraded DNA we re-extracted and re-sequenced just over 10% of all identified faecal samples. We also took a conservative approach to scoring microsatellite alleles, modified from Taberlet et al. (1996), whereby we only accepted heterozygous or homozygous genotypes at each locus if scored from three or four separate PCR reactions respectively. Negative PCR controls for both types of marker were used throughout.
Deviations from Hardy—Weinberg (HW) and linkage equilibria in the microsatellite dataset were tested for in GENEPOP 4.0.10 (Raymond & Rousett, 1995) with significance thresholds corrected for multiple tests using the False Discovery Rate (FDR) (Verhoeven et al., 2005). The presence of null alleles, stutter and large allelic drop-out were tested for in MICROCHECKER (Van Oosterhout et al., 2004).
Data analysis
Phylogenetic analysis of the control region haplotypes was employed to confirm species identification and explore intraspecific diversity. Separate analyses were carried out for each species. Trees were built using a maximum-likelihood approach in PhyML (Guindon & Gascuel, 2003) and a Bayesian Monte-Carlo Markov Chain (MCMC) approach in MrBayes 3.2 (Huelsenbeck & Ronquist, 2001). Trees included GenBank/EMBL entries for target taxa together with representative sequences for other genus members. For Philantomba, some published sequences contain a ca.100 bp insertion, as reported by Ntie et al. (2010a), and so were excluded. We were also able to sequence a number of tissue or faecal samples from outside of the study area including C. natalensis (Smith, 1834) from some other African countries (see appendix 2). Outgroup taxa from the sister genus to each of the species concerned, as presented in Hassanin et al. (2012), were used to root phylogenies, i.e. Philantomba for the Cephalophus trees, Cephalophus for the Philantomba tree and Aepyceros for the Neotragus tree.
Maximum-likelihood analysis (ML) was undertaken starting with an initial neighbourjoining tree and with default PhyML settings except for user-defined substitution models and parameters which were selected in jModelTest 0.1.1 (Posada, 2008), see table 1. The final choice of models was based on AIC values (all models were also ranked in the top three for AICc and BIC). The best tree topology was estimated from both NNI and SPR search options and branch-support was calculated with 1000 bootstrap repetitions.
Table 1.
Best-fit evolutionary substitution models selected for Maximum-Likelihood analysis of forest antelope mitochondrial control region haplotypes from the Udzungwa Mountains, selected with jModelTest (Posada, 2008).
The Bayesian analysis implemented two Metropolis coupled MCMC searches each consisting of one cold and three heated chains. Instead of pre-selecting a substitution model we sampled across the entire general time reversible (GTR) model space (Huelsenbeck et al., 2004) by averaging different models according to their posterior probability (lset nst=mixed rates=gamma) and checking for convergence across different runs with the sump command (Ronquist et al., 2011). Convergence of MCMC runs was achieved by allowing the analysis to continue until the standard deviation of split frequencies reached the recommended threshold of 0.01 (between 1 and 1.5 million iterations in our case). The standard burn-in fraction of 25% was applied to the convergence diagnostic and the subsequent parameter and tree summaries. The sump command was used to check effective sample sizes, log-likelihood values and stationarity plots.
Phylogenetic relationships within C. harveyi were also explored using the median-joining algorithm in Network 4.6.1.0 (Fluxus Technology, Clare, UK). For the C. harveyi control region phylogeny we also calculated uncorrected genetic distances between major clades and other species in MEGA5 (Tamura et al., 2011) including additional GenBank sequences (supplementary table 1). We also investigated microsatellite variation between samples grouped by their position in the mtDNA phylogeny (i.e. control region clades) with principal coordinates analysis in GenAlEx 6 (Peakall & Smouse, 2006).
Standard genetic diversity values were calculated in GenAlEx and Arlequin 3.5.1.2 (Excoffier et al., 2005). To account for differences in number of samples per location we recalculated control region haplotype richness, microsatellite allelic richness and private allelic richness by rarefaction to the smallest sample size in Contrib (Petit et al., 1998) and HP-Rare (Kalinowski, 2005).
Partitioning of genetic variation within and amongst sampling locations was tested for using AMOVA, excluding sample sizes less than five. Pair-wise FST values between sampling locations were estimated with FDR correction applied to probability values. To test for an effect of isolation by distance, we carried out Mantel tests for the larger datasets (> 5 sampling locations) in Arlequin.
RESULTS
We extracted DNA from 639 antelope faecal samples and 13 tissue samples from the Udzungwa Mountains, of which all but nine dung samples and one tissue sample yielded a sequence which could be assigned to one of the five traditionally recognised species. Samples that contained many ambiguous bases were excluded leaving a total of 618 Udzungwa sequence reads for genetic analysis plus sequences from other regions (appendix 2).
Distribution
T. ‘sylvaticus’ was only detected on six occasions during our study, three times in Ruipa and once in Iwonde, Lumemo, and New Dabaga. C. harveyi was recorded from all sampling locations and C. spadix and N. ‘kirchenpaueri’ from all but one each. P. ‘lugens’ had the most disparate distribution with no faecal DNA records from the central or eastern forests despite sampling effort being greatest in these areas (figure 4, table 2).
Table 2.
Sample size (N), number of mitochondrial control region haplotypes, haplotypic richness and genetic diversity values for four species of antelope sampled in twelve locations in the Udzungwa Mountains, Tanzania. Haplotypic richness was rarefied to the smallest sample size (min. 5).
Phylogenetic analysis
A total of 111 control region haplotypes were recovered from Udzungwa faecal and tissue samples. Sixty of 68 re-extracted faecal samples yielded identical sequences, while five failed to amplify target DNA on repeat and three samples had different sequences due to a mislabelling error that was subsequently corrected. We did not conduct phylogenetic analysis for T. ‘sylvaticus’ as we only recovered five haplotypes and an extensive CR phylogeny for this species group has been published previously (Moodley et al., 2009). However, we note that our sequences were highly similar to haplotypes from the sylvaticus cluster, as opposed to scriptus, in BLAST (NCBI, Bethesda, MD, USA).
In comparison to published multi-locus phylogenies (Johnston & Anthony, 2012), Bayesian analysis recovered inter-species relationships more accurately than Maximum-Likelihood. Therefore, Bayesian trees are presented with bootstrap values (BS) for nodes supported by ML. Posterior probability values (PP) below 0.5 (or bootstraps below 50%) are not shown and some node values near terminal ends were omitted for clarity.
Three of the four species were potentially monophyletic with respect to other sequences (figure 6a, b, c, d). For P. ‘lugens’ and N. ‘kirchenpaueri’ this pattern is uncertain as the more strongly supported clades also included sequences from elsewhere in Tanzania but because precise localities are unknown these sequences could potentially originate from other taxa also elevated to species by Groves & Grubb (2011). Reciprocal monophyly was strongly supported for C. spadix, as was a sister relationship to C. sylvicultor (figure 6b).
In contrast, C. harveyi was paraphyletic with respect to C. natalensis (figure 6a). C. harveyi haplotypes from the Udzungwas fell into three strongly supported clades (figure 6a clades A–C; PP = 1, BS = 82–98%; figure 5), two of which appeared closer to C. natalensis than the third, although this deeper node was not supported by the ML tree. All three clades were recovered from the majority of sampling locations with the only exceptions being locations with minimal sampling (figure 5; table 2). Clade B was monophyletic with respect to sequences outside of the Udzungwas as were Udzungwa P. ‘lugens’ haplotypes although with less support (PP = 0.86, BS = 56%). Uncorrected genetic distances between clades within C. harveyi were equivalent to those between C. harveyi clades and C. natalensis (4.5–6.3%) but less than inter-species values for other red duiker (table 4).
The median-joining haplotype network for C. harveyi was largely congruent with the treebased analyses although this approach highlights the long branch-lengths typically connecting haplotypes from other regions to those in the Udzungwas (figure 7a). The one sample from the Selous Game Reserve, to the east of the Udzungwas and mapped as part of the range of C. natalensis in East (1999), clustered closely with Clade A in both tree and network analyses (figure 6a and 7a).
Microsatellite variation did not partition strongly between mitochondrial clades in the Udzungwas for C. harveyi (AMOVA: Among clade variation 0.67%, within clade variation 99.33%, global FST = 0.007, P = 0.047) as illustrated by a plot of principal components (figure 7b). Therefore, we conclude that these lineages do not fulfil Moritz (1994)'s criteria for Evolutionary Significant Units whereby mitochondrial monophyletic groups should also exhibit corresponding significant divergence at nuclear loci.
Genetic diversity and differentiation
All eight microsatellite loci were polymorphic in C. harveyi with three loci deviating from Hardy-Weinberg expectations following FDR correction (BRRIBO, BM143 and INRA05). Furthermore, three loci exhibited homozygote excess indicating the potential presence of null alleles (Bm2113, BRRIBO and BM143). There was no evidence of significant linkage disequilibrium within the data set.
Genetic diversity differed markedly between species with the most notable result being the low values for C. spadix. mtDNA results for C. spadix were typically less than half than those of other species (table 2). Analysis of genetic diversity within sampling locations was confounded by small sample size particularly for the more rarely encountered species. However, most locations had similar results within species with no obvious outliers (table 2 and 3).
Small sample sizes reduced the number of pair-wise FST estimates for C. spadix to six (FST = 0–0.16; P > 0.05) and one for P. ‘lugens’ (Luhomero vs. Uzungwa Scarp, FST = 0.37; P < 0.001). Most comparisons were statistically significant for N. ‘kirchenpaueri’ as were half for C. harveyi (table 5). For microsatellite loci, all but two significant comparisons for C. harveyi involved New Dabaga or Uzungwa Scarp (table 5). AMOVA results were highly significant for all species except C. spadix indicating potential population subdivision (table 6).
There was moderate support for isolation by distance across sampling locations for the control region: N. ‘kirchenpaueri’ (r = 0.47; P = 0.04) and C. harveyi (r = 0.25; P = 0.08) but a strong signal for microsatellite variation in C. harveyi (r = 0.54; P < 0.001).
Table 3.
Sample size, polymorphism, unbiased expected heterozygosity (HE) and allelic richness for C. harveyi sampled in nine locations in the Udzungwa Mountains, Tanzania.
Table 4.
Pair-wise mitochondrial control region uncorrected genetic distances between red duiker clades and species. Values along the diagonal (in bold) are mean within-clade distances. Within-species clades (A–C; 1–2) are labelled on figure 4a. Number of haplotypes is given in parentheses.
Table 5.
Pair-wise FST values between sampling locations for two species of antelope in the Udzungwa Mountains, Tanzania. Values below the diagonal are for mtDNA control region and above for eight microsatellite loci (the latter not used for N. ‘kirchenpaueri’).
Table 6.
Analyses of molecular variance among and within sampling locations for four species of forest antelope in the Udzungwa Mountains, Tanzania.
DISCUSSION
Antelope taxonomy and species distribution
Our results are largely consistent with other published phylogenies (Jansen van Vuuren & Robinson, 2001; Hassanin et al., 2012; Johnston & Anthony, 2012) in recovering major monophyletic species groups including the giant duikers [C. spadix, sylvicultor (Afzelius, 1815) and dorsalis Gray, 1846], dwarf duikers [P. maxwelli (C.H.Smith, 1827) and monticola including lugens], West African red duikers [C. callipygus Peters, 1876, ogilbyi (Waterhouse, 1838), weynsi Thomas, 1901 and niger Gray, 1846] and East African red duikers (C. natalensis, harveyi, nigrifrons Gray, 1871 and rufilatus Gray, 1846). Placement of C. leucogaster Gray, 1873, normally allied to the East African red duikers (Johnston & Anthony, 2012), was inconsistent in our trees. There was also only weak support for S. grimmia being sister to all other duiker species or just to Cephalophus (figure 6a, b, d). In fact, both Hassanin et al. (2012) and Johnston & Anthony (2012) place Sylvicapra as sister to the giant duiker clade making Cephalophus paraphyletic (figure 6d). Systematic resolution of this issue requires including grimmia in Cephalophus or splitting the genus into at least three groups with the name Cephalophura suggested for the red duiker clade by Hassanin et al. (2012).
At the species level, we have provisionally classified forest antelope in the Udzungwas according to the taxonomic revision of Groves & Grubb (2011). The name changes for three species reflect the tendency of these authors to split geographically widespread morphologically variable taxa, an approach which has proven controversial (Heller et al., 2013; Zachos et al., 2013; Cotterill et al., 2014). In many cases, Groves & Grubb (2011) admit that a robust classification still requires more research and this may be the case for P. monticola and N. moschatus where populations have been separated based largely on pelage characters. Our genetic results are clearly not suitable for establishing the phylogenetic status of these putative species although they did not contradict our prediction that such taxa would be monophyletic at the mtDNA locus tested. Much more extensive sampling is needed to document morphological and genetic variation across the ranges of these proposed taxa.
The most interesting phylogenetic result was the mitochondrial paraphyly of C. harveyi with respect to C. natalensis (figure 6a). Although this has been reported before, based on a small number of sequences including nuclear loci (Ntie et al., 2010a; Johnston & Anthony, 2012), the present study represents the most extensive genetic assessment of these species to date. Several authors have treated C. harveyi as a subspecies of C. natalensis (Ellerman et al., 1953) or interpreted narrow genetic distance as support for this proposition (Jansen van Vuuren & Robinson, 2001; Hassanin et al., 2012). However, differences between the two taxa in body size and coat pattern have led other authors and the IUCN Antelope Specialist Group to continue to recognise them as separate species (Kingdon, 1997; East, 1999; Groves & Grubb, 2011). Kingdon (1982) suggested that the two forms hybridise in southern Tanzania but, crucially, we lack samples from south and east of the Udzungwas except for the single Selous sample that grouped with harveyi clade A (figure 6a and 7a). Given the geographical and genetic evidence, it appears that the situation may be more complicated than a simple choice between one or two species or subspecies (see below).
Our prediction that these species of forest antelope would be distributed uniformly across the forested Udzungwas was fulfilled for C. harveyi and nearly so for N. ‘kirchenpaueri’ and C. spadix which were detected in all but one forest each. In the case of N. ‘kirchenpaueri’, this is almost certainly a false absence due to under-sampling in Nyumbanitu, whereas the genuine absence of C. spadix from Nyanganje is possible. The exception to this pattern is P. ‘lugens’ which is almost certainly absent from Mwanihana, contra to earlier reports (Dinesen et al., 2001), but may exist at low densities in other forests even though not detected here. This species not only went unrecorded during this study but also during intensive camera-trap surveys of the same areas (Rovero & Marshall, 2009). There is no obvious explanation for the patchy distribution of this duiker in the Udzungwas as the habitat and anthropogenic threats appear similar and blue duiker coexist with suni in the western Udzungwas as well as in east African coastal forests (Fitzgibbon et al., 1995). Interestingly, there are other forest mammals that have highly localised distributions within the Udzungwas (Davenport et al., 2008; Rovero et al., 2008).
Intraspecific lineage diversity in antelope
There was little evidence for forest antelope populations in the Udzungwas constituting potential Evolutionary Significant Units. Udzungwa haplotypes did not form exclusive monophyletic clades, except for P. ‘lugens’ (weak support, figure 6d) and Clade B in C. harveyi (see below). We interpret this to mean that for the majority of species gene flow with other regions is ongoing or if populations are now isolated they have yet to undergo complete lineage sorting.
The lineage diversity within C. harveyi is more puzzling. On the one hand, the three strongly supported clades within the Udzungwas are as different from one another as they are to C. natalensis samples from a wide geographic area and exhibit equivalent genetic variation (table 5). On the other, the geographic and nuclear genetic data do not indicate any separation of individuals from different clades into different populations (or Evolutionary Significant Units).
Unexpected diversity in mitochondrial DNA can be caused by a variety of processes including inadvertent amplification of mitochondrial pseudo-genes from the nuclear genome (Numts), hybridisation with other species or incomplete sorting of ancestral lineages. Given the limited nature of our dataset we cannot rule out any of these explanations. However, in light of the climatic history of Africa (DeMenocal, 2004) and the radiation of Cephalophus during the Pleistocene (Johnston & Anthony, 2012) we speculate that an ancestral red duiker lineage may have diverged whilst populations were isolated during arid inter-pluvial periods and repeatedly undergone secondary contact in the Udzungwas (and elsewhere) when a wetter climate allowed forest refugia to reconnect. This is the opposite pattern to that shown by savannah antelope species (Lorenzen et al., 2012). This secondary contact resulted in contemporary inter-breeding populations that show no segregation at microsatellite loci but retain the mitochondrial (and perhaps nuclear intron) signature of past isolation. This hypothesis requires testing with multi-locus data from throughout the range of C. harveyi and related species together with coalescent dating of lineage divergence, as in Bowie et al. (2006).
Genetic diversity and differentiation
Pair-wise comparisons and significant partitioning of molecular variation among sampling locations indicate substantial population subdivision in the Udzungwas at least for the two most sampled species (table 5 and 6). As the majority of sampling locations represent discrete forest patches this potential structure could be interpreted as reduced gene flow across unsuitable habitat and implicate deforestation as a threat to population persistence. Indeed, the majority of significant FST values for the C. harveyi microsatellite dataset involved the relatively isolated forests of Uzungwa Scarp and New Dabaga, which therefore stand out as priorities for conservation.
However, we also found a moderate-to-strong effect of isolation by distance, which may complicate interpretation of spatial genetic structure (Safner et al., 2011). In our case, we note that many pair-wise FST values were highly significant over relatively short distances and that most comparisons between sampling locations within continuous forest were not significant (e.g. Lumemo and Ruipa). Therefore, we conclude that there is a likely effect of habitat fragmentation, in addition to isolation by distance, operating on forest antelope population structure.
Despite the potential impact of population subdivision, we found little evidence for negative effects on genetic diversity. Diversity values were generally high and microsatellite heterozygosity in C. harveyi was very similar to the mean for healthy bovid populations (Garner et al., 2005). The stark exception to this pattern was the endangered Tanzanian endemic C. spadix for which we consistently estimated the lowest diversity values regardless of sample size (table 2), a similar result to the microsatellite data published elsewhere (Bowkett et al., 2014). Being one of the largest duiker species, C. spadix typically exists at much lower population densities than other species and so may suffer disproportionately from threats such as hunting, e.g. Nielsen (2011).
Conservation and management implications
We suggest that the data presented here may be typical of non-invasive genetic surveys undertaken over a relatively short period, although few such multi-species studies have been published to date. Problems such as small sample size, DNA quality and failure of loci to cross-amplify in related species are likely to feature in similar attempts to non-invasively sample ecological guilds at a landscape scale. Nevertheless, we have attempted to demonstrate that such surveys can yield valuable information that can guide further research and inform conservation management.
The recognition of highland populations of suni and blue duiker as separate species, the latter endemic to Tanzania, attaches a far greater significance to the Udzungwas as a potential stronghold for what could now be considered range-restricted taxa largely confined to protected areas. The Udzungwas are one of the largest remaining areas of forest within either species' range. This designation did not depend on our genetic sampling but is informed by the knowledge that N. ‘kirchenpaueri’ is found throughout the Udzungwas, whereas P. ‘lugens’ is mainly found on the western and southern escarpments. Whether these taxa are referred to as evolutionary species, subspecies or Evolutionary Significant Units, their distinctness adds further weight to the case for effective conservation of the Udzungwas.
In contrast, C. spadix is recognised by all recent authorities as a unique species restricted to Tanzanian highlands and highly endangered (Moyer 2003, Bowkett et al. 2014). While detected in the majority of surveyed forests (Jones & Bowkett 2012), this study illustrates the very low relative abundance of this species. Maintaining this rare species across remote areas without further loss of genetic variation will be a major challenge to conservationists.
Given the potentially negative effect of habitat fragmentation on forest antelope gene flow, conservation management should aim to maintain connectivity between forest patches where it persists and potentially establish wildlife corridors to connect outlying areas as has been recommended for Uzungwa Scarp (MTSN 2007). While this would be very difficult for forests surrounded by agricultural land, many central forests are now within the Udzungwa Mountains National Park and contiguous Kilombero Nature Reserve. Corridors within these protected areas could be established principally by the control of fire and enforcing bans on cutting trees and hunting. For forests outside of the core protected area such as New Dabaga and Uzungwa Scarp the priority for antelope conservation is to control illegal hunting and halt current population declines (Nielsen 2006, Rovero et al. 2010, Nielsen & Treue 2012).
Non-invasive genetic sampling is an increasingly important tool for assessing the evolutionary and conservation status of rare or elusive species (Zhan et al., 2006; Gebremedhin et al., 2009). While expensive and technology-intensive, non-invasive genetic sampling provides a promising method to monitor the response of forest antelope populations to conservation efforts in terms of abundance, distribution and genetic health.
ACKNOWLEDGMENTS
The Tanzania Wildlife Research Institute, Tanzanian Commission for Science and Technology, Tanzanian Forestry Service and Tanzanian National Parks gave permission to conduct the study. The staff of the Udzungwa Mountains National Park are thanked for their support especially Paul Banga and Ponjoli Joram Kabepele. Thanks to S. Ricci, K. Ally, V. Banga, A. Muhundu, R. Mwakisoma, Y. Sama, M. Matola, A. Mtui, A. Mndeme, M. Mlewa, A. Chitita and especially to Richard Laizzer for assistance with sample collection. The Wildlife Conservation Society Tanzania Program and Zoological Society of London supported T. Jones. Bettine Jansen van Vuuren, Nicola Anthony and Stephan Ntie are thanked for their technical support on duiker genetics.
REFERENCES
Appendices
Appendix 1. Microsatellite amplification from Cephalophus harveyi faecal samples.
Fourteen dinucleotide microsatellite markers from other bovid species were selected from those initially screened by Ntie et al. (2010). Only eight loci amplified consistently with Harvey's duiker C. harveyi samples and these were incorporated into two PCR multiplexes (table S1) and amplified under conditions modified from the QIAGEN Multiplex PCR kit (table S2). Difficulty in scoring stuttered peaks in BRRIBO was overcome by pig-tailing the reverse primer (Brownstein et al., 1996). Optimal amplification in multiplexed reactions required adjustment of relative primer concentrations in pre-mixed solutions (table S3) except for MPLX2B primers (BM143 and INRA05), which were added directly to the PCR media at 0.9 μM each (table S2). PCR negative controls were run to test for contamination. Analogous information for Abbott's duiker C. spadix microsatellite amplification is given in the supplement to Bowkett et al. (2014) at www.intres.com/articles/suppl/n024p105_supp.pdf
Table S1.
Microsatellite primers and multiplex design for use with C. harveyi faecal DNA on a Beckman Coulter capillary sequencer platform (Beckman Coulter, Fullerton, CA, USA). F=forward, R=reverse, primer labelled with fluorescent tag in bold.
Table S2.
Polymerase Chain Reaction (PCR) conditions for multiplexed microsatellite markers (MPLX) with C. harveyi faecal DNA.
Table S3.
Preparation of 10x primer mix for use with the QIAGEN Multiplex PCR Kit as recommended by the manufacturer (each primer at 2 μM) and relative primer concentrations used with C. harveyi faecal DNA.