Genetic connectivity and population structure of African savanna elephants (Loxodonta africana) in Tanzania

Abstract Increasing human population growth, exurban development, and associated habitat fragmentation is accelerating the isolation of many natural areas and wildlife populations across the planet. In Tanzania, rapid and ongoing habitat conversion to agriculture has severed many of the country's former wildlife corridors between protected areas. To identify historically linked protected areas, we investigated the genetic structure and gene flow of African savanna elephants in Tanzania using microsatellite and mitochondrial DNA markers in 688 individuals sampled in 2015 and 2017. Our results indicate distinct population genetic structure within and between ecosystems across Tanzania, and reveal important priority areas for connectivity conservation. In northern Tanzania, elephants sampled from the Tarangire‐Manyara ecosystem appear marginally, yet significantly isolated from elephants sampled from the greater Serengeti ecosystem (mean F ST = 0.03), where two distinct subpopulations were identified.Unexpectedly, elephants in the Lake Manyara region appear to be more closely related to those across the East African Rift wall in the Ngorongoro Conservation Area than they are to the neighboring Tarangire subpopulations. We concluded that the Rift wall has had a negligible influence on genetic differentiation up to this point, but differentiation may accelerate in the future because of ongoing loss of corridors in the area. Interestingly, relatively high genetic similarity was found between elephants in Tarangire and Ruaha although they are separated by >400 km. In southern Tanzania, there was little evidence of female‐mediated gene flow between Ruaha and Selous, probably due to the presence of the Udzungwa Mountains between them. Despite observing evidence of significant isolation, the populations of elephants we examined generally exhibited robust levels of allelic richness (mean A R = 9.96), heterozygosity (mean µH E = 0.73), and effective population sizes (mean N e = 148). Our results may inform efforts to restore wildlife corridors between protected areas in Tanzania in order to facilitate gene flow for long‐term survival of elephants and other species.

Our results indicate distinct population genetic structure within and between ecosystems across Tanzania, and reveal important priority areas for connectivity conservation. In northern Tanzania, elephants sampled from the Tarangire-Manyara ecosystem appear marginally, yet significantly isolated from elephants sampled from the greater Serengeti ecosystem (mean F ST = 0.03), where two distinct subpopulations were identified.Unexpectedly, elephants in the Lake Manyara region appear to be more closely related to those across the East African Rift wall in the Ngorongoro Conservation Area than they are to the neighboring Tarangire subpopulations. We concluded that the Rift wall has had a negligible influence on genetic differentiation up to this point, but differentiation may accelerate in the future because of ongoing loss of corridors in the area. Interestingly, relatively high genetic similarity was found between elephants in Tarangire and Ruaha although they are separated by >400 km.
In southern Tanzania, there was little evidence of female-mediated gene flow between Ruaha and Selous, probably due to the presence of the Udzungwa Mountains between them. Despite observing evidence of significant isolation, the populations of elephants we examined generally exhibited robust levels of allelic richness (mean A R = 9.96), heterozygosity (mean µH E = 0.73), and effective population sizes (mean N e = 148). Our results may inform efforts to restore wildlife corridors between protected areas in Tanzania in order to facilitate gene flow for long-term survival of elephants and other species.

| INTRODUC TI ON
Habitat loss and fragmentation are a significant challenge in species conservation and maintaining biodiversity worldwide (Henle, Davies, Kleyer, Margules, & Settele, 2004). Human population growth near protected area (PA) boundaries is often higher than in comparable rural areas (Wittemyer, Elsen, Bean, Burton, & Brashares, 2008). Population growth often brings changes in land use for agriculture and settlements, which leads to loss of buffer zones adjacent to, and corridors connecting PAs. The isolation of PAs decreases their effective size, limits gene flow between populations, and leads to increased human-wildlife conflicts. Elephant While poaching and illegal ivory trade are the most severe and immediate threats to African elephants (referred to as elephants), range and habitat fragmentation remain a significant long-term threat to the species' survival (CITES, 2014;Douglas-Hamilton, 1987;Wittemyer et al., 2014). Habitat fragmentation mainly affects far-ranging species, like the elephant. Elephants have extensive individual home ranges from 10 to 10,738 km 2 (Douglas- Hamilton, 1971; Leuthold & Sale, 1973;Lindeque & Lindeque, 1991;Thouless, 1995Thouless, , 1996Whyte, 1996), and they show high fidelity to their home ranges and corridors over multiple generations (Desai & Baskaran, 1996). Habitat loss and fragmentation resulting from human population growth and habitat conversion are of particular concern in Tanzania (Newmark, 2008) and threatens the connectivity of elephant populations that are becoming confined inside PAs. Riggio and Caro (2017) (McNaughton & Campbell, 1991). However, outside these PAs, there is an increase in human population and the conversion of land to agriculture. For example, in the western Serengeti, human population growth between 1988 and 2002 was 3.5% per year, and the highest rates of agricultural conversion were close to the PA boundary (Estes, Kuemmerle, Kushnir, Radeloff, & Shugart, 2012). This growth is higher than the national average which is 2.7% per year. Human activities affect the movement of elephants within and between ecosystems by reducing the size of wildlife corridors or introducing connectivity barriers such as roads and human settlements. This limits the gene flow between the populations, threatening their long-term sustainability.
In the Tarangire-Manyara ecosystem, expanding cultivation toward Tarangire National Park (TNP) has severely restricted wildlife movements to dispersal areas outside the park by blocking their migratory corridors. Unfortunately, there is an overlap between land suitable for agriculture, migratory wildlife corridors, and wet season dispersal areas (Msoffe et al., 2011). Thus, the rapidity of rangeland conversion to farming presents significant threats to wildlife conservation and disrupts the ecosystem viability (Lowe & Allendorf, 2010).
Similarly, connectivity areas for the Tarangire-Manyara ecosystem (hereafter, TME) and the Serengeti ecosystem (henceforth, SE) may not be viable in the long term because of its increasing isolation by agricultural settlements (Mwalyosi, 1991).
Female elephants are philopatric and remain with their natal herd , whereas males are ejected from the herd upon sexual maturity and subsequently facilitate gene flow between herds (Poole, 1987;Archie et al., 2007;Hollister-Smith et al., 2007).
Male-mediated dispersal has been documented among elephant populations in both Uganda (Nyakaana & Arctander, 1999) and Kenya (Okello et al., 2008). In both studies, there was a lack of congruence between mitochondrial DNA and nuclear-based variation.
Mitochondrial DNA data showed significant differentiation, whereas nuclear data showed low genetic subdivision between populations (Nyakaana & Arctander, 1999;Okello et al., 2008).  (Sinclair, 2012). However, by 1890 elephants were virtually eliminated from the Serengeti and remained very low for six decades as a result of game hunting and the ivory trade (Sinclair, 2012). In 1950 after the Serengeti was gazetted as a protected area, the elephant population increased to about 3,000 in 1975. This increase was followed by heavy poaching across the continent that caused a decline to only 400 elephants in Serengeti in 1986 (Sinclair, 2012). The international ban of the ivory trade by the Convention on International Trade of Endangered Species (CITES) in 1989 resulted in an increase of the elephant population to about 6,000 in 2014 (TAWIRI, 2015). It has been hypothesized that after heavy poaching, elephant recolonization in the Serengeti in 1950s came from the north and south of Serengeti (Sinclair, 2012;Watson & Bell, 1969). Analysis of elephant genetic structure within and between ecosystems in Tanzania can therefore help elucidate historical founder events, recent connectivity between populations, and consequently indicate priority areas to target connectivity conservation between important protected areas. In Tanzania, previous studies (Ahlering et al., 2012;Ishida, Georgiadis, Hondo, & Roca, 2013) suggested that the East African or Gregory Rift Valley separated the distribution of the elephant mitochondrial DNA haplotypes (i.e., east and west of the Rift Valley which represents the Serengeti and Tarangire-Manyara ecosystems, respectively). However, these studies were based on small sample sizes and did not cover all protected areas within these ecosystems.
Furthermore, the genetic relationship between the elephants in northern and southern Tanzania is not known. For example, in recent years, the wildlife corridor between Ruaha and Tarangire National Parks has been blocked completely (Riggio & Caro, 2017) 34,664 in 200934,664 in to 8,272 in 201434,664 in (TAWIRI, 2015. To keep these populations healthy, and enable long-term survival, it has been suggested that it is vital to maintain connectivity between these populations (Jones et al., 2012). There was previous movement of elephants between the Greater Ruaha and Selous ecosystem in 2009, through corridors that connect to Udzungwa and Mikumi National Park, (Nahonyo, 2009). However, these migratory routes, ranging and dispersal areas are threatened by habitat loss through the expansion of agriculture and human settlement, and poaching (Douglas-Hamilton, Krink, & Vollrath, 2005;Galanti, Preatoni, Martinoli, Wauters, & Tosi, 2006;Wittemyer et al., 2008).
On the migratory routes, poaching is more likely to occur than inside the protected areas (Nahonyo, 2009).
In recent years, the wildlife corridors between Selous Game Reserve and Udzungwa Mountains National Parks ( Figure 2) have closed completely (Jones et al., 2012) and measures to restore corridors were proposed (Jones et al., 2012). However, restoration of corridors requires significant resources. For many long-lived species, such as elephants, it is critically important to understand connectivity because the rate at which habitat is being changed is sufficiently rapid and an evolutionary response to the changes is unlikely (Balkenhol, Cushman, Waits, Storfer, 2016). There is a tendency to assume that elephants (and other wildlife) previously moved freely across the landscape and mated in a panmictic manner before habitat loss (Epps, Wasser, Keim, Mutayoba, & Brashares, 2013); however, there is no direct evidence for this. from those of Selous and is the isolation recent and due to habitat loss or were the populations distinct even before the intensification of human activities?

| Description of study areas
To conduct a regional assessment of the genetic structure, we selected four major ecosystems in Tanzania that have the largest elephant populations. These are Serengeti and Tarangire-Manyara in northern Tanzania and Ruaha and Selous ecosystems in central and southern Tanzania (Figure 2). We then compared our data with studies of other elephant populations in Africa.

| The Serengeti ecosystem
The Serengeti ecosystem (SE) is in the north-east of Tanzania between 34°450-35°500E and 2°-3°200S and includes multiple levels of protected area (Ernest et al., 2012). The SE is comprised of

| Tarangire-Manyara ecosystem
The Tarangire-Manyara ecosystem (TME) comprises three protected areas: Tarangire National Park (TNP), Lake Manyara National Park (LMNP), and Manyara Ranch (MANR) which acts a migratory corridor between the two parks ( Figure 2). TNP and LMNP manage wildlife for tourism but, Manyara Ranch (MANR) is a private land conservancy managed for livestock grazing and wildlife tourism.
Most of the land in the TME (~85%) fall under community land managed as open areas, game controlled areas, or wildlife management areas (Morrison, Link, Newmark, Foley, & Bolger, 2016), which allow different combinations of extractive and nonextractive wildlife uses, and human settlements and agriculture. A rift valley wall (500-1,000 m) at the western edge of the TME may prevent the free movement of wildlife to and from SE (Ahlering et al., 2012). The TME elephants were also affected by poaching before a ban on international trade in 1989. Since then, the population recovered to 4,202 individuals (TAWIRI, 2015) with an annual growth rate of 7% due to low poaching pressure and conducive climatic conditions (Foley & Faust, 2010). Matambwe is the only sector within the SGR that conducts photographic tourism. The rest of the SGR is designated for trophy hunting blocks. The Rufiji River separates Matambwe from Kingupira sector.

| Ruaha and selous ecosystems
The current number of elephants in the Selous ecosystem is 15,217 (TAWIRI, 2015). Between Ruaha and Selous ecosystems, there is Udzungwa Mountain National Park (Udzungwas), part of the Eastern Arc Mountains with a high level of species endemism and richness (Burgess et al., 2007;Dimitrov, Nogués-Bravo, & Scharff, 2012).
Between Udzungwas and SGR are the Kilombero Valley (6,650 km 2 ) which is an important bird area and important farming areas especially for rice and sugar cane (Figure 2), and an area of conservation concern due to rapid habitat conversion.

| Field data sample collection
We used DNA extracted from elephant fecal samples to genotype 11 microsatellite loci and mitochondrial DNA sequences of a hypervariable region in cytochrome b in the control region of D-loop. To ensure a higher concentration of DNA, we only collected samples from fresh dung. Scrapings of 5g were taken from the outer layer of the dung bolus, where most of the epithelial cells are found, and stored in Queen's College Buffer in 50 ml conical tubes, following methods established by Ahlering et al. (2012). Where possible, 50 samples were collected from each site, but a minimum of 25 samples in small, isolated populations were obtained and the GPS location of each recorded. Samples were collected over a short time frame trying not to sample closely related individuals (Ahlering et al., 2012).
We used COLONY (Jones & Wang, 2010) to identify resampled individuals that had identical multi-locus genotypes. Samples were shipped to the Pennsylvania State University for genetic analyses.
Each tube containing sample was sealed well using parafilm and placed into a zip-locked bag to avoid cross-contamination or leakage.
Four tissue samples from the Tanzania Wildlife Research Institute (TAWIRI) for elephants that died naturally were used as a positive control for allele scores due to their higher DNA quality than that of fecal samples. All required permits were obtained from the United Republic of Tanzania and the US Fish and Wildlife Service.

| DNA isolation
DNA was extracted and purified using company protocols for commercially available DNA extraction kits (QIAamp DNA Stool Mini Kit) with minor modifications (Eggert, Maldonado, & Fleischer, 2005).
The initial volume of samples was increased from 200 µl to 1,000 µl and added 1,400 µl of buffer ASL and incubated at 55°C while shaking at 225 rpm for 12 hr (Lohay, 2019). Other steps followed QIAamp Stool mini kit protocol.

| PCR amplification and genotyping
We performed PCR amplification of microsatellite loci from fecal samples using QIAGEN reagents (QIAGEN multiplex PCR kit).
The availability of each primer pair to amplify a PCR fragment with predicted size range was validated via agarose gel electrophoresis.
We estimated fragment lengths for each sample (allele) from digital gel images.
Having optimized our primer selection and amplification conditions, we grouped 11 microsatellite loci into four multiplexed panels for genotyping (Appendix S1). Initial microsatellite fragment detection of the multiplex PCR products was performed at the Genomic Core Facility of the Pennsylvania State University using ABI 3730xl DNA analyzer (Applied Biosystems). Individual microsatellite binning and scoring was conducted using GeneMapper ® v. 5 (Applied Biosystems), and individual alleles in each amplified loci were manually verified and compiled into working multi-locus genotypes for each fecal sample collected. In order to ensure quality control standards, individual samples underwent two separate amplification, fragment analysis, and genotyping rounds, to avoid inconsistent allele patterns or weak peak fragment identification ).

| Genetic diversity and differentiation
To account for allelic dropout, null alleles, and scoring error due to stuttering, we used MICRO-CHECKER 2.2.3 (Van Oosterhout, Hutchinson, Wills, & Shipley, 2004). Genotypes were then corrected based on MICRO-CHECKER 2.2.3 results. Identical multi-locus genotypes were identified using COLONY (Jones & Wang, 2010) and were removed from analysis as they were considered to be resampling of the same individuals. The number of alleles and their frequency was determined for all loci across individuals in all the populations using GenAIEx 6.502 (Peakall & Smouse, 2012). GenAIEX was also used to export files into formats compatible with other genetic analyses software. We also calculated population differentiation (F′ST; Bird, Karl, Smouse, & Toonen, 2011;Meirmans & Hedrick, 2011) and fixation indices (F ST ) in GenAIEx version 6.502. We tested for deviations from expectations under Hardy-Weinberg equilibrium (HWE) and for linkage disequilibrium (LD) within protected areas in GENEPOP (Raymond & Rousset, 1995;Rousset, 2008). The Markov chain strategy was used with 1,000 dememorizations, 1,000 for combining independent test results across study location and the number of loci was used to determine the statistical significance test results. A Bonferroni correction for multiple comparisons was applied using a Holm-Bonferroni sequential correction for both HWE and LD tests (Hochberg, 1988;Rice, 1989). Effective population size (Ne) from each location was estimated using the single-simple linkage disequilibrium (LD) method in LDNe version 1.31 (Waples, Antao, & Luikart, 2014;Waples & Do, 2008). To estimate Ne, we used specific settings including a monogamous mating model (Coombs, 2010) under the LD method (Waples, 2006) and the jackknife method option was selected to obtain 95% confidence intervals (Weathers et al., 2019).

| Population structure
The IBD program (Bohonak, 2002) was used to test whether there is evidence for isolation-by-distance (IBD). Individuals that are geographically close to each other tend to be more genetically related than distant individuals due to random mating. Spatial autocorrelation (r) was then plotted against geographic distance (km) to determine if the two variables were significantly correlated. FSTAT (Goudet, 1995) was used to calculate the inbreeding coefficient (F IS ) and allelic richness (A R ) which represents the number of alleles standardized to the smallest sample size in the study area. GenIEx (Peakall & Smouse, 2006, 2012 was used to calculate F ST whereas ARLEQUIN (Excoffier, Laval, & Schneider, 2005)  To determine elephant population structure in our samples, we used STRUCTURE 2.3 (Falush, Stephens, & Pritchard, 2003Pritchard, Stephens, & Donnelly, 2000). This program uses a Bayesian clustering model to assign individuals to a population while simultaneously estimating population allele frequencies. This model aims to determine the number of subpopulations K within the population, where K in most cases is unknown (Pritchard et al., 2000). K was inferred by running ten iterations for each K value from 1 to 10 using an admixture model with a LOCPRIOR option. A burn-in period at 1 × 10 6 and Markov Chain Monte Carlo (MCMC) repetition value of 1 × 10 6 was set. STRUCTURE HARVESTER was used to identify all primary (i.e., uppermost) ΔK estimates (Evanno, Regnaut, & Goudet, 2005) and subsequent nested population clusters using CLUMPAK (Kopelman, Mayzel, Jakobsson, Rosenberg, & Mayrose, 2015).
Following STRUCTURE, individual clusters were assessed for HWE conformance and LD significance since nonconformance in combination with significant levels of LD may indicate recent gene flow has occurred between clusters or that family groups were sampled. Spatial autocorrelation (r) implemented in GenAIEX (Peakall & Smouse, 2006, 2012 was conducted to test for the presence of spatial structure for the genetic and geographic datasets. Evenly spaced lag distance of 30 km was selected based on the sampling distribution and to provide a sufficient number of pairwise comparisons.

| Mitochondrial sequencing and analysis
All 688 samples were sequenced to capture all existing haplotypes in the sampling locations but only 558 individuals were successfully sequenced (Table 1). 622 base pairs (bp) of cytochrome b gene were PCR amplified using forward primers MDL5 5′-TTACATGAATTGGCAGCCAACCAG-3′ and reverse primers MDL3 5′-CCCACAATTAATGGGCCCGGAGCG-3' (Fernando & Lande, 2000). The primers MDL5 and MDL3 amplify a 622 bp region of mitochondrial DNA including ~100 bp of cytochrome b, the transfer RNAs for threonine and proline, and ~350 bp of the control region or d-loop. A different reverse primer, mtCR3 5′-GTC ATT AAT TA B L E 1 Genetic diversity of African savanna elephants based on the sequence of 622 bp of mtDNA CCA TCG AGA TGT CTT ATT TAA GAG G-3′, was used to samples that did not amplify successfully with the MDL3 reverse primer.
PCR reactions were performed with the initial polymerase activa- analysis implemented in PAUP* 4.0b10 (Swofford, 2002). Support for the nodes for each analysis was assessed using 1,000 bootstrap iterations. We constructed a neighbor-joining (NJ) tree from MEGA version 7 (Kumar, Nei, Dudley, & Tamura, 2008) to determine if both approaches produced tree with similar topology. Haplotype diversity (h) and nucleotide diversity (π) were calculated using Arlequin version 3.5 (Excoffier et al., 2005;Excoffier & Lischer, 2010) and constructed median-joining (MJ) network PopArt 4.8.4 (Leigh & Bryant, 2015). Hierarchical STRUCTURE analysis indicated that the uppermost level of population structure was represented by two clusters (K = 2; Elephants in the Lake Manyara region (MAR) appear to be more closely related to those in the NCA across the East African Rift wall than they are to the neighboring Tarangire subpopulation (Figure 3c vs. e). Three subpopulations were identified within the MAR region but there was weak support of this clustering (ΔK = 7; Figure 3g;
Principal coordinate analysis (PCoA) was performed to determine the number of clusters using F ST values and genetic distance matrix between all pairs of individuals. At least two clusters were identified in northern Tanzania as shown in Figure 4. Samples ordinated close to one another are more closely related than those ordinated far away. RNP and TNP elephants were clustered together while MAR was clustered with the SE. While the Nothern Tanzania sites (NSE, SSE, NCA, and MAR) and SGR clustered on axis 2, RNP was closer to them on axis 2 than to TNP. On Axis 2, SGR was seen as different than any other site (Figure 4).
We found no significant correlation between geographic distance and genetic distance for mtDNA markers (r = −.1946, p = .3590) and nuclear loci (r = .9006, p = .1630) between Ruaha and Selous.
All F ST values, except between RNP and TNP for the mtDNA, showed significant genetic differentiation (  However, we also observed remarkable differences in the haplotype distribution between MAR and TNP which are less than 60 km apart. Phylogenetic analyses of the 622-bp mtDNA alignment using MP and NJ produced trees with similar topologies and we only presented MP (Figure 8). We detected a clear subdivision into F and S-clades with 96% bootstrap values on both branches. The S-clade was further subdivided into two subclades: savanna-wide and southeast-savanna (Figure 8). The savanna-wide subclade only has 62% bootstrap and our support for this subclade is not as strong as the other two. Our data also suggest a clear pattern of haplotype distribution between northern and southern Tanzania. One hundred percent of Selous samples carried haplotypes with the southeast-savanna (Figure 8), whereas the majority of elephants in the SE carry haplotypes in the F-clades which is shared with the forest elephants (Ishida et al., 2013) and was not observed in Ruaha and Selous.
Ruaha elephants had only three haplotypes (one unique) whereas in Selous GR there were ten haplotypes (eight unique) (Appendix S3).
In Ruaha, one haplotype was carried by 84% of individuals sampled, and only one haplotype was unique to Ruaha. Only one was shared between Ruaha and Selous. In Selous, eight out of ten haplotypes were unique to Selous, of which two were previously published (Debruyne, 2005;Ishida et al., 2013). In contrast, we found that geographic normalized genetic distance (F ST /km) was twenty-five times lower between Tarangire and Ruaha elephants than for Tarangire and Manyara. This suggests extensive gene flow over the long distance (>400km) between Tarangire and Ruaha prior to the eventual closure of corridors between them that F I G U R E 8 A neighbor-joining tree constructed from 32 mitochondrial DNA sequences from this study and reference sequences from previous studies: GR0015-0042, KE4540 (Ishida et al., 2013), UG1, BO1, TA1, MO1, and SA3 (Debruyne, 2005), Addo5 (Eggert et al., 2002), Samburu (Ahlering et al., 2012). Forest elephant (Eggert et al., 2002) and Asian elephant (Fernando et al. 2003) were used as outgroups. Blue = East-central, Orange = Savanna-wide, purple = Southeast-savanna. Numbers on the branches represent the percentage bootstrap values > 50% for 1,000 replicates occurred over the past 30 years (Riggio & Caro, 2017;Caro, Jones, & Davenport, 2009). It takes multiple generations for the effects of genetic differentiation to reflect in populations that were once connected. For this reason, we still see low or no genetic differentiation between Ruaha and Tarangire. The discrepancy between F ST values based on mtDNA versus microsatellites is notable. The "mtDNA loci are generally a more sensitive indicator of population structure than are nuclear loci, and mtDNA estimates of F ST -like statistics are generally expected to exceed nuclear loci" (Zink & Barrowclough, 2008).

| D ISCUSS I ON
Differences in the mutation rates of nuclear versus mtDNA markers may explain this discrepancy. Nuclear mutation rates are generally lower than mtDNA mutation rates (~ a factor of 10; Brown, 1983;Zink & Barrowclough, 2008), and F ST estimates are products of events occurring over a period four times longer than for mtDNA and effective population (N e ) size for nuclear loci will frequently be four times that of mtDNA loci (Zink & Barrowclough, 2008). However, nuclear markers such as microsatellites have more resolving power than does mtDNA (Edwards & Bensch, 2009). Interestingly, Ruaha elephants which are more than 400 km away from Tarangire, share the same haplotypes with Tarangire elephants, and they are distinct from Selous GR, which is much closer geographically. This suggests historical female-mediated gene flow between Ruaha and Tarangire, and the similarity of nuclear loci supports both high gene flow by both sexes. Also, there is evidence for gene flow between elephants in Serengeti ecosystem with Ruaha. While the Northern Tanzania sites and SGR clustered on axis 2, RNP was closer to them on axis 2 than to TNP. On Axis 2, SGR was seen as different than any other site (Figure 4). Probably elephants previously moved from southern Serengeti to Ruaha without passing through Tarangire.

West of the
Unfortunately, wildlife corridors between RNP and TNP and between RNP and Serengeti are now entirely blocked (Riggio & Caro, 2017).

| Tarangire-Manyara corridor
Within the Tarangire-Manyara ecosystem, we detected significant genetic differentiation: Manyara formed one cluster, and the Tarangire elephants formed a second cluster. Both nuclear and mitochondrial DNA data showed significant genetic differentiation. Our results suggest limited gene flow between Tarangire and Manyara probably due to increased habitat loss along wildlife corridors that previously connected these ecosystems, and which have been known to be under extreme threat for decades (Mwalyosi, 1991 Tarangire and Manyara. This is also unexpected because the Manyara population is relatively low compared with Tarangire. Perhaps the high diversity in Manyara is due to both gene flow with the NCA and a much higher population size in the 1970s (Douglas-Hamilton, 1973).
One would predict that this diversity will decline significantly over the next few generations as a consequence of inbreeding due to loss of corridors to the NCA (O'Brien et al., 1985).
Furthermore, when we compared haplotype distributions between the east and west of the Rift wall, we found that most Manyara elephants shared haplotypes with Ngorongoro elephants. We detected a significant difference in haplotype distribution. For example, in Tarangire, only four elephants were carrying a haplotype in the Eastcentral subclade, but these haplotypes were common in Manyara. In Tarangire, there were no elephants carrying haplotypes in the southeast-savanna subclade, but haplotypes in the East-central subclade were observed in Manyara ( Figure 6). It appears there is limited female-mediated gene flow between Manyara and Tarangire although there is no landscape barrier between them, other than an increasing human footprint. Alternatively, there could be cultural or behaviorally mediated barriers to genetic connectivity among subpopulations.
Females' social behavior of remaining in their natal groups might drive discontinuities among subpopulations. The only viable migratory route between Tarangire and Manyara Ranch based on GPS data is the Kwa Kuchinja corridor (Kikoti, 2009

| Ngorongoro-Manyara corridor
Wildlife corridors between the Tarangire-Manyara and Serengeti ecosystems are under critical risk of being blocked completely (Lobora, Mduma, Foley, & Jones, 2010). The corridors between Lake Manyara National Park (LMNP) and the Ngorongoro Conservation Area (NCA) are essential for connecting the two ecosystems. Effects of habitat fragmentation on wildlife corridors between NCA and LMNP can be traced back to the 1940s when tsetse eradication allowed the expansion of human settlement in the Mbulu areas blocking the forest corridors (Homewood & Rodgers, 1991). Concerns about loss of corridors have been ongoing in the ecosystem since the 1970s. Increased human activities, especially agricultural settlement around Lake Manyara (Borner, 1985;Douglas-Hamilton, 1973), TA B L E 4 A summary of main results relevant to stakeholders to show where the focus will be required to restore or protect wildlife corridors in Tanzania (Caro et al., 2009;Mduma et al., 2010)

Protected areas Main results Conservation actions/recommendations
Serengeti NP vs. Grumeti GR and Loliondo GCA High genetic similarity between these protected areas There is enough evidence of gene flow between these areas. No major conservation issues identified (Kikoti, 2009) Ngorongoro CA vs. Maswa GR and Mwiba Ranch (SSE) Three clusters identified but there was weak support of genetic differentiation There were a lot of shared mtDNA haplotypes Areas south-west of the Ngorongoro particularly Endulen, Esere, Kakesio are essential for connecting elephants with Mwiba Ranch and Maswa GR. The Ngorongoro Conservation Area Authority should increase protection in these areas and reduce human settlement in some areas that are frequently used by elephants.
Ngorongoro CA and Manyara NP Both mtDNA and microsatellite data show high genetic similarity between Ngorongoro and Manyara A corridor between Lake Manyara to Ngorongoro through Karatu is completely blocked. The best option is to use the corridor south of Lake Manyara NP through Marang Forest to South of Ngorongoro. This corridor was gazettted but human encroachment has occurred in these areas (Kisui, Bernard per.comm). There is evidence that the Silela corridor (north of LMNP) is still open, and there is likely movement along the escarpment by Eyasi. Elephants can still move up through Marang Forest, but they are then blocked by agriculture in the highlands before they can reach the Northern Highland Forest Reserve. Silela corridor is also open and being used but under threat (Morrion& Bolger 2014; Morrison et al., 2016) Lake Manyara and Manyara Ranch Weak support for genetic differentiation Genetic evidence suggests recent or ongoing gene flow between these two areas. However, increasing number of settlements and expansion of Mto wa Mbu town between Manyara Ranch and Lake Manyara could pose a significant threat to this corridor. The government should discourage farming around these areas and keep the traditional pastoralism practiced by Maasai people for many years or include these villages in a WMA.

Manyara and Tarangire
Limited gene flow between these protected areas Only two haplotyepes were shared between Manyara and Tarangire. Manyara had significantly higher haplotype diversity than Tarangire Historically there were at least 9 identified corridors (Mwalysosi, 1991). Now there are about three wildlife corridors connecting these protected areas: Kwakuchinja Corridor through Burunge WMA, Mswakini Chini and Mswakini Juu. Apart from Kwakuchinja which is currently secured after establishing a WMA, other wildlife corridors are threatened. There is evidence for movement of elephants between Tarangire and Manyara Ranch but our data suggest limited gene flow. Immediate action needs to be taken now to protect these wildlife corridors. These corridors are essential to connect Tarangire and Serengeti ecosystems.

Tarangire and Ruaha
Both microsatellite loci and mtDNA show genetic connectivity between Tarangire and Ruaha Wildlife corridors between these protected areas seem to be blocked completely (Riggio & Caro, 2017). More research needs to be done to determine if there are any remaining movements of elephants between these ecosystems Ruaha and Selous Only one haplotype was shared between Ruaha and Selous. Three clusters were identified between Ruaha and Selous. The Eastern Arc Mountains seem to act as a barrier between these two ecosystems A more intensive genetic study needs to be done between Ruaha and Selous. We support recommendations provided by Jones et al. (2012) to restore corridors between the two ecosystems, particularly between Selous and Udzungwa through Kilombero reserve threaten the viability of wildlife corridors and the expansion of Mto wa Mbu agricultural settlement is of particular concern (Mwalyosi, 1991). Wildebeests in Tarangire-Manyara are distinct from the Serengeti without mixing for thousands of years, separated by the rift, whereas wildebeest have been migrating between Tarangire-Manyara and Lake Natron to the North (Morrison & Bolger, 2014;Morrison et al., 2016).
The Ngorongoro elephants show admixture with elephants from Lake Manyara NP, which is also geographically very close. This admixture indicates recent gene flow between these subpopulations.
However, most corridors have likely been lost due to intense agriculture on top of the rift (Table 4). Our study provides some insights into historical and contemporary gene flow between the two protected areas. There are limited number of natural corridors across the rift but these natural corridors are also compatible with agriculture.
Consequently, human settlements have rapidly expanded into these limited natural corridors and blocked seasonal migration of elephants between the Ngorongoro/Serengeti and Lake Manyara. Although the Rift wall may impede the movement of elephants, there is enough evidence that there was gene flow between the two protected areas, including recent telemetry data from at least one male which moved over the rift from Natron to Loliondo GCA (Kikoti, 2009). In con- trast, it appears that there was little mixing between Tarangire and Manyara. Now that the corridors between the NCA and Manyara are largely blocked, Lake Manyara elephants are likely completely isolated.

| Connectivity between Ruaha and Selous
Our genetic data show that elephants from Ruaha and Selous are divided into two divergent mtDNA lineages: savanna-wide and Male-mediated dispersal has been documented among elephant populations in both Uganda (Nyakaana & Arctander, 1999) and Kenya (Okello et al., 2008). In both studies, there was a lack of congruence between mitochondrial DNA and nuclear-based variation.
Mitochondrial DNA data showed significant differentiation, whereas nuclear data showed low genetic subdivision between populations (Nyakaana & Arctander, 1999;Okello et al., 2008). This difference between markers was attributed to differences in the modes of mutations between mitochondrial and nuclear markers. Female elephants stay in family groups while males are ejected from groups after sexual maturity. The home range for elephant family groups is between 15 and 52 km 2 (Douglas-Hamilton, 1973). Thus, mitochondrial markers would likely remain restricted to specific localities (Nyakaana & Arctander, 1999). Although this difference in haplotype frequency was expected between Ruaha and Selous, lack of shared haplotypes could be attributed to the presence of the Eastern Arc Mountains, separating the two ecosystems. Furthermore, there was no significant correlation between genetic and geographic distance.
Genetic differences between Ruaha and Selous have likewise been noted in other species. A recent study on lions (Panthera leo) in Tanzania uncovered significant genetic differentiation between lions of Selous and Ruaha (Smitz et al., 2018). This differentiation could be a combined effect of both anthropogenic and environmental/climatic factors (Smitz et al., 2018). The presence of the Eastern Arc Mountain chains associated with the land use patterns may present a significant biogeographical barrier to lion dispersal (Smitz et al., 2018). Similarly, phylogenetic analyses based on mitochondrial DNA of sable antelope (Hipotragus niger) identified unexpected clear, distinct lineages between Ruaha and Selous which was attributed to the presence of the Eastern Arc Mountains (Pitra, Hansen, Lieckfeldt, & Arctander, 2002). Furthermore, Pitra et al. (2002) found that the initial allopatric fragmentation is geographically consistent with the discontinuous distribution of miombo habitats inside and outside of this montane circle in East Africa. There is also a clear difference in the vegetation cover between the western (Ruaha) and eastern  (Lindsay & Lee, 2006). However, elephants are also known to avoid areas with steep slopes (Wall, Douglas-Hamilton, & Vollrath, 2006).
For that reason, mountain ranges may act as barriers for elephants in some cases (Epps et al., 2013;Wall et al., 2006). Elephants can negotiate relatively steep slope over the short distances, but long-distance movement over steep terrain may be restricted by energetic limitations (Wall et al., 2006). These barriers are semipermeable to the movement of elephants (Sawyer et al., 2013).
Female elephants are philopatric and remain with their natal herd  whereas males are ejected from the herd upon sexual maturity and subsequently facilitate gene flow between herds . Using nuclear markers, there was significant genetic differentiation but low F ST between Matambwe in the Selous and Ruaha, suggesting some amount of gene flow. Nuclear genes are not subject to the same inheritance limitations as mitochondrial DNA because males disperse nuclear markers (Brandt et al., 2014). Our data indicate male-biased gene flow, with little evidence for female-mediated gene flow between Ruaha and Selous.
One of the haplotypes that Selous shares with other sites (SSNG01) is relatively common, being found in 5 other sites, while the second one (SSNG02) is found in 2 other sites and is more common in NCA than in Selous. While one explanation for this may be unidirec-

| History of elephant recolonization in Northern Tanzania
Mitochondrial DNA has been used to infer the ancestry of populations because it is inherited in haploid fashion. There is a hypothesis that elephants which recolonized the Serengeti came from two sources, one from the south and the other one from the north (Sinclair et al., 2008) (Ishida et al., 2013). Interestingly, these genetic differences inherited from the founder populations in the 1950s are still evident in the genetic structure of Serengeti elephants today, despite there being no geographic barriers that separate the northern and southern Serengeti populations.

| Implications for conservation
The full consequences of habitat fragmentation on population structure for species with long-life spans and generations may take time to be observed, because there is a time lag between changes to habitat and when the full implications of these changes are experienced by wildlife (Bennett, 1998(Bennett, , 2003. However, Lobora et al. (2018) identified early signs of genetic differentiation among young cohorts of elephants in south-western Tanzania due to habitat fragmentation of miombo woodland. In recent years, the analysis of the structural connectivity of protected areas has been done at the national level (Riggio & Caro, 2017 Newmark, 2008;Riggio & Caro, 2017). However, minimal mitigation efforts have yet to take place. Likewise, our study has shown high genetic similarity between Ruaha and Tarangire elephants, but the corridors between them are now completely blocked (Table 4).
The possible detrimental effects of this recent isolation on these two previously intermingle populations will take many years to become manifest due to the slow rate of genetic variation decay and the ani-  (Gaynor et al., 2018). In the long run, constructing wildlife overpass crossings, for example, across the Arusha-Babati road, may be the best option to increase genetic connectivity between subpopulations in northern Tanzania although the larger threat is land conversion through agriculture. Traditional protected area systems have long been considered the most effective way of protecting wildlife in Tanzania. Indeed, most wildlife species are found within these protected areas. However, the role of more human-dominated landscapes, especially those adjacent to protected areas, is essential for dispersal areas for large mammals like wildebeest and elephants.
Here, we want to emphasize other categories of protected areas which can accommodate both human use and wildlife conservation, such as WMAs and wildlife ranches, like Manyara Ranch.
There is a need to recognize the importance of conserving wildlife in human-dominated areas (Ogutu, Kuloba, Piepho, & Kanga, 2017).
Formal protected areas such as national parks and game reserves are not enough for the conservation of far-ranging species such as elephants. Communities that have provided some portion of their land have helped provide habitats for many species and villagers have started benefiting financially from conservation projects. Wildlife Management Areas, for example, have been a source of income for some villages. These funds can be used for social development projects such as building schools, health centers, and water supply.
However, a proper land use plan and community rights of occupancy should be considered before implementing these projects. We recommend village leaders to consult organizations such as the Ujamaa Community Resource Team (UCRT) which has empowered many villages in northern Tanzania to secure rights of their natural resources and land. UCRT also help these communities by representing their land rights, advocating on their behalf to local and national government, and securing legal ownership of their traditional lands (http:// www.ujama a-crt.org). Indeed, the future of wildlife conservation in Tanzania, particularly for far-ranging species, and in the face of increasing isolation of protected areas, is reliant upon participation of communities whose livelihoods depend on these same areas. Every effort must be made to ensure it is a mutually beneficial arrangement as recommended in Table 4.

CO N FLI C T O F I NTE R E S T
None declared.