Introduction

The outstanding geographic features and the diversified climate of the tropical eastern Africa have resulted in a biodiversity hotspot. The pattern of species distributions currently observed in this region results from a combination of historical factors, and the high level (80%) of species endemism observed on the tropical mountains of eastern Africa indicates that these plants have been in isolation for a long period (Hedberg 1969; Assefa et al. 2007; Linder 2014). Besides geographical isolation, environmental diversity in these regions had considerably influenced the distribution of both plant and animal species. Surprisingly, there is a significant similarity in biodiversity on different mountain massifs (Hall 1973; Kadu et al. 2013), suggesting the existence of historical links that connected some of the populations, probably during the Pleistocene era when climatic changes favored species dispersal across existing patches of flora, and the subsequent extinction of these patches as they sought refuge in the mountains (Hewitt 1999; White 1981). Presently, most of the species inhabiting these mountains such as Hagenia abyssinica (Bruce) J. F. Gmel, (Rosaceae), display local endemism and exist in isolated population patches.

Several studies have been conducted on widespread and rare endemic species of montane forests of Ethiopian highlands and the tropical mountains of eastern Africa, e.g., Arabis alpina (Koch et al. 2006; Ehrich et al. 2007; Assefa et al. 2007), Lobelia giberroa (Kabede et al. 2007), Prunus africana (Kadu et al. 2011, 2013; Mihretie et al. 2015), and Erica spp. (Gizaw et al. 2013). Among the remarkable findings of these studies are the effects of past climate and environmental heterogeneity on species genetic diversity, phylogeographic and recolonization histories, and implication on genetic conservation. During the Pleistocene epoch, climate fluctuations caused expansion and contraction of the montane forests, eventually resulting into more fragmented montane forests (Kabede et al. 2007). Recently, the incorporation of ecological niche models (ENM) in phylogeographic and genetic structure studies to map the past and current and to predict the future geographic distribution of species has led to better understanding of evolutionary history and the role played by combined historical processes in shaping genetic structure in present populations. ENM is based on interactions between biotic and abiotic factors within specified known habitats of a given species (Guisan and Zimmermann 2000; Peterson 2003a).

In Africa, only a few studies have applied ENM to investigate the influence of climate fluctuations on species distribution, e.g., in Vitellaria paradoxa (Allal et al. 2011) and in Chasmanthera dependens Hochst (Iloh and Ogundipe 2016), highlighting probable refugial areas and potential future distribution patterns for species in West Africa. However, despite eastern Africa being a diversity hotspot and a potential Quaternary glacial refugia, ecological niche models are yet to be conducted on species inhabiting the “sky islands.” Besides climate changes, anthropogenic activities (e.g., expansion of agricultural lands) caused by tremendous increase in human population over the last century have also had an impact on the patterns of vegetation distribution. Habitat loss led to rapid decline in population sizes, prompting Hedberg (1969) to observe that human impact caused disjunction of east African montane forests on Mt. Elgon, Cherang’ani hills, Aberdare Ranges, and Mt. Kenya.

H. abyssinica, a long-lived tree species in a monotypic genus, is found in most Afromontane forests, from the northern parts of Ethiopia and Sudan to the southern regions of Malawi and Zambia, displaying a disjunct distribution (Hedberg 1969). This species is dioecious, anemogamous, and anemochorous (Negash 1995). Its medicinal benefits, mainly anthelmintic, have been relied on for decades by most African communities for treating gastrointestinal parasites in both man and livestock (Woldemariam et al. 1992; Scantlebury et al. 2013; Feyssa et al. 2015). Other uses include source of fuel and timber for construction (Feyissa et al. 2007). H. abyssinica is ecologically linked to a rich biodiversity with a record of stable coexistence with other Afromontane flora (Seburanga et al. 2014). However, overexploitation and habitat destruction have led to reduction in its population sizes and is now classified as an endangered species in Ethiopia and Rwanda (Negash 1995; Feyissa et al. 2007; Seburanga et al. 2014). It is endemic to Afromontane forests, some of which are separated by large plains and grasslands, with several being on the opposite sides of the Great Rift Valley. Therefore, it is an interesting species to study the patterns of gene flow and population history.

Previously, H. abyssinica has been studied in effort to outline its genetic diversity, using randomly amplified polymorphic DNA (RAPD) (Kumilign 2005), inter-simple sequence repeats (ISSRs) (Feyissa et al. 2007), chloroplast microsatellites (cpSSR) (Ayele et al. 2009), and amplified fragment length polymorphism (AFLP) (Ayele et al. 2011). However, these studies focused only on the Ethiopian highlands. The current study sampled 11 natural populations of H. abyssinica distributed in Afromontane forests in Ethiopian highlands and tropical mountains of east Africa. The principal objectives were to quantify levels of gene flow and current genetic diversity and to outline the population structure of natural populations of H. abyssinica across the region using microsatellite markers. We estimated historical and current factors that shape population structure, e.g., correlation between genetic and geographic distances, and used ENM to map the potential distribution zones of Hagenia during the LGM and present. Information on population genetic patterns of this species lays a foundation for further studies on endemic species and deepens our understanding on the roles of historical dynamics on species distribution and dispersal processes.

Materials and methods

Sampling

H. abyssinica has a disjunct distribution in the scattered high mountains of East Africa (Hedberg 1969). Our focus was on Ethiopian highlands and eastern Africa (Kenya, Tanzania, and Uganda), where anthropogenic activities (farming and logging) have negatively impacted on the populations. We collected samples from fragmented natural populations on major mountains where this species is known to grow at a narrow distribution range. Leaf samples were obtained from 168 individuals in 11 natural populations (Fig. 1, Table 1). Individuals in each population were sampled at a distance of between 100 and 250 m depending on the size of the population. Due to inaccessibility of some individuals in certain populations, our sample sizes varied among populations. A voucher specimen was collected from each sampling site and deposited at the Herbarium of Wuhan Botanical Garden, CAS (HIB). Young leaves were put in manila envelopes, dried in silica gel, and stored at −20 °C until DNA was extracted.

Fig. 1
figure 1

A map showing 11 sampling sites of Hagenia abyssinica. The sites are represented by circles while the colors represent the genetic groups to which individuals were assigned to based on the STRUCTURE results (K = 3)

Full size image
Table 1 Localities, sample size, and genetic diversity of each sampled population
Full size table

DNA extraction and SSR genotyping

Total genomic DNA was extracted using the MagicMag Genomic DNA Micro Kit (Sangon Biotech Co., Shanghai, China) following the manufacturer’s instructions. Eleven species-specific SSR markers (Gichira et al. 2016) were used to genotype the 168 samples. The forward primers were labeled with 6-FAM fluorescent dye on the 5′ end. The 25 μL PCR mixture contained 2 μL 50 ng genomic DNA, 2.5 μL 10× Taq buffer (plus Mg2+), 0.25 mM of each dNTP, 0.25 μM of each primer, 0.2 U Taq polymerase (TaKaRa Bio, Dalian, China), and PCR water. PCR amplification was conducted using T100™ Thermal Cycler (Bio-Rad, CA, USA) with the following cycling conditions: an initial step of 4 min at 95 °C, followed by 30 cycles of 30 s at 95 °C, 30 s at 53–58 °C, 30 s at 72 °C, and a final extension step for 10 min at 72 °C. The quality of PCR products was checked on 2.0% agarose and then separated using an ABI 3730 XL automated sequencer (TsingKe Biotech, Beijing, China). Alleles were called using GeneMarker 2.2.0 system (Soft Genetics).

Data analyses

Genetic diversity

Micro-Checker 2.2.3 (Van Oosterhout et al. 2004) was used to detect null alleles. The polymorphic information content (PIC) and deviations from the Hardy-Weinberg equilibrium (HWE) with the Bonferroni correction for each locus were checked in Cervus 3.0 (Kalinowski et al. 2007). GenePop 4.0 (Rousset 2008) was used to check for deviations from the HWE at population level. Genetic diversity details including observed heterozygosity (H O), expected heterozygosity (H E), the number of private alleles(A P), observed alleles (N a), effective number of alleles (N e), Shannon’s diversity index (I), and inbreeding coefficient(F IS) were done in GeneAlex 6.5 (Peakall and Smouse 2012).

Population structure

Global genetic differentiation coefficient (F ST) and the analysis of molecular variance (AMOVA) were conducted in Arlequin 3.5 (Excoffier and Lischer 2010). The principal coordinate analysis (PCoA) and pairwise genetic matrix were done using GeneAlex 6.5 (Peakall and Smouse 2012). A Mantel’s test for isolation by distance was conducted using the IBDWS 3.23 (Jensen et al. 2005), a web-based program using Nei’s genetic distance (Nei 1972) and geographic distance (km) with 999 permutations.

Population structure and individual-based differentiation analyses were carried out using the Bayesian clustering method implemented by STRUCTURE 2.3.3 (Pritchard et al. 2000). STRUCTURE was run using the admixture model, assuming uncorrelated allele frequencies (independent) and without prior information about the populations. Preliminary simulations for K ranged from 1 to 10, each with 10 iterations. One hundred thousand replicates were used for the Markov chain Monte Carlo with a burn-in period of 50,000. Population clusters (K) were calculated using Structure Harvester (Earl 2012), based on ∆K statistics (Evanno et al. 2005).

Gene flow estimation

We used the Bayesian inference strategy with thermodynamic integration implemented in MIGRATE-n program, which uses coalescent and Markov chain Monte Carlo (MCMC) (Beerli 2006) to estimate the rates of gene flow between the three clusters outlined by STRUCTURE. In this analysis, we applied the Brownian approximation model assuming a constant mutation rate across all loci and turned on MCMC static heating in four chains, with no swapping. The different temperatures were set at 1, 1.5, 3, and 104 while the initial parameters for θ and M were generated by the program’s inbuilt F ST calculations (Beerli and Felsenstein 1999). The analysis was set to run in a single long chain with 1,000,000 burn-in, 500,000 sampled steps, and 5000 recorded steps. MIGRATE-n calculates the mutation-scaled effective population size Θ (4N e μ; N e is the effective population size and μ is the mutation rate per locus in a generation), and the mutation-scaled migration rate M (m/μ; m is the immigration rate among populations in a generation) (Beerli and Palczewski 2010). To get the actual effective number of migrants (Nm) per generation, the mutation-scaled migration rate (M) is multiplied with the mutation-scaled effective population size (Θ) and then divided by four (Nm j→i = [Θ i × M j→i]/4).

Environmental data

Nineteen bioclimatic variables (BIOCLIM) were obtained from the WorldClim database (http://www.worldclim.org), derived from annual and monthly mean temperatures and precipitation for the period between 1950 and 2000 (Hijmans et al. 2005). The 19 environmental variables were subjected to a jackknife analysis in MAXTENT, in order to account for correlation among data and to select variables that showed considerable influence on the distribution of the species. Eventually, the following five environmental variables: temperature seasonality (Bio4), temperature annual range (Bio7), isothermality (Bio3), mean temperature of the wettest quarter (Bio8), and annual precipitation (Bio12) were selected for the present and Last Glacial Maximum (LGM, −21 ka, MIROC 3.2) analysis at 2.5 arcmin resolutions. The occurrence data of H. abyssinica was obtained during our field survey, from Ayele et al. (2009) and from the potential natural vegetation (PNV) map (www.vegetationmap4africa.org; van Breugel et al. 2015). In total, 109 presence points (Online Resource 1) were recorded from the northern parts of Ethiopia to the southern parts of Africa in Zambia and Malawi.

Ecological niche modeling

ENM approaches require continuous layers of environmental variables and geographic positioning of species occurrence data (Stockwell 2006). Based on these bioclimatic variables and the obtained presence data, ENM was conducted in MaxEnt program 3.3.3 (Phillips et al. 2006), a species distribution modeling tool which has been credited on its simplicity in use and ability to analyze data from relatively small sample sizes (Wisz et al. 2008), resulting in high model accuracy (Merrow et al. 2013). The parameters were set at default with 500 iterations without duplicates. We selected the random seed option and used 70% of the data in training and 30% in model testing.

MaxEnt output is in the form of a cumulative prediction map that is interpreted based on receiver-operator curve (AUC). AUC is an independent measure used to assess model suitability and it ranges from 0.5, interpreted as random accuracy, to 1.0, perfect discrimination (Elith et al. 2006). ENM models with AUC values >0.7 are considered to be strong and ideal for the data. The predictions were mapped using DIVA-GIS 7.5 (http://www.diva-gis.org).

Results

Genetic diversity

Null alleles were detected at loci Hgssr1, Hgssr5, Hgssr6, and Hgssr10 in some populations. The average PIC was 0.601. Except for Hgssr 3 and Hgssr 11, significant deviations from the HWE were observed at all the other loci (P < 0.001) (Table 2). However, at the population level, only two populations significantly deviated from the HWE. In total, 71 alleles were recorded at all the 11 microsatellite loci. The number of alleles observed per locus (N a) varied from 2 in Hgssr5 and Hgssr10 to 13 in Hgssr6 and Hgssr8 (average = 6.5); values of gene diversity per locus were recorded in Table 2. The mean observed heterozygosity (H O) and mean expected heterozygosity (H E) were estimated at 0.477 and 0.439, respectively. Details of genetic diversity including the number of alleles observed (N a), effective number of alleles (N e), Shannon’s diversity index (I), and inbreeding coefficient (F IS) per population are shown in Table 1. Nine private alleles were discovered in five populations at varying frequencies. The Simiens population had the highest number of private alleles (4) - three at locus Hgssr8, with frequencies of 0.038, 0.154, and 0.077, and one allele at locus Hgssr6, with a frequency of 0.077. Two private alleles were found in the Elgon population at loci Hgssr2 and Hgssr8.

Table 2 Genetic variation statistics of 11 microsatellite loci in individuals of Hagenia abyssinica
Full size table

Population structure and gene flow

Hierarchical AMOVA revealed strong population structure (F ST = 0.32) and there was limited inbreeding among populations (Online Resource file 2 Table S1). PCoA shows clear differentiation among geographic regions along the first three axes accounting for 69.89% of the variation (the scatter plot of coordinates 1 and 2 is shown in Fig. 2). The pairwise F ST values between pairs of the 11 analyzed populations ranged from 0.024 (BLI and BLII) to 0.379 (MK1 and RW). Generally, Ethiopian populations appeared to be less differentiated, with the least F ST, while Rwenzori (RW) was more differentiated against most of the other populations (Table 3). Bayesian clustering, based on ∆K statistics, defined three genetic clusters (K = 3) of H. abyssinica, with few individuals displaying mixed ancestry. This clustering placed RW-BLI-BLII-SM sample sites in one cluster (P1), KM-MR in a second cluster (P2), and CH-EL-AB-MKI-MKII in a third cluster (P3) (Fig. 3, Online Resource 2; S2, Fig. S2.1).

Fig. 2
figure 2

A scatter plot of the first two PCoA axes showing the relationship of the 168 individuals from 11 populations of Hagenia abyssinica. The first and the second coordinates account for 30.41 and 22.69% of the variation, respectively

Full size image
Table 3 Pairwise fixation index (F ST) values, between populations of Hagenia abyssinica, based on 11 microsatellite loci
Full size table
Fig. 3
figure 3

Population genetic structure of Hagenia abyssinica, showing the proportion of membership of the 168 individuals in each of the three inferred clusters based on Bayesian probabilities. PI (red) had populations BLI-BLII-SM-RW, PII (green) KM-MR, and PIII had populations AB-MK I-MKII-CH-EL

Full size image

Mantel’s test revealed significant correlations between Nei’s genetic distance and geographic distances (km) r = 0.425, P < 0.05 (Fig. 4). We estimated the mutation-scaled migration rate (M) as well as the mutation-scaled effective population size (Θ) for the three genetic clusters, based on an unconstrained migration model which allowed genes to flow towards any direction. The values of M and Θ were used to calculate the effective number of migrants (Nm) per generation. Θ values for genetic clusters I, II, and III were estimated at 0.0985 (97.5%, confidence interval 0.0958–0.1000), 0.0965 (0.0836–0.1000), and 0.0152 (0.0075–0.0195), respectively. Accordingly, the migration rates between the genetic clusters in each generation were recorded as follows: P2–1 = 0.5, P3–1 = 0.9, P1–2 = 0.4, P3–2 = 0.6, P1–3 = 0.4, and P2–3 = 0.1 (Online Resource 2; Table S3, Fig. S2.2).

Fig. 4
figure 4

Scatterplot showing positive correlation between the geographic distance (km) and Nei’s genetic distance (Mantel test P < 0.05)

Full size image

Past and present potential distribution of H. abyssinica

Jackknife test results indicated that “temperature seasonality” (Bio4) had the highest gain and therefore contributed most of the information to the ENM. The environmental variable “annual precipitation” (Bio12) had the most unique information and, therefore, eliminating it from the analysis resulted in a reduction in the model’s gain. The model’s evaluation test displayed high efficiency and robustness based on the algorithm’s values (AUC 0.992).

MAXTENT proved the suitability of the used occurrence zones for H. abyssinica, whose predicted geographical distribution is consistent with the currently known habitation zones. Based on the past climate (LGM) model, high species density was observed in the eastern African region (Fig. 5a). This region is dominated by some of the highest and largest mountains in Africa and represents our sampling sites for population genetic study. There were slight but noteworthy changes on species distribution in some areas, specifically the southern regions of eastern Africa, which were unfit for Hagenia during the LGM but became suitable under the current climate models (Fig. 5b).

Fig. 5
figure 5

The potential geographical distribution zones of Hagenia abyssinica in the past (a) and present (b) climates, based on ecological niche modeling using MaxEnt. Red and orange colors indicate zones with high probability of occurrence

Full size image

Discussion

Gene diversity (H E) varied among collection sites with the highest observed in Meru (0.536) and the lowest in Rwenzori (0.316), revealing variation in levels of diversity in populations on different geographic regions. Despite the recent reduction and the highly fragmented populations, H. abyssinica has managed to maintain a significant level of genetic diversity (H E = 0.439 and H O = 0.477) in our area of study. This discovery is contrary to what was expected; while populations of this species have had a history of isolation and its populations are greatly fragmented, it is probable that the effects of genetic drift and excessive inbreeding could have negatively affected the genetic diversity of this species. Comparably, high levels of genetic diversity have been recorded in other species inhabiting the Afromontane forests in tropical Africa, e.g., in P. africana H E = 0.66 (Kadu et al. 2013), and generally in isolated Afromontane forest patches. This observation is in line with the Aguilar et al. (2008) report that negative effects of recent population reduction and habitat fragmentation on genetic diversity may not be felt on the current mature individuals but will be reflected in their future generations.

Although endemic species are ecologically restricted, they may tend to exhibit almost the same levels of genetic diversity as widespread species (Hamrick and Godt 1989). The low and negative inbreeding coefficient (F IS) values observed in most of the populations support the fact that H. abyssinica, being a dioecious and a long-lived species, is primarily an outcrossing species. Species exhibiting these life forms tend to display a higher mean of genetic diversity (Hamrick and Godt 1996). In general, most of the populations displayed heterozygote excess or deficit although only populations MR and RW deviated significantly (P < 0.05) under heterozygote deficit. Homozygosity excess could be a result of biased sampling or existence of null alleles which were notable in four of the 11 loci used.

High levels of within-population diversity (68.01%), compared to the variation expressed among the populations (31.99%), may be attributed to the life history of this species. It is a dioecious and predominantly outcrossing, long-lived, anemophillic, and anemochory tree species, all of which are characteristics that favor high genetic variation within populations (Aguinagalde et al. 2005). AMOVA results reflected a strong variation structure and were supported by a high mean value of genetic differentiation (F ST = 0.32) recorded among all collection sites. The genetic relationship among the 168 individuals was shown by the scatter PCoA. The first three axes accumulatively showed a total of 69.89%, with axis 1 accounting for 30.41% while the second axis accounted for 22.69%, supporting the existence of a strong population structure.

Genetic isolation by distance is a dynamic process taking place in space and time, altering the species’ genetic makeup (Epperson and Li 1996). The Mantel test revealed the existence of positive correlations between genetic and geographic distances among the sampled populations (r = 0.425), indicating that isolation by distance (IBD) has an effect on the genetic differentiation of H. abyssinica. Generally, genetic variation is influenced by natural processes such as natural selection, drift, hybridization, and gene flow. The larger the geographic distance between two populations, the higher the genetic differentiation due to restricted gene exchange. This is further accelerated by the presence of obstacles to gene dispersal, e.g., mountains, rivers, and lakes. IBD is widely accepted as one of the causes of population divergence.

STRUCTURE analysis revealed the existence of a significant genetic structure among the populations by clustering the 11 populations into three distinct groups. The mean admixture α was less than 1, indicating that an individual belonged to a specific population as defined by Falush et al. (2003). However, a few individuals displayed mixed ancestry. This mode of clustering, where populations from distant mountain massifs group together, is common in Afro-alpine species in this region (Wondimu et al. 2014; Gizaw et al. 2013). The life history of H. abyssinica and the existence of geographical barriers, e.g., long distances and high mountains, might explain the strong genetic differentiation and structure displayed by this species.

We further conducted analyses to determine the amount of gene exchange across the genetic clusters defined by STRUCTURE. Although each of the genetic group had a few admixed individuals, the values of gene flow (N m ) estimated by MIGRATE-n were not significantly different from 0 and the highest flow of migrants (0.9) was recorded from cluster III to cluster I followed by 0.6 from cluster III to cluster II. Cross-mountain dispersals in eastern African mountains are mainly facilitated by air (wind) and birds through seed ingestion or which may also be embedded on their feathers and in the mud on their feet (Hedberg 1970). Long distance dispersal is feasible over many kilometers and most plant species have seeds that are highly specialized for wind dispersal. H. abyssinica is a wind-dispersed (seeds and pollen) dioecious tree species and although there is reduced gene flow recorded, a fundamental pattern was observed in that dispersal appears to have occurred in the north-south direction and rarely in the opposite direction. Gene exchange has pronounced implications on genetic differentiation among populations. Greater levels of gene flow (N m  > 1) reduce genetic diversity among populations and prevent genetic drift (Slatkin 1985). However, and as revealed by the relatively low gene flow level in this study, dispersal of both pollen and seeds by wind is ineffective among populations separated by large areas of land (Loveless and Hamrick 1984). Therefore, the displayed genetic structure of this species might be due to the existence of remnant populations that took refuge in these mountains during the Pleistocene era.

Using the jackknifing procedure, we realized that predictions of the distribution of H. abyssinica are strongly influenced by both temperature and precipitation variables. ENM revealed slight but notable changes in distribution of this species during the LGM and the present. For instance, during the LGM, the model predicted strong (0.8–1.0) occurrence of the species in the southwestern region although this appear to have presently shifted towards the southeastern zone. In addition, the species has experienced expansions towards the southern part of our study site where areas that were unsuitable during the LGM displayed a 0.2–0.4 probability of occurrence. The LGM was characterized by low temperatures and reduced precipitation (Wu et al. 2007). Therefore, it is possible that the increased dryness barred this species from spreading all over the lowlands and instead it inhabited only some areas (may be along riverbanks, swamps, and/or lakes); this could explain the sparse patches.

The ENM distribution results appear to be consistent with recorded palynological data which show that H. abyssinica existed in the southwestern parts (Burundi) of our study area from as early as 34,000 years before present (BP) (Bonnefille et al. 1995); in the southeastern parts (Mt. Kenya), it existed 33,350 years BP (Olago et al. 1999) while in the north (Ethiopian highlands), records suggest a recent occurrence (Lamb et al. 2004). These pollen records were, in most places, found at altitudes lower than where this species currently inhabits. Consequently, the patterns of population clustering observed in East Africa (MKI-MKII-AB-CH-EL) could be attributed to gradual migrations along the bottom of the GRV, probably during the ice ages when most high-altitude species survived at the low altitude (McCarty 2001; Gottelli et al. 2004), and before the anthropogenic effects when the Afromontane forests were more widespread (Hedberg 1969). Dispersal bridges could have been more effective for gradual dispersal of species between populations on close mountain massifs such as Mt. Kenya and the Aberdares. However, gene exchange between populations at long distances apart, such as Kilimanjaro-Meru (south of our study region) and Rwenzori (southwest), could have been a result of random dispersal events, e.g., by cyclones (Hedberg 1969).

Our findings strongly suggest that past climate oscillations during the interglacial and the LGM did not break down genetic structuring of H. abyssinica in Ethiopian highlands and eastern Africa. There appear to be three major distribution zones with a high probability of occurrence (0.8–1.0) of Hagenia during the LGM (Fig. 5a), that correspond to areas with major mountains in eastern Africa (Young and Hastenrath 1987). ENM results show that climate variations since the LGM have had little impact on the present distribution of this species on these mountains (Fig. 5); therefore, we hypothesize that these “sky islands” served as glacial refugial area for Hagenia. To further support this hypothesis, the results from our genetic differentiation analyses appeared to group H. abyssinica’s populations into three genetic clusters based on geographical regions which lie within the distribution zones that existed since the LGM. Furthermore, the results revealed high genetic connectivity among populations from the same distribution zone and strong genetic (high F ST) differentiation among distanced populations, as revealed by the matrix of pairwise F ST, which is a product of long-term isolation and restricted gene flow. The presence of private alleles in some of the sampled sites and the Mantel test suggested that differentiation and population structuring observed in H. abyssinica are results of long-term isolation facilitated by long distances and a complex topography.

Conclusion

Climate fluctuations, geographical complexity, and physical distances are the main drivers of genetic diversity in tree species (Gugger et al. 2013; e Souza et al. 2016). By integrating genetic diversity analysis with ecological niche modeling, we observed that the population structure of Hagenia results from a combination of geographical factors, isolation by distance, and characteristic life-forms of this species. Reduced/absence of gene flow limits the chances of gene exchange among populations and may also hinder the colonization of new suitable habitats in the event of further climate change. Currently, although H. abyssinica is yet to be assessed for the IUCN Red List (www.iucnredlist.org 29/09/2016), it is endangered in some regions and has been mentioned under the Red List of Endemic Trees (Vivero et al. 2005). Anthropogenic impingements, e.g., deforestation to create land for cultivation and illegal charcoal burning, are on the rise in eastern Africa and, therefore, it is expected that further loss of rare alleles will lead to reduction in genetic diversity. Consequently, appropriate conservation measures are needed to effectively conserve this species and preserve its natural habitats.

The precision and exactness of genetic diversity (heterozygosity) are mainly affected by polymorphism of individual loci (Hale et al. 2012). In addition, F ST values were observed to be the approximate determinants of genetic distances; when F ST values are >0.05, sampling 20 or less individuals is enough to estimate population genetic distances (Kalinowski 2005). Therefore, although the sample size used in this study was less in some populations and may not have adequately outlined the allele diversity of the populations, it was efficient to capture the population structure and differentiation in H. abyssinica. We also focused on a smaller region compared to the entire distribution zone of this species and thus, in order to outline a conclusive genetic structure of this endangered endemic tree, we propose further future studies with more samples [25–30 individuals per population are ideal to efficiently reveal the genetic diversity of a species (Hale et al. 2012)] from all regions where H. abyssinica is known to grow.