Discussion

Through combining ecological niche modelling, ABC inference of demographic history and landscape genetics we show how historic climatic changes during the Quaternary have shaped the population structure and patterns of genetic variation of the biodiversity in Afromontane and Afroalpine sky islands. We further show the impact of recent anthropogenic land-use change and habitat degradation on population declines and loss of genetic diversity in species that are also threatened by disappearing suitable niche under future climate change in one of the most endemism rich and endangered ecosystem in the world (Williams et al., 2004).

Effects of climate change past and future

Ecological niche models support the restricted distribution of P. balensis in Afroalpine moorland and Afromontane grassland and woodland high elevation regions of the Ethiopian highlands, and show that its distribution is primarily limited by temperatures. Models projected across temporal scales show that climatic suitability for P. balensis has been progressively decreasing since the LGM and will continue to decrease with increasing temperatures over the next few decades, with a direct effect on population fragmentation and isolation. The genus Plecotus is of Palearctic origin and is found primarily across temperate and Mediterranean habitats (Spitzenberger et al., 2006). As such, it is not surprising that climatic conditions in the tropics were more suitable for this bat when temperatures were lower during the LGM. Increased climatic suitability during LGM and displacement of the Afroalpine and ericaceous zone to lower elevations (Bonnefille, Roeland, & Guiot, 1990) allowed high altitude species to extend their range (Gottelli, Marino, Sillero-Zubiri, & Funk, 2004) and likely led to contact between isolated sky island populations (McCormack et al., 2009). This can explain the observed limited differentiation between mountain ranges not separated by the substantially lower Rift Valley.
Sky islands are sensitive to future climate change because the high montane and alpine habitats could contract into higher elevations with even minor temperature increases, reducing the habitat available for their associated endemic taxa (McCormack et al., 2009). Tropical montane forests, in particular, are some of the most threatened habitats under climate warming (Moritz & Agudo, 2013). Tropical montane regions are predicted to experience highest levels of disappearing climates, and consequently species extinctions and community disruptions (Williams, Jackson, & Kutzbach, 2007). In Ethiopia, Nyssen et al. (2014) already identified evidence of upper shifts of the high elevation treeline in response to climatic changes in the past century. This ecotone is located between the Afromontane and Afroalpine zones and is particularly rich in biodiversity (Kidane et al., 2012). Indeed our models predict that only a quarter of the current range of P. balensis will remain climatically suitable by the end of the century. These predictions are relevant to other high altitude Afromontane biodiversity, and therefore are particularly worrying given the high levels of endemism found in this ecoregion, in particular among vertebrates (Williams et al., 2004) and vascular plants (Gizaw et al., 2016).

Effects of geographical barriers – Rift Valley and sky islands

The Ethiopian Rift Valley was identified as the main barrier to contemporary gene flow shaping the population structure of P. balensis . The Great Rift Valley, extending from Lebanon through the Red Sea to the Zambezi River, is less than 100 km wide across the Ethiopian central highlands (Ebinger et al., 2000), dividing them into two massifs from northwest to southeast. It played a major role in structuring biodiversity across Eastern Africa, from the Afroalpine plantsArabis alpine (Asefa, Ehrich, Taberlet, Nemomissa, & Brochmann, 2007) and Lobelia giberroa (Kebede, Ehrich, Taberlet, Namomissa, & Brochmann, 2007) to most of Ethiopian anurans (Freilich et al., 2016), including the high altitude African clawed frogs, Xenopussp. (Evans, Bliss, Mendel, & Tinsley, 2011) and the forest tree frogs (Reyes-Velasco et al., 2018). Interestingly, amphibians found at elevations below 2500 masl did not show differences between the northern and southern highland massifs (Freilich et al., 2016). As P. balensis is associated with low temperatures and montane ecosystems above 2000 m (Benda et al., 2004), the much warmer and drier savannah-like habitats at low elevation along the Rift Valley are very likely to be inhospitable for this bat. In fact, according to our data, the main split along the Rift Valley, can be traced to the last interglacial period when conditions were even warmer and drier than in the Holocene (Adams, Maslin, & Thomas, 1999). Evidence of limited allele sharing between both sides of the Rift Valley is most probably related to the short time that elapsed since the last contact between Afromontane habitats across the Rift Valley, rather than present day gene flow.
The complex geological history of the Ethiopian highlands is partially responsible for its high biodiversity and endemism. The highlands do not constitute a single unit, but instead are formed by various massifs of different areas, origins, ages and degrees of isolation (Reyes-Velasco et al., 2018). Genomic differentiation in highland species, like the river frog, Amietia nutti , is explained not only by the Rift Valley but also by other major geographical barriers, the Blue Nile Valley and the Omo River (Manthey, Reyes-Velasco, Freilich, & Boissinot, 2017). This geographical and historical complexity is expected to affect the phylogeography of P. balensis . The isolation of sky islands is supported by the deep genetic structure found in the mtDNA markers and the fact no haplotypes were shared between mountain ranges and all mountain ranges formed separate clades in the mtDNA tree. All clades, except the Bale population that included deeply divergent basal lineages, could be the result of isolation processes relating to older glacial cycles during the Pleistocene. These episodes of allopatric isolation appear as the main driver of differentiation in Ethiopian highland vertebrates (Reyes-Velasco et al., 2018). The phylogeographic structure of P. balensis is consistent with that of the Ethiopian wolf (Gottelli et al., 2004), whereby three main clusters are present corresponding to the three well-defined mountain areas: the southern Chilalo/Bale, the central Guassa/Abune Yosef and the northern isolated Siemen Massifs. This strong structure at the mtDNA level indicates that sky islands have acted as historic barriers to gene flow in P. balensis . Sky islands are strong isolating mechanism in limited dispersal species. For example, the endemic high altitude spider Microhexura montivaga in the southern Appalachians is divided into separate clades corresponding to mountain ranges, though this species shows strong population subdivision also at the nuclear level (Hedin, Carlson, & Coyle, 2015). In contrast, in P. balensis the nuclear microsatellite dataset only identified a split into two clusters, corresponding to the populations on either side of the Rift Valley.
More limited population structure based on bi-parentally inherited markers suggests that gene flow between mountain ranges is primarily male-mediated, a common pattern in temperate bat species (Moussy et al., 2013; Razgour, Salicini, Ibáñez, Randi, & Juste, 2015). However, as extensive contemporary gene flow between mountain ranges is unlikely given the extent of habitat conversion across the Ethiopian highlands (Lemenih & Kassa, 2014), the shallow genetic structure at the nuclear level, may instead indicate habitat connectivity during the Holocene, possibly through montane forest bridges during episodes of increased moisture (Umer et al., 2007). However, other Ethiopian highland species do show strong structure and limited gene flow in bi-parentally inherited markers, including the mountain nyala (Atickem et al., 2013) and Ethiopian wolf, in which current gene flow is limited to geographically close populations (Gottelli, Sillero-Zubiri, Marino, Funk, & Wang, 2013). Alternatively, genetic differentiation may be driven by local adaptations and environmental dissimilarity and therefore its signature cannot be identified in neutral markers (Manthey & Moyle, 2015).

Effects of anthropogenic land-use change

Although ENMs identified that P. balensis is primarily associated with Afroalpine moorland and grassland and Afromontane woodland ecoregions, reproductive females and sub-adults were only found in high altitude Afromontane woodlands, highlighting the importance of this habitat for the species’ reproductive success. A similar pattern of elevational segregation was observed in other Palearctic bats such as the European noctules (Ibáñez et al., 2009) and the congeneric alpine long-eared bat Plecotus macrobullaris (Alberdi, Aizpurua, Aihartza, & Garin, 2014), whereby maternity colonies in colder regions of the distribution are found below the treeline. Alberdi et al. (2014) attribute this segregation to restrictions on facultative heterothermy and use of torpor in pregnant and lactating females.
The Afromontane woodland habitat is highly threatened and disappearing due to extensive deforestation and overgrazing (Yalden et al., 1996). Deforestation rates in the Ethiopian Highlands are severe following centuries of landscape changes due to subsistence farming, settlement and demands for fuelwood (Lemenih & Kassa, 2014). Native forest cover has declined in the past 100 years from 45% to only 5% of the country, and currently most of the native forest cover in the highlands is concentrated in small patches surrounding orthodox churches, known as church forests (Abbott, 2019). The mountain forest belt has been pushed upwards and fragmented due to extensive agriculture (Kebede et al., 2007). Some of the forests where bats were caught south of the Rift Valley were degraded, fragmented and showed limited evidence of recruitment due to overgrazing of ground vegetation cover. This represents a broader trend of forest degradation in the Bale Mountains region, including inside the National Park, due to logging, fuelwood collection and livestock grazing (Asefa, Mengesha, Shimelis, & Mamo, 2015), which has been linked with the recent declines of most large wild mammals populations within the park (Stephens et al., 2001).
The recent decline of the south-eastern P. balensis population identified in our ABC model-based inference appears to coincide with a period of accelerated forest loss and degradation and the expansion of agriculture, though causal mechanisms behind the decline were not tested here. No evidence of decline in the north-western population could be attributed to the general higher density of montane forests in the wetter north, as well as the buffering effect of the thousands of church forests containing native trees that are spread across the north-western highlands (Klepeis et al., 2016). However, small sample sizes may have limited the inference power of the ABC analysis when it comes to identifying more moderate bottlenecks. Genetic diversity in the Ethiopian wolf was also higher in the northern highlands (Gottelli et al., 2004), indicating that habitat degradation in the south is having a detrimental effect on Afroalpine and Afromontane mammals in general. Although native forest loss has been more limited in the northern highlands over the past 150 years, degradation is on-going in the more recently settled Simien mountains (Nyssen et al., 2014), suggesting that the northern highlands Afromontane biodiversity is also at risk.
Despite the strong effect of anthropogenic land-use change on genetic diversity and population decline in P. balensis , indicators of anthropogenic impact, like artificial lights at night and human footprint index, did not affect gene flow. Low sample size (five populations) may have reduced the power of the landscape genetics statistical analysis, in particular when considering the combined impact of more than one variable, and therefore this limitation may have obscured the importance of some drivers of landscape resistance. Although native church forest patches and small community-managed forest reserves offer some level of landscape connectivity across the agricultural and human settlement landscape, percent tree cover did not affect genetic connectivity. However, this may be attributed to the tree cover map used, which does not distinguish between native forests and eucalyptus plantations (Hansen et al., 2013), and therefore could not capture the effect of the landscape on the movement of P. balensis . Instead, the landscape genetics analysis highlights the importance of altitude and Afroalpine and Afromontane habitats. The best-supported model gave lowest resistance costs to the high altitude alpine moorland ecoregion, rather than the montane forest ecoregion, perhaps due to the more restricted distribution of native montane forest than the ecoregion classification suggests because of extensive deforestation in the past century. Alternatively, this may reflect temperature or other climatic limitations on the movement of P. balensis , as is the case with other cold-adapted high altitude mammals, like the American pika, Ochotona princeps (Castillo, Epps, Davis, & Cushman, 2014). Altitude was also identified as the main landscape element limiting gene flow in high altitude salamanders, alongside pond network structure. However, genetic connectivity in these more limited dispersal species was primarily affected by geographic distance presumably due to the rarity of dispersal events in the extreme high altitude environment (Savage, Fremier, & Bradley, 2010).

Conclusions

Using a combination of ecological niche models, landscape genetics and ABC model-based inference of demographic history we show how historic climate change and geographic barriers interact with recent anthropogenic habitat loss and degradation to shape the population size, structure, diversity and connectivity of tropical montane biodiversity. Focusing on the endemic Ethiopian Highlands bat, P. balensis , we found that despite strong associations with high altitude environments and mtDNA pattern associated with sky island structure, some level of genetic connectivity is maintained among sky islands, and only substantially lower altitudes, like the Rift Valley, form a true barrier to gene flow. Of particular concern are evidence of recent population decline, likely in response to deforestation and land conversion, the decline in genetic diversity with increasing arable land cover, and the importance of Afroalpine and Afromontane ecoregions for both range suitability and genetic connectivity between sky islands. Given that similar patterns of genetic population structure have been recorded in other high altitude mammals and amphibians, such losses of genetic diversity and population declines may be widespread in tropical montane ecosystems. The situation described for this bat epitomises a wider trend in high altitude Afromontane and Afroalpine biodiversity, which is under threat from extensive deforestation, agricultural conversion and overgrazing following rapid human population expansion (Yalden et al., 1996). These threats are not likely to be reversed in the future in countries like Ethiopia where accelerated human population growth is projected to continue until the end of the century (UN, 2019). These immediate threats will interact with future climate change and the projected disappearance of suitable climates and their associated ecoregions in tropical montane regions (Williams et al., 2007), which will increase the likelihood of extinction of high altitude species. This study presents alarming predictions for the fate of tropical montane and alpine biodiversity, projected to be squeezed to higher altitudes (or even disappear) due to climate change while already losing genetic diversity and suffering population declines due to anthropogenic land-use change. The combination of these different threats can push these ecosystems beyond their resilience limits. We conclude that assessments of threats to biodiversity under global change should adopt a holistic approach, simultaneously studying the effects of multiple threats and considering the impacts of past events, present stressors and future projections based on genetic, ecological and geographic information.