Aim: Forest relicts in the mountainous regions of Africa represent one of the most diverse ecosystems on our planet, but the processes that have generated this remarkable diversity are still poorly understood. We estimate divergence times for an endemic, flightless grasshopper family and reconstruct a potential scenario for their colonization of Africa to test the hypothesis that the diversity of these mountain-top endemics has been generated by multiple fragmentations and reconnections of tropical rain forests in parallel with climatic fluctuations.

Location: Sub-Saharan Africa.

Taxon: Lentulidae (Orthoptera)

Methods: We inferred the phylogeny of 7 genera and 28 species of the Lentulidae based on COI, 16S and Histone 3 sequences by using a Bayesian approach, and we also estimated divergence dates. Based on our molecular phylogeny as well as the available information on the relationship of 6 additional genera and local occurrence records for 120 lentulid species across Africa, we reconstruct a potential colonization scenario for most species of this group.

Results: Our findings indicate that the forest-bound lentulids of East Africa represent a monophyletic group that originates from South Africa. We show that major splits in the phylogeny of the Lentulidae coincide with three known fragmentation events of the African rain forests (27, 16 and 9 million years ago) and that lentulids subsequently diversified rapidly in parallel with the aridification and strong geological activity in East Africa.

Main conclusions: Our results corroborate the diversification patterns reported for several endemic African forest-bound animal taxa at small scales and endemic African plant taxa at larger scales, highlighting the finding that much of the biodiversity presently found in the forest relicts of the Eastern Arc Mountains biodiversity hotspot has been generated by the interplay between humid periods that allowed the spread of forest-bound lineages across Africa and periods of aridity-driven isolation of forests and their associated fauna.

Molecular phylogenetic analysis

Samples

Specimens of lentulid species were collected in Uganda, Kenya, and Tanzania, as well as in KwaZulu-Natal, (South Africa) between 2007 and 2016. Specimens were identified with the keys published by Jago (1981) and were compared with material from the entomological collections of the National Museums of Kenya in Nairobi, the Natural History Museums in London, and Berlin. Specimens were preserved in 70 % alcohol or dried in the field. For each specimen, the locality, collection dates and altitude were noted. For long-term preservation, insect material was transferred to 96 % ethanol and stored at 4 °C.

DNA extraction, amplification, purification and sequencing

DNA was extracted from the muscle tissue of the hind femur using the QIAamp^® DNeasy (QIAGEN, Germany) and the standard protocol for blood and tissue. The extracted genomic DNA was used as a template for PCR amplification with insect primers modified for Orthoptera. Primers were (forward first and reverse second): 16a: 5′-CGC CTG TTT ATC AAA AAC AT-3′ and 16b: 5′-CCG GTC TGA ACT CAG ATC ACG T-3′ for the 16S rDNA (Kocher et al., 1989); H3F: 5′-ATG GCT CGT ACC AAG CAG ACG GC-3′ and H3R: 5′-ATA TCC TTG GGC ATG ATG GTG AC-3′ for the Histone H3 gene (Colgan et al., 1998); and LCO1490: 5′-GGT CAA CAA ATC ATA AAG ATA TTG G-3′) and HCO2198 (5′-TAA ACT TCA GGG TGA CCA AAA AAT CA-3′) for the COI gene (Vrijenhoek, 1994). Amplification was performed under the following conditions: initial denaturation was 5 minutes at 94 °C; 38 cycles of 45 seconds at 94 °C, 45 seconds at 52 °C, and 80 seconds at 72 °C; and a final extension step at 72 °C for 7 minutes. The amplification product (2 µl) was loaded onto 1.5 % agarose gels for size-fractionation in gel electrophoresis to check the amplification results (Green, Sambrook, & Sambrook, 2012). Purification of the amplification products was performed using the standard protocol of the QIAquick PCR purification kit (QIAGEN, Germany). Amplified DNA templates were sequenced at LGC Genomics GmbH (Berlin, Germany). For the final editing of sequences, we used the software CODONCODE ALIGNER. For the phylogenetic analyses, we selected rapidly evolving mitochondrial genes [16S rRNA gene and the barcoding gene cytochrome oxidase subunit I (COI)] and a slowly evolving nuclear gene (Histone 3) to facilitate the recovery of both deep and recent nodes of the phylogeny.

We used PYLOGENERATOR (Pearse & Purvis, 2013; accessed 1 November 2016), as well as a recent checklist of all Lentulidae and Acridoidea (catalogue of life; accessed 25 October 2016), to automatically search GenBank repositories and download additional sequences of 16S, COI and Histone 3. Sequences for ten genera and 15 species of Acridoidea were used as the outgroup. Taxon sampling and GenBank accession numbers are given in the Supporting Information (see Appendix 1 Table S1).

Alignment preparation and phylogenetic analysis

The molecular data were aligned with the MUSCLE algorithm and by hand in MEGA (Tamura, Stecher, Peterson, Filipski, & Kumar, 2013). For each alignment, we ran separate model tests in MEGA. Two different models of nucleotide substitution were selected according to the highest scores of the Bayesian Information Criterion (BIC; Appendix 1 Table S2). For 16S and COI, the most complex model, GTR + Γ + I was selected; for Histone 3 the HKY + Γ model was used. To evaluate the robustness of our phylogenetic reconstruction, we compared the results between two different approaches. First, we used the best-fit substitution model to generate phylogenetic trees separately for the three regions 16S, COI and Histone 3, using a maximum likelihood, as implemented in MEGA (Tamura et al., 2013; see trees in Appendix 1 Figs S1-3). Bootstrap replications were set to 1,000 trees, whereas the default settings were used for all other parameters. After adjusting the taxon similarity among alignments by replacing missing sequences with blanks, we used the interface BEAUTI (Drummond, Suchard, Xie, & Rambaut, 2012) to link the three regions 16S, COI and Histone 3 and to create the xml-file for the BEAST analysis. The three alignments added up to 1,627 bp (16S: 557 bp; COI: 688 bp; Histone 3: 382 bp). Both separate clock tests (in MEGA) for each region, as well as the standard deviation of the coefficient of variation of the multi-gene phylogeny (checked in TRACER), indicated that the assumption of rate heterogeneity can be rejected. Therefore, branch lengths were inferred with an estimated strict molecular clock for 250 million MCMC simulations (Yule birth process) with a Bayesian approach implemented in BEAST (Drummond et al., 2012). To assess the posterior distributions of the parameter estimates and to determine the burn-in — based on the point of stationarity on the log-likelihood curves as well as split-frequencies — tree statistics were sampled every 10,000^th iteration and subsequently checked with the diagnostics of TRACER. To exclude an appropriate burn-in of 10 % from the posterior sample, to summarize the statistics of the remaining trees and to annotate this information to the tree with the highest clade credibility, we used functions of ANNOTATOR (Drummond et al., 2012). The molecular clock was calibrated using three events of volcanism that represent the origin of Mount Hanang, Kenya and Kilimanjaro (Nonnotte et al., 2008) and hence a starting point for their colonization by ancestors of Usambilla chlorophrygana Jago and U. hanangensis Hemp (normal distribution with mean ±SD = 2.25 myr ±0.12 myr), ancestors of Altiusambilla keniensis Hemp and A. modicicrus Karsch (normal distribution with mean ±SD = 2.50 myr ±0.20 myr) and populations of A. modicicrus on Mount Kilimanjaro and Mount Meru (normal distribution with mean ±SD = 1.75 myr ±0.12 myr) that led to the differentiation of these taxa. Phylogenetic trees were visualized and formatted in FIGTREE (Drummond et al., 2012).