Genetic analysis

We took samples with a sterile biopsy punch (Stiefel Laboratories; diameter: 2 mm) of 317 bats from 10 locations (six in Brunei Darussalam, four in Sarawak). Samples were stored in 90% ethanol or dried with silica gel until DNA extraction (Silica Gel Orange, Carl Roth GmbH)⁴⁵. DNA was extracted from wing biopsy punches using a modified ammonium acetate extraction protocol (Strauss 2001), eluted in Low TE and stored at −20 °C. We used DNA samples at final concentrations of at least 2 ng μl⁻¹ (quantified from extracted samples on a NanoDrop ND-1000 Spectrophotometer, Thermo Fisher Scientific).

We sent genomic DNA to the Max-Planck-Institute for Evolutionary Biology in Plön that created a microsatellite library using high-throughput shotgun 454-sequencing. Using the programme MISA (Microsatellite Identification Tool; http://pgrc.ipk-gatersleben.de/misa/misa.html) we found 66,289 potential microsatellite sequences from which we developed 40 unlabelled primer pairs using the programmes Nucleic Acid Sequence Massager (http://www.attotron.com/cybertory/analysis/seqMassager.htm) for cleaning the sequences and Primer 3 v. 4.0.0 (http://sourceforge.net/projects/primer3/)^46,47 to design the primers. Using pooled DNA from two individuals we tested these primer pairs for amplification and polymorphism at four different annealing temperatures (56–62 °C; ABI 3130xl Genetic Analyser, Applied Biosystems).

Based on amplification success and polymorphism, a set of 16 loci was selected (Supplementary Table ^S3) and optimised in two multiplex reactions (MP1/MP2) which were conducted for each individual in 8 μl (MP1) and 5 μl (MP2) reaction volumes, each consisting of 1.0 μl DNA, 1 × Multiplex PCR Master Mix (Qiagen) and primer concentrations as indicated in Table ^S3. We used the following amplification conditions: 95 °C for 15 min; 32 cycles of 94 °C for 30 s, 60 °C for 90 s, 72 °C for 60 s; 60 °C for 30 min. All PCR products were run on an ABI 3130xl Genetic Analyser (Applied Biosystems) and sized with an internal lane standard (GeneScan™ 500 LIZ™ dye Size Standard, Thermo Fisher) and the software GeneMapper v. 5 (Applied Biosystems).

To check for genotyping consistency, 23.0% of samples were amplified and genotyped twice. We could not detect departures from Hardy-Weinberg and linkage equilibrium at the site level after Bonferroni correction using Genepop v. 4.1.4 (except for individuals of study site “Labi 31” where we had 33 significant linkages between markers probably due to the presence of close relatives). We also found no evidence for the presence of null alleles, large allelic drop-out or possible scoring errors across populations within our dataset (tested with Micro-checker v. 2.2.3)⁴⁸.

To investigate if there is a correlation between the populations’ pairwise genetic distance and pairwise geographic distance matrices, we conducted a Mantel test (99,999 permutations) with the R package ecodist⁴⁹. We calculated F_ST with GenoDive v.2. Ob27 to measure pairwise population differentiation. With STRUCTURE v. 2.3.4^50,51 we investigated the population structure using a burn-in length of 20,000 and a run length of 200,000 without prior population information. The admixture model and the correlated allele frequencies between population options were selected. After an initial test we chose the burn-in and run length by looking at the convergence of the values of summary statistics and consistency between runs. All other parameters were left as by default. We undertook ten independent runs for K-values ranging from one to ten, which reflects the minimum and maximum number of populations suspected. The number of populations was inferred from the corrected posterior probability and four new estimators that have been shown to outperform other estimators, namely MedMeaK, MaxMeaK, MedMedK and MaxMedK⁵². Additionally, we conducted a Principal Component Analysis (based on individuals) using the adegenet v. 1.3-9⁵³ and ade4 v. 1.4–14⁵⁴ packages in R (Fig. 3).

For all bats of the study site “Airport” (which is the only study site where bats were roosting in pitcher plants and in furled leaves) we calculated pairwise relatedness (triadic likelihood relatedness estimate (TrioML) with Coancestry v. 1.0.1.5⁵⁵. With Monte Carlo we tested the null hypothesis that the pairwise relatedness of bats did not differ in relation to the preferred roost type (pitchers, furled leaves). Therefore, we randomly selected (1,000 times) seven individuals roosting in furled leaves and combined them with the seven individuals roosting in pitcher plants. We compared the mean pairwise relatedness of bat pairs roosting in pitchers, of bat pairs roosting in furled leaves, and of bat pairs with differing roost preference (one in pitchers, one in furled leaves) to the distribution of values expected under the null hypothesis. We obtained the null hypothesis distribution by randomly assigning roost preferences and then calculating mean difference for pairs roosting in furled leaves, in pitchers, or in both. This procedure was repeated 10,000 times. To calculate the P-values, we compared the observed mean values of relatedness to the null distributions. To visualize the observed pairwise relatedness (TrioML) between the individuals at the study site “Airport”, we constructed an unweighted and undirected network of the bats using the R package igraph v. 0.7.1⁵⁶. To focus on very closely related pairs of bats (parent-offspring or full-sibling pairs), we kept only links with TrioML relatedness >0.44 (Fig. 4).