Termite diversity in Neotropical dry forests of Colombia and the potential role of rainfall in structuring termite diversity

Termites are ecosystem engineers that play an important role in the biotransformation and re‐distribution of nutrients in soil. The dry forests are endemic repositories, but at same time, they are most threatened by extensive livestock and crop farming, fires, and climate change. In Colombia, the best‐protected dry forests are located in the north. The termite fauna of dry forests are poorly known. The aim was to identify the termite species occurring in tropical dry forests of the Colombian Caribbean coast in relation to diet and precipitation, temperature, elevation, and soil properties. A total of 32 species in 1,103 occurrences were found. Termitidae accounted for 78% of the species richness with the Anoplotermes‐group, Microcerotermes, and Nasutitermes being the dominant genera. Differences in species composition and abundance were found across sites. These differences may be linked to anthropogenic disturbance and polygyny and polydomy. Strikingly, our highest elevation site (334 m) had the highest species richness much higher than the two lower elevation sites. This implies an inversion of the common elevation‐diversity gradient, also found for termites which can be explained by increasing precipitation with elevation in the dry forest. An analysis of termite species richness at the global scale confirms that termite species richness correlates positively with rainfall. Hence, rainfall seems to positively affect termite diversity. In line, the studied Colombian tropical dry forests had low diversity compared to rain forests. A decline of species‐rich soil‐feeding termites with increasing aridity may explain why the highest termite diversity occurs in humid tropical rain forests.

Environmental factors can influence termite communities. For instance, precipitation, elevation, temperature, and soil properties are known to affect termite diversity (Bourguignon, Drouet, Šobotník, Hanus, & Roisin, 2015;Gathorne-Hardy, Syaukani, & Eggleton, 2001;Holt & Lepage, 2000). Studying the association of termite occurrences with different environmental factor is a first, essential step for a better understanding of termite ecology in tropical dry forests.
The aims of this study were (a) to identify the termite species occurring in tropical dry forests of the Colombian Caribbean coast; (b) to describe associations of termite communities with environmental factors; (c) to test whether there is a correlation between species richness and rainfall at the regional and the global scales.

| Study sites and termite sampling
The study sites were located in the Coraza Forestry Reserve "Colosó" Colosó is a regional forestry reserve created in 1983, with important primary and secondary tropical dry forest, bordered by small plots of agriculture and livestock farming. Its mean annual temperature is 26. 7°C (min: 25.8; max: 27.8) with an annual precipitation of around 1,337 mm (Hijmans, Cameron, Parra, Jones, & Jarvis, 2005;Instituto Nacional de los Recursos Naturales Renovables y del Ambiente, 1983). Ceibal is a private remnant of primary and secondary tropical dry forest, surrounded by large livestock farms with small rural settlements. In 2013, it was declared a protected area. It has a mean annual temperature of 27.6°C (min: 26.9; max: 28.1) and an annual precipitation of around ~1,080 mm (Corporación Autónoma Regional del Canal del Dique, 2013;Hijmans et al., 2005). Tayrona is a Natural Park with primary and secondary forest created in 1963 without human settlements or agriculture and the best-protected and conserved dry forest in Colombia. Its mean annual temperature is 27.8°C (min: 25.8; max: 29.0) with an annual precipitation of about 713 mm (Carbonó, 2010;Hijmans et al., 2005; Instituto Colombiano para la Reforma Agraria, 1964; Instituto de Investigación Alexander von Humboldt, 2014). All study sites were located belong to the Caribbean coast.
Termites were sampled in Colosó in July 2014, in Ceibal in August 2014, and in Tayrona in August 2015, after the first seasonal rains. The standardized belt transect sampling protocols of Jones and Eggleton (2000) and Hausberger and Korb (2015) were used.
Five 100 m × 2 m transects belts were arbitrarily located at each site. Each transect was divided into 20 contiguous sections (each 2 m × 5 m). Two trained people searched each section for a total of 30 min systematically for termites, including all microhabitats such as leaf litter, dead wood, trunks, foraging galleries nest, and twigs. The transect sampling was supplemented by eight soil scrapes (15 cm × 15 cm, 10 cm depth) per transect section to collect soildwelling termites. Whenever termites were encountered, several individuals were collected (if possible including soldiers) and stored as a sample. Collected termite samples were individually labeled and stored in 100% ethanol for DNA analysis and 80% ethanol for museum curation.
Additionally, soil samples (approx. 1 kg per sample) were taken from the top horizon (0-15 cm) at a distance of 1 m parallel to each belt transect with three replicates per transect belt (one each at the start, in the middle, and at the end of a belt transect), resulting in a total of 15 samples per site. Samples were cooled after the field trips, and subsequently, they were dried and sealed in plastic bags, and immediately sent for analysis to the Soil Laboratory of the Geoscience School of the Universidad Nacional in Colombia. Soil texture (sand, clay, and silt content), pH, cationic exchange capacity, organic matter, calcium, magnesium, potassium, sodium, and phosphorus were determined using in house protocols (http://www.unalmed.edu. co/~esgeocien/metodologia_quimica.html).
For molecular species identification, we isolated DNA and sequenced fragments of three mitochondrial genes, cytochrome oxidase subunit II (COII; total length 740 bp), 12S rDNA (~385 bp), and 16S rDNA (~480 bp), as described elsewhere (Doyle & Doyle, 1987;Hausberger et al., 2011). These sequences were used to reconstruct phylogenetic trees (see below) to delimitate and identify species as in former termite studies (Eaton, Jones, & Jenkins, 2016;Hausberger et al., 2011;Legendre et al., 2008;Roy et al., 2014). COII was most useful for "barcoding" (i.e., assigning species to samples) because it amplified well and gave appropriate resolution for species identification. To delimitate species, we constructed phylogenies comprising all species occurring in Colosó, Ceibal, and Tayrona. Samples forming a well-supported cluster were named identically. Species names were carefully checked with available literature and correspond to those reported for Neotropical termites (Bourguignon et al., 2017;Cameron, Lo, Bourguignon, Svenson, & Evans, 2012;Casalla et al., 2016a,b;Eaton et al., 2016;Inward et al., 2007;Roy et al., 2014). Voucher specimens of all species are held at the University of Freiburg, Germany, and will be deposited at the Natural History Museum of the Alexander von Humboldt Institute of Bogotá (MIAvH).

| Phylogenetic analysis
COII, 12S, and 16S rRNA sequences were checked manually and then aligned with MUSCLE as implemented in MEGA v7.0 (Kumar, Stecher, & Tamura, 2016). Sequences were submitted to NCBI and provisional numbers access were gathered (Supporting Information Table S1). We inferred a phylogenetic tree based on a Bayesian approach using MrBayes 3.2.1. (Ronquist & Huelsenbeck, 2003; 10 7 generations with every 1000th tree sampled, using the default of four chains). For COII, the bestfitting model was TIM3+G+I, for 12S rRNA GTR+G+I, and for 16S rRNA TIM2+I+G (Posada, 2008). After checking for convergence, we discarded 50% as burn-in. For phylogenetic tree inference based on the maximum-likelihood (ML) approach, we applied IQTREE version 1.4.3 (Nguyen, Schmidt, Von Haeseler, & Minh, 2015). We performed partitioned, non-parametric bootstrapping with 10,000 replicates. Finally, we plotted all bootstrap replicates on the ML tree with the best log-likelihood values. For all phylogenetic tree analyses, we choose the cockroach Blatta orientalis Linnaeus, 1758 as outgroup taxon. The resultant tree from Mr. Bayes and IQTREE were visualized using FigTree version 1.4.2 (Nguyen et al., 2015). Additionally, we also used MEGA 7.0 (Kumar et al., 2016) to calculate p-distances between putative species for the combined COII, 12S, and 16S nucleotide sequences (3,000 bootstrap replications, Gamma Distributed rates among sites and Transitions + Transversions substitution model). As complementary data to the 32 sequences in our study (Supporting Information Table S1), we downloaded 194 COII barcode sequences (longer than 600 bp in length) from NCBI, from closely and distantly related Kalotermitidae, Rhinotermitidae, and Termitidae. We built a phylogenetic tree based on a Bayesian approach using MrBayes (see above) and used the OTUs clusters formed to infer species.

| Diversity analyses
To compare alpha-and beta-diversity across sites, the Shannon (H) and the Simpson's (1-D) indices as well as rarefaction and extrapolation curves of richness were calculated using EstimateS v9.1.0 (Colwell, 2013) and iNext (Chao et al., 2014). For the most abundant and richest genera (Microcerotermes, Nasutitermes, Anoplotermesgroup, and Amitermes), Kruskal-Wallis tests, followed by multiple comparison post hoc Dunn-Šidák corrections, were used to test for differences between sites.

| Determination of feeding group
Feeding groups were identified based on Donovan, Eggleton, and Bignell (2001) and Eggleton and Tayasu (2001). Accordingly to Neotropics, four feeding groups are distinguished: dead wood and/ or grass feeders with flagellates in guts (group I); dead wood, grass, leaf litter feeders (group II); upper layer organic-rich soil feeders (group III), and soil feeders (group IV). For Apicotermitinae, enteric valve and molar plates were checked to identify feeding groups.

| Determination of other environmental variables
Climate data were downloaded from WorldClim v 1.4 (http://www. worldclim.org). We calculated the annual median, maximum and minimum temperature, and precipitation over the last 10 years (Hijmans et al., 2005). Kruskal-Wallis tests, followed by multiple comparison post hoc Dunn-Šidák corrections, were used to test for differences in both climate and soil variables between sites.

| Canonical correspondence analysis
A canonical correspondence analysis (CCA) (constrained, unimodal, and ordinary scale) was done to reveal potential associations between environmental variables (annual precipitation, and annual median temperature, elevation, soil properties; calcium, clay, cationic exchange capacity, magnesium, pH, and sand) and termite community composition. Canonical correspondence analysis performs quite well for skewed species distributions with highly intercorrelated quantitative environmental variables (Palmer, 1993

| Other statistical analyses
Richness and rainfall values across four continents were obtained through a compilation of research articles on termite diversity.
Intra-genera resolution was high in most cases, and two main clustered groups formed by Nasutitermitinae and Apicotermitinae were found (>56% BPP, >45 BV). Overall genetic p-distance was 13.2%; the lowest distance was observed between Nasutitermes dasyopsis Thorne, 1989 and Nasutitermes sp1 (p-distance 0.56%; Figure 2, Supporting Information Table S2). Yet the latter two species were clearly morphologically distinct (Supporting Information Figure S1).    Table S3). The mean species richness and occurrences across all sites were 10.1 (±SD 3.0) and 73.5 (±SD 31.8), respectively.
However, we found significant differences between the observed richness and the Chao 2 estimator (Mann-Whitney test Z = −3.62, p < 0.001).

| Habitat preferences and feeding groups
Most termites were encountered in twigs/litter on the soil surface (55.7%), followed by soil scrapes (37.2%) and trees (6.6%), while only TA B L E 1 Comparison of termite species richness and occurrence for selected genera between study sites. Values are the median values and IQR (interquartile range a few arboreal and mound nests were found (0.7%). Among all sites, feeding group II was the most dominant (66.9%; Figure 2, Supporting Information Figure S5), followed by feeding group I (15.3%), feeding group IV (14.5%), and feeding group III (2.8%). Feeding group II had significantly more occurrences than the other feeding groups (ANOVA: F 3,56 = 171.73, p < 0.001; Tukey test: p < 0.001, Supporting Information Figure S5), while both feeding groups I and IV differed from III (Tukey test: p = 0.001, Supporting Information Figure S5).
We analyzed the contribution of environmental factors on termite community composition. Only five variables were chosen: precipitation, temperature, elevation, magnesium, and pH (Supporting Information Tables S5A,B and S6). The CCA showed site-specific clusters, which were associated with certain termite species and environmental factors (Figure 5a,b) Tayrona formed a cluster along the y-axis, with the lowest richness and encounter rates of Anoplotermes-group and Nasutitermes. It was associated with the occurrence N. dasyopsis and drywood termites (Kalotermitidae), high temperatures, sandy soils, and organic matter (Figures 2 and 5a,b).

| D ISCUSS I ON
Our study revealed differences in termite community composition across the three study sites. In the following, we discuss how environmental conditions, disturbance, and biological traits might have influenced species richness and community composition.

| Termite diversity and the potential role of rainfall
Species diversity in Neotropical forests is highly variable, ranging from 10 termite species in the semi-arid savannah region of Caatinga-Brazil to 100 species in a natural rain forest (Alves, Mota, Lima, Bellezoni, & Vasconcellos, 2011;Bourguignon et al., 2011;Couto, Albuquerque, Vasconcellos, & Castro, 2015;Dambros, 2015;Palin et al., 2011;Vasconcellos, 2010;Vasconcellos et al., 2010;Viana et al., 2014). With 32 species, our data fall at the lower range. Many of the other studies worked with morpho-species, so data are not completely comparable.
Besides, sampling effort, forest type, disturbance, or geography could account for the different numbers. We might have missed a few species, as accumulation curves did not reach an asymptote for Colosó and Tayrona and we found significant differences between the observed richness and the Chao 2 estimator (Mann-Whitney test Z = −3.62, p < 0.001).
However, additional non-standardized sampling revealed few new species. Hence, we missed a few species, but this cannot explain the huge difference to some other rain forest studies. Hence, our results suggest that these dry forests are less species rich than lowland rain forests.
Although there were differences in estimated richness, we ob-  Figure S3, Table 2).
High species richness was related to rainfall, which is generally considered to be an important driver of ecosystem dynamics and productivity (Weltzin et al., 2003). Temperatures also decreased with increasing elevation and hence could also explain species richness. Yet we consider temperature alone as less important because (a) temperatures were still very similar across all sites and (b) they were all at the optimal range for tropical termites (review in Scheffrahn et al., 2015;e.g., Ackerman et al., 2009;Eggleton et al., 1999).
Dry forests have a large biomass of leaf litter with up to seven tons per hectare (Murphy & Lugo, 1986). Decomposition of leaf litter by bacteria and fungi is limited by humidity and temperature (Powers et al., 2009;Witkamp, 1966). Thus, while plenty of food is available for wood-feeding and leaf litter-feeding termites (feeding groups I and II), less exist for soil feeders (feeding groups III and IV). In line, we found fewer soil feeders such as Anoplotermes-group at the low elevation site Tayrona, despite its high content of organic matter. This site had many feeding group II Microcerotermes and Nasutitermes, and it had most feeding group I Kalotermitidae ( Figure 2). By contrast, our high elevation site Colosó had many soil feeders, in addition to Nasutitermes and Microcerotermes. These results imply that rainfall may be an important environmental factor affecting termite diversity and community composition as has been hypothesized by Lepage and Darlington (2000), Sugimoto, Bignell, and Macdonald (2000), Gathorne-Hardy et al. (2001) and Davies, Rensburg, Eggleton, and Parr (2013). Low rainfall can also explain the overall low diversity of termite species in these tropical dry forests compared to humid rain forests. To test this hypothesis, we did an additional analysis using published studies across the globe (Supporting Information Table S6). Even when excluding all other factors, such as biogeography, we found a significant increase in termite species richness with precipitation (Spearman's rho = 0.713; p < 0.001) worldwide, covering humid areas, savannahs, dry forests, and xerophilous ecosystems (Figure 3b). Our conclusions are supported by several studies, which all showed that soil feeders are especially rich under warm, humid conditions (Bourguignon et al., 2011;Couto et al., 2015;Eggleton et al., 1996;Eggleton et al., 1999;Eggleton, Davies et al., 2002;Isra et al., 2008;Palin et al., 2011;Roisin et al., 2006;Valladares, 2016).

| Other factors which may affect termite communities
Besides rainfall, our data indicate three other factors that may affect termite diversity and community composition. First, as in other studies Dosso et al., 2013;Hausberger & Korb, 2016), anthropogenic disturbance seems to influence termite community composition. Across all studies, strong disturbance associated with a transformation of natural ecosystems into agricultural fields leads to a decline in termite richness and a change in species composition (Attignon et al., 2005;Dosso et al., 2013;Leponce et al., 1997;Luke et al., 2014 which were only present in this site and had high abundances. The first two are invasive pests, and the last often occurs in pastures (Constantino, 2002;Scheffrahn, 2010). Ceibal only became a protected area in 2013 (Corporación Autónoma Regional del Canal del Dique, 2013), and it has a small size of 144 ha, strongly influenced by surrounding livestock farms and pastures. Hence, despite the fact that it has remnants of primary and secondary dry forest, the influence of silvopastoral systems and anthropogenic activity seems still evident.
A second factor potentially influencing termite composition, especially in Tayrona, might be high sand content. Sandy soils provide less stability for building tunnels and underground nests by termites (Lee & Wood, 1971), and the water infiltrates faster through the soil sandy matrix. Termites prefer moist areas (Cornelius & Osbrink, 2010 The third factor that may explain the high abundance of some genera might be linked with their biology. Several Microcerotermes spp. as well as N. corniger and N. dasyopsis have polycalic arboreal nests with several queens that are interconnected by networks of galleries (Adams, 1991;Hartke, 2010;Roisin, 1990;Roisin & Pasteels, 1986;Thorne, 1984;Thorne & Levings, 1989;Vasconcellos & Bandeira, 2006; Supporting Information Figure S4A,B). These attributes allow them to reach large colony sizes (Buschini & Lenonardo, 1999) that occupy and defend huge territories (Adams & Levings, 1987;Roisin & Pasteels, 1986), which makes them ecological dominant (Thorne, 1984;Vasconcellos & Bandeira, 2006) and can explain their high abundance in our study.
To conclude, our study revealed that termite diversity in Colombian tropical dry forests is less species rich than that of humid tropical rain forests. A decrease in termite diversity was associated with an inverse elevational gradient from the highest site with most species, more rainfall and low temperatures (Colosó) to the driest place, at low elevation (Tayrona) that harbored the least termite species. Along this gradient also the species composition changed with Nasutitermes and soil feeders (especially Anoplotermes-group) occurring at higher elevation and drywood termites (Kalotermitidae) being restricted to the low elevation sites. Thus, our data support the view that environmental filtering plays a role along elevation gradients, but in contrast to other studies, we found an inverse elevational gradient with most species occurring at high elevation. Based on these results, a meta-analysis revealed a positive correlation between termite species diversity and rainfall across the globe.

ACK N OWLED G M ENTS
We thank Universidad del Norte Barranquilla, Colombia, for a re- Roisin for help in species identification, to Karen Meusemann for comments on the text, and all those who indirectly helped in conducting this work.

DATA AVA I L A B I L I T Y
The data used in this study are archived in GenBank (accession numbers MH90825-MH90914).