Genetic Evidence for Landscape Effects on Dispersal in the Army Ant Eciton Burchellii

Molecular Ecology (2014) 23, 96–109 doi: 10.1111/mec.12573

Genetic evidence for landscape effects on dispersal in the army ant Eciton burchellii

THOMAS W. SOARE,* ANJALI KUMAR,*† KERRY A. NAISH‡ and SEAN O’DONNELL*§ *Animal Behavior Program, Department of Psychology, University of Washington, Seattle, WA 98195, USA, †Massachusetts Institute of Technology, Cambridge, MA 02139, USA, ‡School of Aquatic and Fishery Sciences, University of Washington, Seattle, WA 98195, USA, §Department of Biology, Drexel University, Philadelphia, PA 19104, USA

Abstract Inhibited dispersal, leading to reduced gene flow, threatens populations with inbreeding depression and local extinction. Fragmentation may be especially detrimental to social insects because inhibited gene flow has important consequences for cooperation and competition within and among colonies. Army ants have winged males and per- manently wingless queens; these traits imply male-biased dispersal. However, army ant colonies are obligately nomadic and have the potential to traverse landscapes. Eciton burchellii, the most regularly nomadic army ant, is a forest interior species: colony raiding activities are limited in the absence of forest cover. To examine whether nomadism and landscape (forest clearing and elevation) affect population genetic structure in a montane E. burchellii population, we reconstructed queen and male genotypes from 25 colonies at seven polymorphic microsatellite loci. Pairwise genetic distances among individuals were compared to pairwise geographical and resistance distances using regressions with permutations, partial Mantel tests and random forests analyses. Although there was no significant spatial genetic structure in queens or males in montane forest, dispersal may be male-biased. We found significant isolation by landscape resistance for queens based on land cover (forest clearing), but not on elevation. Summed colony emigrations over the lifetime of the queen may contribute to gene flow in this species and forest clearing impedes these movements and subse- quent gene dispersal. Further forest cover removal may increasingly inhibit Eciton burchellii colony dispersal. We recommend maintaining habitat connectivity in tropical forests to promote population persistence for this keystone species.

Keywords: deforestation, habitat fragmentation, isolation by distance, landscape genetics, sex-biased dispersal Received 21 November 2012; revision received 30 September 2013; accepted 15 October 2013

fragmentation can inhibit gene flow by isolating popu- Introduction lations, create inbreeding depression within subpopula- Dispersal has profound evolutionary consequences for tions, result in local extinction (Saccheri et al. 1998; many levels of biological organization (Broquet & Petit Segelbacher et al. 2010) and change species biology 2009). In continuous populations, the interaction (Fischer & Lindenmayer 2007). In tropical forest ecosys- between dispersal and genetic drift may lead to isola- tems, habitat fragmentation has altered species richness, tion by distance (Slatkin & Maddison 1990; Hardy & abundances and interactions, and affected ecosystem Vekemans 1999) and the balance between dispersal and processes such as nutrient cycling (Bierregaard et al. selection can influence adaptive evolution and specia- 1992; Laurance et al. 2002). tion (Turelli et al. 2001). Dispersal restricted by habitat Fragmentation may have significant effects on Hyme- noptera, which are haplodiploid and generally have a Correspondence: Thomas W. Soare, Fax: +1 206 685 3157; lower effective population size and less molecular vari- E-mail: [email protected] ation than diploids (Hedrick & Parker 1997). Gene flow

© 2013 John Wiley & Sons Ltd LANDSCAPE AFFECTS DISPERSAL IN ARMY ANTS 97 in social insects may have important consequences for colonies at high elevations are probably living at the cooperation and competition within colonies, among lower limit of their thermal tolerance (O’Donnell & colonies and among populations (Pamilo et al. 1997; Kumar 2006; O’Donnell et al. 2011; Soare et al. 2011), Ross 2001). Evolutionary processes are of particular high elevations may inhibit dispersal uniformly. Alter- interest in nomadic army ants (Eciton burchellii) because natively, colonies may be locally adapted to and may they are top predators and ecological keystones in Neo- preferentially disperse within their natal elevation band; tropical forests (Franks 1982; Franks & Bossert 1983; elevation bands are coarsely correlated with Holdridge Kaspari et al. 2011). life zones (Holdridge 1967; Haber 2000). In undisturbed ant populations, dispersal is primarily The specific aim of this study was to determine determined by the mode of colony founding: indepen- whether landscape variables (forest fragmentation and dent founding by flying queens or dependent founding elevation) affect recent gene flow in army ants. To where queens are accompanied by flightless workers achieve this aim, we determined whether colony emi- (Pamilo et al. 1997; Peeters & Ito 2001; Ross 2001). Dis- grations contributed to gene dispersion by comparing persal of dependent foundresses is determined by the spatial genetic structure (SGS) across castes (workers, distance over which they can walk and therefore is queens and males) sampled after dispersal. We then more limited than that of independent foundresses compared relatedness patterns among queens with (Pamilo et al. 1997; Peeters & Ito 2001; Sundstrom€ et al. competing isolation by landscape resistance models to 2005). Male ants are winged in most species (Holldobler€ examine the effects of forest clearing and elevation on & Wilson 1990); therefore, male-biased dispersal is more colony dispersal. We used a landscape genetics strongly associated with dependent colony founding in approach and sampled at the level of individuals with- ants (Pamilo et al. 1997; Sundstrom€ et al. 2005). out identifying populations in advance, enabling detec- Regular colony emigrations have the potential to con- tion of SGS (e.g. isolation-by-distance patterns: Rousset tribute to gene flow in nomadic army ants. Eciton burch- 2000; Hazlitt et al. 2004; Hardy et al. 2008; Banks & ellii army ant colonies are monogynous and reproduce Peakall 2012; Ivens et al. 2012) and comparison of rela- by fission. Two giant wingless queens each walk away tive influence of landscape variables on contemporary with half of the worker force (Schneirla 1971; Gotwald gene flow (Manel et al. 2003; Segelbacher et al. 2010; 1995; Kronauer 2009). Virgin queens mate immediately Storfer et al. 2010; Short Bull et al. 2011). with multiple males and store this sperm for the rest of their lives (Kronauer et al. 2006; Kronauer & Boomsma Methods 2007). Because army ants reproduce by colony fission but males are winged, dispersal may be male-biased. We sampled 25 Eciton burchellii colonies in Neotropical Previous genetic studies of army ants have detected montane forest around Monteverde, Costa Rica, from 15 high rates of polyandry (Kronauer et al. 2006) and evi- Jul 2006 to 16 Sep 2006 over 4.6 km E-W and 9.3 km N- dence for male-biased dispersal (Berghoff et al. 2008; S and across 600 m of elevation. We located colonies Perez-Espona et al. 2012a). However, army ant colonies through systematic trail walks and opportunistic have a regular two week nomadic period tied to their encounters (Vidal-Riggs & Chaves-Campos 2008; five-week reproductive cycle while the larvae are grow- Kumar & O’Donnell 2009; Soare et al. 2011), thus ing (Schneirla 1971; Franks & Fletcher 1983) and cover assuming a continuously distributed population and hundreds of metres during this time (Franks & Fletcher independence of observations (Storfer et al. 2007). Addi- 1983; Willson et al. 2011). tional details regarding the study site, sample collection Landscape variables may primarily affect queen dis- and genotyping can be found in the Supporting infor- persal by restricting colony emigrations. Despite regular mation (Data S1, Tables S1, S2). obligate nomadism and a wide geographical range, the army ant Eciton burchellii is a forest interior species Genetic analysis (Schneirla 1971; Meisel 2006) and populations are vulnerable to local extinction in forest fragments (Partridge We reconstructed queen and male genotypes from 13– et al. 1996; Boswell et al. 1998; Meisel 2004) and agroeco- 20 worker genotypes from each of 25 colonies (Table systems (Roberts et al. 2000). In lowland rainforest, Eci- S1) in the program COLONY 2.0 (Jones & Wang 2010). ton burchellii avoids entering open areas (Meisel 2006) To test the effects of genotyping error and to confirm and deforestation might inhibit dispersal (Perez-Espona reliability, we repeated COLONY analyses with error et al. 2012a). Colonies increasingly forage in open areas rates ranging from 0 to 5% and multiple random num- at high elevations (Kumar & O’Donnell 2009), and thus ber seeds. We calculated overall and within locus fragmentation in montane forest may or may not inbreeding among queen genotypes (FIq) in SPAGeDi impede gene flow. Furthermore, because Eciton burchellii 1.3 (Hardy & Vekemans 2002) and tested for linkage

© 2013 John Wiley & Sons Ltd 98 T. W. SOARE ET AL. disequilibrium among all pairs of loci in queen geno- In order to further evaluate the power of our types using FSTAT 2.9.3.2 (Goudet 2002) with 420 per- approach, we compared our data to a recent landscape mutations (Bonferroni adjusted nominal P = 0.0024). genetics study of a population of Eciton burchellii foreli The effective mating frequency [a corrected estimate of Mayr (1886) in Panama (Perez-Espona et al. 2012a,b; see the number of mates of a given queen (Nielsen et al. publications for details on sample collection and geno- 2003), useful for situations with skewed paternity or typing). Although six common microsatellite loci were high mating frequency] of army ant queens in Montev- amplified in both populations, we could not combine erde, Costa Rica, was compared with published data on the genetic data from Monteverde with that from San queens from Barro Colorado Island, Panama (Kronauer Lorenzo, Panama, into a single analysis because mark- et al. 2006); Henri Pittier National Park, Venezuela (Kro- ers were redesigned for the San Lorenzo population (S. nauer & Boomsma 2007); and Chiapas, Mexico (Jaffe Perez-Espona personal communication), and there are et al. 2009) with a Kruskal–Wallis ANOVA in the stats general difficulties associated with standardizing allele package in R 2.14.1 (R Development Core Team 2011). scores across different platforms (Seeb et al. 2007). To To examine the ability of the relatedness measures to facilitate a direct comparison of SBD across populations, detect known relationships with the observed allele fre- we regressed pairwise kinship (Fij) for queens and quencies, we conducted a power analysis on simulated males (genotypes obtained from 15 colonies: Perez- genotypes (see Data S2, Supporting information, for Espona et al. 2012b) on the natural logarithm of Euclid- details). ean distance and restricted distances (0–9.3 km, 0–7 km, 0–4 km, 0–2 km, 0–1 km and 0–0.5 km) in SPAGeDi 1.4 (Hardy & Vekemans 2002). The proportion of individu- Comparison of dispersal distances for queens, males als represented in each distance interval and the coeffi- and workers cient of variation (CV) of the number of times each To examine whether there was sex-biased dispersal individual was represented in each interval did not (SBD) in our system, we tested the relationship between comply with the suggestions of the software authors pairwise kinship (Fij: Loiselle et al. 1995) among queens, for the three smallest distance intervals (Hardy & males and workers and the natural logarithm of Euclid- Vekemans 2002). ean distance between colonies using regression and jackknifing over loci in SPAGeDi 1.3 (Hardy & Veke- Spatial analysis and evaluating effects of landscape mans 2002; Hardy et al. 2008; Sanllorente et al. 2010). We obtained two-tailed P values by permuting locations To determine whether land cover (forest clearing) or of individuals 10000 times and comparing the observed elevation impede colony migrations (queen dispersal), slope of the regression line to the mean permuted we evaluated six competing models: isolation by value, which is the equivalent of a Mantel test (Hardy Euclidean distance (Model 1), isolation by land cover & Vekemans 2002; Hardy et al. 2008). The level of resistance (Model 2, Fig. 1), isolation by elevation resis- significance obtained from such permutation tests (or tance (Model 3, Fig. 2), isolation by combined land Mantel tests as below) can be considered as a test of the cover and elevation resistance (Model 4), isolation by significance of the observed data compared to the null elevation band resistance (decreasing ability to disperse hypothesis of no relationship. Because our sampling out of natal elevation band, Model 5) and isolation by scheme was two-dimensional, pairwise kinship between combined land cover and elevation band resistance individuals should scale with the natural logarithm of (Model 6). To compare these models of dispersal, we Euclidean distance (Rousset 1997; Hardy & Vekemans constructed friction maps based on the five resistance 1999; Vekemans & Hardy 2004; McRae & Beier 2007; models (see Data S2, Supporting information, for Hardy et al. 2008), and thus, the transformed distance details) and calculated pairwise resistance distances was used in all analyses (hereafter ‘Euclidean distance’). among collection locations with the software CIRCUIT- Similarly, we also regressed pairwise kinship between SCAPE 3.5.4 (McRae & Beier 2007; Goulson et al. 2011; individuals in SPAGeDi over the following preselected Devitt et al. 2013). Resistance distances are based on cir- spatial extents: 0–7 km, 0–4 km, 0–2 km, 0–1 km and 0– cuit theory and simultaneously consider all possible 0.5 km. The proportion of individuals represented in pathways between nodes, with multiple or wider con- each distance interval and the coefficient of variation necting corridors allowing greater gene flow (McRae & (CV) of the number of times each individual was repre- Beier 2007). sented in each interval conformed to the suggestions of We predicted that landscape processes would affect the software authors, with the exception of the maxi- patterns of queen dispersal because queens disperse mum distance interval (7–9.3 km), which had a CV terrestrially. Males would be unlikely to be affected greater than one (Hardy & Vekemans 2002). because they disperse by flying. We therefore evaluated

Fig. 1 Location of study site and map of land cover classes with locations of sampled colonies. Five land cover classes are primary forest, secondary forest, edge habitat, agriculture/pasture, urban/bare earth. Decreasing shading indicates increasing habitat resistance with white being a complete barrier to dispersal.

01230.5 4 km

012340.5 km

Collection points < 850 m asl 851 – 1050 m asl 1051 – 1250 m asl 1251 – 1450 m asl Collection points 1451 – 1650 m asl Primary forest > 1650 m asl Seconday forest Edge habitat Fig. 2 Map of elevation bands with locations of sampled colonies. Elevation bands are <850, 851-1050, 1051-1250, 1251-1450, Agriculture/pasture 1451-1650 and > 1650 m asl. Decreasing shading indicates increasing habitat resistance with white being a complete bar- Urban/bare earth rier to dispersal.

© 2013 John Wiley & Sons Ltd 100 T. W. SOARE ET AL. isolation by resistance scenarios on queen genotypes, a Results method powerful to detect recent gene ﬂow and ﬁne- scale SGS (Hazlitt et al. 2004; Vekemans & Hardy 2004; Quality of genetic data Clemencet et al. 2005; Hardy et al. 2008; Kronauer et al. 2010; Sanllorente et al. 2010; Segelbacher et al. 2010; Two workers were removed for having genotypes Latch et al. 2011; Banks & Peakall 2012). incompatible with the majority matriline (i.e. likely the

Pairwise relatedness (rij: Queller & Goodnight 1989; daughter of a previously replaced queen: Kronauer and Fij: Loiselle et al. 1995) among queens was et al. 2006), and two workers due to poor amplifications regressed on pairwise resistance distances for Models (missing data at four of seven and five of seven loci, 2-6 in SPAGeDi as described for the SBD analyses above respectively). We also removed four workers from the (Sundstrom€ et al. 2003; Hazlitt et al. 2004). In a continuous data set for likely 2-bp mutations (two at Eb24 and two two-dimensional landscape, resistance distance is at Eb51); taking these mutations into account would expected to scale linearly with genetic similarity (McRae have made the genotypes consistent with the majority & Beier 2007). Coefficients of determination (r2) were matriline. This finding reflects the previously reported compared among the various models to determine high mutation rate of these loci (Kronauer et al. 2006). which predictor distance (Model) accounted for the All remaining workers (n = 413) were genotyped at five most variation in the genetic distance among queens or more loci (390 workers had full genotypes at all loci). (highest r2 value) (McRae & Beier 2007; Perez-Espona et al. 2008; Devitt et al. 2013). To account for testing six Population descriptive statistics a priori distance models, we adjusted the alpha level with the Benjamini and Yekutieli false discovery rate COLONY reconstructed 25 female and 246 male geno- procedure (B-Y FDR: Benjamini & Yekutieli 2001; types contributing to workers from the 25 collections; Narum 2006; Latch et al. 2011). Due to the challenge of thus, each collection used in the genetic analysis was a assigning friction values when specific landscape effects unique colony. The average number of mates per queen are unknown (Spear et al. 2010), a range of friction val- was 9.8 Æ 0.4 SE (range: 5–14) and the effective mating Æ ues or resistance distances was tested for each model frequency (me: Nielsen et al. 2003) was 11.0 1.5 SE.

(see Table S3, Supporting information, for details). The effective mating frequency (me) of army ant queens To determine whether habitat variables affect queen did not significantly differ among populations studied dispersal after removing any effect of Euclidean dis- to date: Barro Colorado Island, Panama (12.9 Æ 1.1 SE: tance, we performed partial Mantel tests in R 2.14.1 (R Kronauer et al. 2006), Henri Pittier National Park, Development Core Team 2011) using the package Venezuela (9.4 Æ 0.2 SE: Kronauer & Boomsma 2007), ECODIST 1.2.5 (Goslee & Urban 2007) and obtained and Chiapas, Mexico (12.7 Æ 0.5 SE: Jaffe et al. 2009) two-tailed P values based upon 10000 permutations of (Kruskal-Wallis ANOVA v2 = 5.77, P = 0.12). pairwise genetic similarity among all pairs (Goslee & The overall and locus-specific values for FIq, a measure Urban 2007; Goulson et al. 2011; Short Bull et al. 2011). of inbreeding in queens, were not significantly different Partial Mantel tests detect effects of landscape on from zero (overall: À0.02 Æ 0.05 SE, P = 0.503; and each genetic structure while removing any effects of Euclid- locus: P > 0.05). We did not detect any significant linkage ean distance (McRae & Beier 2007; Storfer et al. 2010; between any pairs of loci for the queen genotypes (each Goulson et al. 2011; Short Bull et al. 2011), and are P > 0.0024, the nominal P after Bonferroni adjustment for appropriate for use on distance matrices (Legendre & multiple comparisons). Males mated to the same queen Fortin 2010). We reported uncorrected P values for all were related: pairwise kinship among a queen’s mates, Æ possible partial Mantel tests (including inverse partial Fmm, was significantly greater than zero (0.0553 Mantel tests) and adjusted the alpha level using the B-Y 0.0076 SE, P < 0.0001). The two measures of genetic simi- FDR procedure to account for multiple tests. Testing all larity between queens, the pairwise kinship coefficient possible Mantel and partial Mantel tests in this manner (Fij: Loiselle et al. 1995) and the pairwise relationship is equivalent to a causal modeling framework (Cushman coefficient (rij: Queller & Goodnight 1989), were highly et al. 2006; Cushman & Landguth 2010a). correlated (r = 0.924, P < 0.0001). Performing multiple partial Mantel tests can lead to inflation of the Type I error rate (Balkenhol et al. 2009), Power analysis especially when the predictors (distance measures) are highly correlated (Cushman & Landguth 2010a). There- Both the pairwise kinship coefficient (Fij) and the pair- fore, we also performed random forests analyses to wise relationship coefficient (rij) significantly predicted confirm the reliability of our results (see Data S2, true pairwise relatedness (rtrue) among genotypes simu- Supporting information, for details).

© 2013 John Wiley & Sons Ltd LANDSCAPE AFFECTS DISPERSAL IN ARMY ANTS 101 lated using allele frequencies from Monteverde (Adj In the 15 colonies sampled by Perez-Espona et al. 2 = < 2 = < R 0.474, P 0.0001; and Adj R 0.414, P 0.0001; (2012a,b) in San Lorenzo, Panama, pairwise kinship (Fij) respectively; Fig. S1, Supporting information). Values of among queens significantly declined with the natural the pairwise kinship and relationship coefficients were logarithm of Euclidean distance over 16.1 km (b = significantly greater for the highest relatedness class À0.021, r2 = 0.0640, P = 0.018; Table 1, Fig. 4). Pairwise = (rtrue 0.5) than for the other classes of relatedness kinship (Fij) among males also significantly declined over (Tukey’s HSD P < 0.0001 in each case; Fig. S1). Both the same distance (b = À0.014, r2 = 0.0060, P = 0.005). genetic similarity measures for intermediate levels of When pairwise kinship was regressed over a distance = = relatedness (rtrue 0.25 and rtrue 0.125) were signifi- range restricted to 9.3 km (the maximum pairwise cantly greater than those of the lowest levels of related- distance sampled in Monteverde, Costa Rica), there = = < ness (rtrue 0.0313 and rtrue 0, Tukey’s HSD P 0.05 was still evidence of spatial genetic structuring in both in all cases; Fig. S1). sexes (queens: b = À0.024, r2 = 0.0652, P = 0.033; males: b = À0.011, r2 = 0.0033, P = 0.047; Table 1). Dispersal Effects of landscape The regression of the pairwise kinship coefficient (Fij) on the natural logarithm of Euclidean distance (Model To compare effects of landscape variables on colony 1) showed no significant isolation by distance for any dispersal, we regressed the pairwise relationship coeffi- caste over the spatial extent of 9.3 km measured in cient (rij) among queens on resistance distances. Pair- Monteverde (queens: b = À0.002, r2 = 0.0004, P = 0.687; wise relatedness among queens significantly declined males: b = À0.0007, r2 < 0.0001, P = 0.684; workers: with land cover resistance (Model 2: b = À0.0020, b = 0.0003, r2 < 0.0001, P = 0.932; Table 1, Fig. 3). There r2 = 0.0469, P = 0.005; Table 2). Queen pairwise related- was also no significant spatial genetic structure (SGS) ness significantly increased with increasing elevation over distance categories up to 7 km, 4 km, 2 km, 1 km resistance (Model 3: b = 0.0025, r2 = 0.0247, P = 0.015). or 500 m for any caste (Table 1). There was no significant effect of elevation band (Model

Table 1 Pairwise genetic similarity (kinship coefficient, Fij) within castes is not significantly predicted by Euclidean distance over any restricted distance in E. burchellii parvispinum (Monteverde, Costa Rica) but is for E. burchellii foreli (San Lorenzo, Panama) within 16.1 km, 9.3 km, 7 km (males only) and 0.5 km. Sample size (N) is the number of pairwise comparisons among individuals (exclud- ing within-colony comparisons) over designated spatial extent. Slope (b), coefficient of determination (r2) and significance level (P, two-tailed) demonstrate strength of relationships. Bold values represent significant relationships

Monteverde, Costa Rica San Lorenzo, Panama Spatial Caste extent (km) N br2 PbrN 2 P

Queens 0–16.1 —— — — 105 À0.0207 0.0640 0.018 Males —— — — 22685 À0.0137 0.0060 0.005 Workers ———————— Queens 0–9.3 300 À0.0022 0.0004 0.687 69 À0.0243 0.0652 0.033 Males 28999 À0.0007 <0.0001 0.684 14458 À0.0111 0.0033 0.047 Workers 81794 0.0003 <0.0001 0.932 ———— Queens 0–7 266 À0.0059 0.0023 0.415 54 À0.0231 0.0479 0.076 Males 25409 0.0008 <0.0001 0.764 11260 À0.0154 0.0056 0.032 Workers 72901 0.0003 <0.0001 0.942 ———— Queens 0–4 193 À0.0054 0.0014 0.566 26 À0.0375 0.0587 0.089 Males 18302 À0.0008 <0.0001 0.755 5458 À0.0139 0.0030 0.204 Workers 53413 À0.00004 <0.0001 0.959 ———— Queens 0–2 98 0.0111 0.0043 0.493 12 À0.0080 0.0015 0.703 Males 9304 À0.0026 0.0001 0.590 2518 À0.0057 0.0004 0.662 Workers 26686 À0.0020 0.0002 0.702 ———— Queens 0–138À0.0032 0.0003 0.900 7 0.0812 0.0769 0.046 Males 3492 À0.0003 <0.0001 0.973 1492 0.0806 0.0424 0.003 Workers 10620 À0.0084 0.0025 0.382 ———— Queens 0–0.5 11 À0.0025 0.0003 0.988 1 ——— Males 1105 À0.0002 <0.0001 0.993 221 ——— Workers 2998 À0.0302 0.0159 0.237 ————

0.18 Fig. 3 There was no spatial genetic struc-

ture (pairwise kinship coefﬁcient, Fij) for Queens queens, males and workers in a popula- Males tion of Eciton burchellii parvispinum in 0.12 Workers Monteverde, Costa Rica. Distance inter- ) ij vals are 0 (within colony), 0–0.5 km (not including within colony), 0.5–1 km, 1– 0.06 2 km, 2–4 km, 4–7 km and 7–10 km, graphed on the natural logarithm of Euclidean distance. Error bars repre-

Pairwise kinship (F sent Æ 2SE. Values offset to facilitate 0 visualization.

–0.06 00.5 124710 Pairwise Euclidean distance (km)

0.18 Fig. 4 Spatial genetic structure (pairwise

kinship coefﬁcient, Fij) for queens and Queens males in a population of Eciton burchellii 0.12 Males foreli in San Lorenzo, Panama. Distance intervals are 0 (within colony), 0–0.5 km ) ij (not including within colony), 0.5–1 km, 0.06 1–2 km, 2–4 km, 4–7 km, 7–11 km and 11–16 km, graphed on the natural logarithm of Euclidean distance. Error bars 0 represent Æ 2SE. Values offset to facili-

Pairwise kinship (F tate visualization. –0.06

–0.12 0 0.5 1 2 4 7 11 16 Pairwise Euclidean distance (km)

Table 2 Resistance distances based on land cover and elevation significantly predict pairwise genetic similarity (relationship coeffi- 2 cient, rij) among queens. Slope (b), coefficient of determination (r ) and significance level (P, two-tailed) demonstrate strength of relationships. Bold values represent significant relationships (B-Y FDR adjusted alpha = 0.0161)

Pairwise relationship (rij) Pairwise kinship (Fij)

Caste Model br2 Pb r2 P

Queens 1. ln(Euclidean) 0.0119 0.0023 0.475 À0.0022 0.0004 0.687 Males 1. ln(Euclidean) ———À0.0007 0.00001 0.684 Workers 1. ln(Euclidean) 0.0068 0.0007 0.469 0.0003 0.00001 0.932 Queens 2. Land cover À0.0020 0.0469 0.005 À0.0004 0.0093 0.080 Queens 3. Elev 0.0025 0.0247 0.015 0.0003 0.0019 0.440 Queens 4. Land + Elev À0.0006 0.0052 0.323 À0.0002 0.0044 0.232 Queens 5. Elev band 0.0024 0.0067 0.202 0.0004 0.0007 0.681 Queens 6. Land + Elev band À0.0013 0.0232 0.039 À0.0003 0.0060 0.161

5: b = 0.0024, r2 = 0.0067, P = 0.202) or elevation com- FDR adjusted alpha = 0.0161). None of the resistance bined with land cover (Model 4: b = À0.0006, r2 = distances were correlated with the pairwise kinship = = À 2 = 0.0052, P 0.323; and Model 6: b 0.0013, r 0.0232, coefﬁcient (Fij) over the spatial extent measured (Table 2). P = 0.039) on pairwise relatedness among queens (B-Y However, land cover (Model 2) accounted for the

© 2013 John Wiley & Sons Ltd LANDSCAPE AFFECTS DISPERSAL IN ARMY ANTS 103 greatest amount of variation in pairwise kinship relative hoc partial Mantel test controlling for land cover resis- to the other resistance distances tested (b = À0.0004, tance (Model 2), the significance of elevation (Model 3) 2 = = r 0.0093, P 0.080). Tests of the effects of different as a predictor of pairwise queen relatedness (rij) disap- friction values or resistance distances on model results peared (r = 0.1172, P = 0.098). Alternatively, land cover revealed few qualitative changes. Statistical significance resistance (Model 2) remained a significant predictor of was lost in two models predicting the pairwise relation- pairwise relatedness (rij) among queens after controlling ship coefficient (rij) among queens: when the least for elevation resistance (Model 3), in a second post hoc vegetated class (urban/bare earth) and the highest partial Mantel test (r = À0.1902, P = 0.008). elevations (> 1650 m asl) were not considered to be In random forests (RF) analyses, land cover ranked complete barriers to dispersal (Models 2.g and 3.f: Table highest in variable importance for predicting both the

S3, Supporting information). pairwise relationship coefﬁcient (rij) and pairwise

Some of the distance measures were significantly kinship coefficient (Fij) among queens (Fig. 5). Land correlated with each other (Table S4, Supporting infor- cover remained the highest ranked for variable impor- mation). After controlling for Euclidean distance, land tance in analyses with randomly chosen variable sub- cover (Model 2) significantly predicted values of the sets (mtry) of size 4 and 1 (Figs. S2, S3, Supporting pairwise relationship coefficient (rij): pairwise relat- information). edness among queens decreased with increasing land cover resistance (r = À0.2252, P = 0.001; Table 3). Discussion Euclidean distance remained a nonsignificant predictor of pairwise relatedness among queens after controlling Our primary aims were to examine sex-biased dispersal for land cover resistance (r = 0.0793, P = 0.226). After (SBD) of army ants and to determine whether forest controlling for Euclidean distance, elevation (Model 3) clearing and elevation affected gene flow in tropical also remained a significant predictor of the pairwise montane forest. The mating frequency of army ant relationship coefficient such that relatedness among queens was comparable with other sites. We did not queens increased with increasing elevation (r = 0.1930, detect any evidence for reduced diversity within indi- P = 0.008). Euclidean distance remained a nonsignifi- viduals, nor linkage disequilibrium, suggesting that cant predictor of queen relatedness after controlling for there is no inbreeding in this population. Current levels elevation resistance (r = À0.1229, P = 0.078). In a post of forest clearing and development in Monteverde,

Table 3 Resistance distances based on land cover and elevation significantly predicted pairwise genetic similarity (relationship coef- ~ ficient, rij) after controlling for Euclidean distance. Each partial Mantel test was modelled as Y X1 |X2 and measured the correlation of X1 & Y after controlling for (given) X2. Correlation coefficient (r) and significance level (P, two-tailed) demonstrate strength of relationships. Bold values represent significant relationships (B-Y FDR adjusted alpha = 0.0139)

Y Model X1 X2 rP

À rij 2 Land cover ln(Eucl) 0.2252 0.001 ln(Eucl) Land cover 0.0793 0.226 3 Elev ln(Eucl) 0.1930 0.008 ln(Eucl) Elev À0.1229 0.078 4 Land cover + Elev ln(Eucl) À0.1502 0.047 ln(Eucl) Land cover + Elev 0.1404 0.039 5 Elev band ln(Eucl) 0.0718 0.252 ln(Eucl) Elev band À0.0254 0.682 6 Land cover + Elev band ln(Eucl) À0.1941 0.008 ln(Eucl) Land cover + Elev band 0.1307 0.053 À Fij 2 Land cover ln(Eucl) 0.0948 0.077 ln(Eucl) Land cover À0.0061 0.916 3 Elev ln(Eucl) 0.0943 0.080 ln(Eucl) Elev À0.0857 0.115 4 Land cover + Elev ln(Eucl) À0.0753 0.179 ln(Eucl) Land cover + Elev 0.0402 0.484 5 Elev band ln(Eucl) 0.0650 0.241 ln(Eucl) Elev band À0.0623 0.273 6 Land cover + Elev band ln(Eucl) À0.0773 0.156 ln(Eucl) Land cover + Elev band 0.0177 0.754

(a) Fig. 5 Ranking of variable importance in random forests (RF) analyses for predicting (a) the pairwise relationship coefﬁ-

cient (rij) and (b) the pairwise kinship coefﬁcient (Fij). Mean percent increase in mean squared error (MSE) is the normal- ized difference in MSE for predicting the out-of-bag (OOB) data when the variable is included in the model as observed ver- sus that when the variable is randomly permuted, averaged over all trees. Ran- (b) dom permutations of an important variable will lead to large relative increases in MSE. Mean increase in accuracy is the reduction in node impurities when split- ting on the variable, as measured by residual sum of squares, averaged over all trees.

Costa Rica, have not yet restricted access to mates nor intermediate and low levels of relatedness among simu- generated inbreeding in army ants. lated genotypes based on observed allele frequencies. The spatial extent of the current study may have been insufficient for detecting isolation by distance. We Dispersal patterns across sexes detected isolation by distance in queens and males We found no SGS in queens (which disperse by walk- sampled over 16.1 km in Panama (Perez-Espona et al. ing) or males (which disperse by flying) up to 9.3 km in 2012b), suggesting that larger geographic distances may Monteverde. In contrast, Berghoff et al. (2008) found be important for studying SBD (Storfer et al. 2007; Oy- weak but significant genetic differentiation among sub- ler-McCance et al. 2013). However, the distance over populations separated by 5–10 km in the Panama Canal which colonies were sampled in Monteverde (9.3 km) Zone, suggesting a dispersal limitation for one or both represents a greater distance than previous army ant sexes. SGS in both sexes was readily detected from sam- dispersal estimates (Berghoff et al. 2008; Jaffe et al. ples collected in San Lorenzo, Panama (Perez-Espona 2009). In other ant species examined to date, dispersal et al. 2012b). In San Lorenzo, the slope of the regression distance of winged reproductives has been estimated to of queen pairwise relatedness on distance was more be up to 2 km (Vogt et al. 2000; Doums et al. 2002; negative than that of males, suggesting stronger spatial Clemencet et al. 2005; Hardy et al. 2008). Furthermore, genetic structuring in queens (male-biased dispersal). we performed regressions of queen and male pairwise Comparing pairwise relatedness estimates among indi- relatedness that were restricted to a distance of 9.3 km viduals at nuclear loci across the sexes supported differ- in the San Lorenzo data set (Perez-Espona et al. 2012b), ences in nuclear and mitochondrial DNA differentiation and detected the same significant patterns of queen and reported in the original study (Perez-Espona et al. male spatial genetic structuring as those found over 2012a). This result demonstrates the differences between 16.1 km. Thus, the spatial extent of 9.3 km should have the two populations (Monteverde and San Lorenzo), the been sufficient to detect SGS in data from Monteverde. utility of the individual-based approach in detecting The respective geographies of the two study sites SBD and the power of the current study design (same may have led to differences in levels of SGS. San Lore- opportunistic sampling scheme, same number of loci nzo is located on a peninsula created by the construc- utilized and larger sample size than that of San Lore- tion of the Panama Canal (see Fig. 1 in Perez-Espona nzo; Perez-Espona et al. 2012a,b). However, undersam- et al. 2012a) and Monteverde consists of multiple frag- pling of pairs of colonies within certain distance ments of formerly continuous forest (Nadkarni & intervals in both data sets may have introduced bias Wheelwright 2000). An army ant population on a pen- into the computation of the average genetic similarity insula may be more likely to display stronger spatial for those intervals. Power analyses revealed that both genetic structuring than those in sites unconfined by measures of genetic similarity used in this study (pair- water. On the other hand, colony emigration patterns wise kinship coefficient, Fij, and pairwise relationship do not seem to differ between a confined population on coefficient, rij) were able to discriminate among high, Barro Colorado Island and populations in continuous

© 2013 John Wiley & Sons Ltd LANDSCAPE AFFECTS DISPERSAL IN ARMY ANTS 105 forest (Franks & Fletcher 1983; Willson et al. 2011; but tested) and significantly predicted pairwise relatedness see Califano & Chaves-Campos 2011). The relatively after controlling for any effects of Euclidean distance recent history of development in the Monteverde area and elevation. This result was supported by the fact (since the 1950s: Nadkarni & Wheelwright 2000), com- that the random forests analyses showed land cover pared to the construction of the Panama Canal (1910s: was the most important variable in predicting pairwise Berghoff et al. 2008), may mean that the natural popula- queen relatedness. Therefore, summed migrations over tion in Monteverde is not yet at equilibrium (Hardy & a colony’s lifetime may contribute to gene flow in this Vekemans 1999) or that insufficient generations have species and forest clearing may inhibit dispersal. The passed to create isolation by distance (Vekemans & effects of land cover on army ant dispersal in Montever- Hardy 2004; Landguth et al. 2012). However, SGS and de agree with those in San Lorenzo: resistance distances SBD have been detected in simulations after a single based on deforestation explained most of the pairwise generation (Landguth et al. 2010; Banks & Peakall 2012). differentiation among colonies at nuclear loci (Perez- Dispersal patterns may differ between populations due Espona et al. 2012a). to ecology or evolutionary histories (Spear et al. 2010). Cleared areas might represent a barrier to male dis- The Monteverde and San Lorenzo populations consist persal. Perez-Espona et al. (2012a) found deforestation of morphologically distinct subspecies (E. b. parvispinum to be correlated with genetic distance based on nuclear and E. b. foreli, respectively), and any biological DNA but not mtDNA and concluded that open areas differences between the two may affect spatial genetic inhibit male but not queen dispersal. However, mtDNA structuring. has a relatively low mutation rate and may lack the The lack of SGS in E. burchellii queens might be spatial and temporal resolution to detect recent gene explained by regular colony emigrations. If a colony flow (Foitzik et al. 2009; Anderson et al. 2010; Storfer continued marching over the landscape for the queen’s et al. 2010; Wang 2010, 2011). Because males probably lifetime (at least 4 years: Rettenmeyer 1963), then fly across the waters of the Panama Canal and Chagres summed migrations may separate related queens by River (Berghoff et al. 2008; Perez-Espona et al. 2012a), several kilometres. Successive statary bivouacs are we assume that males are able to cross deforested areas located about 500 m apart, which is greater than the as well. expected distance if a colony was performing a random The models of isolation by resistance may not have walk during the nomadic phase (Franks & Fletcher captured potentially relevant factors to army ant dis- 1983; Willson et al. 2011). Colonies also follow a single persal such as precise Holdridge life zone (Holdridge compass bearing during entire nomadic phases (Franks 1967; Haber 2000), microclimatic differences (Meisel & Fletcher 1983; Willson et al. 2011; but see Califano & 2006; O’Donnell & Kumar 2006; Soare et al. 2011) or Chaves-Campos 2011). If a colony pursued the same fine-scale topography (Murphy et al. 2010; Latch et al. compass bearing for four years, and moved 500 m in 2011) because the friction maps were based on satellite each nomadic phase, it would travel approximately data. We also assumed no other barriers to dispersal 21 km. Long-term observations in continuous habitat, (e.g. roads, swamps, streams) because we have or repeat genetic sampling, may elucidate lifetime observed E. burchellii colonies crossing dirt roads and movements of E. burchellii colonies. streams (T. W. Soare, A. Kumar., and S. O’Donnell, Resource heterogeneity, combined with low emigra- personal observation). In line with these observations, tion costs (e.g. low intercolony aggression: Swartz 1997; Perez-Espona et al. (2012a) did not find effects of roads Willson et al. 2011), may explain the high queen or a swamp on army ant dispersal, although streams dispersal rate observed in this study. Long distance had mixed effects. Past ecological studies have found dispersal is favoured under conditions of environmental that exposed areas limit colony movements (Meisel heterogeneity (Johnson & Gaines 1990). E. burchellii col- 2006), although the effect appears to decline with eleva- onies exhaust resources locally (Franks 1982; Otis et al. tion (Kumar & O’Donnell 2009). A land cover classifica- 1986), migrate to new foraging grounds (Franks & tion captures differences in forest cover (Cihlar 2000), Fletcher 1983; Kaspari & O’Donnell 2003) and specialize and because E. burchellii is a forest interior species, on high prey-density patches (Kaspari et al. 2011). resistance distances based on land cover may be most relevant when examining queen dispersal in this species. Effects of landscape on dispersal The power to detect effects of elevation on army ant We found evidence for genetic isolation by landscape dispersal might have been influenced by the fact that resistance in Eciton burchellii queens. Land cover the full range of elevations at which E. burchellii occurs accounted for the greatest amount of variation in pair- (Watkins 1976) was not captured here (Short Bull et al. wise kinship (relative to the other resistance distances 2011). Furthermore, the sampling scheme was not

© 2013 John Wiley & Sons Ltd 106 T. W. SOARE ET AL. sufficient to separate elevation variability and forest 2004). Future research should address which habitat fragmentation. Spatial genetic structuring of mitochon- types and corridor widths permit Eciton gene flow drial DNA in the African army ant Dorylus molestus among isolated populations, and what levels of gene over a 22-km transect was observed in an undisturbed flow are required to allow population persistence across habitat in Mt. Kenya Forest Reserve, Kenya (over a the latitudinal and elevational range of these top preda- 1200-m elevation range: Kronauer et al. 2010). This tors. Supporting army ant populations would increase result might be explained, however, by the fact that the local biodiversity (Rettenmeyer et al. 2011), especially timing and direction of D. molestus emigrations are for ant-following birds (Meisel 2004; Kumar & O’Don- more random than those of E. burchellii although both nell 2007). Increasing development and climate change species occupy similar trophic niches (Gotwald 1995; may exacerbate the inhibition of army ant dispersal by Schoning€ et al. 2005). Although power for detecting forest clearing. Therefore re-evaluations of the effect of effects of landscape on genetic variation also decreases habitat fragmentation on dispersal of this keystone with coarser resolution (Storfer et al. 2007; Cushman & species are vital. Landguth 2010b), a 90-m grain size for elevation is still less than the average distance covered by a colony dur- Acknowledgments ing the entire nomadic phase (c. 500 m: Franks & Fletcher 1983; Willson et al. 2011). Thus, this grain size Yamile Molina and Sebastian Jurado assisted with collections. appears to be sufficient to detect effects of elevation on We thank Sılvia Perez-Espona for allowing the use of the dispersal within the range sampled (Anderson et al. queen and male genotypes ahead of the full release of the San Lorenzo data set. Sean Schoville, Graham Stone, Sılvia Perez- 2010). A further constraint on the power of the current Espona and three anonymous reviewers made helpful com- study is that the elevation bands were not equally rep- ments on the manuscript. We thank Jim Wolfe, the Stuckey resented. One band contained nearly half of the colonies family, the Rockwell family, the Vitosi family, the Salazar fam- sampled and many collections were geographically ily, the Monteverde Conservation League, the Monteverde proximate. However, if elevation (or elevation band) Cloud Forest Reserve, the Monteverde Butterfly Garden, inhibited dispersal, resistance distance should trend Ecolodge San Luis and the University of Georgia for allowing negatively with genetic relatedness. The opposite pat- us to work on their lands. Various residents of Monteverde, especially the Joyce-van Dusen family, and the Monteverde tern was observed here, suggesting that land cover Institute provided logistical support. We thank Bruce Godfrey seems to be the only landscape variable examined that and members of the Naish lab for genetic troubleshooting, inhibits army ant dispersal in the Monteverde area. especially Todd Seamons for help with allele scoring. Nick If queens are capable of migrating long distances Cuba made helpful comments on the spatial analyses. Funding (including hundreds of metres in elevation) over their was provided by a grant from the University of Washington lifetimes, then E. burchellii colonies may encounter vari- Royalty Research Fund and NSF grants IBN 0347315 and IOS ation in climate and life zone ecology (Holdridge 1967; 1209072 to S.O’D. and funding from the Organization for Trop- ical Studies to A.K. Field research was conducted under per- Haber 2000) over the broad latitudinal and elevational mits from the Costa Rican government (MINAE scientific range of the species (Watkins 1976). Raiding behaviour passport #0387) in accordance with the laws of Costa Rica. (O’Donnell & Kumar 2006; Kumar & O’Donnell 2009) and bivouac site selection (Soare et al. 2011) change with elevation. Perhaps colony-level behaviours are References phenotypically flexible, such that colonies respond to local environmental conditions. Experimental colony Anderson CD, Epperson BK, Fortin MJ et al. (2010) Consider- ing spatial and temporal scale in landscape-genetic studies relocations (e.g. Franks & Fletcher 1983) could be used of gene flow. Molecular Ecology, 19, 3565–3575. to test this hypothesis. Balkenhol N, Waits LP, Dezzani RJ (2009) Statistical Given our finding that land cover correlated with approaches in landscape genetics: an evaluation of methods queen relatedness, we recommend maintaining habitat for linking landscape and genetic data. Ecography, 32, connectivity to facilitate E. burchellii gene flow across 818–830. the landscape. The negative effects of forest clearing on Banks SC, Peakall R (2012) Genetic spatial autocorrelation can army ant mobility are known to decline at higher eleva- readily detect sex-biased dispersal. Molecular Ecology, 21, 2092–2105. tions, as foraging activity in open areas increases with Benjamini Y, Yekutieli D (2001) The control of the false discov- elevation (Kumar & O’Donnell 2009). Thus, the effects ery rate in multiple testing under dependency. Annals of of land cover on emigration may be more pronounced Statistics, 29, 1165–1188. in lowland forests (Perez-Espona et al. 2012a). Army Berghoff SM, Kronauer DJC, Edwards KJ, Franks NR (2008) ants avoid open areas in lowlands (Meisel 2006), and Dispersal and population structure of a New World preda- populations are vulnerable to extinction in forest frag- tor, the army ant Eciton burchellii. Journal of Evolutionary – ments (Partridge et al. 1996; Boswell et al. 1998; Meisel Biology, 21, 1125 1132.

Bierregaard RO, Lovejoy TE, Kapos V, Dossantos AA, Hutch- Gotwald WH (1995) Army Ants: The Biology of Social Predation. ings RW (1992) The biological dynamics of tropical rainforest Cornell University Press, Ithaca. fragments. BioScience, 42, 859–866. Goudet J (2002) FSTAT, A Program to Estimate And Test Gene Boswell GP, Britton NF, Franks NR (1998) Habitat fragmenta- Diversities And Fixation Indices 2.9.3.2. Department of Ecology tion, percolation theory and the conservation of a keystone & Evolution, University of Lausanne. URL http://www2. species. Proceedings of the Royal Society of London Series B-Bio- unil.ch/popgen/softwares/fstat.htm. logical Sciences, 265, 1921–1925. Goulson D, Kaden JC, Lepais O, Lye GC, Darvill B (2011) Pop- Broquet T, Petit EJ (2009) Molecular estimation of dispersal for ulation structure, dispersal and colonization history of the ecology and population genetics. Annual Review of Ecology garden bumblebee Bombus hortorum in the Western Isles of Evolution and Systematics, 40, 193–216. Scotland. Conservation Genetics, 12, 867–879. Califano D, Chaves-Campos J (2011) Effect of trail pheromones Haber WA (2000) Plants and Vegetation. In: Monteverde: Ecology and weather on the moving behaviour of the army ant Eciton and Conservation of a Tropical Cloud Forest(eds Nadkarni NM burchellii. Insectes Sociaux, 58, 309–315. & Wheelwright NT), pp. 39–94. Oxford University Press Inc., Cihlar J (2000) Land cover mapping of large areas from satel- New York. lites: status and research priorities. International Journal of Hardy OJ, Vekemans X (1999) Isolation by distance in a contin- Remote Sensing, 21, 1093–1114. uous population: reconciliation between spatial autocorrela- Clemencet J, Viginier B, Doums C (2005) Hierarchical analysis tion analysis and population genetics models. Heredity, 83, of population genetic structure in the monogynous ant Cata- 145–154. glyphis cursor using microsatellite and mitochondrial DNA Hardy OJ, Vekemans X (2002) SPAGeDi: a versatile com- markers. Molecular Ecology, 14, 3735–3744. puter program to analyse spatial genetic structure at the Cushman SA, Landguth EL (2010a) Spurious correlations and individual or population levels. Molecular Ecology Notes, 2, inference in landscape genetics. Molecular Ecology, 19, 618–620. 3592–3602. Hardy OJ, Pearcy M, Aron S (2008) Small-scale spatial genetic Cushman SA, Landguth EL (2010b) Scale dependent inference structure in an ant species with sex-biased dispersal. Biologi- in landscape genetics. Landscape Ecology, 25, 967–979. cal Journal of the Linnean Society, 93, 465–473. Cushman SA, McKelvey KS, Hayden J, Schwartz MK (2006) Hazlitt SL, Eldridge MDB, Goldizen AW (2004) Fine-scale Gene flow in complex landscapes: testing multiple hypothe- spatial genetic correlation analyses reveal strong ses with causal modeling. American Naturalist, 168, 486–499. female philopatry within a brush-tailed rock-wallaby col- Devitt TJ, Cameron Devitt SE, Hollingsworth BD, McGuire JA, ony in southeast Queensland. Molecular Ecology, 13, Moritz C (2013) Montane refugia predict population genetic 3621–3632. structure in the Large-blotched Ensatina salamander. Molecu- Hedrick PW, Parker JD (1997) Evolutionary genetics and lar Ecology, 22, 1650–1665. genetic variation of haplodiploids and X-linked genes. Doums C, Cabrera H, Peeters C (2002) Population genetic Annual Review of Ecology and Systematics, 28,55–83. structure and male-biased dispersal in the queenless ant Holdridge LR (1967) Life Zone Ecology. Tropical Science Center, Diacamma cyaneiventre. Molecular Ecology, 11, 2251–2264. San Jose, Costa Rica. Fischer J, Lindenmayer DB (2007) Landscape modification and Holldobler€ B, Wilson EO (1990) The Ants. Belknap Press, Cam- habitat fragmentation: a synthesis. Global Ecology and Biogeog- bridge. raphy, 16, 265–280. Ivens ABF, Kronauer DJC, Pen I, Weissing FJ, Boomsma JJ Foitzik S, Bauer S, Laurent S, Pennings PS (2009) Genetic diver- (2012) Reproduction and dispersal in an ant-associated root sity, population structure and sex-biased dispersal in three aphid community. Molecular Ecology, 21, 4257–4269. co-evolving species. Journal of Evolutionary Biology, 22, Jaffe R, Moritz RFA, Kraus FB (2009) Gene flow is maintained 2470–2480. by polyandry and male dispersal in the army ant Eciton bur- Franks NR (1982) Social insects in the aftermath of swarm chellii. Population Ecology, 51, 227–236. raids of the army ant Eciton burchelli. In: The Biology of Johnson ML, Gaines MS (1990) Evolution of dispersal: theoreti- Social Insects: Proceedings of the Ninth Congress of the Interna- cal models and empirical tests using birds and Mammals. tional Union for the Study of Social Insects (eds Breed MD, Annual Review of Ecology and Systematics, 21, 449–480. Michener CD, Evans HE), pp. 275–279. Westview Press, Jones OR, Wang J (2010) COLONY: a program for parentage Boulder. and sibship inference from multilocus genotype data. Molec- Franks NR, Bossert WH (1983) The influence of swarm raiding ular Ecology Resources, 10, 551–555. army ants on the patchiness and diversity of a tropical leaf Kaspari M, O’Donnell S (2003) High rates of army ant raids in litter ant community. In: Tropical Rain Forest: Ecology and the Neotropics and implications for ant colony and commu- Management(eds Sutton SL, Chadwick AC & Whitmore TC), nity structure. Evolutionary Ecology Research, 5, 933–939. pp. 151–163. Blackwell Scientific Publications, Boston. Kaspari M, Powell S, Lattke J, O’Donnell S (2011) Predation Franks NR, Fletcher CR (1983) Spatial patterns in army ant for- and patchiness in the tropical litter: do swarm-raiding army aging and migration: Eciton burchelli on Barro Colorado ants skim the cream or drain the bottle? Journal of Animal Island, Panama. Behavioral Ecology and Sociobiology, 12, Ecology, 80, 818–823. 261–270. Kronauer DJC (2009) Recent advances in army ant biology Goslee SC, Urban DL (2007) The ecodist package for dissimilar- (Hymenoptera: Formicidae). Myrmecological News, 12,51–65. ity-based analysis of ecological data. Journal of Statistical Kronauer DJC, Boomsma JJ (2007) Do army ant queens re-mate Software, 22,1–19. later in life? Insectes Sociaux, 54,20–28.

Kronauer DJC, Berghoff SM, Powell S et al. (2006) A reassess- O’Donnell S, Kumar A (2006) Microclimatic factors associated ment of the mating system characteristics of the army ant with elevational changes in army ant density in tropical Eciton burchellii. Naturwissenschaften, 93, 402–406. montane forest. Ecological Entomology, 31, 491–498. Kronauer DJC, Schoning€ C, d’Ettorre P, Boomsma JJ (2010) Col- O’Donnell S, Kaspari M, Kumar A, Lattke J, Powell S (2011) ony fusion and worker reproduction after queen loss in Elevational and geographic variation in army ant swarm raid army ants. Proceedings of the Royal Society B-Biological Sciences, rates. Insectes Sociaux, 58, 293–298. 277, 755–763. Otis GW, Santana E, Crawford DL, Higgins ML (1986) The Kumar A, O’Donnell S (2007) Fragmentation and elevation effect of foraging army ants on leaf-litter arthropods. Biotro- effects on bird-army ant interactions in neotropical montane pica, 18,56–61. forest of Costa Rica. Journal of Tropical Ecology, 23, 581–590. Oyler-McCance SJ, Fedy BC, Landguth EL (2013) Sample Kumar A, O’Donnell S (2009) Elevation and forest clearing design effects in landscape genetics. Conservation Genetics, 14, effects on foraging differ between surface - and subterranean 275–285. - foraging army ants (Formicidae: Ecitoninae). Journal of Pamilo P, Gertsch P, Thoren P, Seppa€ P (1997) Molecular popu- Animal Ecology, 78,91–97. lation genetics of social insects. Annual Review of Ecology and Landguth EL, Cushman SA, Schwartz MK et al. (2010) Quanti- Systematics, 28,1–25. fying the lag time to detect barriers in landscape genetics. Partridge LW, Britton NF, Franks NR (1996) Army ant Molecular Ecology, 19, 4179–4191. population dynamics: the effects of habitat quality and reserve Landguth EL, Fedy BC, Oyler-McCance SJ et al. (2012) Effects size on population size and time to extinction. Proceedings of of sample size, number of markers, and allelic richness on the Royal Society of London Series B-Biological Sciences, 263, 735– the detection of spatial genetic pattern. Molecular Ecology 741. Resources, 12, 276–284. Peeters C, Ito F (2001) Colony dispersal and the evolution of Latch EK, Boarman WI, Walde A, Fleischer RC (2011) Fine- queen morphology in social hymenoptera. Annual Review of scale analysis reveals cryptic landscape genetic structure in Entomology, 46, 601–630. desert tortoises. PLoS ONE, 6, e27794. Perez-Espona S, Perez -Barberıa FJ, McLeod JE et al. (2008) Laurance WF, Lovejoy TE, Vasconcelos HL et al. (2002) Ecosys- Landscape features affect gene flow of Scottish Highland red tem decay of Amazonian forest fragments: a 22-year investi- deer (Cervus elaphus). Molecular Ecology, 17, 981–996. gation. Conservation Biology, 16, 265–280. Perez-Espona S, McLeod JE, Franks NR (2012a) Landscape Legendre P, Fortin M-J (2010) Comparison of the Mantel test genetics of a top neotropical predator. Molecular Ecology, 21, and alternative approaches for detecting complex multivari- 5969–5985. ate relationships in the spatial analysis of genetic data. Perez-Espona S, McLeod JE, Franks NR (2012b) Data from: Molecular Ecology Resources, 10, 831–844. landscape genetics of a top neotropical predator. Dryad Loiselle BA, Sork VL, Nason J, Graham C (1995) Spatial Digital Repository, doi:10.5061/dryad.k2qc1. genetic structure of a tropical understory shrub, Psychotria Queller DC, Goodnight KF (1989) Estimating relatedness using officinalis (Rubiaceae). American Journal of Botany, 82, genetic markers. Evolution, 43, 258–275. 1420–1425. R Development Core Team (2011) R: A Language and Environ- Manel S, Schwartz MK, Luikart G, Taberlet P (2003) Landscape ment for Statistical Computing. R Foundation for Statistical genetics: combining landscape ecology and population genet- Computing, R Development Core Team, Vienna, Austria. ics. Trends in Ecology & Evolution, 18, 189–197. ISBN 3-900051-07-0, URL http://www.R-project.org/ McRae BH, Beier P (2007) Circuit theory predicts gene flow in Rettenmeyer CW (1963) Behavioral studies of army ants. plant and animal populations. Proceedings of the National University of Kansas Science Bulletin, 64, 281–465. Academy of Sciences of the United States of America, 104, Rettenmeyer CW, Rettenmeyer ME, Joseph J, Berghoff SM 19885–19890. (2011) The largest animal association centered on one spe- Meisel JE (2004) The influence of microclimate and habitat area cies: the army ant Eciton burchellii and its more than 300 on the ecology of the army ant Eciton burchellii in tropical associates. Insectes Sociaux, 58, 281–292. forest fragments. PhD thesis, University of Wisconsin, Roberts DL, Cooper RJ, Petit LJ (2000) Use of premontane moist Madison. forest and shade coffee agroecosystems by army ants in wes- Meisel JE (2006) Thermal ecology of the neotropical army ant tern Panama. Conservation Biology, 14, 192–199. Eciton burchellii. Ecological Applications, 16, 913–922. Ross KG (2001) Molecular ecology of social behaviour: analyses Murphy MA, Evans JS, Storfer A (2010) Quantifying Bufo boreas of breeding systems and genetic structure. Molecular Ecology, connectivity in Yellowstone National Park with landscape 10, 265–284. genetics. Ecology, 91, 252–261. Rousset F (1997) Genetic differentiation and estimation of gene Nadkarni NM, Wheelwright NT (2000) Monteverde: Ecology and flow from F-statistics under isolation by distance. Genetics, Conservation of a Tropical Cloud Forest. Oxford University 145, 1219–1228. Press Inc., New York. Rousset F (2000) Genetic differentiation between individuals. Narum SR (2006) Beyond Bonferroni: less conservative analy- Journal of Evolutionary Biology, 13,58–62. ses for conservation genetics. Conservation Genetics, 7, Saccheri I, Kuussaari M, Kankare M et al. (1998) Inbreeding 783–787. and extinction in a butterfly metapopulation. Nature, 392, Nielsen R, Tarpy DR, Reeve HK (2003) Estimating effective 491–494. paternity number in social insects and the effective number Sanllorente O, Hammond RL, Ruano F, Keller L, Tinaut A of alleles in a population. Molecular Ecology, 12, 3157–3164. (2010) Extreme population differentiation in a vulnerable

slavemaking ant with a fragmented distribution. Conservation Wang IJ (2011) Choosing appropriate genetic markers and ana- Genetics, 11, 1701–1710. lytical methods for testing landscape genetic hypotheses. Schneirla TC (1971) Army Ants: A Study in Social Organization (ed Molecular Ecology, 20, 2480–2482. Topoff HR). WH Freeman and Company, San Francisco. Watkins JF (1976) The Identification and Distribution of New Schoning€ C, Njagi WM, Franks NR (2005) Temporal and spatial World Army Ants (Dorylinae: Formicidae). Baylor University patterns in the emigrations of the army ant Dorylus (Anom- Press, Waco. ma) molestus in the montane forest of Mt Kenya. Ecological Willson SK, Sharp R, Ramler IP, Sen A (2011) Spatial move- Entomology, 30, 532–540. ment optimization in Amazonian Eciton burchellii army ants. Seeb LW, Antonovich A, Banks AA et al. (2007) Development Insectes Sociaux, 58, 325–334. of a standardized DNA database for Chinook salmon. Fisher- ies, 32, 540–552. Segelbacher G, Cushman SA, Epperson BK et al. (2010) Applications of landscape genetics in conservation biology: T.W.S., A.K., K.A.N., and S.O’D. designed the study. concepts and challenges. Conservation Genetics, 11, 375–385. T.W.S., A.K., and S.O’D. performed field and laboratory Short Bull RA, Cushman SA, Mace R et al. (2011) Why work. T.W.S., A.K., and K.A.N. analysed the data. T.W.S. replication is important in landscape genetics: American black wrote the first version of this manuscript. All authors bear in the Rocky Mountains. Molecular Ecology, 20, 1092–1107. contributed to and approved the final manuscript. Slatkin M, Maddison WP (1990) Detecting isolation by distance using phylogenies of genes. Genetics, 126, 249–260. Soare TW, Tully SI, Willson SK, Kronauer DJC, O’Donnell S (2011) Choice of nest site protects army ant colonies from environmental extremes in tropical montane forest. Insectes Data accessibility Sociaux, 58, 299–308. Microsatellite genotypes uploaded as DRYAD entry Soare TW, Kumar A, Naish KA, O’Donnell S (2013) Data from: genetic evidence for landscape effects on dispersal in the doi:10.5061/dryad.j33k3. army ant Eciton burchellii. Dryad Digital Repository, doi:10. 5061/dryad.j33k3. Supporting information Spear SF, Balkenhol N, Fortin M-J, McRae BH, Scribner K (2010) Use of resistance surfaces for landscape genetic Additional supporting information may be found in the online studies: considerations for parameterization and analysis. version of this article. Molecular Ecology, 19, 3576–3591. Storfer A, Murphy MA, Evans JS et al. (2007) Putting the ‘land- Data S1 Methodological details. scape’ in landscape genetics. Heredity, 98, 128–142. Data S2 Storfer A, Murphy MA, Spear SF, Holderegger R, Waits LP Analytical details. (2010) Landscape genetics: where are we now? Molecular Table S1 Date, collection number (coll_no), and number of – Ecology, 19, 3496 3514. workers (no_workers) of army ant colonies used in genetic € Sundstrom L, Keller L, Chapuisat M (2003) Inbreeding and analyses. sex-biased gene flow in the ant Formica exsecta. Evolution, 57, – 1552 1561. Table S2 Annealing temperatures (Ta) for seven primers, € € Sundstrom L, Seppa P, Pamilo P (2005) Genetic population number of alleles (Na), range of allele sizes, and observed (Ho) structure and dispersal patterns in Formica ants - a review. and expected (He) heterozygosities at each locus. Annales Zoologici Fennici, 42, 163–177. Swartz MB (1997) Behavioral and population ecology of the Table S3 Alternative models of landscape resistance predicting army ant Eciton burchelli and ant-following birds. PhD thesis, pairwise queen relatedness. University of Texas, Austin. Table S4 Pairwise distance measures among sample locations Turelli M, Barton NH, Coyne JA (2001) Theory and speciation. were correlated. Trends in Ecology & Evolution, 16, 330–343. Vekemans X, Hardy OJ (2004) New insights from fine-scale Fig. S1 Plot of estimates of genetic similarity of simulated spatial genetic structure analyses in plant populations. Molec- dyads of known relationships. ular Ecology, 13, 921–935. Vidal-Riggs JM, Chaves-Campos J (2008) Method review: esti- Fig. S2 Ranking of variable importance in random forests (RF) mation of colony densities of the army ant Eciton burchellii in analyses for predicting (a) the pairwise relationship coefficient Costa Rica. Biotropica, 40, 259–262. (rij) and (b) the pairwise kinship coefficient (Fij) with random Vogt JT, Appel AG, West MS (2000) Flight energetics and variable subsets (mtry) of size 4. dispersal capability of the fire ant, Solenopsis invicta Buren. Journal of Insect Physiology, 46, 697–707. Fig. S3 Ranking of variable importance in random forests (RF) Wang IJ (2010) Recognizing the temporal distinctions between analyses for predicting (a) the pairwise relationship coefficient landscape genetics and phylogeography. Molecular Ecology, (rij) and (b) the pairwise kinship coefficient (Fij) with random 19, 2605–2608. variable subsets (mtry) of size 1.

All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately.