Study area and sample collection
The Rivas Isthmus is a narrow land corridor connecting the Pacific slope of Central America (Fig. 2). As in all of the dry forests in Mesoamerica, the Rivas Isthmus has been populated for centuries and consequently has a long history of anthropogenic disturbance. It is estimated that less than 0.1% of the original old growth, Mesoamerican dry forest remains (Janzen 1988; Gillespie et al. 2000), and this forest type is currently a high priority for conservation (Miles et al. 2006). However, Rivas retains significant cover of closed canopy dry forest and therefore has the potential to support a diverse animal and plant community.
Figure 2. Distribution of the 15 sampling sites or putative social groups of spider monkeys (Ateles geoffroyi) on the Rivas Isthmus, Nicaragua used in this analysis. Site locations reflect the centroid of a 3 ha sampling area. Dark gray shading indicates closed-canopy forest cover, light gray is open-canopy, and lines represent permanent roads and the international border (forest cover derived from Sesnie et al. 2008).
Download figure to PowerPoint
We surveyed for spider monkeys across a 50,000 ha area on the isthmus, bordered on all sides by actual or potential barriers to animal migration. The Costa Rican border forms the southern boundary and the northern boundary is a major two-lane road. Both features are buffered by deforested land and nonnative, planted forests that do not provide habitat. Spider monkeys may have been extirpated north of this study area possibly because it is more heavily populated and urbanized. Spider monkeys are found in forest patches south of the study area, along the southern border of Lake Nicaragua. Reconnecting these patches could enable gene flow between spider monkeys on the Pacific and Atlantic coasts.
We stratified sample collection based on expected social group structure as spider monkeys are not continuously or randomly distributed across the landscape. They likely have localized patterns of genetic similarity that result from the size and distribution of social groups, female-biased dispersal, and the species' preference for closed canopy forest. As such, we searched for animals or fresh samples within 300 ha plots, or within the expected home range size for a social group (95–390 ha; Wallace 2008; Spehar et al. 2009). Most sampling locations were chosen based on the presence of large, closed-canopy forest stands (coverage described in Sesnie et al. 2008). Informal interviews with local residents and behavioral observations were also used to confirm the likelihood that samples came from a single social group. Spider monkeys fission into small foraging parties and actively avoid human contact, which makes locating animals difficult. Hence, we used a concentrated searching technique (Pruetz and Leasor 2002) to maximize the number of genotypes potentially represented in the fecal samples, focusing on riparian areas and large, fruit-bearing trees. We collected fecal samples within 8 h of defecation and stored them in a 1:1 solution with RNALater for 1–3 months in ambient temperatures prior to DNA extraction or long-term storage at −20°C. Most samples were collected during the dry season (January–May).
We collected fecal samples from 2007 to 2010 to describe genetic variation within each social group. We retained groups for population-level analysis that included >10 samples; observed group sizes in Ateles spp. are 15-55, in A. geoffroyi 18-42 (Wallace 2008). This resulted in separate global and restricted datasets for both of the genomes reported below. Not all sites were sampled in all years, and sampling was suspended at any site that was impacted by significant land cover change. For example, forest was cleared on several private properties near Sapoá that likely resulted in a change in the home range of spider monkeys at two sites (DM and GU; Fig. 2). Hence, these sites were sampled between 2007 and 2008, before these changes took place. Because of the slow reproductive rate in this species, it is unlikely that we sampled multiple generations; in dry forests, females do not reproduce until approximately 7 years of age, and interbirth intervals can exceed 48 months (Fedigan and Rose 1995).
DNA extraction and species identification
We extracted DNA from fecal samples using QIAmp DNA Stool Mini Kits with an extended proteinase K digestion step, to maximize DNA yield. We confirmed the source species of all mammalian DNA by analyzing two overlapping, 400–500 bp segments of the mitochondrial cytochrome-b region using recommended protocols (Janczewski et al. 1995; Verma and Singh 2003). It is necessary to confirm the species identity for all fecal samples as fecal morphology can be misleading, particularly if there are sympatric animals that have a similar diet. In our study, this procedure removed morphologically similar samples of primates (A. palliata and C. capucinus) and kinkajou (Potos flavus). The polymerase chain reaction (PCR) products were purified using ExoSap-It (USB) and sequenced using the standard BigDye Terminator 3.1 (Applied Biosystems, Foster City, CA) or stepped elongation cycle sequencing protocols (Platt et al. 2007). Sequencing reactions were purified using the recommended ethanol precipitation method and analyzed on an ABI 3730 Genetic Analyzer. All samples that matched the Ateles genus in the GenBank database were retained for analysis; that is those with identity scores of 100% and e-value >0 using nucleotide BLAST within Geneious Pro 5 (Biomatters, Auckland, New Zealand).
Analyses of the mtDNA control-region
We used information from Collins and Dubach (2000) and Ascunce et al. (2003) to develop primers within the first hypervariable portion of the mitochondrial control region (HVRI). Our forward (5′ GTGCATTATTGCTTGTCCCC) and reverse (5′ GTTGGTTTCACGGAGGATGG) primers are similar to those in Ascunce et al. (2003), but with minor sequence shifts to better match spider monkey template DNA, minimize the risk of hairpinning, and increase GC content in the 5′ end. This primer pair results in a 221 bp amplicon that overlaps with those produced by Collins and Dubach (2000). PCR reactions were optimized in 20 μL reactions at the following concentrations: 0.5 units of Taq polymerase, 1× PCR buffer, 1.5 mmol/L MgCl2, 0.2 mmol/L dNTP mix, 0.5 μg BSA, and 0.1μmol/L of each primer with 4 μL of the extracted DNA solution. We used touchdown PCR profiles for the HVR1 and microsatellite markers (described below) to ensure amplification. Annealing temperatures ranged from 68 to 51°C with starting and denaturation at 94°C and extension at 72°C. The profile was as follows: 94°C (5 min), 17 cycles [94°C (30 sec)], 68-51°C (30 sec each, −2°C/cycle), 72°C (30 sec)], 23 cycles [94°C (30 sec)], 50°C (30 sec each), 72°C (30 sec)], and 70°C (5 min). Sequencing protocols follow those used for species identification above. The resulting sequences were manually edited, aligned, and checked against accessioned sequences in Geneious Pro 5. Samples with identical mtDNA haplotypes that also displayed identical multilocus nuclear microsatellite genotypes (described below) were considered to be duplicate samples of the same individual and were removed. We calculated diversity statistics and tests of neutrality and variance in DNASP (Librado and Rozas 2009) and Arlequin 3.5.1 (Excoffier and Lischer 2010) and used these to construct a haplotype network in HapStar (Teacher and Griffiths 2011). We further tested the significance of differentiation among sampling sites using on Hudson et al. (1992) method for high diversity datasets (for all tests P < 0.05, 10,000 permutations).
Analyses of nuclear microsatellite genotypes
We screened 23 nuclear microsatellite primer pairs previously amplified in platyrrhine primate species and chose eight easily scored and polymorphic loci to avoid downstream genotyping errors (DeWoody et al. 2006; Supporting Information). We tested both M13 universal tail (Schuelke 2000) and directly labeled primers. PCR was prepared in 10 μL volumes with 3 μL of template, and a final concentration of 0.3 units of taq polymerase, 1× PCR buffer, 2 mmol/L MgCl2, 0.2 mmol/L dNTP mix, 0.5 μg BSA, and 0.02 μmol/L of each primer. Some sample DNA was more successfully amplified by increasing the MgCl2/dNTP ratio. We used touchdown PCR around the optimal annealing temperatures to ensure amplification of all alleles: 95°C (5 min), 35 cycles [95°C (45 sec)], Ta (45 sec), 72°C (45 sec)], and 72°C (7 min). All alleles were analyzed on an ABI 3730 and were scored in GeneMapper 4.0 (Applied Biosystems).
We took several precautions to avoid genotyping errors, which is a particular concern when using noninvasive samples to amplify nuclear DNA (e.g., DeWoody et al. 2006). A blood sample from a captive A. geoffroyi (Hogle Zoo, Utah) and human buccal swabs were genotyped for all loci as positive controls, to predict allele size, and to assess contamination. Alleles were discarded if they resembled those found in human DNA rather than the spider monkey control. Two independent observers scored alleles visually (without automated binning) and we replicated PCR reactions to confirm our results. Following replication trials, samples that were missing data for more than two of the eight loci were removed from the analysis. We checked for indications of null alleles, allelic dropout, and stuttering based on patterns of homozygosity and allele size using MicroChecker (van Oosterhout et al. 2004). We matched multi-locus genotypes, reviewing the chromatograms for all genotypes that differed by fewer than three alleles. We used this final dataset to estimate the probability of identity across loci and genotypes, using both the unbiased estimator and the conservative P(ID)sib (in Gimlet; Valière 2002; Waits et al. 2001).
Population-level statistics were calculated in Genepop 4.0 (Rousset 2008), SMOGD (Crawford 2010), and SPAGeDi (Hardy and Vekemans 2002), unless otherwise noted. We tested for linkage disequilibrium in the eight loci using the log-likelihood G-test. We assessed allelic and genotypic diversity via allele and private allele richness (rarefaction calculations based on one minus the minimum number of alleles at a locus in Hp-rare 1.0; Kalinowski 2005), expected and observed heterozygosity with Levene's correction for small samples sizes, and diversity and differentiation statistics (Hs, Nei's Gst, and Jost's D as estimated in SMOGD; 1000 bootstrap). Whereas Nei's “coefficient of differentiation” (Gst) may represent the effect of migration rates, Jost's D may better illustrate differentiation based on mutation rate and allele identity (Jost 2008, 2009). To determine if the observed level of heterozygosity was significantly lower than expected, we tested for homoscedasticity (Bartlett test) and conducted a paired t-test for deviations greater than zero (P < 0.05; in the R package adegenet 1.2-6; Jombart 2008). To identify deviations from Hardy Weinberg Equilibrium (HWE) we performed Raymond and Rousset's (1995) multisample test for heterozygote deficiency and Fisher exact tests with MCMC sampling (100 batches, 1000 iterations per batch; Guo and Thompson 1992).
To describe population structure and differentiation, we calculated Fst using the standard method of moments estimation (Weir and Cockerham 1984) and Jost's D. Although Hedrick's G'st as a standardization of Nei's Gst is a robust metric for highly variable markers or when comparing across markers, it does not correct for bias due to small sample or population size (Meirmans 2006). Furthermore, heterozygosity-based statistics may be biased if both diversity and population structuring are high, as both affect the partitioning of variance. Jost's D specifically incorporates effective alleles and genetic identity, information which is lost when using heterozygosity alone (Jost 2008). We tested for isolation by distance (IBD) in the population using Mantel tests in the vegan R package (Oksanen et al. 2005). We used two genetic distances among sites, the linearized Fst (to standardize Fst under IBD; Rousset 1997) and differentiation (D; Jost 2008). Spatial distances were calculated as Euclidean distances from a central location in each study site (measured in ArcGIS 9.3; ESRI, Redlands, CA), and were log-transformed for analysis (Rousset 1997).
To further investigate localized mating patterns, we compared relatedness coefficients between individuals among and within sampling sites and geographic distance classes (Hardy and Vekemans 2002). We chose Loiselle et al.'s (1995) kinship coefficient as it is independent of HWE and robust to localized, discontinuous mating patterns, and low frequency alleles (Vekemans and Hardy 2004). Kinship was calculated among individuals in predefined subpopulations (i.e., the putative social groups based on sampling site) and among 10 distance classes, ranging from 2.5 to 25 km (the longest distance between sites). Distance classes were designed to include an equal frequency of pairwise comparisons (Aspi et al. 2009).We used our entire sample of individuals (not just those within groups of >10 individuals) to analyze distance classes, in order to use all available information. SPAGeDI's jackknife procedure was used to summarize over loci and estimate standard errors. We ran 10,000 permutations of individual spatial locations for all analyses (Hardy and Vekemans 2002).