Microsatellites
We measured population genetic structure by Analysis of Molecular
Variance (AMOVA) (Excoffier, Smouse, &
Quattro, 1992) and by pairwise Fst values between populations in
Arlequin v3.5.2.2 (Excoffier & Lischer,
2010). Significance was tested using 1,000 permutations. We generated
several subsets to examine signatures of population differentiation
related to the varying ecologies and potential barriers to dispersal
across our collection sites. Specifically, our subsets were: a) all
regions; b) all regions with the island excluded; c) ANF panhandle
regions only (i.e., ARD and WRD); d) peninsular regions only (i.e.,
North Central near Gainesville and Central near Orlando); and e) ANF
regions as one collective region and peninsular regions as one
collective region. For each subset, we adjusted the allowed missing data
(from 0.15 to 0.5) in order to include all nine loci in all analyses.
To further assess population genetic structure, we used the Bayesian
clustering algorithm implemented through STRUCTURE v2.3.4
(Pritchard, Stephens, & Donnelly, 2000),
which uses allele frequency data to assign individuals into genetic
clusters (K). These analyses were executed with no prior population
information and under the admixture model with correlated allele
frequencies. In order to determine K, we assessed K = 1 through K = 10
with default parameters. To explore each K value, we performed 10
replicates with a burnin value of 50,000 and 500,000 MCMC iterations. We
determined the optimal K value by assessing the maximum likelihood
values and ΔK (Evanno, Regnaut, & Goudet,
2005) implemented in Structure Harvester v0.6.94
(Earl & von Holdt, 2012). We then
averaged the 10 STRUCTURE replicates for the optimal K through the
CLUMPAK pipeline (Kopelman, Mayzel,
Jakobsson, Rosenberg, & Mayrose, 2015) and visualized the results in
DISTRUCT v1.1 (Rosenberg, 2004).
We performed Mantel tests to test whether genetic distances correlate
with geographic distance (isolation by distance; IBD). We generated
several subsets to test for IBD. Specifically, we considered the
following: a) all sampling sites across all regions; b) all sampling
sites excluding the island; c) all sampling sites within the ANF; and d)
all sampling sites within the peninsula. For measures of genetic
distances, we used pairwise Fst values obtained from Arlequin and
linearized them as Fst/(1-Fst) (Rousset,
1997). Negative Fst values, if present, were set to zero. Otherwise,
the absolute difference between values would be inflated, when in
actuality they are effectively zero (i.e., no genetic differentiation).
We used Geographic Distance Matrix Generator v1.2.3
(Ersts, 2013) to generate pairwise
geographic distances (in km) from the GPS coordinates of the sampling
sites. We implemented Mantel tests using the ade4 package
(Dray & Dufour, 2007) in R
(Team, 2018) with 9,999 permutations.