Propagule pressure and hunting pressure jointly determine genetic evolution in insular populations of a global frog invader

Islands are often considered to be more susceptible to biological invasions and to suffer greater impacts from invaders than mainland areas, and this difference is generally attributed to differences in species introductions, ecological factors or human activities between islands and mainland areas. Genetic variation, as a good estimate of evolutionary potential, can influence the invasion process and impacts of alien species. However, few studies have compared the genetic diversity of alien species between islands and a corresponding mainland. Here, we examined the genetic variation and differentiation in feral populations (30 sampled individuals/population) of a globally invasive species (the American bullfrog, Lithobates catesbeianus) that was extensively farmed on 14 islands in the Zhoushan Archipelago of China and in three nearby regions on the mainland. We quantified the relative importance of propagule pressure and hunting pressures on the genetic variation of bullfrog populations and found that insular populations have greater genetic variation than their mainland counterparts. Although genetic differentiation between the populations was observed, no evidence of recent bottlenecks or population expansion in any of the tested population was found. Our results suggest that the propagule pressures of bullfrogs escaping from farms, multiple releases and hunting pressure influence the genetic variation among bullfrog populations. These results might have important implications for understanding the establishment and evolution of alien species on islands and for the management of invasive species.

Questionnaire survey method and forms used to assess the bullfrog residence time. Table S2 Information of all the primers used in the study. Table S3 Population sampling localities, GPS coordinates, and sample sizes (n).  Table S8 The best models (i.e. ΔAIC ≤ 2) containing factors influencing the genetic variation of L. catesbeianaus in the Zhoushan Archipelago. Table S9 Summary of model averaging results based on multiple linear regression models. Table S10 Coefficients and probabilities of the Spearman correlation analysis among predictors. Figure S1 Genetic clusters (K) obtained from the STRUCTURE analysis of three invasive regions of the mainland and 14 islands. Figure S2 Coefficients and probabilities (in parentheses) of the Spearman correlation analysis among predictors.

Table S1
Questionnaire survey method on bullfrog residence time and Questionnaire Form Methods A questionnaire was sent to residents of houses on or close to the banks of water bodies, or to farmers that live near a water body and had raised bullfrogs on their farms before, or to patrolmen who are responsible for the safety of a water body. The residents were most likely first to sight bullfrog individuals or hear bullfrog calls when bullfrogs invaded water bodies in close proximity; the farmers were likely to notice if their bullfrogs had escaped from their farms, and to where; the patrolmen usually patrolled the banks of a water body once every one to two weeks, and so were likely to notice any unusual events at the water bodies, including sight or sound of bullfrogs. People canvassed were usually at least 45 years old (this means that they were already adults in the period when local farms started raising bullfrogs), and long-term workers or residents in the village where the water bodies investigated were located. Most of the people knew farmers when the farmers in their village were raising bullfrogs.
People were first asked if they recognized bullfrogs by describing the morphology and let them see pictures of bullfrogs or if they recognized calls of bullfrogs by play backing the calls. Those who did not recognize bullfrogs or their calls were excluded from the further survey. People were then asked if they had sighted bullfrog individuals or heard bullfrog calls, and for how many years, in a sampled water body that they lived close to or that they patrolled. They were also asked to describe distinct events linked to the time when they first saw bullfrogs. This would help to clarify if the accuracy of their answers could be confirmed by independent events. Answers that could not be confirmed by events were excluded from the analysis. D. When did you first see bullfrogs or hear bullfrog calls from the water body investigated? E. Why are you sure the year that you saw bullfrogs or heard bullfrog calls from the water body? Were there some significant events occurring that year (significantly political events, large natural disasters, the year that your child went to some schools, birth dates or death date of your family members or your neighbor hoods, periods that someone lost a job that was related to you, and so on)?

Table S5
Asymmetric migration rates between Lithobates catesbeianaus populations inferred in MIGRATE. Populations in bold and italics are the source and sink populations respectively. Mean value (±95% CIs). Migration rates (M=m/μ, where m is the immigration rate per generation) have been estimated between populations using a coalescent based Monte Carlo Markov Chain method. We used a Brownian approximation model, and mutation was considered to be constant for all loci. We implemented Fst estimates and a UPGMA tree as starting parameters for the estimation of M and performed five independent runs using one long chain with a run of 10 7 recorded parameter genealogies after discarding the first 10 5 genealogies as burn-in for each locus. Daishan   •, displays that a factor is included in the model; ΔAICc, the difference between each model and the highest ranked model; AICc, the second-order Akaike information criterion; Wi (Akaike weights), the probability that the predictor is a component of one of the best models; R 2 , R-squared.  •, displays that a factor is included in the model; ΔAICc, the difference between each model and the highest ranked model; AICc, the second-order Akaike information criterion; Wi (Akaike weights), the probability that the predictor is a component of one of the best models; R 2 , R-squared.  •, displays that a factor is included in the model; ΔAICc, the difference between each model and the highest ranked model; AICc, the second-order Akaike information criterion; Wi (Akaike weights), the probability that the predictor is a component of one of the best models; R 2 , R-squared.

Figure S1
The approach was run using an admixture model estimating the most likely number of genetic clusters (K) populations that have different allele frequencies at a number of independent loci. The burn in was set to 500,000 iterations followed by 1,000,000 iterations of MCMC. All runs were repeated ten times for each number of possible genetic clusters (K) ranging from 1 to n+1(n is number of sampling sites in different regions of mainland or islands). The optimal number of clusters (the best K) was determined using method of 1 and was implemented in STRUCTURE HARVESTER 2 . The ΔK method was not applicable in this dataset because mean lnP(X|K) decreased and its variance increased with increasing values of K 1 .

Figure S2
The relationships among predictors: A, number of bullfrogs raised and number of farms for insular populations (Spearman correlation, r = 0.627, P = 0.016); B, number of bullfrogs raised and residence time for insular populations (r = 0.655, P = 0.011); C, number of bullfrogs raised and hunting pressures for insular populations (r = -0.574, P = 0.032).