2. Materials and methods
(a) Capture-recapture analysis
We uniquely marked and recaptured juvenile and adult Rana luteiventris from 21 ponds in two replicate basins, Keeler Creek (9 ponds) and Marten Creek (12 ponds), in northwestern Montana, USA (figure 1a). Ponds were separated by a maximum straight-line distance of approximately 7km in each of these basins, which are sixth code hydrologic units (Seaber et al. 1984). Most of the ponds used by R. luteiventris in Keeler Creek and Marten Creek are beaver ponds adjacent to the creeks and connected to them by small inlet and outlet streams. Frogs were caught using dip-nets during capture sessions of approximately three weeks in July and August of each year for four consecutive years starting in 2000. We made a total of 15008 captures of 10443 uniquely marked frogs during these 4 years.
We marked frogs by clipping a unique combination of 3-7 toes using an alphanumeric coding system (Waichman 1992; Donnelly et al. 1994). Thumbs were not cut because they are used by males for clasping females during breeding. We tested whether there was an effect of toe-clipping on return rate using logistic regression with the number of toes clipped and the year first marked as independent variables (Parris & McCarthy 2001). The regression coefficient for the number of toes clipped was not significant for Keeler Creek (b=-0.075, n=2563, p=0.407), but was significant for Marten Creek (b=-0.206, n=7879, p
Movement distributions were compared among stages, sexes and basins using Kolmogorov-Smirnov tests (Sokal & Rohlf 1981). Upstream or downstream bias in movement was examined by testing whether movement distributions were significantly skewed (Zar 1984). Site-specific capture histories were then used to estimate annual stage-specific movement probabilities between the lower and upper group of ponds in each basin using multistate capture-recapture analysis. Basins were divided into lower and upper groups of ponds at the elevational midpoint between the lowest and highest pond in each basin. In Keeler Creek, the upper group was pond A and the lower group comprised ponds B-I (figure 1a(i)). In Marten Creek, the upper group included ponds A-D and ponds E-L were considered the lower group (figure 1a(ii)).
We analysed capture-recapture models with stage-, annual- and site-specific variation in movement, survival and capture probabilities in the program MARK (White & Burnham 1999). A step-down modelling approach (Lebreton et al. 1992) was used to reduce sources of variation in survival and capture probabilities and then test hypotheses about variation in movement probabilities. Sixty-four models were analysed to examine variation in survival and capture probabilities, and 16 were used to analyse variation in movement probabilities in each basin. Akaike's information criterion adjusted for sample size (AICc) was used to identify the best models in terms of a trade-off between parsimony and fit to the data. Because no generally agreed upon method exists for independently testing the fit of multistate models, we followed the recommendation of Cooch & White (2001) to increase the variance inflation factor from one to assess confidence in the best model. Increasing favoured models with fewer parameters, as expected, but did not qualitatively change our finding that juvenile dispersal rates are high in both Keeler and Marten Creeks. (The best-supported capture-recapture models are found in tables 1 and 2 in Electronic Appendix A.)
(b) Microsatellite analysis
We also analysed genetic variation in five ponds from Keeler Creek (ponds A, D, F, H and I) and six ponds from Marten Creek (ponds B, C, E, G, H and K) at six microsatellite loci to estimate gene flow (figure 1a). These ponds were chosen because they supported the largest numbers of breeding adults. We genotyped a total of 312 adult frogs (mean=28 frogs per pond) sampled during spring breeding seasons. Primer sequences, DNA extraction methods, microsatellite DNA amplification conditions and Hardy-Weinberg proportion and gametic disequilibrium analyses are found in Funk et al. (2005). Fst averaged over loci was estimated using FSTAT v. 1.2 (Goudet 1995). The five ponds sampled in Keeler Creek are equivalent to ponds 1-5, and the six ponds sampled in Marten Creek are equivalent to ponds 7-12 in Funk et al. (2005), respectively.