Results
Genetics
All loci were highly polymorphic, ranging from 21 alleles atGbr 12 to 32 at Gvu 05. Population meanH o ranged from 0.64 to 0.83 (mean 0.76) andH e from 0.57 to 0.82 (mean 0.75; Table 1). Allelic richness across all loci tended to be lower in lake-developing populations, ranging from 2.8 in Lake Christabel to 4.9 in Lake Wanaka, while coastal populations ranged from 3.9 to 6.3. Departures from Hardy-Weinberg equilibrium were found in 49 out of 240 locus-population combinations before stepwise Bonferroni correction. As 11 of these occurred at Gbr 130, and 21 occurred at Gbr 140, these loci were excluded. Bonferroni corrections were then applied with only 1 of 192 remaining out of equilibrium after stepwise Bonferroni corrections (p < 0.0003). No locus was affected by null alleles in any sample, so all remaining loci were used in the study.
All landlocked populations exhibited a higher degree of genetic structuring on a lake by lake basis, when compared to diadromous populations (Fig. 2a), regardless of distance from coastal populations (Mantel p = 0.07; see Supplementary Tables A1, A2 for full results). Landlocked systems had pairwise F STvalues of approximately 0.03–0.06 when compared to immediate downstream sites less than 5 km away (Fig. 2a,b). Lakes located in the same catchments (L. Rotoiti, L. Rotoroa), had lowerF ST values, resulting in a significant relationship with increasing distance (Mantel p = 0.04). Comparisons among coastal streams tend to have lowerF ST values, but significantly increase with geographic distance (Mantel, p = 0.04). High structuring levels were present in these comparisons even when geographic distance was relatively small and sites were located in the same catchment, such as an F ST value of 0.05 for Smokey and Stoney Creeks, 18 km apart. Structuring within the respective tributaries of L. Wanaka and L. Wakatipu was relatively low, comparable to that of among coastal stream comparisons (F ST = 0.01–0.03), and no structure was shown by STRUCTURE analysis (Supplementary Fig. A2; Supplementary Table A3).
Patterns present in F ST scores were corroborated by patterns identified using STRUCTURE (Fig. 3). STRUCTURE initially split populations in the tributaries of east coast lakes (Wanaka and Wakatipu) from all sites at K = 2. At K = 3, the tributaries of west coast lakes were split from coastal west coast sites. Tributaries of L. Paringa and L. Moeraki split at K = 4 with tributaries of Lake Cristabel beginning to split at K = 5. Tributaries of L. Wanaka and L. Wakatipu are clearly split at K = 6, which was determined as the most likely population number using Evanno’s method. At K = 9 both tributaries of both L. Moeraki and L. Cristabel are clearly split form the other west coast lakes (Fig. 3). Within lake STRUCTURE results did not show distinct population clustering for tributaries of L. Wakatipu or L. Wanaka (Supplementary material, Fig. A2).
Otoliths
The larval development period of all otoliths from juvenile G. brevipinnis collected from lake tributaries lacked the high Sr:Ca ratios indicative of a diadromous life-history, whereas those from coastal populations all displayed high Sr:Ca and low Ba:Ca. That is, all individuals in lake tributaries had developed in freshwater, while all those caught from coastal streams had developed in the ocean. Linear discriminant analysis suggested that (in order of importance): Rb, Ba, Sr, Mg, Cu, Ni were useful in stock discrimination, with linear discriminant (LD1) explaining > 67% of variation in all systems (Table 2).
At the largest (system) spatial scale, evidence of population structuring was clear, with coastal, Wanaka, and Wakatipu samples all forming distinct clusters (Fig. 4a), with reclassification to system of capture 96% successful (Table 2). At a finer scale (the a prioriregions within each lake and along the coast), regional clusters were generally well defined (Fig. 4b), and reclassification success rates were again high: 94% for coastal regions, 86% within Wanaka and 71% within Wakatipu (Table 2). At the finest spatial scale, individual sampling sites still show distinct, though less well-defined, clustering (Fig. 4b), and reclassification success rates for coastal, Wanaka, and Wakatipu sites were 62.5%, 78.6%, and 60% respectively (Table 2). Reclassification success was reflected in the LDA plots, with clear clustering present at high reclassification levels (e.g. Wanaka), and greater overlap occurring with lower reclassification levels (e.g. Wakatipu) (Fig. 3b).
Trawling
Trawling in the lakes collected a total of 1,991 larvae, with 95% being collected in river plumes, 2% in nearshore areas and 3% in offshore areas. All large river plumes had significantly higher densities of larvae than trawls collected from offshore or nearshore sites. There were no significant density differences among small river sites, although the highest densities were recorded in the plumes associated with Buckler Burn and Albert Burn (Fig. 5.) Statistical outputs relating to larval trawling are available in Supplementary material (Appendix 1, Table A4).