FIGURE 3 Comparison of the percentage of correctly assigned
samples (A, D), false negatives (B, D) and false positives (C, E) using
lake–specific GB-RFLPs (A–C) and species-specific GB-RFLPs. On the
left side of the plots are always estimates based on bootstrapping data
using allele frequencies from a previously generated genomic dataset are
on the left, data from GB-RFLP analyses on the right. Color-code in D–E
indicates the populations of Crater Lake Apoyo (red), Xiloá (blue) and
the great lake species A. citrinellus and A. labiatus(grey).
Most markers showed similar accuracy as in the bootstrapping dataset
that was based on allele frequencies in the respective lakes with 1/18
markers performing much worse (>20% fewer correctly
assigned samples; Fig. S3R) in the RT-RFLP assay than in the
bootstrapping dataset (Fig. 3A–C). We also tested whether a combination
of two markers would improve accuracy. For two populations, Great Lake
Nicaragua (69% instead of 62.5%) and Crater Lake Masaya (100% instead
of 94%), we found a marginal improvement of correct assignments (Table
S4).
TABLE 1 Tested RFLP markers, their location in the reference
genome, used restriction enzyme, quality of the marker, and correctly
assigned individuals (in %). Quality was assessed based on a
combination of the predicted and tested number of correctly assigned
specimens (++++: >99%, +++: >95%, ++:
>90%, +: >80%, –: <80%). Ingroup
means “within test population”, outgroup “not within test
population”. For lake markers (above line) the outgroup contains all
samples, for species markers (below line) only sympatric species within
the same respective crater lake.