#### Mitochondrial sequencing

Sequences of the *CO1 *mitochondrial gene fragment in five
individuals across islands showed almost no variation, with two
individuals (one Santa Cruz highland and one Floreana highland) having
an identical single nucleotide substitution (T-G). This supports the
existence of one sampled species across the three islands.

#### Genetic diversity and differentiation

Probability of identity (PI) analyses showed that the microsatellite
loci had sufficient power and resolution for the analyses. The unbiased
PI value was 1.333^{-06}, and the PI for sibs was 2.610^{-3}.
This equates to one individual in approximately 751 880 having a
non-unique genotype where individuals are unrelated (unbiased), and one
individual in approximately 383 individuals having a non-unique
genotype if all individuals are siblings.

The total number of alleles observed at each locus was as follows;
Pd1 = 4; Pd2 = 3; Pd4 = 4; Pd6 = 5; Pd7 = 3; Pd8 = 4; Pd9 = 3; Pd10 = 3
(Table 2). There was significant genotypic differentiation across the three islands (Fisher's Exact method: *X*^{2 }=
72.75; df = 16; P < 0.001). Mean allelic richness across loci was
almost identical on each island (Santa Cruz: 3.50; Floreana 3.63;
Isabela: 3.5) and the range of observed heterozygosity across loci was
also similar (Santa Cruz: 0.45–0.70; Floreana: 0.45–0.73; Isabela:
0.44–0.89). The number and size of alleles from each island population
were the same at each locus with two exceptions: there was a unique
allele at locus Pd6 on Isabela (allele frequency = 0.055), and at locus
Pd7 on Floreana (allele frequency = 0.012), which were each detected
only in a single individual. Pairwise *F*st analysis showed low, but significant levels of genetic differentiation between Santa Cruz and Floreana (*F*st = 0.02, P < 0.02) Isabela and Floreana (*F*st = 0.04, P < 0.02), but not between Santa Cruz and Isabela (*F*st
= 0.01, P > 0.1). The low genetic differentiation found between
islands was reflected in an AMOVA, which showed that just 2% of the
molecular variance was attributable to variation among islands (sum of
squares (SS) = 14.23; df = 2; variance components (V) = 0.052), 4%
among individuals (SS = 413.06; df = 155; V = 0.133), and 94% within
individuals (SS = 385.5; df = 158; V = 2.44).

**Table 2.** Allele frequencies for eight microsatellite loci in *P. downsi *within two genetic clusters.

#### Bottleneck analysis

Combining individual from all islands (n = 158), a clear excess of
heterozygosity (He) relative to the equilibrium heterozygosity (H_{eq})
was observed, indicative of a population bottleneck under the TPM model
(Wilcoxon sign-rank test; P < 0.01) and under the SMM model (P <
0.01). A mode-shift distortion in the distribution of allele
frequencies was evident (Figure 2).

**Figure 2.** **Distribution of allele frequencies indicating a mode-shift**.
Bars represent the proportion of alleles found in each allele frequency
class. Deviation from an L-shaped distribution is indicative of a
mode-shift in allele frequency due to a recent genetic bottleneck.

#### Bayesian clustering analysis

Individual-based cluster analysis using STRUCTURE[48] did not detect any genetic structuring in *P. downsi *collected across the three islands (Figure 3a), with individual assignment being evenly proportioned across variable numbers of *k*.
This implies high levels of inter-island ancestry brought about by
frequent dispersal and subsequent gene flow across the three islands
sampled. However, when incorporating geographic coordinates of sampling
locations into Bayesian analyses using GENELAND[49], two distinct genetic clusters were consistently found across runs (Figures 3b and 4).
The first cluster includes all individuals sampled from Santa Cruz and
Isabela Islands (n = 62), while the second cluster includes all
individuals sampled from Floreana Island (n = 76). Assignment
probabilities were between 0.98 and 1.0 across all individuals.

**Figure 3.** **Estimated number of populations from STRUCTURE (a) and GENELAND (b) analyses**. (a) Mean ( ± SD) probabilities of the data (LnPr [X|*k*]) over three replicate STRUCTURE runs plotted as a function of the putative number of clusters (*k*); (b) Posterior density distribution of the number of clusters estimated from GENELAND analysis in three replicates.

**Figure 4.** **Genetic assignment of ***P. downsi *individuals across three islands using Bayesian clustering analysis.
Two genetic clusters are identified: (a) including all individuals from
Santa Cruz (n = 62) (bottom left) and Isabela (n = 9) (centre top), and
(b) all individuals from Floreana Island (n = 76) (bottom right). Black
dots represent independent geographic sampling points (i.e. location of
bird nests). Note that two geographic sampling points on Isabela Island
were within 5 m of each other and are not distinguishable.

#### Genetic diversity and differentiation among inferred clusters

The two clusters identified by GENELAND displayed
comparable genetic diversity with regard to allelic richness and
differed slightly in heterozygosity across loci (Tables 2 and 3). Although two clusters were detected, measures of genetic differentiation (*F*st) between them demonstrated the low divergence between individuals on Floreana Island and those on Santa Cruz and Isabela (*F*st
= 0.024; 95% Confidence Interval (CI) = 0.014 – 0.034; P < 0.05).
Tests of departure from HW equilibrium showed no significant deviation
in either of the two clusters across all loci.

**Table 3.** Genetic variation at the eight microsatellite loci for the two *P. downsi *populations inferred from cluster analysis in GENELAND.