Introduction

Species invasions are occurring more frequently as globalization increases (Hulme 2009; McGeoch et al. 2010). The economic impact of these invasions is estimated to exceed $120 billion annually (Pimentel et al. 2005) and the ecological ramifications of species invasions, which occur at ecosystem, community, and population levels, are perhaps even more staggering (Wilcove et al. 1998; Mooney 2005; Raizada et al. 2008; Vila et al. 2011; Simberloff 2013; Hejda et al. 2017). Invasive species can disrupt ecosystem services (Charles and Dukes 2008; Walsh et al. 2016) and lower biodiversity through predation, competition, parasitism, and disease (Reaser et al. 2007; Rowles and O’Dowd 2007; Vilcinskas et al. 2013). The loss of biodiversity as a consequence of invasive species has been especially well documented on islands, posing a particular threat to endemic species that are unique to these ecosystems (Sax and Gaines 2008; Tershy et al. 2015; Doherty et al. 2016).

Management of target populations of invasive species on islands can alleviate negative effects on native ecosystems (Vreysen et al. 2000; Causton et al. 2005; Cruz et al. 2005; Jones et al. 2016). However, defining target populations based on faulty presumptions of population structure can mislead management efforts and reduce their efficacy. The identification of “eradication units” focuses instead on the eradication or management of genetically isolated clusters of populations (Robertson and Gemmell 2004), such that the risk of reinvasion is minimized. This approach has increased the success of efforts aimed at eradicating rat populations in several insular systems (Abdelkrim et al. 2005; Savidge et al. 2012; Ragionieri et al. 2013). For example, Abdelkrim et al. (2005) used a population genetics approach to show that Norway rats (Rattus norvegicus) are able to regularly migrate between some islands in Archipelagos off the Brittany Coast of France, such that eradication of a “population” on a single island was likely to be highly ineffective. Importantly, this study demonstrated the necessity of understanding population connectivity prior to implementing management strategies.

Philornis downsi is a parasitic nest fly of birds and is invasive to the Galápagos Islands, Ecuador (Kleindorfer and Dudaniec 2016; Fessl et al. 2018; McNew and Clayton 2018). Adult females lay eggs in the nests of 22 of the 28 land bird species that nest there, including 11 of the 17 species of Darwin’s finches (Fessl et al. 2018). The fly larvae are semi-hematophagous parasites of nestlings and adult female birds (Fessl et al. 2006; Cimadom et al. 2016; Common et al. 2019). Nestling mortality can be extremely high but varies by host species and rainfall patterns (Dudaniec et al. 2007; Koop et al. 2013; Cimadom et al. 2014; Heimpel et al. 2017). Philornis downsi has been implicated in the decline of several endemic Galápagos birds, including the Mangrove Finch (Cactospiza heliobates), the Medium Tree-Finch (Camarhynchus pauper), and the Warbler Finch (Certhidia olivacea) (Dvorak et al. 2004; Grant et al. 2005; Cunninghame et al. 2017; Peters and Kleindorfer 2018; Bulgarella et al. 2019). While first recorded in the Galápagos Islands in 1964, P. downsi was not observed in the nests of land birds until 1997 (Fessl and Tebbich 2002). Currently, the fly has been observed on 15 of the 17 major islands sampled within the archipelago (Wiedenfeld et al. 2007; Jimenez-Uzcategui et al. 2011). To date, the source of colonization remains unknown though P. downsi has been documented in Trinidad and Tobago (Dodge and Aitken 1968), Argentina (Silvestri et al. 2010), Brazil (de Carvalho and Couri 1999), and mainland Ecuador (Bulgarella et al. 2015). The Galápagos National Park and the Charles Darwin Foundation have prioritized management efforts for P. downsi in the Galápagos, including exploration of biological control and sterile insect techniques (Bulgarella et al. 2017; Fessl et al. 2018; Boulton et al. 2019). However, these programs can greatly benefit from knowledge of the population genetics of the targeted species.

A major goal of this study is to describe population structure in the native and invasive ranges of this parasite. Very little genetic work has been done on this system, largely due to the inherent difficulties of collecting samples. However, an important study by Dudaniec et al. (2008) used mitochondrial and microsatellite data to assess population structure among three island populations of P. downsi in the Galápagos. This study confirmed that a single species of P. downsi was present on all of the islands tested, as was recently supported by Common et al. (2020) who found no morphological differences between island populations of flies. Importantly, Dudaniec et al. (2008) showed that there were low, but significant levels of genetic differentiation between some of the islands, though clustering analyses did not differentiate between island populations. Furthermore, this study found evidence of a recent genetic bottleneck, consistent with the hypothesis that P. downsi invaded the Galápagos Islands recently, which is also supported by historical considerations (Kleindorfer and Sulloway 2016; Fessl et al. 2018). Our study builds on this work by examining current population structure among island and mainland populations of P. downsi. Here, we use a genomics approach including over 2000 genome-wide SNP markers obtained from restriction-site associated DNA sequencing (RAD-seq) and a phylogenetics approach using two gene sequences to further estimate population structure, phylogenetic history, and possible connectivity patterns for P. downsi collected from the Galápagos Islands and mainland Ecuador.

Methods

A genomics approach: RAD sequence processing and population genetics analyses

Adult P. downsi were collected from five islands within the Galápagos archipelago (Santa Cruz, San Cristobal, Floreana, Isabela, Santiago) between March 2015 and April 2017 (Fig. 1). On Floreana, Isabela, and Santiago, flies were sampled from several locations within each island. For all island sampling, adult flies were caught using McPhail traps (BioQuip Products, California, USA), baited with fermented papaya juice, that were set out for 3–5 days as in Causton et al. (2019). Following collection, flies were stored in 70% ethanol until being transported to the United States for further processing. In addition, P. downsi were collected at a site in mainland Ecuador. This was the Reserva Ecológica Loma Alta in the Province of Santa Elena, Ecuador (hereafter referred to as Loma Alta), where 38 P. downsi pupae were collected from 5 nests between April 5 and 25, 2017; each were identified as P. downsi based upon pupal characteristics (Couri 1999). These pupae were transported to a U.S. government-approved Quarantine Facility at the University of Minnesota under Ecuadorian export and United States import permits, where they were reared to adulthood (as in Bulgarella et al. 2015). These adults were used in laboratory studies prior to being preserved in 70% ethanol. DNA was extracted using DNeasy Blood and Tissue Kit (Qiagen, Valencia, California, USA) on whole fly samples. Samples that returned low yields of DNA (< 10 ng/μl) were not submitted for sequencing and were removed from the study.

Fig. 1
figure 1

Map of Ecuador, South America and the Galápagos Islands showing sampling locations of P. downsi. For the genomics study (solid circles), samples were collected from a single location from Santa Cruz Island, San Cristobal Island, and Loma Alta, mainland. On three islands, flies were collected from multiple locations and combined for analyses (Floreana Island, 2 locations; Isabela Island, 2 locations; Santiago Island, 2 locations). For the phylogenetics study (open circles), samples were collected from 2 locations on the mainland and 2 locations on Santa Cruz Island. Hashed circles represent locations where samples were collected for both the genomics study and the phylogenetics study

Full size image

In total, 89 samples were submitted for sequencing. These included P. downsi samples from Santiago (n = 3), Santa Cruz (n = 14), San Cristobal (n = 5), Floreana (n = 8), Isabela (n = 25), and mainland Ecuador, Loma Alta (n = 34) (Table S1). DNA extractions were normalized to 10 ng/μl, loaded individually in 100 μl quantities on a 96-well plate and submitted for sequencing. Single-digest restriction site associated DNA sequencing (RAD-seq) and library preparation were performed by Floragenex (Eugene, Oregon, USA) using the protocol by Etter et al. (2011). Genomic DNA was first digested with the restriction enzyme SbfI (New England Biolabs, Inc., Ipswich, Massachusetts, USA) followed by ligation of a P1 adapter. Multiplexing barcodes were adhered to each sample followed by pooling, shearing and size selection of samples, ligation of a P2 adapter and PCR enrichment. Finally, pooled samples were sequenced on a single lane of a HiSeq 2000 Illumina sequencer (single end 100 bp sequencing).

The software program Stacks v 2.41 was used to process the raw data, call variants, and perform some population-level analyses (Catchen et al. 2013). Raw sequences were demultiplexed, barcodes removed, and low-quality reads were filtered using process_radtags (Catchen et al. 2013). All samples had a sequencing read depth > 20 × (as estimated with the denovo_map.pl program in Stacks); therefore, no samples needed to be removed to avoid misrepresentation of heterozygote diversity (Rochette and Catchen 2017).

There currently is no reference genome for P. downsi; therefore, we used the wrapper script denovo_map.pl within Stacks to execute a pipeline of commands. The wrapper implemented ustacks (builds loci de novo), cstacks (catalogs and matches loci across samples), sstacks (matches samples against the catalog), tsv2bam (transposes data such that it is organized by RAD locus), and gstacks (calls SNPs for each locus) (Catchen et al. 2013). A subset of individuals were analyzed by varying the parameters for maximum number of nucleotide mismatches allowed between stacks within (M) and between (n) individuals (M = n, ranging from 1 to 9) in order to assemble putative alleles and loci, respectively (Paris et al. 2017; Rochette and Catchen 2017). Optimal conditions were determined as the minimum value of M and n that resulted in an asymptotic number of loci being shared by \(\ge\) 80% of the subset individuals. Thus, for final analyses of the full dataset, a maximum of two nucleotide mismatches were allowed between stacks within (M = 2) and between (n = 2) individuals, and the number of reads required to form a stack was set at 3 (M = 3) (Rochette and Catchen 2017).

The populations function in Stacks was used to further filter the data such that the final data set contained loci that were in \(\ge\) 60% of individuals overall and \(\ge\) 80% of individuals within each population and could be found in 3 or more populations. Additionally, the minimum minor allele frequency was set to 0.05 and the maximum observed heterozygosity required to process a SNP was set at 0.70 (Rochette and Catchen 2017). Analyses were restricted to a single random SNP per locus, which can act to reduce linkage disequilibrium. Output files from Stacks were converted using PGDSpider v. 2.1.1.5 (Lischer and Excoffier 2012), then analyzed with BayeScan v.2.1 (Foll and Gaggiotti 2008) to identify loci under selection. Loci with an FDR > 0.1 were removed from further analyses (n = 4 loci). All default options were used, except a prior odds value was set at 100 (default is 10) to avoid an excess of false positives, especially with limited sampling of individuals in some populations (Foll and Gaggiotti 2008). In addition, VCFtools (Danecek et al. 2011) was used to test for SNPs deviating from Hardy–Weinberg Equilibrium in each population. Following all of these filtering steps, 2222 of 328,992 loci were retained for downstream analyses. We explored the data using both less restrictive and more restrictive filtering parameters, none of which drastically altered the main conclusions of our study. Estimates of genetic diversity, including observed heterozygosity (HO) and expected heterozygosity (HE), were calculated with the hierfstat package (Goudet 2005) in R (version 3.4) (R Development Core Team 2013). The number of private alleles per population was calculated with the poppr package in R (Kamvar et al. 2014). Pairwise FST distances [with associated p-values following correction by false discovery rate (FDR)] and measures of variance within and between populations (AMOVA) was calculated using GENODIVE (version 3.0) (Meirmans 2020).

The level of genetic differentiation relative to the geographic distance between populations (i.e. isolation by distance) was tested by comparing FST values to linear geographic distances (km) estimated using Google Earth. Significance was determined using Mantel tests performed in R with the ecodist package (Goslee and Urban 2007), with 10,000 permutations.

The number of genetically distinct clusters needed to describe the data was first estimated using the K-means method generated in STRUCTURE v.2.3.4 (Evanno et al. 2005). Five independent runs (10,000 length burn-in, 100,000 iterations, admixture allowed) were conducted for each K value (K = 1–6). The CLUMPAK online server (https://clumpak.tau.ac.il/contact.html) was used to average iterative runs of K and determine an optimal estimate of the number of distinct genetic clusters using the delta K method (Evanno et al. 2005; Kopelman et al. 2015; Meirmans 2015, 2019; Janes et al. 2017). We complemented this analysis by performing discriminant analysis of principal components (DAPC), which optimizes variance between groups while minimizing variance within groups to identify populations. These analyses were done using the dapc function (Jombart et al. 2010) within the adegenet package in R (Jombart 2008). DAPC performs a discriminant analysis on principal components retained through the cross-validation step implemented using the xvalDapc function.

To address limitations in sample size at three of our island sites, we ran all of these same clustering analyses with a limited data set that included only individuals from Santa Cruz Island (BG), Isabela Island (IS), and the mainland (LA). We hereafter refer to these data sets as the full and limited data sets, respectively. We ran all of the same filtration steps as described above, which resulted in 2193 variable loci being used in the limited dataset, after 29 loci that were non-variant among these 3 populations were removed.

Fastq sequence files have been deposited in the NCBI Short-Read Archive (SRA) database (BioProject Number PRJNA614982; Table S1).

A phylogenetics approach: gene sequencing and phylogenetics analyses

In a separate study, we extracted, amplified, and sequenced 1005 bp of the mitochondrial cytochrome oxidase 1 gene (CO1) and 442 bp of the nuclear ITS2 rDNA gene region (ITS2) for 31 P. downsi flies collected during the 2014/2015 season from two sites in mainland Ecuador (Bosque Protector Cerro Blanco, Guayas Province, n = 11, and Reserva Ecológica Loma Alta, Santa Elena Province, n = 1) and from two sites in Santa Cruz Island, Galápagos (Los Gemelos, n = 14, and Playa El Garrapatero, n = 5) (Fig. 1; Table S1). We used the DNeasy Blood and Tissue Extraction kit (Qiagen, Valencia, California, USA) for DNA extraction and samples were amplified using the forward primer M202 (C1-J-1751, Simon et al. 1994) paired with the reverse primer M70 (UEA10, Lunt et al. 1996) for CO1. For ITS2, we used the primer pair developed for Philornis spp. by Monje et al. (2013). PCR amplifications and Sanger sequencing followed standard protocols (Bulgarella et al. 2015). Sequences from opposite strands were reconciled and verified for accuracy using Sequencher v.5.2.4 (Gene Codes, Ann Arbor, Michigan, USA). We aligned the sequences for each gene using the Geneious v.10.2.3 alignment algorithm (https://www.geneious.com) with the default parameter settings. We concatenated the CO1 and ITS2 data, partitioned by gene region, to obtain a combined dataset of 1529 bp. We conducted maximum likelihood and Bayesian analyses on the combined dataset using a single, best-fit model of sequence evolution as determined by AIC (Akaike 1973) implemented in Modeltest v.3.7 (Posada and Crandall 1998). For maximum likelihood analyses, we completed 1000 full heuristic tree searches in the PAUP*4.0b10 (Swofford 2002) plug-in for Geneious, each with random addition of taxa. Statistical support for nodes was evaluated with nonparametric bootstrapping (Felsenstein 1985) implementing 2000 replicates in PAUP. Trees were rooted with Philornis falsificus as a paraphyletic group from the DNA of a single P. falsificus individual collected at Cerro Blanco in May, 2015 (Bulgarella et al. 2017).

Clade probabilities for the combined 1529 bp dataset were also obtained from the posterior distribution using the MrBayes v.3.2.6 (Huelsenbeck and Ronquist 2001) plug-in for Geneious. Bayesian analyses were replicated twice, each with 4 Markov chains of 2 million generations. Trees were sampled every 2500 generations, of which the first 150,000 generations were discarded as burn-in. Sequences have been archived in GenBank (accession numbers for P. downsi are MT186607–MT186637, MT198952–MT198979, KP730049; for P. falsificus MT229984 and MT240486; Table S1).

Results

Genomics approach

In the full data set (including 6 locations and 2222 variant loci) observed heterozygosity (HO) was similar across all island and mainland populations, ranging from 0.21 to 0.24 (Table S2). Private alleles varied widely across populations and were highest in the mainland population (Loma Alta: 478 private alleles) compared to the islands (private alleles ranged from 1 to 14 per island population) (Table S2). Pairwise genetic distances (FST) between populations ranged from 0.001 to 0.282, and were highest in comparisons between mainland and island populations (Table 1). Analysis of molecular variance comparing all six populations showed that 14.8% of the molecular variance was attributable to variation among populations [FST = 0.148, standard deviation (SD) = 0.005], 14.7% was attributable to variation among individuals (FIS = 0.173, SD = 0.004), and 70.5% was attributable to variation within individuals (FIT = 0.295, SD = 0.005). There was a significant isolation-by-distance effect across all populations (Mantel, r = 0.98, p = 0.02). However, when the mainland population was excluded, there was no longer a significant signature of isolation-by-distance among island samples (Mantel, r = 0.60, p = 0.19).

Table 1 Pairwise FST values for each population included in the full dataset genomics study including 2222 variant loci
Full size table

The K-means method as applied through STRUCTURE indicated an optimal number of K = 2 genetic clusters with the full dataset (Fig. S1). Given documented issues with K = 2 being identified more often than by chance when using the delta K method (Janes et al. 2017), we show the results for K = 2–5 in Fig. 2a. With K = 2, the mainland and island populations differentiate. Then, with increasing values of K, additional genetic variation is found largely within the mainland population cluster. The DAPC method was used to complement analyses using STRUCTURE. In the first step, 100 principal components were retained (Fig. S2a), which accounted for over 98% of the variance. The Bayesian Information Criterion (BIC) was used to identify an optimal number of clusters as K = 2 (Fig. 2b, left). Then, 11 principal components were retained for the DAPC following cross-validation analysis (Fig. S2b) and 1 discriminant function. DAPC shows two main clusters (Fig. 2b, right), and the assignment plot shows that the clusters correlate with mainland versus island individuals (Fig. S2c).

Fig. 2
figure 2

Results of STRUCTURE and DAPC analyses of Philornis downsi sampled from the Galápagos Islands and mainland Ecuador. a Graphical representation of Bayesian clustering approach using STRUCTURE and CLUMPAK to graph a consensus of five iterative runs at K values 2–5, following permutations testing. Each vertical line represents a single individual. Sampling locations are denoted below each plot, representing samples from the Galápagos and mainland. Various colors within the plot represent inferred membership to a given genetic cluster. Posterior probability methods identified K = 2 as the optimal number of distinct genetic clusters within the total sample set. b Plot of the number of clusters in the DAPC analyses using Bayesian Information Criterion (BIC) to select an optimum. The “elbow joint” in the plot indicates the optimal number of clusters to select as it represents the lowest BIC value. c DAPC scatter plot of 2 main clusters based on 11 retained PCs and 1 discriminant function. As with the STRUCTURE analysis, the distinction between these clusters is based on geographic origin of the samples (i.e. mainland or islands)

Full size image

To address low sample size for 3 island populations, we performed these same clustering analyses on a limited dataset that included only locations with a sample size \(\ge\) 14 individuals [i.e. the islands of Santa Cruz, Isabela, and mainland Ecuador (Loma Alta)] with 2193 variant loci. The K-means method as applied through STRUCTURE again indicated an optimal number of K = 2 genetic clusters (Figs. S3, S4a). At K = 2, there is clear differentiation between samples from the Galápagos Islands and those from the mainland. Increasing values of K show increasing genetic variation largely within the mainland population. Eighty PCs were retained in the first step of the DAPC method (Fig. S4b, top left). The BIC was used to identify an optimal number of clusters as K = 2 (Fig. S4b top right). Following cross-validation (Fig. S4b, bottom left), 10 PCs were retained and 1 discriminant function, which resulted in 2 clusters (Fig. S4b, bottom right).

Phylogenetics approach

For the combined dataset, the best fit model of nucleotide substitution was a transition model (TIM) with variable base frequencies, variable transition rates and two transversion rates. The percent of identical sites for the 31 P. downsi flies sampled from all sites was 99.6% showing little genetic variability. Maximum likelihood analyses showed that P. downsi collected in mainland Ecuador and the Galápagos Islands form a monophyletic group with 100% bootstrap support (Fig. 3). We found the same result when we performed a Bayesian analysis. The monophyly of P. downsi was supported by a posterior probability of 1 (not shown).

Fig. 3
figure 3

Maximum likelihood tree of 31 P. downsi specimens from mainland Ecuador (Guayaquil and Loma Alta), the Galápagos Islands (Santa Cruz Island), rooted with the closely related species, Philornis falsificus collected in Guayaquil, Ecuador. Support values above branches correspond to nonparametric bootstrapping

Full size image

Discussion

We used genomic and phylogenetic approaches to assess the population structure of P. downsi sampled from its native and invasive ranges. Our primary goal was to better understand patterns of genetic differentiation between mainland and island populations of flies and between island populations of flies. We consistently found genetic differentiation between mainland and island populations using the genomics approach. Phylogenetic comparisons using CO1 and ITS2 genes showed almost no differentiation between mainland and island populations of flies, though this is likely due in part to the more conserved nature of these genetic markers. In addition, we found very little genetic differentiation between island populations of flies and no pattern of isolation by distance across island populations. Together, our results suggest that population connectivity between mainland and island populations is likely limited, but there is the potential for continued gene flow among island populations that warrants further investigation. It is important to note that in regard to the lack of differentiation found between island populations, we did not explicitly test whether this pattern is the consequence of a recent invasion or ongoing migration of flies between islands. Thus, we will discuss the consequences of either or both being true.

Multiple lines of evidence support the hypothesis that P. downsi invaded the Galápagos Islands relatively recently. The first report of this species in the Galápagos dates from 1964 (Kleindorfer and Dudaniec 2016; Fessl et al. 2018) and multiple prior insect surveys failed to report this species (Linsley and Usinger 1966; Fessl et al. 2018). The first observation of P. downsi in a finch nest was made in 1997 (Fessl et al. 2001), despite inspections of nests prior to this (Lack 1947; Grant 1999). In addition, Kleindorfer and Sulloway (2016) inspected museum specimens of Small Ground-Finches (Geospiza fuliginosa) from Floreana Island and only found evidence of beak scarring after 1960. First-instar larvae of P. downsi can feed in the naris of nestlings, leading to scarring and malformations that persist into adulthood (Kleindorfer and Sulloway 2016). Finally, Dudaniec et al. (2008) found minimal genetic differentiation between island populations and an excess of heterozygosity across all individuals using eight microsatellites markers, a pattern consistent with a very recent genetic bottleneck.

Our study found further evidence supporting the hypothesis that P. downsi recently invaded the Galápagos. We found that mainland flies clustered as a separate population from island flies across multiple genomic analyses. The mainland fly population had significantly higher levels of genetic diversity (e.g. number of private alleles) than did the Galápagos populations, which is expected if mainland Ecuador represents the native range. Phylogenetic comparisons showed almost no differentiation between mainland and island populations of flies, and this result is expected under the premise of a recent invasion, given the conserved nature of these genetic markers. Here, we chose sites on the mainland for their close proximity to the city of Guayaquil, Ecuador. Guayaquil is and has served as the port city for most of the air and boat traffic to the Galápagos, which is a suspected mechanism of introduction for P. downsi (Bulgarella et al. 2015, 2017; Fessl et al. 2018). Further sampling of mainland populations of P. downsi could alter the patterns of diversification shown by our study, though they are unlikely to refute the recent invasion hypothesis. That is to say, additional sampling on the mainland may reveal a population that is more genetically similar to island populations than what we have sampled here. Significant further study is needed to explore potential mainland source populations and resolve a complete invasion history of this parasite.

Low levels of genetic differentiation can be the result of a recent invasion in which genetic drift or selective sweeps have had minimal time to fix alleles within populations. Similarly, we expect to find low levels of differentiation when gene flow is persistent among populations. While our study was not designed to differentiate between these two non-mutually exclusive hypotheses, our results suggest that ongoing gene flow between island populations may be occurring. Most of the genetic variation observed in our study was explained at the level of individuals, consistent with analyses showing relatively low pairwise genetic differentiation between island populations. Under conditions of limited gene flow, a pattern of isolation by distance is expected to occur relatively quickly, yet we did not find evidence of this pattern among island populations. Our results, particularly those using the limited dataset, are largely consistent with Dudaniec et al. (2008), who used eight microsatellite markers in a landscape genetic analysis to show limited, but significant genetic differentiation (FST values) between Floreana Island and Santa Cruz Island, Floreana Island and Isabela Island, but not between Santa Cruz Island and Isabela Island. Similarly, cluster analyses of this same dataset did not detect structure among the three island populations (Dudaniec et al. 2008). It is important to note that sampling from several of our island populations was limited, making our conclusions about population connectivity between these other islands more speculative. A more detailed study on Floreana Island showed varying levels of genetic relatedness among flies based on habitat type (Dudaniec et al. 2010), demonstrating that additional sampling has the potential to reveal more fine scale genetic structure than was shown by our full or limited datasets. Thus, additional sampling within and among islands will facilitate more accurate estimates of gene flow within the archipelago and is highly recommended.

The mechanism by which flies may be dispersing between the islands remains unknown. Adult flies are strong fliers and may be able to move between islands on air currents. Transport via air and boat vessels is also quite likely. Philornis downsi has been documented on boats moving between islands (Lomas 2008), though very infrequently. Boat and air traffic between islands and between islands and the mainland are frequent and increasing (Toral-Granda et al. 2017). Four of the five islands we sampled are inhabited with permanent human residents (Santa Cruz, San Cristobal, Isabela, Floreana) and the fifth island (Santiago) is a popular tourist destination with multiple boats arriving daily for day-hikes. Therefore, ample opportunities exist for interisland transport of flies via boats. The introduction of P. downsi from the mainland to the islands also presumably occurred via boats or airplanes, although further details remain unknown. Biosecurity measures aimed at reducing alien species introductions to the Galápagos have been put into place relatively recently, including spraying pesticides on arriving planes and boats. While the Galápagos now relies on its own biosecurity agency, considerable investment is still needed to have a significant impact on reducing the risks of new or repeated introductions (Toral-Granda et al. 2017).

Management implications

Philornis downsi presents an imminent threat to the endemic Galápagos avifauna with modelling studies suggesting the possibility of extinction of even the more common bird species within a century (Koop et al. 2015). Treatments with insecticides, either through direct application to nests or through inoculated cotton that the birds place in nests themselves, are approaches that have been demonstrated to effectively reduce parasite abundance and improve fledging success (Koop et al. 2011; Knutie et al. 2014). However, these approaches are only a short-term solution for protecting birds in small areas and may introduce risks to birds and other non-target organisms (Causton et al. 2019). Thus, long-term integrated management techniques are under development, and consideration of aspects of P. downsi population genetics and phylogeography could greatly aid these efforts.

One such management technique involves the potential introduction of highly-specialized parasitoid wasps to control P. downsi (Bulgarella et al. 2017; Heimpel 2017; Boulton et al. 2019). Having an understanding of the location within the native range from which the Galápagos populations originated can inform such efforts by directing parasitoid collecting to appropriate mainland locations and gaining information on the coevolution of P. downsi and its parasitoids (Hufbauer and Roderick 2005; Heimpel and Mills 2017). A sterile insect program is also being considered, which involves the field-release of many lab-reared sterilized male flies with the aim that they breed with wild females, who then fail to produce offspring. These programs have been highly successful in the control of a number of other invasive fly species (Hendrichs et al. 1995; Vreysen et al. 2000; Klassen and Curtis 2005). For the sterile insect technique, an understanding of P. downsi population genetics would be crucial since the method is very sensitive to factors such as repeated introductions and multiple mating (Krafsur 2005). Finally, the development of attractants for use in fly traps is also underway (Cha et al. 2016; Mieles 2018) and an understanding of dispersal patterns—such as the possibility of migration from highland to lowland sites on a single island—would greatly enhance the interpretation of trap catch data (Causton et al. 2019).

The intention of our study was to provide insight on the population structure and connectivity of P. downsi in the Galápagos Islands and mainland Ecuador. Our results suggest that with respect to managing P. downsi, the archipelago may need to be treated as a single management unit (Robertson and Gemmell 2004), as several studies have now found little evidence of isolation among island populations of flies (Dudaniec et al. 2008). Efforts to eliminate or manage populations, irrespective of the specific technique, would need to involve simultaneous treatment of islands since efforts to control populations on a single island could be thwarted by recolonization from a neighboring island. The use of modern genetic resources to indirectly assess patterns of gene flow between populations has great potential to improve management efforts across biological systems. In the Galápagos, further sampling will allow for more detailed estimates of gene flow between the islands and perhaps, even within individual islands (Dudaniec et al. 2010). Furthermore, these same genetic tools can be used to identify the source population(s) and the most likely mechanism of invasion, allowing managers to reduce the likelihood of reinvasion from the mainland.