Comparative studies of invasive species in human-inhabited versus truly uninhabited habitats, particularly on their genetic structure, remain scarce. Sub-Antarctic islands provide an ideal system to study invasions in such contrasting environments as they represent semi-pristine conditions in highly remote areas that are accessible only through a small number of introduction routes. Here we studied the invasion genetics of annual bluegrass Poa annua on the Prince Edward Islands (PEI) that include the inhabited Marion Island and the uninhabited Prince Edward Island. We analysed variation at nuclear microsatellite loci and performed flow cytometry analyses to compare the genetic diversity and structure of this widespread invasive grass. We also used ecological niche models to estimate currently suitable habitat in these islands. We found high levels of genetic diversity and evidence for extensive admixture between genetically distinct groups of P. annua on Marion Island. In contrast, Prince Edward Island populations showed low levels of genetic diversity and no apparent admixture. Higher genetic diversity was apparent at the human entry points and around human settlements on both islands, suggesting that these areas received multiple introductions and acted as both initial introduction sites and secondary sources for invasive populations within the archipelago. Over 70 years of continuous human activity on Marion Island have led to the invasive spread of this species around human settlements and along footpaths, facilitating ongoing gene flow among geographically separated populations. In contrast, this was not the case for Prince Edward Island. The high levels of genetic variation, admixture, and habitat suitability in invasive P. annua facilitated by human actions, may increase the adaptive potential of the species, which could further enhance the species’ invasiveness.

Microsatellite-containing sequences were isolated by Ecogenics GmbH (Balgach, Switzerland). Size selected fragments from P. annua genomic DNA were enriched for nuclear microsatellite repeats using magnetic streptavidin beads and biotin-labelled tri- and tetra-mer repeat oligonucleotides. The microsatellite-enriched library was sequenced on a Roche 454 platform using the Roche GS FLX Titanium technology (Roche Diagnostics Corporation). This resulted in 861 reads containing microsatellite sequences with at least six tri- or tetra-nucleotide repeat units. Primers were designed for 24 of these loci, that showed both amplification and polymorphism. Of these, 12 loci were discarded due to amplification failure in most samples or poor electrophoretic profiles. Primers for the  remaining 12 microsatellites are shown in Table S1.

Genomic DNA was extracted from all samples using a modified cetyltrimethylammonium bromide (CTAB) method (Doyle & Doyle, 1987) with the addition of 0.2 M sodium sulphite to the extraction and wash buffers. DNA quality and quantity was measured using a Nanodrop spectrophotometer (Infinite 200 PRO NanoQuant, Tecan Group Ltd, Männedorf, Switzerland), and all DNA samples were diluted to 10 ng/μL⁻¹ prior to PCR amplification and stored at −80°C until further use. Amplification of the 12 retained nuclear microsatellites was performed in two multiplex PCR reactions (Table S1). All PCR reactions were carried out in 15 μL reaction volumes containing 1.5 μL template DNA (20 ng/μL, 7.5 μL KAPA2G Fast Multiplex Mix (Kapa Biosystems, Cape Town, South Africa), 1.5 μL primer mix (2μM), and 4.5 μL distilled H₂O. Samples were amplified using the following PCR conditions: 3 min of denaturation at 95°C, 30 cycles of 15 sec of denaturation at 95°C, 30 sec multiplex-specific annealing (Table S1), 25 sec of elongation at 72°C, and a final extension for 10 min at 72°C. Each 96‐well PCR plate contained 93 samples plus two randomly selected technical replicates and one negative control (H₂O). Gel capillary electrophoretic separation of amplified fragments was carried out at the Central Analytical Facility, Stellenbosch University (Stellenbosch, South Africa). All microsatellites were scored using GeneMarker software (version 2.6.4; SoftGenetics LLC, State College, Pennsylvania, USA) with the LIZ 500 size standard. We applied semi‐automatic genotype scoring for each allele, with visual inspection of each sample, following Dewoody, Nason, and Hipkkins (2006), to reduce scoring errors. After this, three more loci were eliminated (Poa5, Poa6 and Poa12) because of low levels of variation or band stuttering.

We evaluated data quality by testing for the rate of meiotic error, the presence of null alleles and homoplasy between isoloci. For this, we analysed genotypes under the assumption of random segregation and assigned alleles to isoloci in POLYSAT (Clark & Schreier 2010). The algorithm processDatasetAllo indicated significant positive correlations between alleles at locus Poa1. However, we ruled out the possibility of scoring errors at this locus because the positively correlated alleles did not have similar amplicon sizes (tetranucleotide motif). After scoring of genotypes, some loci (Poa 1, 3, 8, 9 and 11 in Table S1) were split into two isoloci, resulting in a final dataset of 14 loci.

Missing data: -9