Immigration allows population persistence and maintains genetic diversity despite an attempted experimental extinction

Widespread fragmentation and degradation of habitats make organisms increasingly vulnerable to declines in population. Immigration is a key process potentially affecting the rescue and persistence of populations in the face of such pressures. Field research addressing severe demographic declines in the context of immigration among interconnected local populations is limited due to difficulties in detecting such demographic events and the need for long-term monitoring of populations. In a 17-subpopulation metapopulation of the butterfly, Parnassius smintheus, all adults observed in two adjacent patches were removed over eight consecutive generations. Despite this severe and long-term reduction in survival and reproduction, the targeted populations did not go extinct. Here, we use genetic data to assess the role of immigration versus in situ reproduction in allowing the persistence of these populations. We genotyped 471 samples collected from the targeted populations throughout the removal experiment at 171 single nucleotide polymorphisms. We found no reduction of genetic diversity of the targeted populations over time, but a decrease in number of loci in Hardy-Weinberg equilibrium, consistent with a high level of immigration from multiple, surrounding populations. Our results highlight the role of connectivity and movement in making metapopulations resilient to even severe and protracted localized population reductions.

Methods
The Jumpingpound Ridge metapopulation occurs within a network of 17 patches of alpine meadow habitat ranging in size from 0.2 ha to 22.7 ha. This system has been the subject of long-term population monitoring and genetic sample collection since 1995, with mark-recapture performed during the annual flight season to monitor population size and movement. The removal experiment on patches P and Q was conducted from 2001 to 2008 to investigate the effects of severe and continuous population reduction in patches P and Q on the population dynamics in surrounding, neighbouring patches . Through the course of the experiment, patches P and Q were surveyed every one to three days during the flight season each year, and all observed butterflies were captured by hand netting and removed from the site. In total, 4,830 butterflies were removed from P and Q over the eight-year period. The removed individuals were labelled with the date and location of capture, and then stored, dried, and pinned in a collection housed at the University of Cincinnati.

We sampled from this collection of pinned specimens from patches P and Q, removing one leg per individual across each year of the removal experiment. For years in which the number of captured individuals in a patch was lower than 30, we sampled all available individuals. Otherwise, we sampled between 30-60 individuals per patch per year.  Samples from 2001 represent the basal or initial state of the populations, as they would not yet have responded to the experimental manipulation. Additionally, leg tissues from 42 frozen whole butterflies, removed during the 2008 flight season were also used; in combination with pinned samples from 2008 these represent the final year of the experimental removals. In total, we sampled a total of 471 individuals from patches P and Q across eight years.

Our SNP panel contained 171 unlinked SNP loci and was developed based on a double digest restriction site associated DNA sequencing (ddRADseq) dataset for the species [44]. Briefly, multiple ddRADseq libraries were constructed from 80-100 individuals each (for a total of 501 individuals, sampled from Jumpingpound Ridge and the surrounding region and not overlapping with the individuals removed from patches P and Q) by digesting genomic DNA using the restriction enzymes NlaIII and EcoRI, then selecting and amplifying fragments between 200 and 500 bp and sequencing the libraries on an Illumina HiSeq 2500 sequencer. We identified SNPs from the resulting sequences using the program STACKS with the following parameters: minimum stack depth (m) = 3, mismatches allowed between putative catalog loci (n) = 3, mismatches allowed between putative alleles (M) = 2, mismatches allowed to align secondary reads (N) = 4, and a maximum allowed missing data of 50%. This resulted in a ddRADSeq dataset of 8814 SNP loci. From the ddRADSeq sequences, we filtered SNPs with at least 40 bp upstream and downstream flanking sequence to allow space for designing primers for the iPlex Gold MASSarray assay, and used the software PLINK to identify and remove statistically linked loci. To identify SNPs that might putatively be functional, we aligned the filtered ddRADSeq sequences to a transcriptome of thorax tissue of adult butterflies captured during flight using Magic-BLAST. We defined fragments with at least a 90% sequence match as putatively expressed loci. Potential functions of these putatively expressed loci were identified using Trinotate, and 35 loci were chosen to reflect a variety of cellular functions such as metabolism, transcription regulation, transmembrane proteins, protein modification, and insect development. An additional 130 loci having both less than 10% sequence match to the transcriptome and a greater than 90% sequence match to a shotgun sequenced genome were chosen at random to include in the final SNP panel. Finally, an additional six non-synonymous SNPs were included from the coding region of phosphoglucose isomerase (Pgi) , a gene associated with dispersal and flight metabolism in other insects, including the Glanville fritillary butterfly.

Of the 171 SNPs targeted for genotyping, assays for two SNPs failed altogether, seven SNPs failed across more than 70% of samples, and nine SNPs were monomorphic across all samples in all years from both patches P and Q. These 19 SNPs were removed, and we used the genotype data from the remaining 152 polymorphic SNPs for further analyses.  Of the 477 individuals genotyped from P and Q, six individuals were successfully genotyped at less than 80% of all SNPs and were also removed from the dataset, with data from 471 individuals total remaining.