Reinforcement translocations are increasingly utilised in conservation with the goal of achieving genetic rescue. However, concerns regarding undesirable results, such as genetic homogenisation or replacement, are widespread. One factor influencing translocation outcomes is the rate at which the resident and the introduced individuals interbreed. Consequently, post-release mate choice is a key behaviour to consider in conservation planning. Here we studied mating, and its consequences for genomic admixture, in the North Island brown kiwi Apteryx mantelli population on Ponui Island which was founded by two translocation events over 50 years ago. The two source populations used are now recognised as belonging to two separate management units between which birds differ in size and are genetically differentiated. We examined the correlation between male and female morphometrics for 17 known pairs and quantified the relatedness of 20 pairs from this admixed population. In addition, we compared the genetic similarity and makeup of 106 Ponui Island birds, including 23 known pairs, to birds representing the source populations for the original translocations. We found no evidence for size-assortative mating. On the contrary, genomic SNP data suggested that kiwi of one feather did not flock together, meaning that mate choice resulted in pairing between individuals that were less related than expected by random chance. Furthermore, the birds in the current Ponui Island population were found to fall along a gradient of genomic composition consistent with non-clustered representation of the two parental genomes. These findings indicate potential for successful genetic rescue in future Apteryx reinforcement translocations, a potential that is currently underutilised due to restrictive translocation policies. In light of our findings, we suggest that reconsideration of these policies could render great benefits for the future diversity of this iconic genus in New Zealand.

Blood sampling and morphometric measurements of the birds were conducted in accordance with the Kiwi Best Practice Manual (Robertson & Colbourne 2017), the Massey University Animal Ethics Committee (MUAEC) permits 06/05, 07/144, 16/92, and 18/83, and the Department of Conservation Wildlife permits AK-14969-RES, AK-21519-FAU, 50249-FAU and 70875-RES. Samples from the parental populations were collected in 2020 and 2021. For birds fitted with transmitters (most of the Ponui Island birds analysed, all the studied pairs, and most of the Hauturu birds from the Remutaka Forest and the Pūkaha National Wildlife Centre), blood sampling and measuring occurred together with the annual transmitter replacement. A licenced kiwi dog assisted with catching the remaining birds, which were caught specifically for this study. The density of kiwi has been estimated to be around one kiwi per three ha both in Trounson and on Hauturu (in 2007 and 2008, respectively; (Holzapfel et al. 2008)(Craig et al. 2011, Craig 2019)). Age of the birds was mostly unknown, hence individuals were considered adults if they were known to have bred or, when the breeding status was unknown, based on their sex, size, and weight combined; females were considered adult if weighing > 2000 g, or > 1700 g if having a tarsus width (TW) > 11 mm or a bill > 113 mm; males were considered adults if weighting > 1700 g, or > 1400 g if having a tarsus length (TL) > 90 mm or a bill > 90 mm.

Bill length, tarsus length, and the ratio between these measurements were used to explore differences in size distribution among adult birds from Hauturu, Trounson, and Ponui Island. Linear modelling (lm) in R (R core team version 3.6.2) was used for this analysis. To analyse size assortative mating, Pearson correlations between female and male morphometric values were investigated. The measurements analysed were TL, TW, bill length, weight to TL ratio, bill length to TL ratio, and body condition was used for 17 Ponui Island pairs. Body condition was calculated based on weight and TW following Taborsky and Taborsky (1999).

DNA was extracted from 10-50 µl thawed A. mantelli blood using a High Pure PCR Template Preparation Kit (Roche, Basel, Switzerland). The manufacturer’s instructions were followed with the exception that the DNA was eluted twice using 50 µl of elution buffer for each centrifugation round. The DNA extraction success and quality were validated using agarose gel electrophoresis (1% (w/v) agarose in 1x TAE buffer) and the concentration of DNA was measured with a Qubit 2.0 fluorometer using the dsDNA High Sensitivity assay (Life Technologies, CA, USA).

Pair-ended Genotype-by-sequencing (GBS) library, sequencing preparation, and associated quality checks were done by The Elshire Group Limited. GBS libraries were constructed using 100 ng of genomic DNA, 1.44 ng of adapters, the restriction enzyme EcoT22i, and 18 PCR cycles, and otherwise following the protocol presented in Elshire et al. (2011). For this study, sequencing of 150 unique birds across three 96-well plates were analysed. Sample location within plates was randomised and each plate contained one positive and one negative control. Sequencing was performed on an Illumina HiSeq XTen with 2 x 150bp paired-end reads.

The previously published A. mantelli genome (Le Duc et al. 2015) was deemed suboptimal as a reference genome for our study based on being (1) highly fragmented and (2) a composite genome of multiple individuals representing two separate management units within the species. Raw short-read data generated by Le Duc et al. (2015) was accessed from the European Nucleotide Archive (Study Assession RJEB6383) and the ERR519283_1.fastq.gz date file was used to re-assemble a reference genome based on a single individual. Raw reads were aggressively trimmed using trim_galore (https://github.com/FelixKrueger/TrimGalore) before assembly with Meraculous 2.2.5.1 (Chapman et al. 2011) using a kmer size of 35 and heterozygous mode 1. The resulting reference genome utilised for our analyses was haploid, with heterozygous regions collapsed into a single haplotype chosen at random between the two possible sequences.

Processing of raw read data, including filtering, trimming, alignment, and SNP calling was conducted by Tea Break Bioinformatics. The 1 538 639 658 raw sequencing reads were demultiplexed using Axe (axe-demux; Murray & Borevitz 2018), adapters and barcodes were trimmed using the batch_trim.pl script (https://github.com/Lanilen/GBS-PreProcess) using default parameters. Forward and reverse reads were pair-matched and aligned to the reference genome using Bowtie 2 (Langmead & Salzberg 2012) using default parameters.

SNP-calling was conducted in STACKS 2.5 (Catchen et al. 2013) using the populations program set for the EcoT22i enzyme, bootstrapping, and site merging. Initially, this was done for each plate separately. The graphical output from Kinship-using-GBS-with-Depth-adjustment program (KGD; Dodds et al. 2015) and Tensorflow Projector (https://projector.tensorflow.org/) were then used to verify the absence of bias or batch effects after which Stacks 2.5 was rerun for the combined dataset of all three plates using the following command line: --vcf -r 0.1 --min-maf 0.1 -e ecoT22I --ordered-export --bootstrap --merge-sites --genepop --structure --fasta-loci --fasta-samples --fasta-samples-raw --write-single-snp. These settings allowed for up to 90% missing data per locus (-r 0.1) to maximise the number of individuals included in the resulting dataset. SNPs derived from different cut sites that had overlapping read coverage were combined into single loci (--merge-sites). To ensure robust SNP identification and to restrict noise, the minimum minor allele frequency was set to 10% (min-maf 0.1). The output was filtered to only include the first variable site per locus (--write-single-snp) resulting in 51691 SNPs that were utilised for analyses with an average depth of 24 reads (with a standard deviation of 12), and an average call rate of 0.6 (with a standard deviation of 0.07). The generated output provided as text files formatted for genepop (see below), and fasta files for loci and samples (--genepop --fasta-loci --fasta-samples).

For relatedness analyses, known first-degree relatives (known offspring and siblings) were excluded and these analyses contained 20 known pairs and a total of 74 Ponui Island individuals. Unscaled pairwise relatedness values were derived using KGD (Dodds et al. 2015). This matrix of relatedness values is based on pairwise proportional allelic similarity but accounts for read depth (including missing data) which can lead to values greater than one and below zero. These values were used to evaluate whether paired individuals were more or less related than expected under a scenario of random mating in two ways. Test one explored whether the relatedness of paired individuals to their partner was higher or lower than the average relatedness of the paired individuals to all other Ponui Island birds in the data set. This approach provided an indication of whether paired individuals were more closely related than expected by chance (a prediction of assortative mating) or more distantly related than expected by chance (a prediction of disassortative mating). To account for possible bias caused by cryptic clusters (in particular if a few paired individuals belonged to a cluster very distantly related to the rest), test one was complemented with test two. Test two involved ranking the relatedness of the female to the male and vice versa for each pair and then comparing these results to expectations for ranked relatedness from a computer simulation scenario of random mating. A ranking of 1 suggested that the partner represented the least related bird within the dataset and a ranking of 73 that the partner was the most related bird in the dataset. Random mating was simulated using 10*40 randomly drawn ranks between 1 and 73. T-tests were used for both these tests.

In addition, the average relatedness of the male and female in each pair on Ponui Island (n = 23 Ponui Island pairs) to the birds from Hauturu, was used to categorise pairs as being ‘similar’ or ‘different’ in their genetic makeup). Based on the average relatedness to the Hauturu birds for 106 Ponui Island, the relatedness space was split into four quadrants to categorise pairs. Birds were classified as ‘similar’ when both male and female were found to be less related or both more related to Hauturu than the average value. Birds were classified as ‘different’ when the male was more and the female less related than the average or vice versa.

The following data files accompany the publication “Mixed mating in a multi-origin population suggests high potential for genetic rescue in North Island brown kiwi, Apteryx mantelli”. 
The data are from the A. mantelli populations in the Trounson Kauri Park, on Hauturu-o-Toi (also known as Little Barrier Island), and on Ponui Island.
Provided as CSV files are: 

1. Morphometric data compared between adults of three populations. This file identifies sample origin, taxon identity, sex, tarsus length, tarsus width, bill length, and bill length to tarsus length ratio.
2. Morphometric data compared among 17 pairs of birds. This file identifies 17 pairs and the female and male bill length to tarsus length ratio, the tarsus length, the weight to tarsus length ratio, the bill length, the weight, the tarsus width, and the body condition score using sex specific or non-sex specific conversion. 
3. A relatedness matrix including all 150 individuals analysed. This file is the output from Kinship-using-GBS-with-Depth-adjustment program, with metadata added to indicate known firstdegree relatives, taxon identity, population origin, and birds ID.
4. Relatedness values and ranked relatedness values among 20 analysed pairs. This file includes the allelic relatedness between paired birds, the ranked relatedness between each paired birds and its partner, the average relatedness of each bird to all birds in the dataset, the ranked relatedness based on random simulation, the average value for all paird relatedness in the population, the average ranked relatedness based on simulation. 

Requests for raw reads from the underlying GBS sequence library should be made to Richard Witehira: richardwitehira@xtra.co.nz or Andre Witehira: andre.witehira@gmail.com.  Permission would then be sought from the kaitiaki (guardians) representing hapu (sub-tribes) that affiliate with the areas of sample collection. Kiwi are taonga (treasures) to the indigenous Māori people of Aotearoa New Zealand. All individuals, samples, or genomic data obtained from taonga species have whakapapa (genealogy, connections, and belonging) and are considered taonga in their own right. Tikanga Māori (Māori customary practices) determines their use.