As a baseline for genetic assignment and mixed-stock analysis, we drew on the respective mtDNA and microsatellite data of Jensen et al. and Roden et al. (2023), comprising 20 rookeries throughout the Pacific (Table 1; Figure 1). The samples from these two papers overlap and represent the most up-to-date compilation of new and previously published population genetic datasets for Pacific green turtles. Green turtle bycatch samples, with curved carapace lengths (CCL) of 27-91 cm, were acquired from longline fishing vessels based out of Hawaii and American Samoa (29 and 24 turtles, respectively). Tissue for DNA extraction was sampled at sea by trained observers between 1996 and 2017. Storage and transport of tissue collections, as well as laboratory processing, was performed using the same methods as Jensen et al. and Roden et al. (2023). Final products included a 770-bp fragment of the mitochondrial control region (see Dutton et al., 2014), and ten unlinked microsatellite loci.

Any turtle that was missing data at more than three loci (microsatellite or mtDNA) was removed prior to analysis. Loci were removed if they had more than 30% missing data in either the genetic baseline or fishery bycatch samples. Diversity statistics for each baseline rookery and bycatch collection were calculated in the R package DiveRsity v. 1.9.90 (Keenan et al., 2013). Tests of Hardy-Weinberg equilibrium and the presence of null alleles for microsatellites were performed previously by Roden et al. (2023). To help inform the organization of baseline samples into reporting units for genetic assignment and mixed-stock analysis, we computed population-level genetic distances between all pairs of rookeries (microsatellite loci only) using the R package ADEGENET 1.5 (Jombart et al., 2011). Two different computations were used: Nei’s genetic distance (Nei, 1978) which is a heterozygosity-based metric, and Edwards genetic distance (Edwards, 1971) which is an allele frequency-based metric. Both types of genetic distance were plotted as networks using the R package NETWORK v. 1.16.1 (Csardi and Nepusz, 2006) for visualization and interpretation. We also took into account the population genetic patterns in Jensen et al. and Roden et al. (2023) for combining rookeries into reporting units.

After considering genetic distances and other available results, the two primary criteria for determining reporting unit boundaries were mean simulated reporting accuracy and self-assignment rate, calculated using the R package RUBIAS v. 0.3 (Moran and Anderson, 2019). Self-assignment was the rate at which nesting females could be correctly grouped with other individuals from the same reporting unit given their genotypic similarity and using the leave-one-out method of Anderson et al. (2008). Reporting accuracy was assessed by simulating stock mixtures from baseline allele frequencies and was the proportion of the simulated mixture that was correctly identified by analysis. We simulated 10,000 mixtures, with a mixture size of 50, sampling all reporting units (repunit) uniformly from a flat Dirichlet distribution with a default sampling parameter of 1.5 per unit, allowing the proportion of reporting units in the mixture to vary randomly.

We aimed for accuracies of > 75% in at least one of the two approaches and tested different combinations of nesting populations iteratively, starting at the largest geographic scale (West-central Pacific vs. Eastern Pacific) and gradually splitting groups into the smallest units possible per our predefined criteria. If any nesting population had less than 20 individuals these were automatically combined with the most genetically similar population, or populations, into a single reporting unit. Pre-defined distinct population segments (DPSs) as recognized for C. mydas under The United States Endangered Species Act (Seminoff et al., 2015) were not used to determine reporting units, and neither were other proposed management units for this species (Wallace et al., 2010). To gauge the power of different marker types for genetic assignment and mixed-stock analysis, we simulated stock mixtures with the mtDNA and microsatellite loci separately as well as jointly. To assess whether the stock proportions of our different bycatch collections biased the analysis, we also retrospectively simulated 10,000 mixtures where the means of the Dirichlet sampling distributions for each repunit were the same as the mean mixing proportions estimated for fisheries bycatch, making the simulated stock mixtures more realistic for our data.

For our final mixed-stock analysis we ran RUBIAS with 10,000 iterations of the MCMC, discarding the first 1,000 samples as burn-in. Each mixture inference was also parametrically bootstrapped 1,000 times as a measure to reduce reporting unit biases (Moran and Anderson, 2019). The analysis was run ten times with different starting seed to ensure that mixing proportions remained consistent across runs. To assess whether any bycatch turtles could have originated from populations not represented by our baseline, we consulted the z-score values calculated by RUBIAS, which are the log-likelihoods for each individual compared to an expected distribution of log-likelihoods, while accounting for missing data (Clemento et al., 2014). The range of acceptable z-scores was determined by calculating the same statistic for nesting females from each reporting unit. Overall, an individual assignment to a reporting unit was considered plausible if the assignment posterior was greater than 0.70 and the z-score was larger than the reference minimum. An assignment was considered strong if the posterior was greater than 0.9 (Anderson et al., 2008) and the z-score was within the interquartile range of the reference.

A total of 27 out of 29, and 19 out of 24, green turtle bycatch samples, respectively from the Hawaiian and American Samoan longline fisheries, were retained for analysis after filtering for missing data. For our baseline, 1,042 out of 1,111 nesting females were retained (Table 1). Out of ten original microsatellite loci, one was removed because more than 30% of bycatch samples had missing data at this locus. For comparison with Roden et al. (2023), the locus removed was A6. Missing genotypes made up 6.2% of the baseline data, with the Coral Sea rookery having the most missing data at 15% and the Revillagigedo Islands having the least at 2.3%. In general, the West Pacific was more genetically diverse than the East Pacific, with the highest diversity estimates belonging to southwestern rookeries (Table 1). Microsatellite genetic distances between rookeries were consistent at most scales (Figure 2), and mostly consistent with population structure reported in Jensen et al. and Roden et al. (2023).

Based on a minimum self-assignment rate or simulated reporting accuracy threshold of 0.75, we resolved ten baseline reporting units from 20 green turtle nesting rookeries using mtDNA and nuclear microsatellite loci jointly (Tables 1, 2; Figure 1). Six of these units were represented by single nesting populations, and the rest grouped somewhat according to tectonic boundaries. The six Southwestern Pacific rookeries from Australian plate populations formed a reporting unit, as did the four rookeries from islands running the eastern boundary of the Philippine Plate. The two rookeries from the Central American coast formed a unit, while the last included American Samoa and western French Polynesia. With the exception of the Hawaiian Islands, all other Pacific Plate rookeries had some genetic similarities, sharing many of the same mtDNA haplotypes, and might have been combined into one reporting unit. However, the Marshall Islands also had similarities with the other Micronesian unit, could be considered intermediate between the Pacific Plate rookeries and the Philippine Plate rookeries, and had high enough accuracy scores to stand alone. American Samoa had high enough accuracy scores to be its own reporting unit as well, but French Polynesia was too small (n = 11) and was geographically close enough that combining the two rookeries made biological sense (Figure 2).

The Hawaiian reporting unit had the highest mean scores for self-assignment and simulated accuracy at 0.97 for both tests, while the Taiwanese Wan-An unit and the Galápagos unit had the lowest (Table 2). Notwithstanding that the Galápagos unit only had a mean self-assignment rate of 0.65, the median self-assignment score for this reporting unit was 0.81 and median simulated reporting accuracy was 0.96. Furthermore, all individual assignments of bycatch to this reporting unit had high statistical support (Table 3), suggesting enough unique genetic diversity in this rookery to justify being separate from the other Eastern Pacific stocks.

Accuracy was greatly improved when both marker types were included, and every reporting unit showed some improvement (Table 2). For the Revillagigedo reporting unit, the combined self-assignment rate was almost 40 percentage points higher than using microsatellites alone. For the Marshall Islands unit, the combined self-assignment rate was over 50 percentage points higher than for the mtDNA alone. These results argue that mixing marker types has an obvious benefit for genetic assignment and mixed-stock analysis of sea turtles, and that nuclear DNA markers generally enhance the signal of natal philopatry in the mtDNA rather than detract from it.

The results of mixed-stock analysis and genetic assignment were consistent across ten independent runs, suggesting that the MCMC was able to adequately sample from the posterior distribution. The mean of all assignment posteriors for bycatch samples was 0.94, indicating that most assignments were highly supported. The mean z-score across samples was -0.39, which was within the interquartile ranges of all reference populations (Figure 3). Only four bycatch turtles were given assignments that were statistically unreliable, having a posterior less than 0.70, or a z-score below the reference minimums (Table 3).

Turtles captured in the Hawaii-based fishery were mostly of Eastern Pacific origin. The Galápagos Islands reporting unit, in particular, represented over half of the total stock mixture with a mean scaled posterior of 0.52. Hawaiian green turtles composed about a third of the mixture (mean scaled posterior = 0.32) and all of these were encountered north of the main Hawaiian Islands (Figure 4). Other reporting units in the stock mixture were present only at low frequencies, but the individual assignments of turtles from each of these units were strong. The one turtle assigned to the Central American reporting unit had a mtDNA haplotype that is private to the Colola rookery (CmP5.1), indicating that this individual came from mainland Mexico (Table 3). The one turtle assigned to the Southwestern Pacific unit (Australia, Coral Sea, and New Caledonia) had a haplotype that is private to the Northern Great Barrier reef (CmP98.1), suggesting that this individual may have come from Raine Island, the largest green turtle rookery in the world (Seminoff et al., 2015). We also looked at private microsatellite alleles to help identify the origins of turtles within their assigned reporting units, but private alleles in our baseline were present only at low frequencies and none of our bycatch samples, apart from a few individuals assigning to the Hawaiian reporting unit, exhibited private alleles. One unassigned individual (125862) had a haplotype known previously only from Hawaii, was 91 cm in curved carapace length, and large enough to be considered an adult (Figure 3).

The three most prevalent reporting units in the bycatch from the American Samoa collection were: American Samoa and French Polynesia (mean scaled posterior = 0.33), Southwestern Pacific (mean scaled posterior = 0.32), and the Galápagos islands (mean scaled posterior = 0.21). The other two units with top-five mixture proportions were both from Micronesia but the lower 95% boundary of the credible interval was zero in both cases (Figure 5). The one turtle in the American Samoa bycatch collection that was individually assigned to the Marshall Islands also showed genetic similarity to the Micronesian reporting unit that included Palau (Table 3), suggesting that this individual has a common Micronesian genotype and mtDNA haplotype. Of the seven turtles assigned to the Southwestern Pacific reporting unit, two had the same rare haplotype (CmP31.1) that is known only from the Coral Sea and New Caledonia, four had haplotypes CmP47.1 and CmP80.1 that are not known from the Gulf of Carpentaria and are uncommon in the Northern Great Barrier reef, suggesting that these individuals probably also came from the Coral Sea, New Caledonia, or the Southern Great Barrier Reef. The last mtDNA haplotype for this repunit (CmP97.1) is not known from any nesting population but the microsatellite data suggests that it comes from the Southwestern Pacific. One other bycatch turtle from the American Samoa longline fishery had an orphan mtDNA haplotype, but it was missing data at five out of the nine microsatellite loci so it was not suitable for genetic assignment. Bycatch turtles excluded from genetic assignment for missing data at too many microsatellite loci are also listed in Table 3, with their mtDNA haplotype for comparison with Jensen et al. (this issue).

All turtles in our bycatch collections had mtDNA haplotypes that were typical of the reporting unit they were assigned to, except for one: Turtle 100690 had the CmP22.1 haplotype that is known from American Samoa and the Marshall Islands, but this individual was assigned to the Galápagos Islands. We are unable to say whether this is a dubious assignment or if this turtle simply has an unusual genotype/haplotype combination, but the assignment to the Galápagos was strong with an assignment posterior greater than 0.99 and a positive z-score of 1.32. Therefore, in this one sample alone the signal in the microsatellites seems to be overpowering the signal in the mtDNA.

Post hoc tests of reporting unit accuracy revealed a linear relationship between simulated stock proportions and the estimation of those stock proportions (Figure 6), adding credence to our results in spite of the fact that our analysis was conducted using only nine microsatellite loci and one mtDNA marker. The Galápagos and Southwestern Pacific units had slight downward biases, and the Central American unit had a slight upwards bias at low mixing proportions, but all were nonetheless informative. Furthermore, there were no discernable differences in accuracy between reporting units present in both the Hawaiian and American Samoan fisheries, suggesting no biases between bycatch collections.

During their early years, green turtles live a cryptic existence in the open ocean and are not frequently encountered by humans, except as fisheries bycatch. Even so, they are caught less often than other species and typically only when young. For these reasons the distributions of young green turtles in pelagic habitats have been largely unknown, particularly in the Pacific which has received less research attention than the Atlantic. Our results show that at least seven out of the ten baseline units designated in this study were present in the green turtle bycatch, suggesting that the pelagic distributions from many different rookeries across the Pacific basin overlap broadly in the Central Pacific.

Green turtles from the Eastern Pacific have a different color phenotype, being darker than their western counterparts, and are previously known from Hawaiian waters (and as far west as New Zealand: Godoy et al., 2012), but mtDNA has been unable to precisely assign most of these turtles to their nesting populations (Parker et al., 2011). Our mixed-marker assay better resolves genetic differences among Eastern Pacific rookeries and indicates a Galápagos origin for the majority of these individuals. The Galápagos also contributed substantially to the American Samoa bycatch, suggesting a strong presence of this reporting unit throughout the Central Pacific. Turtles assigned to the Galápagos were larger on average than those from other locations (Figure 7), with several individuals exceeding 70 cm CCL (Table 3), well above the 35-45 cm range at which greens typically transition to herbivorous diets and recruit to neritic habitats (Reich et al., 2007), and overlapping with the 60 cm CCL minimum size reported for Galápagos nesting females (Zarate et al., 2003). Parker et al. (2011) proposed several hypotheses for why eastern green turtles in the north Central Pacific tend to be larger, including an extended pelagic stage as an alternate life history, and facultative pelagic foraging by post-nesting females (see Seminoff et al., 2008). Whatever the case, our results show that eastern green turtles in the South Pacific also attain large sizes, and that this could be peculiar to the Galápagos population.

Satellite tagging studies conducted in the Gulf of Mexico show no evidence that ocean currents influence the dispersal trajectories of 14-30 cm green turtles (Putman and Mansfield, 2015, see also Polovina et al., 2006), but with so many individuals in our study assigned to the Galápagos, the possibility of turtles riding the large North and South Equatorial Currents into the Central Pacific is difficult to ignore. It’s also worth pointing out that six turtles were caught in the North Equatorial Counter Current that spans 3-10 degrees north of the equator (Figure 4), one of these was assigned to the Southwest Pacific, possibly indicating current-use as a means of reaching the Central Pacific from the west. Young green turtles may not be passive propagules, but the algal rafts that provide them with food and cover are subject to ocean circulations (Thiel and Gutow, 2005) and could motivate turtles to go with the flow. Some evidence suggests that adult green turtles sometimes use ocean currents to migrate between foraging areas and nesting beaches (Luschi et al., 2003; Bass et al., 2006; Nishizawa et al., 2013), but at what point they learn to use currents is unknown.

The combined mixed-marker data set was considerably more powerful for genetic assignment than either marker type alone (Table 2). Therefore, mixing marker types allowed a finer partitioning of baseline units and greater assignment precision, without compromising much accuracy. In addition, because the two marker types have different modes of inheritance, and different substitution rates, together they offer a more complete representation of genetic diversity, not confined to strictly matrilineal gene genealogies but not excluding them either. Most assignments were consistent with the known distributions of mitochondrial haplotypes, indicating that microsatellite variation was largely in agreement with the mtDNA, while also increasing resolution. However, there were several ambiguous assignments in our results that highlight shortcomings in our marker panel and overall study design.

First, there were four individuals with assignment posteriors below 0.7 (Table 3). Considering the limited number of markers used, only four such individuals could be considered a success but all of them involved assignments to rookeries from the Western Pacific genetic cluster (Figure 2). Ambiguous assignments made up nearly a quarter of all Western Pacific assignments from this study, suggesting that our marker panel did not characterize the genetic diversity of this region very well, though sample size was small. Confusion with the Marshall Islands was a feature of every ambiguous assignment, perhaps because this reporting unit had the weakest mtDNA genetic fingerprint (Table 2). The obvious solution to this problem is to increase the number of molecular markers used, specifically selecting for loci that better resolve the Western Pacific (see next section). Increasing resolution through larger marker sets also means increasing the number of reporting units, which will enhance assignment precision.

Large combined reporting units are undesirable for several reasons, aside from being imprecise for genetic assignment and mixed stock analysis. In our data, the combined reporting units had z-score distributions much lower than single-rookery units (Figure 3), making z-scores a less reliable statistic for assessing the quality of genetic assignments. Conglomerate reporting units also introduce bias into mixed-stock analysis because larger more diverse reporting units receive a larger mixing proportion when the information content of the genetic data is poor (Moran and Anderson, 2019). In our analysis we took measures to reduce reporting unit bias (see methods) but combining so many rookeries together, such as in the case of the Southwestern Pacific, should still be avoided when possible.

Second, based on all known green turtle nesting locations (Wallace et al., 2010; Seminoff et al., 2015) the baseline used in this study is largely comprehensive for Hawaii and the East Pacific populations, but significant gaps exist in the Western Pacific. Many nesting sites in New Guinea, Indonesia, the Philippines, Japan, mainland Asia, Melanesia, and elsewhere, could have contributed to green turtle bycatch assignment in the Central Pacific. Therefore, genotyping samples from an increased number of baseline rookeries will give context to mixed-stock analysis and help resolve ambiguities in the Western Pacific.

Genetic assignment and mixed-stock analysis are useful tools for conservation and management, especially when molecular markers are sensitive enough to finely delineate the stock structure of baseline samples (Beacham et al., 2009; Seeb et al., 2011; McKinney et al., 2017; Beacham et al., 2020). For sea turtles, it is recognized that new molecular assays are needed to better describe the genetic diversity of nesting populations and organize them into precise demographic units (Komoroske et al., 2017). Some labs, including ours, are currently exploring genome-wide SNP panels for sea turtle conservation genetics (Komoroske et al., 2018; Banerjee et al., 2020; Driller et al., 2020; Hamabata et al., 2020; Horne et al. 2023), but the results of the present study show that not all green turtle populations require extensive panels of molecular markers for reliable stock discrimination. Furthermore, new marker development may not be fruitful without increased baseline sampling (Komoroske et al., 2017). When designing new sets of genetic markers for sea turtles it should be remembered that programs such as RUBIAS accept all bi-parentally inherited loci, including SNPs, microsatellite variants, and indels, as well as maternally inherited mtDNA, or any combination thereof. As such, mixtures of different marker types, each with its own advantages (SNPs for genomic coverage and abundance, microsatellites for allelic diversity, and mtDNA haplotypes for their association with female natal homing) should not be overlooked as best options for conservation genetics, both in terms of accuracy and cost effectiveness.

Inasmuch as the Central Pacific appears to be a crossroads for young green turtles, bycatch in this area has the potential to impact a large number of nesting populations. Turtles from nearby rookeries make up a significant portion of the bycatch (Hawaiian turtles in the Hawaiian fishery and American Samoan turtles in the American Samoan fishery; Figures 4, 5), so proximity to nesting beaches appears to influence mixed-stock proportions to some degree. But since many of the common reporting units in the bycatch, such as Hawaii, the Galápagos, mainland Mexico, American Samoa, and the Southwest Pacific, represent populations thought to be stable or increasing in numbers of nesting females (Seminoff et al., 2015), stock proportions could also be a reflection of reproductive output from major rookeries. However, turtles from struggling populations, such as the Marshall Islands (Seminoff et al., 2015), were also present in both fishery collections, in spite of small sample sizes. Therefore, the overall bycatch is collectively a well-mixed migrant pool consisting of individuals from diverse origins and represents an international marine management issue.

Though the data were limited, it may be worth noting that all bycatch individuals assigned to the Hawaiian reporting unit were encountered above 24 degrees latitude. If turtles from this rookery are primarily using pelagic habitat north of the main Hawaiian Islands during their early years, then this information could be helpful in managing the Hawaiian unit, which is currently recognized as its own DPS. Naro-Maciel et al. (2014) also show that adults foraging at Palmyra Atoll, just south of the main Hawaiian Islands (5.88˚N, 162.1˚W) do not come from the Hawaiian rookery, possibly supporting the idea that Hawaiian green turtles stay north. Without a larger mixture sample, it is difficult to draw such conclusions concretely, but given that the ranges of Pacific green turtles during their pelagic life history stages remain largely unknown, even incomplete patterns could provide clues for spatial management planning.