The 710-kilometer-long Uruguayan coast is part of a complex hydrological system comprising the frontal zone of the Río de la Plata estuary and the Atlantic Ocean ( Figure 2 ). The Malvinas/Falkland current prevails during the austral winter and the Brazilian current dominates during the austral summer ( Figure 1 ) (Garcia, 1998; Ortega and Martínez, 2007). This causes variations of >15°C in sea surface temperature (range 10–27°C) throughout the year (Acha et al., 2004). The study area also includes the coastal waters between Colonia (34° 28’ S, 57° 49’ O) and “Barra del Chuy” in the Department of Rocha (33° 44’ S, 53° 22’ W). The site comprises six coastal marine protected areas (Humedales de Santa Lucia, Isla de Flores, Laguna Garzón, Laguna de Rocha, Cabo Polonio, and Cerro Verde e Islas de La Coronilla ( Figure 2 ), all of which are important marine turtle feeding grounds. Cerro Verde is one of the most significant in terms of green turtle numbers (López-Mendilaharsu et al., 2016). The coastline is characterized by a succession of sandy beaches separated by rocky outcrops rich in macroalgae, a prominent part of the green turtle diet, and mussels (Borthagaray and Carranza, 2007; Scarabino et al., 2015; Vélez-Rubio et al., 2021).

The study area, based on differences in hydrological characteristics, was divided into three zones: 1) an inner estuarine zone (IEZ, ca. 350 km) from Nueva Palmira to Montevideo, characterized by a fluvial-marine salinity regime (salinity <12 psu) influenced mainly by the Rio de la Plata discharge; 2) an outer estuarine zone (OEZ, ca. 130 km) from Montevideo to Punta del Este, representing a transition between oceanic and estuarine characteristics with an intermediate salinity range (13–25 psu); and 3) an oceanic zone (OZ, ca. 230 km) from Punta del Este to Barra del Chuy (salinity >26 psu) ( Figure 2 ).

Turtle tissue was sampled according to permitting requirements by professionally trained technicians of the Karumbé Non-Government Organization using two methodologies, recovery of stranded turtles and scientific captures. In total, 201 green turtle tissue samples were collected in two sampling periods: 2003–2005 and 2009-2014. Muscle and epidermis samples were preserved in 70% ethanol, using standard protocols following Dutton (1996). All sampled turtles were juveniles well under the minimum size of nesting females from the closest nesting colonies (Curved Carapace Length=90 cm in Ascension Island; Weber et al., 2014). Following standard methods (Dutton, 1996), green turtles were caught alive over rocky bottoms in shallow waters. Set nets (nylon monofilament, 50 m length × 3 m depth, 30 cm stretched mesh size) were deployed perpendicular to wave direction and monitored constantly to ensure turtle well-being. A standard epidermal biopsy was collected, and Curved Carapace Length (CCL, notch to tip) was measured for each turtle using flexible tape (± 0.1 cm). All turtles were tagged with inconel flipper tags (Style # 681, National Band and Tag, Kentucky, USA) before release at the capture site (López-Mendilaharsu et al., 2016; Vélez-Rubio et al., 2018b). Alternately, stranded turtles were registered by the Uruguayan Sea Turtle Stranding Network and muscle tissue was sampled from fresh carcasses by Karumbé members (Vélez-Rubio et al., 2013).

DNA extractions were conducted following protocols outlined by Allen et al. (1998) with modifications. Two sets of PCR reactions were performed using overlapping primers. For samples collected between 2003 and 2005, primers LTCM2 and HDCM2 (Lahanas et al., 1994) were used to amplify and sequence ~481 bp of the mtDNA control region using standard conditions and negative controls. These are referred to here as shorter segments or haplotypes. For samples collected between 2009 and 2014, to expand coverage, a ~780 bp mtDNA control region fragment containing the original ~481 bp segment was amplified using the primers LCM15382 and H950 (Abreu-Grobois et al., 2006), referred to here as longer segments or subhaplotypes. The polymerase chain reaction (PCR) cocktail consisted of ~90 ng of DNA, 1X Phusion HF Buffer, 400 mM of dNTPs, 0.5 uM of each primer, and 0.5 U of Phusion High Fidelity DNA polymerase (New England Biolabs) in a total volume of 40 µL. The amplification cycle followed Vilaça et al. (2013). PCR products were sequenced using a 3730 xl DNA analyzer (Applied Biosystems) by the Macrogen sequencing service (Macrogen Europe, Netherlands). All samples were sequenced in both directions.

Sequences were edited and aligned using Bioedit V 7.0 (Hall, 1999). Mitochondrial haplotypes were classified according to the widely-utilized standardized designation provided by the Marine Turtle Sequences website, maintained by the Archie Carr Center for Sea Turtle Research at the University of Florida (ACCSTR; http://accstr.ufl.edu/genetics.html). In this system, a standardized haplotype designation, such as CM-A8, is used for the shorter ~481 bp segments, with CM standing for Chelonia mydas, A for Atlantic, and 8 for a unique identifier of that haplotype. The subhaplotype designation for the longer ~780 bp fragment, which overlaps with and extends the shorter ~481 section, addresses any additional variation beyond the overlapping part with a decimal point followed by a unique number, such as CM-A8.1 or CM-A8.2.

Variability among the three sampling areas within Uruguay (inner estuarine zone (IEZ), outer estuarine zone (OEZ) and oceanic zone (OZ), Figure 2 ), as well as between years and sampling periods, was evaluated. The statistical analysis was done with RStudio 2023.06.0 + 421 (RStudio Team, 2020). As no significant differences were found (see Results for details), these data were pooled into one single Uruguay sample for subsequent analyses. The sequence data were also included in a broader analysis of western Atlantic feeding grounds studied to date (Savada et al., 2021).

Haplotype (h) and nucleotide diversity (π) (Nei, 1987) were calculated using Arlequín (version 3.5; Excoffier and Lischer, 2010). The haplotype network was executed in PopART 1.7 (Leigh and Bryant, 2015). Analysis of molecular variance (AMOVA) and pairwise comparisons with Φ_ST (using Kimura 2P genetic distances; Kimura, 1980) were carried out using the ARLEQUIN v3.5 program (Excoffier and Lischer, 2010) to investigate genetic structuring within and among feeding grounds. The AMOVA considered three groups: North Atlantic rookeries, South Atlantic rookeries, and Uruguay. Genetic differentiation between all western Atlantic feeding grounds was further investigated using pairwise comparisons. Statistical significance was tested using 1,000 permutations.

Contributions of nesting colonies to the Uruguayan feeding grounds (UY) were assessed through Mixed Stock Analysis (MSA) using Bayesian methods implemented in the program BAYES (Pella and Masuda, 2001). The 16 previously described Atlantic or Mediterranean nesting colonies were explored as sources in the MSA ( Figure 1 ; Supplementary Table S1 ; Savada et al., 2021). The BAYES program integrates information from the observed data (nesting colonies and feeding grounds) and is not biased by small sample size or rare haplotypes (Pella and Masuda, 2001). Three MSAs were performed as follows: 1) with equal prior probability assigned to each rookery, 2) considering contribution weighted by nesting population size, and 3) taking into account distance from the nesting ground. The straight-line distance estimates are conservative, as sea turtles do not move in straight lines. Size estimates of the nesting populations were taken from: Bellini (1996), Bellini et al. (1996), Seminoff et al. (2002, 2015), and Formia et al. (2006). The analyses were carried out using about 7,500 iterations until Gelman and Rubin diagnostics confirmed convergence of the chains to the desired posterior density, with most shrink factors near 1.0 and below 1.2. The first halves of the chains were discarded as “burn-in” and estimates were based on the second halves only as recommended (Pella and Masuda, 2001).

A ~481-bp-long fragment was successfully amplified in 201 samples collected from 2003–2005 (n=144) and 2009–2014 (n=57) ( Table 1 ). In total 12 haplotypes (GenBank: PP378115–PP378126), defined by 14 variable sites and a 4 bp insertion/deletion, or indel, were detected ( Table 1 ; Figure 3A ). The most frequent haplotype was CM-A8, occurring in 78% of Uruguay samples (n=157; Figures 2 , 3A ). CM-A5, the second-most common sequence, was found in 9% of samples (n=17). The remaining haplotypes, including CM-A6, CM-A9, CM-A10, CM-A44, CM-A45, and CM-A33 were observed at low frequencies (≤ 5%). Haplotype CM-A33 was confirmed in one Uruguayan individual for the first time. This haplotype, although previously described (Bjorndal et al., 2006), had been reported erroneously on Trindade Island, Brazil (Shamblin et al., 2015) and withdrawn from consideration.

The longer fragment (~780 bp) obtained from the same 57 samples collected between 2009–2014 revealed additional polymorphism in the extended segments. A total of 10 subhaplotypes was found (GenBank: PP425333-PP425342), the most frequent being CM-A8.1 (74%), followed by CM-A5.1 (7%), CM-A42.1 (5%) and CM-A9.1 (4%). CM-A8.2, CM-A6.1, CM-A10.1, CM-A44.1, and CM-A45.1 were found at frequencies of 2% ( Table 1 , Figure 3B ). The only haplotype that split into two subhaplotypes was CMA-8, divided into CMA-8.1 and CMA-8.2 ( Table 1B ).

As no significant differences were found when comparing haplotype frequencies between the different sampling areas in Uruguay (inner estuarine, outer estuarine, and oceanic zones; Figure 2 ; X ² (11, n = 201) = 13.686, p = 0.251) or among sampling years or periods there (X ² (22, n = 201) = 14.523, p = 0.002), data from various coastal locations and time periods were pooled into one single Uruguay sample.

For comparison to other areas, the shorter sequences were used, as corresponding longer sequences remain unpublished. For these 201 samples, haplotype diversity (h) was 0.392 +/− 0.043 and nucleotide diversity (π) was 0.002 +/− 0.001.

Analysis of Molecular Variance (AMOVA) based on divergence between haplotypes revealed significant overall differentiation among western Atlantic feeding grounds (Φ_ST = 0.730, P < 0.001). A gradient was observed, where the North Atlantic populations presented differences from the South American feeding areas. Pairwise comparisons revealed significant differentiation between Uruguay and most other western Atlantic feeding grounds, with the exception of Cassino, Paranagua and Ubatuba located relatively nearby in southern Brazil ( Figure 1 , Table 2 ).

Correspondingly, mitochondrial D-loop sequences from Uruguay, such as the dominant CM-A8, had been previously reported in other foraging areas along the South American coast, and CM-A5 is common in the Caribbean ( Supplementary Table S3 ). The rare haplotype CM-A42 detected in Uruguay had been previously reported among turtles foraging in Argentina and Brazil (GenBank: JF308481.1). Further, sub-haplotypes CM-A8.2 (GenBank: JF308473.1), CM-A42.1 (GenBank: JF308481.1), CM-A44.1 (GenBank: PP379906), and CM-A45.1 (PP429908) have been described in foraging areas in Brazil and Africa and/or African rookeries.

The juvenile green turtle developmental feeding area of Uruguay was significantly differentiated from each regional nesting beach, rejecting the hypothesis of single rookery origins and qualifying the site for Mixed Stock Analysis. The three MSAs analyzed (1: with equal prior probability assigned to each colony, 2: considering the contribution weighted by the size of the nesting population, and 3: taking into account the geographic distance from the nesting site), demonstrated that Ascension Island is the main source of the Uruguay feeding grounds (43–47%), followed by Guinea Bissau (36–37%), and Trindade Island, Brazil, (6–9%), and with lesser contributions from the Guadeloupe, Suriname and French Guiana rookeries ( Figure 4 ; Supplementary Table S4 ). The 95% confidence intervals were fairly broad, as previously reported in the literature. The results of the Gelman and Rubin diagnosis were between 1.0 and 1.02, indicating chain convergence.

The Mixed Stock Analysis findings were generally consistent with haplotype and subhaplotype geographic patterns. The CM-A8 haplotype (and corresponding CM-A8.1 subhaplotype) dominant in the Uruguay feeding ground is a common sequence dominant throughout the South Atlantic at varying frequencies including at African (Guinea Bissau, Sao Tome e Principe, and Bioko) and Brazilian (Trindade and Atol das Rocas/Fernando de Noronha) rookeries, as well as Ascension Island ( Supplementary Table S1 ). Among nesting colonies, the second most common sequence, CM-A5 (and corresponding CM-A5.1 subhaplotype), has been reported most frequently in northern hemisphere rookeries of Costa Rica, Aves Island in Venezuela, and Suriname. The remaining rare haplotypes were previously detected among rookeries in Ascension Island, Brazil, and Africa, in addition to Mexico, Costa Rica, Cuba, Block Island, French Guiana, Guyana, and Suriname ( Figure 1 ). In particular, the rare haplotype CM-A42, unique among rookeries to Guinea Bissau (Patrício et al., 2017), was revealed in Uruguay.

The Mixed Stock Analysis indicated that these juvenile green turtles migrate from multiple areas of the Atlantic Ocean, mostly from a combination of Ascension Island, Guinea Bissau, and Trindade Island, Brazil. The reproductive connection to distant Ascension Island on the Mid-Atlantic Ridge ( Figure 1 ) requires impressive island-finding and navigation abilities. These are known to depend upon navigation using the Earth’s magnetic field, as well as olfactory and chemical cues (Lohmann and Lohmann, 1996). This migration, supported by satellite tracking and tag returns of adults departing Ascension Island to the coast of South America, is celebrated as one of the most remarkable of marine vertebrate voyages (Luschi et al., 1996).

These natal origins are based on analysis of the shorter segments (~481bp) for comparison to other sites, and are consistent with findings from newly characterized longer (~780bp) sequences. The CM-A8.1 subhaplotype that dominates Uruguay is widespread in the South Atlantic, supporting those linkages. Further, the rarer CM-A42.1, CM-A44.1, and CM-A45.1 subhaplotypes link Uruguay to Ascension Island and/or African nesting and feeding areas (GenBank JF308473, JF308481; Patrício et al., 2017 respectively; Supplementary Table S3 ). Similar connections between South Atlantic foraging groups in Brazil and Argentina and rookeries of Ascension Island, Trindade Island (Brazil), and West Africa have been previously reported from genetic and mark-recapture studies (Naro-Maciel et al., 2007; 2012; Torezani et al., 2010; Proietti et al., 2012; Prosdocimi et al., 2012; Jordão et al., 2015; Coelho et al., 2018; Savada et al., 2021).

However, the confidence intervals in all of these Mixed Stock Analysis studies are high. This is primarily attributed to the recent population expansions out of glacial refugia after the Last Glacial Maximum approximately 20,000 years ago, as reflected in the shallow haplotype network and prevalence of common shared haplotypes like CM-A8 ( Figure 3 ) (Naro-Maciel et al., 2014; Reid et al., 2019). Guinea Bissau, for instance, consists of 99.6% CM-A8 in addition to 0.4% CMA-42 ( Supplementary Table S1 ), and its source contribution to Uruguay could range from 0 to about 60% with 95% confidence ( Figure 3 ). This lack of resolution highlights the need for further geographic sampling, particularly of uncharacterized African foraging areas to determine their connectivity, as well as genomic sampling to narrow down confidence intervals if possible. Testing alternative methods such as stable isotope analysis of dietary composition, as well as expanding traditional tagging and satellite telemetry studies, are recommended to determine connectivity in the context of oceanographic studies like particle modeling that predict hatching movements. Undoubtedly, success in the management of these areas requires multidisciplinary studies that allow understanding of their dynamics.

The results of our study do not support the hypothesis that geographic distance is the main factor determining rookery contributions to the Uruguayan foraging ground, although population size and ocean currents may play important roles. The Mixed Stock Analyses confirm that the two nesting beaches that contribute around 80% of individuals to Uruguay, are Ascension Island and Poilão, Guinea Bissau, the largest in the South Atlantic. These areas are located 5,000 - 6,000 km away from Uruguay and have almost 4,000 nesting females each per year (Seminoff et al., 2015). Ascension Island is an important regional source and is also the largest contributor to the other Southwestern Atlantic Ocean feeding areas ( Figure 1 ) (Naro-Maciel et al., 2007; Proietti et al., 2012; Prosdocimi et al., 2012; Jordão et al., 2015; Coelho et al., 2018; Savada et al., 2021). This island contributes more individuals to the Southwestern Atlantic coasts than Trindade Island in Brazil, which is 2,000 km closer but is also a much smaller rookery. Although the closest rookeries are clearly not the largest contributors to Uruguay, geographic distance has been implicated in juvenile natal homing, whereby green turtles progressively move closer to their source rookeries to forage as they get older, as hypothesized for Brazil (Naro-Maciel et al., 2007, 2012). On average the green turtles foraging further north, closer to Ascension Island for example, are much larger and closer to reproductive age than the small juveniles developing in Uruguay. As noted above, the latter are likely recent recruits from the pelagic area given their size, and are probably not old enough to start the juvenile natal homing process.

Especially for hatchlings and small pelagic turtles, oceanic circulation patterns likely play a significant role in dispersal. Currents can contribute to determining the representation of different rookeries, such as Ascension Island, at various feeding and development regions. For instance, advanced particle modeling research suggests that the South Equatorial Current and its branches, including the Brazil Current, are likely to disperse small turtles (represented by particles) from Ascension and Trindade toward Uruguay and other South American coastal foraging grounds (Putman and Naro-Maciel, 2013).

Within the Southwestern Atlantic, differences among feeding grounds were less significant in comparison to ocean-basin-wide differentiation, suggesting the movement of turtles among populations along the coast. Even so, there is an increased prevalence of haplotype CM-A5 further north in Brazil, reflecting the turtles coming from Suriname and other sites where this primarily northern hemisphere lineage occurs. Thus, although turtles mix along the South American coast, there is still some structure derived from the different source rookeries. This further supports the hypothesis of juvenile natal homing discussed above (Naro-Maciel et al., 2007; 2012).

These patterns were confirmed by other methodologies (e.g. satellite telemetry, mark-recapture) indicating that foraging grounds in the Southwestern Atlantic are shared. Small juveniles recruit from the pelagic then start to move south or north along the coast (Gallo et al., 2006; Santos et al., 2015; Jardim et al., 2016; Vélez-Rubio et al., 2018b), presenting different individual preferences to feeding areas that include seasonal migrations and high fidelity in the region (González Carman et al., 2012; Santos et al., 2015; Vélez-Rubio et al., 2016; Campos and Cardona, 2019). Their occurrence is determined by the seasonal distribution and abundance of prey as well as oceanographic conditions (González Carman et al., 2012; López-Mendilaharsu et al., 2016; Vélez-Rubio et al., 2018b). Most green turtles leave Uruguayan waters by the colder months, performing seasonal migrations, but others remain year round as noted above (Vélez-Rubio et al., 2018a; Velez-Rubio et al., 2022). Additional studies integrating multiple approaches and considering ecological parameters, such as oceanographic data on currents and distribution of food resources, would be valuable for understanding the demographic dynamics in the region. These different sources of information would help determine if the changes in the proportion of recruits are related to intrinsic complex behaviors or transient responses to ecological factors.

The study results support the inclusion, currently being considered, of new Marine Protected Areas spanning green turtle habitats in the National System of Protected Areas in Uruguay, as well as legislative and other measures. Not only will these protect this distinctive group of juvenile green turtles, but also maintain their related ecosystems and food webs. According to Crouse et al. (1987) and Chevallier et al. (2020), the key to improving the prognosis of threatened sea turtle populations lies fundamentally in reducing the mortality of immature stages like juveniles, the stages to which population growth is most sensitive. By increasing the survival of young turtles in their developmental feeding areas such as in Uruguay through protective measures, the future of adult males and females that contribute to the maintenance of the different nesting colonies can also be enhanced. International cooperation is required given the transboundary nature of green turtle populations connected by dispersal and migration. Many of the sites in this study are contained within protected areas, where effectiveness must be maintained. Threats that occur outside of these sites, such as fishery interactions or habitat alteration along reproductive or developmental migrations, must be addressed such as through effective regulation and establishment of new protected areas. In conclusion, this study emphasizes the necessity of implementing protected zones and regulatory measures that encompass the diverse transboundary habitats where these endangered species thrive, with the overarching goal of securing the conservation, and that of their ecosystems, for the long term.