Introduction

Ontogenetic dietary shifts are frequent in aquatic vertebrates because of age-related changes in habitat, body size and characteristics of the feeding organs (e.g., Bethea et al. 2007; Costalago et al. 2012; Vales et al. 2015). As a consequence, species fill different trophic niches during their life cycles, which results into food webs more complex than expected on species richness only (McCann 2012).

Green turtles (Chelonia mydas) exhibit an oceanic–neritic developmental pattern (Bolten 2003). Early research suggested that green turtles shift from an omnivorous diet during their juvenile, oceanic stage of their early life to a primarily herbivorous diet when they are 3–6 years old and recruit to neritic habitats (Bjorndal 1985, 1997; Seminoff et al. 2002). Such an ontogenetic dietary shift to herbivory was described as abrupt and irreversible in some tropical regions (Bjorndal 1997; Reich et al. 2007; Arthur et al. 2008). However, recent research has revealed regional differences in the timing of this process (Cardona et al. 2010), with high levels of omnivory after recruitment (e.g., Cardona et al. 2009, Burkholder et al. 2011, Lemons et al. 2011, Russell et al. 2011, Gonzalez Carman et al. 2013) and the persistence of a carnivorous diet in adults that forage in the open ocean throughout their life (Hatase et al. 2006; Kelez 2011; Parker et al. 2011). Such intraspecific variability can be expected for a species inhabiting a wide diversity of habitats in tropical and warm-temperate waters around the globe (Wallace et al. 2010), as trophic plasticity will ensure survival on a wide range of local conditions (Santos et al. 2015).

Green turtles in the southwestern Atlantic nest primarily on tropical islands (Almeida et al. 2011; Bellini et al. 2013; Weber et al. 2014 and references therein) and forage along the coast of mainland South America. According to stomach content analysis and direct observations, juvenile green turtles from southern Brazil often have omnivorous diets, but the relative abundance of vegetal material increases quickly with turtle size (Nagaoka et al. 2012; Reisser et al. 2013; Morais et al. 2014). Conversely, stomach content and stable isotope analysis indicate that juvenile green turtles captured during the warm season off Argentina feed primarily on gelatinous plankton (Gonzalez Carman et al. 2013). Juvenile green turtles occur in Uruguayan waters year round, but their abundance peaks are in late austral summer and early austral fall (López-Mendilaharsu et al. 2006; Gonzalez Carman et al. 2012; Vélez-Rubio et al. 2013; Martinez Souza 2015), probably because of a seasonal latitudinal migration (Gonzalez Carman et al. 2012). The main foraging grounds of green turtles off Uruguay are found in the rocky outcrops scattered along the east coast (López-Mendilaharsu et al. 2006; Vélez-Rubio et al. 2013), but little is known about their diet. Previous research suggested that green turtles inhabiting warm-temperate regions with wide fluctuations in sea surface temperature (SST) exhibit a delayed ontogenetic dietary shift and high levels of omnivory (Cardona et al. 2009, 2010), a pattern recently confirmed also for the species in Brazil (Santos et al. 2015). A similar situation may occur off Uruguay, where SST ranges annually 9–27 °C (Acha et al. 2004).

Esophageal lavage and stomach contents analysis provide with direct information about diet composition, although they are often highly biased and represent only a short time window (days to weeks), which usually results in an incomplete description of an animal’s diet (Hyslop 1980; Burkholder et al. 2011; Gonzalez Carman et al. 2013). On the other hand, stable isotope analysis of tissues with a low turnover rate integrates dietary information during longer periods (Hobson 1999; Dalerum and Angerbjörn 2005). Furthermore, stable isotope analysis gives information about assimilated prey and not only on those consumed (Post 2002). Dietary inference from stable isotope analysis of animal tissue is possible because animal tissue isotopic composition is ultimately derived from that of its complete diet over time, plus the effect of trophic discrimination from predator to prey (DeNiro and Epstein 1981; Arthur et al. 2008). Stable isotope ratios in sea turtle skin integrate diet during two or three months, and the prey-to-consumer trophic discrimination factor for the epidermis of green turtles has been determined experimentally (Seminoff et al. 2006). This offers an opportunity not only to study the ontogenetic dietary change in green turtles occurring off Uruguay, but also to test hypothesis about the movements of green turtles between southern Brazil and Uruguay, namely that (1) most turtles migrate from southern Brazil to coastal waters off Uruguay to forage during the warm season and (2) some turtles stay year round in Uruguayan waters, possibly overwintering along coastal habitats. In this paper, we used esophageal lavage, stomach contents analysis and stable isotope analysis to test these hypotheses.

Materials and methods

Study area

The Uruguayan coast (710 km length) is part of a complex hydrological system that comprises the frontal zone of the Rio de la Plata estuary and the Atlantic Ocean (Fig. 1). This system is characterized by a strong salinity gradient, affected by seasonal and episodic variations in the outflow of the estuary (Ortega and Martínez 2007; Campos et al. 2008; Horta and Defeo 2012). The estuarine plume flows from the west and mixes with the oceanic waters of the subtropical convergence zone. The cold Falkland/Malvinas current influences this zone in the austral winter and the warm Brazilian current during the austral summer. This results in SST variations greater than 15 °C (range 9–27 °C) throughout the year (Acha et al. 2004). Three zones can be distinguished in Uruguayan coastal waters (Fig. 1b) based on the differences in hydrological characteristics (e.g., Defeo et al. 2009): the inner estuarine zone (Zone 1) and the outer estuarine zone (Zone 2) are influenced directly by the Rio de la Plata discharge, whereas the oceanic zone (Zone 3) has a remarkable but variable oceanic regime. The Uruguayan coast is a succession of sandy beaches of variable extension (2–20 km long) separated by rocky outcrops, scarcer in the estuarine zones than in the oceanic zone. As in waters of the southeastern coast of Brazil, Uruguayan coastal waters host few brown algae and not allow the development of seagrass meadows probably due to a high turbidity (Oliveira 1984; Coll and Oliveira 1999). The most common macroalgae species in shallow rocky bottoms are the chlorophytes Ulva lactuca, Ulva fasciata, Codium cf. decorticatum, Chaetomorpha sp., Cladophora cf vagabunda and the rhodophytes Grateloupia cf filicina, Grateloupia cuneifolia, Chondracanthus teedei, Pterocladiella capillacea, Criptopleura ramosa, Corallina officinalis, Polysiphonia sp., Hypnea musciformis and Rhodymenia sp. (Coll and Oliveira 1999). Mussels (Mytilus edulis) are the most conspicuous invertebrates in the community (Borthagaray and Carranza 2007).

Fig. 1
figure 1

a, Map of the southwestern Atlantic and the study area indicated (solid line); b, map of the Uruguayan coast. Dark circles indicate the stranding locations of turtles included in the study ([Zone 1] inner estuarine zone, [Zone 2] outer estuarine zone and [Zone 3] oceanic zone). The arrows indicate the locations where green turtles were captured for the study (a, Coastal-Marine Protected Area of Cerro Verde e Islas de La Coronilla and b, Coastal-Marine Protected Area of Cabo Polonio)

Full size image

Field data collection and process of samples

From 2003 to 2005, 74 green turtles were captured for esophageal lavage in two Coastal-Marine Protected Areas (CMPA) of Uruguay (Cerro Verde e Islas de La Coronilla and Cabo Polonio, Fig. 1b, Table 1), as part of the long-term study on abundance and habitat use of the species in eastern Uruguay conducted by the nongovernment organization (NGO) Karumbé. Turtles were captured alive over rocky bottoms in very shallow waters. Set nets (nylon monofilament, 50 m length × 3 m depth, 30 cm stretched mesh size) were deployed perpendicular to wave direction and were monitored constantly to avoid drowning of turtles caught. Curved carapace length (CCL, notch to tip) was measured for each turtle using a flexible tape (±0.1 cm). All turtles were tagged with inconel flipper tags (Style # 681, National Band and Tag, Kentucky, USA) before release in the same site of capture. Since marine turtles from different populations may reach sexual maturity at different sizes, we used the minimum size of nesting females from the closest nesting colonies (CCL = 90 cm in Trindade island Brazil; Almeida et al. 2011) to classify the turtles captured as putative juveniles or putative adults. The turtles included in the present study are really small compared to those putative adults, as clarified latter in the results section.

Table 1 Resume of the three methodologies for green turtle diet analysis in the present study
Full size table

Esophageal lavage was used to provide a sample of food resources that the turtles had ingested in the previous hours before being captured (Forbes and Limpus 1993). Methodology for esophageal lavage followed Makowski et al. (2006) protocol. Turtles were placed on their carapace, with the head positioned downward, keeping the mouth open with a pry bar. The water injection tube had 4 mm ID (internal diameter) with a wall thickness of 1.5 mm and 3 m in length. The tube was lubricated with vegetable oil and passed down the esophagus until resistance was met and clean seawater was gently pumped into the esophagus. The lavage was sieved (0.2 mm) to retrieve food items regurgitated by the turtle. All food material obtained was preserved in a 4 % formalin solution with seawater.

Samples for stable isotope analysis were collected from 126 green turtles captured in the same locations in 2012 and 2013 (Fig. 1b, Table 1). Skin biopsies were taken from the dorsal side of the inguinal region of the left hind flipper. The skin area was cleaned for disinfection and to remove ectoparasites. A thin layer of epidermal tissue (~1 cm2) was collected with a scalpel and preserved in a NaCl solution until analysis.

Stomach content samples and skin samples were also collected from the carcasses of dead turtles stranded along the entire Uruguayan coast during the period 2009–2013 (Fig. 1b, Table 1). These stranded turtles were recorded during beach surveys conducted by technicians of the marine turtle stranding network (see Vélez-Rubio et al. 2013). Fifty-two digestive track contents were analyzed and preserved separately in esophagus, stomach and intestine sections. Stomach contents were rinsed and preserved in a 4 % formalin solution in seawater. For skin samples, we used the same protocol as with the captured green turtles (see above). All the turtles were freshly dead as a result of bycatch, and controlled experiments have demonstrated that δ13C and δ15N values in the skin of sea turtles do not change during the first weeks after death (Payo-Payo et al. 2013).

Samples of macroalgae were collected in Uruguay (17 species; in CMPA of Cerro Verde and CMPA of Cabo Polonio) and southern Brazil (5 species; in Santa Catarina). Sampling was designed to collect most of the species more frequently in the gut contents of green turtles off Uruguay (present study) and southern Brazil (Reisser et al. 2013). Furthermore, two species of gelatinous macrozooplankton were collected: Chrysaora lactea from Uruguay and Velella velella from Brazil. All the samples were rinsed with water and frozen until analysis or stored dry as herbarium samples. The description of the variability of the stable isotope values of the turtles and their prey within the δ13C–δ15N space, hereafter called the isospace, was complemented with data from the literature about the squid Loligo sanpaulensis from Brazil and Uruguay (Drago et al. 2015) and the seagrass Halodule wrightii from Brazil (Lazzari 2012).

Gut contents analysis

Dietary items retrieved by esophageal lavage and stomach content analysis were identified to the lowest possible taxonomic level. Dietary item groups were quantified by frequency of occurrence (FO) and relative volume (RV) (Hyslop 1980). For relative volume measure, the entire sample volume and the relative sample volume of each diet group were calculated by means of water displacement in a graduated cylinder. Any item with a relative volume >5 % in at least one sample was considered a major diet component (Garnett et al. 1985; López-Mendilaharsu et al. 2005).

To determine the variation in diet between individual animals, RV of each major prey group was calculated for each individual turtle as follows:

$$\% {\text{RV}}_{i} = \frac{\text{Total volume of diet item in each turtle}}{\text{Total volume of each turtle}} \times 100$$

FO of each dietary species and RV of each major dietary item group (macroalgae, gelatinous macrozooplankton and marine debris) were determined as follows:

$$\begin{aligned} \% {\text{FO }} = \frac{\text{Number of samples containing diet item}}{\text{Total mumber of samples}} \times 100 \hfill \\ \% {\text{RV}} = \frac{\text{Total volume of diet item in all samples}}{\text{Total volume of all samples}} \times 100 \hfill \\ \end{aligned}$$

We estimated the importance of the main dietary groups through the Index of Relative Importance (%IRI, Pinkas et al. 1971) as follows:

$$\begin{aligned} {\text{IRI}} = \% {\text{RV}} \times \% {\text{FO}} \hfill \\ {\text{\% IRI = }}\left( {\frac{IRI}{{\sum\nolimits_{n - 1}^{n} {IRI} }}} \right) \times 100 \hfill \\ \end{aligned}$$

We studied the ontogenetic dietary shift of green turtles through stomach contents analysis (collected during 2009–2013, N = 52) and for esophagus analysis (collected during 2003–2005, N = 74) of specimens classified in three size classes A: CCL < 35 cm; B: 35 ≥ CCL < 45 cm; and C: CCL ≥ 45 cm (Table 1).

Stable isotope analysis

A total of 126 epidermis green turtle samples were available for stable isotope analysis. Samples were dried at 60 °C for three days, grounded to a fine powder, and lipids extracted with a chloroform/methanol (2:1) solution (Bligh and Dyer 1959). Lipids are depleted in 13C in comparison with other molecules, which could bias δ13C values; it is thus desirable to remove lipids prior to stable isotope analysis (DeNiro and Epstein 1978). After the chloroform/methanol treatment, the C:N ratio of turtle epidermis was always lower than 4, thus confirming that lipids had been removed efficiently or were naturally scarce. Macroalgae are often covered with epibionts containing carbonates, which may also bias δ13C values. Accordingly, macroalgae samples were split into two subsamples. One of them was treated with O.5 N hydrochloric acid (HCl), to remove carbonates and with a chloroform:methanol (2:1) solution to remove lipids. However, acidification may modify the δ15N value, so the second bulk subsample was used to determine the δ15N value (Cardona et al. 2012). Samples were weighed into tin cups with a microbalance (0.3 mg for skin and animal prey samples and 0.5 mg for algae samples, because they differ in their N contents), combusted at 1000 °C, and analyzed in continuous flow isotope ratio mass spectrometer (Flash 1112 IRMS Delta C Series EA; Thermo Finnigan) at the Centres Científics i Tecnològics of the Universitat de Barcelona (Spain).

Stable isotope abundances were expressed in δ notation according to the following expression:

$$\delta X = \left[ {\left( {R_{\text{sample}} /R_{\text{standard}} } \right) - 1} \right] \times 10^{3}$$

where X is 13C or 15N, Rsample is the heavy to light isotope ratio of the sample (13C/12C or 15N/14N), respectively, and Rstandard is the heavy to light isotope ratio of the reference standards, which were VPDB (Vienna Pee Dee Belemnite) calcium carbonate for 13C and atmospheric nitrogen (air) for 15N. International isotope secondary standards of known 13C/12C ratios, as given by the IAEA (International Atomic Energy Agency IAEA), namely polyethylene (IAEA CH7, δ13C = –31.8 ‰), graphite (IAEA USGS24, δ13C = –16.1 ‰) and sucrose (IAEA CH6, δ13C = –10.4 ‰), were used for calibration at a precision of 0.2 ‰. For nitrogen, international isotope secondary standards of known 15N/14N ratios, namely (NH4)2SO4 (IAEA N1, δ15N = +0.4  ‰ and IAEA N2, δ15N = +20.3  ‰) and KNO3 (IAEA NO3, δ15N = +4.7  ‰), were used to a precision of 0.3  ‰.

Statistical analysis

Results are reported as mean ± standard deviation, unless otherwise stated.

The relationship between δ13C and δ15N values and turtle size (curved carapace length, CCL), date of capture or stranding (Julian day), distance to the inner part of Rio de la Plata Estuary, and type of record (captured or stranded) were modeled using generalized additive models (GAM) with package mgcv (package version 1.7–5, Wood 2011) in R 2.11 (R Development Core Team 2010). This method is based on the use of nonparametric smoothing functions that allow flexible description of complex species responses to environment (Leathwick et al. 2006). The GAM approach is an extension of the generalized linear model (GLM). This technique enables robust analysis of regression models with nonlinear covariate functional form and a range of nonnormal error terms (Hastie and Tibshirani 1990). The degree of smoothness of model terms was estimated as part of fitting using penalized cubic regression splines. A Gaussian error model was used in the GAM analysis, with a link identity function. Model selection was guided by Akaike information criterion (AIC). For GAMs plots, the y-axis is a relative scale, with a positive value on the plots, indicating a positive effect of that explanatory variable on the dependent variable and a negative y-value indicating a negative effect of that variable. The range of the smoothed function indicates the relative importance of each predictor.

One-way ANOVA and Tukey’s post hoc test were used to compare the stable isotope ratios of macroalgae species simultaneously collected from Brazil and Uruguay and to test for differences in the isotope baseline of both regions.

We used the Bayesian stable isotope mixing model in the Stable Isotope Analysis in R (SIAR) package (Parnell et al. 2010) to estimate the relative contributions of the different dietary items (macroalgae, cnidarians, seagrasses and squids) from Uruguay and Brazil to turtle diet. SIAR assumes that the variability associated with food sources and trophic enrichment is normally distributed (Parnell et al. 2010). To better restrict our model, we used elemental concentrations (%C and  %N) measured for each organic basal source in this study (Coelho Claudino et al. 2013). Although SIAR incorporates uncertainty about diet–tissue isotopic discrimination factors in the form of standard deviation, we conducted a sensitivity analysis running SIAR for green turtle epidermis with diet–tissue isotopic discrimination factors of +0.17 ± 0.03 for δ13C and +2.80 ± 0.11 δ 15N (Seminoff et al. 2006). The criteria to include putative dietary items into the model were that they (1) had been observed in the gut contents in Brazil (Reisser et al. 2013) or Uruguay (this study) and (2) are common species in the Uruguayan coast (Coll and Oliveira 1999; Borthagaray and Carranza 2007). Accordingly, the model included the macroalgae Ulva sp. and Grateloupia sp., the cnidarian Chrysaora lactea and the squid Loligo sanpaulensis from Uruguay and the macroalgae Ulva sp. and Codium decorticatum, the seagrass Halodule wrightii, the cnidarian Velella velella and the squid Loligo sanpualensis from southern Brazil.

Results

Gut content analysis

Seventy-four esophageal lavage samples were collected from healthy green turtles that were captured while feeding off east Uruguay. All turtles were considered juvenile, ranging from 32.6 to 58.4 cm CCL (mean ± SD = 41.6 ± 5.8 cm, N = 74). We identified 18 types of dietary items: 10 species of rhodophytes, 4 species of chlorophytes, 2 mollusk species and 2 cnidarian species (Table 2). Macroalgae occurred in all the samples and represented the bulk of the dietary items collected from the esophagus lavages.

Table 2 Comparison of the main dietary groups found in the digestive track analysis of esophageal lavage (N = 74) and stomach contents (N = 52)
Full size table

All the stranded turtles sampled for stomach contents analysis were considered juvenile, as they ranged 29.8–62.0 cm CCL (mean ± SD curved carapace length = 40.0 ± 7.0 cm, N = 52). These turtles presented an omnivorous diet, with a high occurrence of both macroalgae (FO = 69.2 %) and gelatinous macrozooplankton (FO = 48.1 %). Furthermore, marine debris occurred in 51.9 % of the stomach contents analyzed (Table 2). Relative volume data revealed a similar picture, with macroalgae prevailing on the average diet (RV = 46.6 ± 44.6 %), followed by gelatinous macrozooplankton (23.0 ± 30.8 %) and marine debris (13.3 ± 19.5 %). However, the broad standard deviations associated with these means revealed a large variability of individual diet. The IRI values revealed that the relevance of gelatinous zooplankton decreased and the relevance of macroalgae increased with turtle size (Fig. 2). Actually, macroalgae were the only significant food item (%IRI > 96.3) for turtles larger than 45 cm CCL. Although we identified twenty different species of macroalgae (Table 2), five taxa occurred in more than 35 % of the samples: Ulva sp. (FO = 82.5 %), Grateloupia sp. (FO = 62.5 %), Chondracanthus sp. (FO = 47.5 %), Codium decorticatum (FO = 40.0 %) and Pterocladiella capillacea (FO = 40.0 %).

Fig. 2
figure 2

Ontogenetic dietary change in green turtles. Frequency of occurrence, FO, (white bars), relative volume, RV, (black bars), and Index of Relative Importance, IRI, (gray bars), of the main dietary categories of items (macroalgae, gelatinous macrozooplankton and marine debris) in the stomach contents of stranded green turtles from Uruguayan according to size classes (curved carapace length (CCL): <35 cm (n = 10); ≥35 <45 (n = 29); and ≥45 (n = 8))

Full size image

Stable isotope values

The macroalgae species collected in Brazil and Uruguay (Codium decorticatum, Pterocladiella capillacea and Ulva sp.) differed significantly in their δ15N values (ANOVA, F = 69.31, df = 2, p value < 0.001), with those from Uruguay enriched in 15N as compared with those from Brazil. Conversely, the δ13C values of macroalgae from Brazil and Uruguay were not statistically different (ANOVA, F = 0.95 df = 2, p value = 0.400). Hence, differences in the δ15N baseline existed for both areas.

CCL of green turtles for stable isotope analysis ranged 27.8–66.8 cm (mean ± SD CCL = 39.4 ± 6.4 cm, N = 126). The stable isotope ratios in the epidermis of turtles ranged from −18.2 to −13.6  ‰ for δ13C and from 6.7 to 15.6  ‰ for δ15N (Table 3). Figure 3 shows the distribution of green turtle isotope values within the isospace defined by the potential prey from Uruguay and Brazil, after accounting for the trophic discrimination factor.

Table 3 Stable isotope values (δ13C and δ15N) in the epidermis of green turtles from Uruguay and their potential prey from Uruguay and Brazil
Full size table
Fig. 3
figure 3

Values of δ15N and δ13C in the epidermis of juvenile green turtles (intentionally captured and stranded) from Uruguayan coastal waters, and those of potential prey. Values have been corrected using the trophic discrimination factors from Seminoff et al. (2006). Symbols show the average stable isotope ratios of potential dietary species from Uruguay (black symbols) and Brazil (gray symbols): macroalgae (circles; Co: Codium decorticatum, Gr: Grateulopia sp., Pt: Pterocladiella capillacea, Ul: Ulva sp.), gelatinous macrozooplankton (triangle; Ch: Chrysaora lactea, Ve: Velella velella), cephalopods (Loligo sanpaulensis), seagrass (diamond; Halodule wrightii). a, summer to winter: green turtles classified by size (n = 126); colored symbols represent the stable isotope ratios of turtles for each size group: <35 cm (n = 20, black open circles); 35 ≥ CCL > 40 cm (n = 46, red open triangles); 45 ≥ CCL < 50 cm (n = 42, green tails); >50 cm (n = 6, blue crosses); b, spring: green turtles classified by size (n = 12); colored symbols represent the stable isotope ratios of turtles for each size group: 35 ≥ CCL > 40 cm (n = 5, red open triangles); 45 ≥ CCL < 50 cm (n = 6, green tails); >50 cm (n = 1, blue crosses)

Full size image

Based on the AIC values of competing models, the GAM that best explained the variability of δ13C values in turtle epidermis with the lowest AIC value also included CCL, Julian day and distance from the estuary as the best explanatory variables (Table 4, Fig. 4a). The explanatory power of this model was 45.1 %, and the adjusted R-square was 0.36. There was a nonlinear relationship between epidermis δ13C and CCL: positive for turtles smaller than 35 cm CCL, negative for turtles between 35 and 40 cm CCL and almost linear for turtles larger than 40 cm CCL. Epidermis δ13C values were not affected by Julian day from January to July (summer, fall and early winter) and negatively from August to December (late winter to spring). Finally, the relationship between the distance from the estuary and epidermis δ13C values was nonlinear, with peaks of the δ13C values at approximately 375 and 575 km from the estuary and a decreasing trend in between them.

Table 4 Parameter estimates from the best GAM to describe the variation in values of δ13C of green turtles epidermic tissue as a function of covariate (CCL, day and distance from the estuary)
Full size table
Fig. 4
figure 4

Model terms for the generalized additive model (GAM) of the variation in values of δ13C [A] and δ15N [B] of green turtles epidermic tissue. Estimated smooth functions (solid lines) with 95 % confidence interval (dashed lines) are shown for each explanatory variable: A.1/B.1, curved carapace length (CCL); A.2/B.2: day (in Julian); A.3/B.3, distance from the inner part of the Rio de La Plata Estuary (in Km). Y-axis = fitted function with estimated degrees of freedom in parenthesis; x-axis = variable range with rug plots indicating sampled values. Note the difference in y-axes scale

Full size image

The GAM that best explained the variability of δ15N values in turtle epidermis included CCL, Julian day and distance from the estuary as the best explanatory variables (Table 5, Fig. 4b). The explanatory power of this model was 45.0 %, and the adjusted R-square was 0.41. Epidermis δ15N values increased linearly with CCL and nonlinearly with distance to the estuary and were higher in late winter and spring than the rest of the year (Fig. 4b).

Table 5 Parameter estimates from the best GAM to describe the variation in values of δ 15N of green turtles epidermic tissue as a function of covariate (CCL, day and distance to the estuary)
Full size table

Diet reconstruction with SIAR

According to the GAMs, turtles were divided for diet reconstruction into two seasons (late winter/spring vs. rest of the year) and four size classes (<35 cm CCL, 35–40 cm CCL, 40–50 cm and more than 50 cm). No turtle less than 35 cm CCL was collected in spring.

According to SIAR, turtles less than 50 cm CCL captured/stranded in summer, fall and early winter had omnivorous diets based primarily on Pterocladiella capillacea and Codium decorticatum from Brazil and with a contribution of animal prey up to 30 % of the assimilated nutrients (Fig. 5). Turtles larger than 50 cm CCL also had a macroalgae-based omnivorous diet, but prey from Brazil and Uruguay made similar contributions. The turtles captured or stranded in late winter and spring also had a macroalgae-based omnivorous diet, independently of their carapace length, but with a more balanced contribution of prey from Brazil and Uruguay (Fig. 6).

Fig. 5
figure 5

Feasible contribution of potential prey (macroalgae, gelatinous macrozooplankton, squids and seagrass) from Brazil and Uruguay to the diet of juvenile green turtles sampled in spring according to SIAR (95, 75 and 50 % confidence intervals). Macroalgae: Gr: Grateloupia sp., Ul: Ulva sp, Pt: Pterocladiella capillacea, Co: Codium decorticatum. Gelatinous macrozooplankton: Ve: Velella velella, Cn: Chrysaora lactea. Squids: Loligo sanpaulensis. Seagrasses: Halodule wrightii

Full size image
Fig. 6
figure 6

Feasible contribution of dietary species (macroalgae, gelatinous macrozooplankton, squids and seagrass) from Brazil and Uruguay to the diet of juvenile green turtles sampled in all season (except spring) according to SIAR (95, 75 and 50 % confidence intervals). Turtles were divided into four size groups according to the GAM: ≥ 35 CCL < 40, ≥ 40 CCL < 50, >50 CCL (no turtles of <35 CCL were captured in this season). Macroalgae: Gr: Grateloupia sp., Ul: Ulva sp, Pt: Pterocladiella capillacea, Co: Codium decorticatum. Gelatinous macrozooplankton: Ve: Velella velella, Cn: Chrysaora lactea. Squids: Loligo sanpaulensis. Seagrasses: Halodule wrightii

Full size image

Discussion

The results of the gut contents analysis reported here revealed a macroalgae-based omnivorous diet for the juvenile green turtles occurring year round at the foraging grounds off Uruguay. Green turtles present a rapid, but not abrupt, dietary shift after recruiting to neritic habitats. This shows that green turtles in the region start consuming macroalgae soon after recruiting to neritic habitats, but continue consuming relevant amounts of gelatinous plankton and some squids, even as late juveniles. It should be noted, however, the sharp differences found between the results revealed by esophageal lavage and stomach contents analysis, as hardly any animal prey was obtained from the esophageal lavages. A possible explanation is that turtles were captured while grazing on rocky outcrops, and hence esophageal lavages revealed only the prey consumed immediately before the capture. Conversely, stomach contents probably integrate diet over previous days and hence informed also about the diet consumed while foraging in other habitats. On the other hand, samples from esophageal lavages and stomach contents were dominated by Ulva sp., Chondracanthus spp., Grateloupia spp. and Polysiphonia spp., precisely the four taxa that dominate the macroalgae community along the Uruguayan coastline (Vélez-Rubio et al. In prep.). This suggests little selectivity on macroalgae by green turtles, but further research based on controlled experiments is needed before concluding that green turtles graze in a nonselective way.

Cephalopod beaks occurred in 31.5 % of the stomach contents (Vélez-Rubio et al. 2015) analyzed in this study, but they did not necessarily reveal the recent consumption of cephalopods by neritic green turtles. Cephalopod beaks, composed of hardly digestible chitin, are known to accumulate into the gut of marine vertebrates for years (Hernández-García 1995; Tomás et al. 2001; Xavier et al. 2005), which probably explain why most of the beaks recovered from the stomach and intestine of neritic green turtles corresponded to oceanic species (Vélez-Rubio et al. 2015). This conclusion is further supported by the modest contribution of squids to the diet of green turtles according to SIAR and the absence of fresh squids into the stomachs of turtles.

Nevertheless, dietary reconstructions based on stable isotope faced serious limitations due to the complex variations in the isotope baseline within the region. Previous research has demonstrated the existence of a strong latitudinal gradient in the δ15N baseline across Río de la Plata estuary, with species from southern Brazil typically depleted in 15N as compared to those in Uruguay and northern Argentina (Vales et al. 2014; Drago et al. 2015). Southern Brazil remains away from the Río de la Plata plume because the confluence of the southward Brazil current and the northward Falkland/Malvinas current causes an offshore flux and hence drifts away the plume (Longhurst 1998). Nevertheless, such δ15N gradient is stronger than any regional onshore/offshore gradient and may be related to the sewage input from Buenos Aires and Montevideo (Acha et al. 2004). Such latitudinal gradient had a strong influence on the δ15N values of the potential prey analyzed here, and we also detected a weak, but significant, effect of sampling location along the estuarine plume of the δ15N values of green turtles epidermis. Such variability in the δ15N baseline certainly compromised the performance of SIAR, because the specificity with which mixing models can decipher diet trends is only as powerful as the specificity in the stable isotope values across the various potential diet items. A common solution in this situation is to cluster species with statistically similar stable isotope values, as far as the resulting groups are ecologically meaningful (e.g., Cardona et al. 2012; Zenteno et al. 2015). However, this is hard to undertake in the present situation, because the two species with the largest overall, both in their δ13C and δ15N values, were a jellyfish from Brazil (Velella velella) and a macroalgae from Uruguay (Grateulopia sp.).

As every technique available for dietary studies is somewhat biased, diet description is highly dependent on the method used, with SIA usually revealing a larger contribution of animal prey to the diet of green turtles than gut contents analysis (Table 6). Nevertheless, when the results from this study are combined with previous research conducted in southern Brazil (Nagaoka et al. 2012; Reisser et al., 2013; Morais et al. 2014; Santos et al. 2015) and Argentina (Gonzalez Carman et al. 2012, 2013), a clear regional pattern on the ontogenetic dietary shift of green turtles emerges. Green turtles smaller than 45 cm CCL inhabiting the Atlantic south to latitude 27ºS have an omnivorous diet, based on plant material and with gelatinous macrozooplankton as the main animal prey. On the other hand, green turtles larger than 45 cm CCL are primarily herbivores, but they still consume some animal prey. Only in areas with a scarcity of submerged macrophytes, as in the turbid plume of the Río de La Plata estuary, neritic juvenile green turtles may resume a mainly carnivorous diet (Gonzalez Carman et al. 2013). Omnivorous diet and increasing consumption of plant material with carapace length have been reported for green turtles inhabiting other warm-temperate regions (Cardona et al. 2009, 2010; Lemons et al. 2011) and even from some tropical regions (Amorocho and Reina 2007; Russell et al. 2011). As a consequence, the fast shift to herbivory following recruitment, once thought to be typical of green turtles, apparently happens in only some tropical regions characterized by extensive seagrass meadows (Bjorndal 1997; Reich et al. 2007; Arthur et al. 2008).

Table 6 Comparisons of juvenile and adult green turtle diet studies from different populations worldwide
Full size table

Furthermore, the latitudinal variation in the isotope baseline offers a good opportunity to trace the movements of turtles. Stomach contents analysis revealed a decrease in the consumption of animal prey as the turtles grow, thus ruling out the hypothesis that increasing δ15N values in the epidermis of larger turtles is because of a higher trophic level. Instead, the positive relationship between δ15N values and turtle size is better explained by a longer residence of larger turtles in Uruguay, where macroalgae, gelatinous macrozooplankton and squids are enriched in 15N as compared with those from Brazil (Vales et al. 2014; Drago et al. 2015; this study). The only possible confounding factor here would be that the amount of gelatinous zooplankton in the stomach contents of the largest turtles may had been underestimated because of its faster digestion, especially in large-size dead stranded individuals.

It has to be kept in mind that stable isotope ratios in turtle skin integrate dietary information during several months, although the exact turnover rate has been assessed experimentally for only smaller juveniles (Reich et al. 2008) and hence the time window integrated by the skin of larger juveniles is unknown. In this scenario, a strong isotopic signal of Brazilian macrophytes in the epidermis of the turtles captured off Uruguay from summer to winter indicates that most of these turtles had spent only a short time (weeks to months) there prior to capture. This interpretation is consistent with previous satellite telemetry and mark-recapture studies conducted in Uruguay and reflecting seasonal movements along the coastal waters of Brazil, Uruguay and Argentina (López-Mendilaharsu et al. 2006; Gonzalez Carman et al. 2012; Martinez Souza 2015). Nevertheless, the stable isotope ratios in the skin of green turtles occurring off Uruguay in spring clearly show they have been shifting frequently foraging grounds off Uruguay and southern Brazil and hence are not recent immigrants. Compared with the rest of the turtle groups considered here, turtles sampled in spring exploited other macroalgae species, as Polysiphonia sp. and some Corallinaceae species, probably because of the dominance of this species in the macroalgae community during winter along the Uruguayan coast (G V–R scuba diving observation). Due to the low SST values recorded during the coldest months of the year, a great proportion of the green turtle aggregation seems to migrate north in austral fall, although some turtles remain in the area overwintering in bays, estuaries (e.g., Valizas river, Andreoni channel) or harbors (e.g., Port of La Paloma), where the SST does not drop below 12–15 °C (Vélez-Rubio et al. 2013; Martinez Souza 2015). These turtles probably exhibit periods of brumation or winter dormancy (Witherington and Ehrhart 1989), as a strategy to survive with these cold temperatures (López-Mendilaharsu et al. 2006; Martinez Souza 2015).

Therefore, according to the present study and other studies conducted in the area (Gonzalez Carman et al. 2012; Martinez Souza 2015), we propose that most of the green turtles occurring off Uruguay from summer to winter had dispersed recently from southern Brazil and spend only a few weeks or months off Uruguay and that only a few turtles overwinter in Uruguayan waters along coastal habitats. Finally, some green turtles may overwinter offshore, as noted by some satellite tracked green turtles from Argentina (Gonzalez Carman et al. 2012). The reason why most juvenile green turtles spend only some weeks foraging off Uruguay might be related to the latitudinal pattern of SST. The activity of the microbial flora digesting the plant material ingested by young green turtles is temperature dependent (Bjorndal 1980); hence, remaining for longer periods in Uruguayan waters offers no advantage for them than if they can forage into the warmer waters of southern Brazil. Only those turtles resuming a carnivorous diet may gain any benefits from spending longer periods at high latitude (Gonzalez Carman et al. 2013).

In summary, juvenile green turtles occurring off Uruguay are short-term migrants with a macroalgae-based omnivorous diet, with a decreasing contribution of animal prey to the diet when turtles reach 45 cm CCL. Diet variation reflects regional seasonal migrations in the southwestern Atlantic; hence, cooperation between neighboring countries is mandatory for conservation of this endangered species.