Introduction

Catches of tunas and billfishes have increased dramatically the past 20 years in the western Indian Ocean, very likely altering the structure and functioning of the ecosystems through trophic cascades (Essington et al. 2002; FAO 2006). Concomitantly to these top-down controls, bottom-up effects, via environmental and climatic changes, are also controlling abundance and spatial dynamics of top predators that depend on food availability (Cury et al. 2003; Frank et al. 2006; Frederiksen et al. 2006). Therefore, studies based on trophic ecology and movements of top predators are useful to assess the impact of fisheries and climate on marine resources, and to provide basic elements for an ecosystem approach to fisheries management (Sinclair and Valdimarsson 2003; FAO 2003). Unlike the Pacific and Atlantic Oceans, few studies have investigated the diet of thunniform fishes from stomach content analyses in the Indian Ocean (Watanabe 1960; Kornilova 1981; Roger 1994; Maldeniya 1996; Potier et al. 2004, 2007). Furthermore, stomach content analyses only reflect the composition of the most recent meal and limit our ability to address spatial and temporal variability of feeding behaviours.

Small- and large-scale movements of top predators are now assessed using conventional and electronic tagging programmes combined with catch statistics (e.g. Block et al. 2005). A tagging programme, the Regional Tuna Tagging Programme of the Indian Ocean Tuna Commission (RTTP-IO, http://www.rttp-io.org/en/about/) is underway in the Indian Ocean. So far, few tag returns are available and catch statistics themselves do not reflect the real movement patterns (Hilborn and Walters 1992; Walters 2003). The knowledge of the spatial dynamics of tuna in the Indian Ocean is therefore still minimal.

Additional information on dietary sources, trophic levels, feeding strategies or movement patterns of migratory species can be obtained from stable isotope analyses of animal tissues (Rau et al. 1983; Fry 1988; Kelly 2000; Rubenstein and Hobson 2004; Cherel and Hobson 2005, 2007). The stable isotope composition of an organism depends on its diet, its trophic level, but also on the isotopic signature at the base of the food web (DeNiro and Epstein 1978, 1981; Post 2002; Fry 2006). δ15N measurements mainly serve as indicators of consumer’s trophic position, while δ13C values are used to determine the sources of primary production, inshore versus offshore or pelagic versus benthic contribution to food intake (Hobson 1999). Indeed, different oceanic processes affect isotopic baselines of δ15N and δ13C in marine pelagic food webs (Rau et al. 1982; Altabet et al. 1995; Gruber and Sarmiento 1997; Lourey et al. 2003). δ13C values of phytoplankton decrease from low to high latitudes (Lourey et al. 2003) while δ15N of particulate organic matter (POM) is driven by nutrient utilization and the nitrogen source used by primary producers (nitrate, ammonium, N2 gas; Wada and Hattory 1991). The resulting spatial and temporal variability in the isotopic baseline has been shown to be incorporated and conserved through several trophic levels (up to pelagic consumers) across ocean basins (Takai et al. 2000; Wallace et al. 2006) or within a region of a single basin (Schell et al. 1989; Lesage et al. 2001; Quillfeldt et al. 2005; Cherel and Hobson 2005, 2007; Cherel et al. 2005). Hobson (1999) illustrated this approach by the new maxim “you are what you swim in” that complements the well-known dogma of stable isotope ecology “you are what you eat” (DeNiro and Epstein 1976). Consequently, the stable isotope ratios of animal tissues have the potential of characterizing the isotopically distinct regions crossed by migrating fish and investigating their feeding ecology. Graham et al. (2006) successfully applied this approach to yellowfin tunas (Thunnus albacares) of the Pacific Ocean. One objective of the present study is therefore to investigate the relationships between the isotopic signature of yellowfin tuna and swordfish (Xiphias gladius) versus latitude in the western Indian Ocean, and their relative degree of residency. Indeed, for migratory species, the variability of the isotopic signature in their tissue is supposed to be low if the migration rate is quicker than isotopic tissue turnover. Conversely, for more resident species, stable isotope ratios of tissues would reflect the isotopic patterns at the base of the food web (Fry 2006; Graham et al. 2006; Popp et al. 2007). A second objective was to document the feeding ecology of yellowfin tuna and swordfish, and potential ontogenetic effects on their trophic status. Ontogenetic shifts in tuna and swordfish feeding behaviour are also expected as larger fish usually expand their feeding habitat and exploit a larger size range of prey in the environment (Ménard et al. 2006; Young et al. 2006; Graham et al. 2007).

Inter-specific, spatial and ontogenetic differences in the stable isotope composition of muscle tissues were thus investigated for these two migratory top predators of the western Indian Ocean. Linear models and linear mixed-effects models were developed to test and disentangle potential latitudinal and body length effects on the stable isotope values (δ15N and δ13C) of each species. According to the model predictions, the trophic positions of individuals of different body lengths caught in different oceanic regions can then be compared. It is indeed a prerequisite to understand these geographical and ontogenetic variations before determining the trophic position of these species. In this paper, we implement the isotope approach to gain insight into the feeding ecologies and movement patterns on the studied predators, the first initiative of this nature in the Indian Ocean.

Materials and methods

Sample collection

Fishes were caught by industrial purse seiners with scientific observers onboard; a 20-m research longliner “Amitié” of the Seychelles Fishing Authority, and the French 24-m industrial longliner “Cap Morgane”. Samples were collected from 2001 to 2004 in the western Indian Ocean along a latitudinal gradient (23°S to 5°N, Fig. 1). A total of 245 yellowfin tuna (T. albacares) and 136 swordfish (X. gladius) from various body lengths were collected during these cruises. Fork length (FL) ranged from 40 to 160 cm for yellowfin tuna (mean = 103 cm and median = 108 cm) and Lower Jaw Fork length (LJFL) for swordfish ranged from 68 to 225 cm (mean = 135 cm and median = 133 cm). LJFL is a reliable measure of swordfish body length that allows comparisons with tunas by reducing the bias due to the bill. Table 1 displays all the sample characteristics. White muscle tissues from the dorsal region before the first dorsal fin were collected onboard from freshly caught fishes and were stored frozen at −20°C until processing.

Fig. 1
figure 1

Muscle sample collection sites of swordfish (open circles) and yellowfin tuna (crosses) in the western Indian Ocean

Full size image
Table 1 Sample characteristics of yellowfin tuna and swordfish
Full size table

Sample preparation and analysis

Samples were freeze dried and ground to a fine powder. Lipid extraction was performed using 20 ml of cyclohexane on powder aliquots of about 1 g, and the lipid-extracted sample was dried at 60°C before processing. One milligram sample was then placed into 8 × 5 mm2 tin cups for CF-IRMS analysis, using a Europea Scientific ANCA-NT 20-20 Stable Isotope Analyser with ANCA-NT Solid/Liquid Preparation Module (PDZ Europa Ltd., Crewz, UK). Replicate measurements of internal laboratory standards indicate measurement errors of ±0.2‰ for δ13C and δ15N. Triplicate analyses performed on some samples confirmed that analytical reproducibility was very good (0.2‰ maximum variation). Isotopic ratios are expressed in the conventional δ notation as parts per thousand (‰) deviation from the international standards: atmospheric nitrogen for δ15N and VPDB Belemnite for δ13C:

$$ \delta X = {\left( {R_{{{\text{sample}}}} {\text{/}}R_{{{\text{standard}}}} - {\text{1}}} \right)} \times {\text{1,000,}} $$

where X is 15N or 13C and R the corresponding ratio 15N/14N or 13C/12C.

Lipid content in tuna and swordfish muscles may be high (>50%, unpublished data; Estrada et al. 2005). As lipids are highly depleted in 13C (Tieszen et al. 1983), C/N mass ratios were used to check the lipid extraction process. δ13C outliers (15 for yellowfin tuna and 33 for swordfish) were removed from the analyses according to the distributions of C/N values for each species. We estimated the corresponding thresholds to suppress any relationships between δ13C and C/N mass ratios (3.62 and 3.70 for yellowfin tuna and swordfish, respectively). The resulting distributions of C/N were normal (mean ± standard deviation of 3.36 ± 0.11 and 3.50 ± 0.11 for yellowfin tuna and swordfish, respectively) and the sampling ranges of latitude and body length were not reduced for both species.

Statistical analysis

Linear regressions were used on the δ15N and δ13C data for each species to test the covariates of interest, i.e. latitude and body length. However, all the individuals of one species are not independent and the sampling scheme is clearly unbalanced. The individuals can be grouped according to different factors (e.g. cruise, year, season, etc.; Table 1). We use the two main seasons of the monsoon system to group the individuals of each species caught during the cruises carried out from 2001 to 2004. Indeed, the ocean circulation in the West Indian Ocean is strongly related to the wind monsoon regime, which in turn strongly affects biological productivity (Tomczak and Godfrey 1994; Longhurst 1998; Schott and McCreary 2001). The Northeast (NE) monsoon becomes established in boreal winter (December–March), and is characterized by winds blowing from the Asian continent to the equatorial zone. The Southwest (SW) monsoon becomes established in boreal summer (June–September), and is characterized by a reversal of the winds in the northern Indian Ocean. Therefore, each observation can be classified according to the season on which it was made (NE or SW monsoon, Table 1). The five cruises which took place during the inter-monsoons are relocated in their nearest monsoon: October and May in the SW monsoon, and November in the NE monsoon. A classification based on the four seasons (including the two additional inter-monsoon seasons) was also tested but not retained as the various models fit to the δ15N and δ13C values gave similar results to the monsoon only scenario. To test this grouping, the seasonal effect was treated as random variations around a population mean, and the body length and the latitude were assessed as two fixed continuous covariates, using linear mixed-effects models (lme models; Pinheiro and Bates 2000). These lme models combine a random-effects analysis of variance model (variability amongst seasons) with a linear regression model. Lme models were tested against simple linear regression models using likelihood ratio tests. Population predicted values (obtained by setting the random effects to zero in the lme models) were used to compare latitude and body length effects for yellowfin tuna, for swordfish, and between the two species. All the computations and tests were performed on S-Plus (Insightful 2005).

Results

Muscle δ15N and δ13C values for yellowfin tuna and swordfish plotted versus body length, latitude and season (NE and SW monsoons) are shown in Figs. 2 and 3. The δ15N values for yellowfin tuna ranged from 10.2 to 15.2‰ and from 11.8 to 16.2‰ for swordfish (Fig. 2). The δ13C values for yellowfin tuna ranged from −17.4 to −15.2‰ and from −17.4 to −15.0‰ for swordfish (Fig. 3). The range of variation for the δ13C values is therefore more reduced than the δ15N range (≈2‰ vs. 5‰).

Fig. 2
figure 2

Muscle δ15N values from yellowfin tuna (a) and swordfish (b) plotted versus body length or latitude according to the considered season: SW monsoon (full symbols) and NE monsoon (open symbols). Simple linear regression for δ15N versus body length was not significant for yellowfin tuna (F [1,243] = 0.48, P = 0.49 and r 2 = 0.002) and significant for swordfish (F [1,134] = 58.47, P < 0.0001 and r 2 = 0.30). Both regressions were significant for δ15N versus latitude (F [1,243] = 81.11, P < 0.0001 and r 2 = 0.25 for yellowfin tuna; F [1,134] = 12.06, P < 0.001 and r 2 = 0.082 for swordfish)

Full size image
Fig. 3
figure 3

Muscle δ13C values from yellowfin tuna (a) and swordfish (b) plotted versus body length or latitude according to the considered season: SW monsoon (full symbols) and NE monsoon (open symbols). Simple linear regressions were significant for δ13C versus body length (F [1,228] = 4.32, P = 0.039 and r 2 = 0.02 for yellowfin tuna; F [1,101] = 17.59, P < 0.0001 and r 2 = 0.15 for swordfish), and for δ13C versus latitude (F [1,228] = 25.87, P = 0.0001 and r 2 = 0.10 for yellowfin tuna; F [1,101] = 24.38, P < 0.0001 and r 2 = 0.19 for swordfish)

Full size image

δ15N values were significantly different between seasons for both species (Kruskal–Wallis χ 2 = 13.18, P < 0.001 for yellowfin tuna; Kruskal–Wallis χ 2 = 14.27, P < 0.001 for swordfish). For δ13C, the difference between seasons was significant for swordfish (Kruskal–Wallis χ 2 = 4.81, P = 0.03), but not for yellowfin tuna (Kruskal–Wallis χ 2 = 0.26, P = 0.61). Linear regression models with latitude and body length added sequentially were significant (P = 0.026 for δ15N and P < 0.001 for δ13C for yellowfin tuna; P < 0.001 for δ15N and δ13C for swordfish). However, deviations from the models suggest that other models might be appropriate. For example, Fig. 2 reveals that intercepts of the models for δ15N may differ between seasons.

Linear mixed-effects models (lme models) were fitted to the muscle δ15N values grouped by season. The most parsimonious model for swordfish was obtained with both latitude and body length as fixed-effect covariates (P < 0.0001). For yellowfin tuna, compared with a lme model containing latitude only, the fit was only marginally improved when body length was added (P = 0.092). This P-value evidences a lack of significance of body length for yellowfin tuna δ15N values at a significance level of 5%, a result already exhibited in Fig. 2. According to the likelihood ratio test, lme models for both species provided a much better description of the δ15N data than the linear regression models did (P < 0.0007 and 0.0001 for yellowfin tuna and swordfish, respectively). Figure 4 displays the predicted lines for each season (using the estimated random effects) and the original data for model checking. These plots exhibit the large variability of the δ15N values for each season, and confirm that latitude was the strongest linear fixed-effect for yellowfin tuna, while body length was the most significant fixed-effect for swordfish. For both species, the random effects were associated with the intercepts only. Therefore within-season intercept estimates for δ15N data were different, while slopes were identical. Interestingly, within-season intercepts exhibited a similar pattern whatever the species: the NE monsoon intercept was always greater than the SW monsoon intercept (difference estimated at 0.37 and 0.67‰ for yellowfin tuna and swordfish, respectively). The assumption of normality and independence for the random effect and the residuals were graphically assessed (not shown). On the other hand, lme model fits to the fish muscle δ13C values were not significantly better than the linear regression models (P > 0.50 for yellowfin tuna and P = 0.33 for swordfish).

Fig. 4
figure 4

Within-season predicted muscle δ15N values (solid lines) from the linear mixed-effects models for yellowfin tuna (a) and swordfish (b) plotted versus body length or latitude. The original data (open circles) were superimposed on the predicted lines within each season, SW monsoon (left side) and NE monsoon (right side)

Full size image

Body length data was well balanced along latitude for swordfish, while the distribution for yellowfin tuna displayed a surplus of small individuals in the high latitudes and a deficit of small individuals in the low latitudes (figures not shown). However, linear mixed-effects models provide a flexible and powerful tool for analysing unbalanced grouped data. In addition, the effect of body length on stable isotope ratios was weak for yellowfin tuna. Table 2 lists the coefficients and standard errors estimated by the most parsimonious models fit to the δ15N and δ13C data.

Table 2 Estimated intercepts and slopes (with standard errors) and within-season standard deviations σseason for the random effects for linear mixed-effects models fit to the δ15N values of yellowfin tuna and swordfish, and estimated intercepts and slopes (with standard errors) for linear regression models fit to the δ13C values of yellowfin tuna and swordfish
Full size table

Discussion

Our results provide evidence for a relationship between latitude and body length with δ15N and δ13C values of yellowfin tuna and swordfish. The use of body mass instead of body length may partly explain the observed variability (Fig. 4). Indeed, two fish of the same length can have different masses, due to different overall nutritional states, with possible consequences on the δ13C and δ15N values. However, due to the sampling conditions, we were only able to record body lengths. Linear mixed-effects models were used for δ15N data and provided identical slopes with different intercepts between the two seasons for both latitude and body length. Simple linear models with no seasonal effect were selected for δ13C data. Both body length and latitude influence δ15N values of the two species more strongly than δ13C values. Model predictions at the population level allow us to analyse these effects separately. Figure 5a illustrates δ15N and δ13C predicted values of yellowfin tuna and swordfish as a function of varying body lengths for fish caught at different latitudes (−10° and 0°). In the same way, Fig. 5b represents δ15N and δ13C predicted values for fish of different lengths (80 and 160 cm) as a function of a latitudinal gradient. We now examine the hypotheses supported by our results with respect to the trophic ecology of yellowfin tuna and swordfish, and to the oceanic processes affecting the isotopic baseline of δ15N and δ13C in marine food webs.

Fig. 5
figure 5

Population predicted muscle δ15N and δ13C values from the linear models (linear mixed-effects models for δ15N and simple linear models for δ13C) for swordfish (SWO, dashed line) and yellowfin tuna (YFT, full line) plotted versus body length a considering two different fixed latitudes (0°N or l0°S); and plotted versus latitude b for two different body lengths (80 or 160 cm)

Full size image

Latitudinal effect

The range of variation for the δ15N values is 2.4‰ for swordfish and 1.1‰ for yellowfin tuna along a latitudinal gradient of about 30° (Fig. 5b), whereas those variations are <1‰ for δ13C (0.8 and 0.7‰ for yellowfin tuna and swordfish, respectively). Three hypotheses can be put forward to explain the δ15N and δ13C increase from the Mozambique Channel to the Somali basin: (1) dietary changes, (2) starvation and (3) a shift in δ15N baseline. Trophic level differences or starvation of the northern individuals seem highly unlikely given the regularity of the observed variations, and are not supported by any ecological data. We argue that this spatial pattern results from different oceanic processes at the base of the food web that vary by region in the western Indian Ocean, and that are conserved through different trophic levels up to top predators.

Particulate organic matter δ15N and δ13C isotopic values are not available in the western Indian Ocean to document a latitudinal pattern at the base of the food chain. However, knowledge of nitrogen dynamics in several zones of the western Indian Ocean suggests that differences in δ15N values of POM might occur. In particular, the Somali region should have higher δ15N baseline values compared to Mozambique Channel. This is because the Arabian Sea is a major area of anoxia (Gruber and Sarmiento 1997), and is characterized by intensive denitrification that leads to an accumulation of isotopically enriched nitrate in subsurface waters (Gaye-Haake et al. 2005; Naqvi et al. 2006). Conversely, different tracers and biological indicators in the surface waters of the South Indian subtropical gyre (around 20°S, 57°E) have shown a prevailing N2 fixation, known to generate lower δ15N values for phytoplankton (Capone and Carpenter 1982; Carpenter 1983; Gruber and Sarmiento 1997). Gruber and Sarmiento (1997) also found a latitudinal gradient between the Arabian Sea and 25°S in the western Indian Ocean with decreasing denitrification and increasing N2 fixation from North to South. We did not sample the core of the South Indian subtropical gyre, nor the Arabian Sea; however, the northern and southern edges of our sampling zone are connected through the current system to the two most contrasted zones of denitrification (the Arabian Sea) and N2 fixation (subtropical gyre, Tomczak and Godfrey 1994; Schott and McCreary 2001; Davis 2005). Therefore, the δ15N baseline values of the Somali region are likely to be strongly influenced by the Arabian Sea while those of the Mozambique Channel are under the influence of the subtropical gyre (Davis 2005).

In several studies using stable isotopes to delineate feeding locations of marine predators of the southern Ocean, δ13C displayed strong variations with latitude, whereas δ15N values responded mainly to trophic enrichment (Quillfeldt et al. 2005; Cherel and Hobson 2005, 2007). Indeed, in the southern hemisphere, the geographical δ13C gradient in POM of surface waters is well defined, and ranges from high δ13C values in warm subtropical waters in the North, to low values in cold Antarctic waters in the South (François et al. 1993; Trull and Armand 2001), with abrupt changes at fronts (Subtropical, Subantarctic and Polar fronts). Gradients in terms of sea temperature are much more reduced in our sampling zone (annual mean surface temperatures vary from 25.5 to 28.5°C; see Fig. 2.5 in Tomczak and Godfrey 1994), which could explain the weak δ13C variations in the western tropical Indian Ocean revealed by our study. Broad δ15N gradients, as observed in this study, have been found in other open ocean regions. Comparing leatherback turtle δ15N signatures in the Eastern Tropical Pacific and in the Atlantic Ocean, Wallace et al. (2006) found inter-basin differences of 5‰ between denitrification and N2 fixation zones. In the Equatorial Pacific, Graham et al. (2006) have shown basin-wide δ15N differences as high as 11‰ in tuna muscle tissue. In the Indian Ocean, we detected a much lower intra-basin difference (i.e. maximum of 2.4‰ for yellowfin tuna), most probably because our samples did not cover the core areas mentioned earlier. Consequently, even if yellowfin tuna and swordfish are migrating between the two contrasted regions (the Arabian Sea versus the subtropical gyre), we are probably observing a diluted effect of this general intra-basin difference. Furthermore, isotope values of muscle tissues of these species might never reflect the isotope values of their recent diet plus the corresponding trophic enrichment, because of their continuous movement, their opportunistic feeding behaviour and their muscle tissue isotopic turnover (half-life around 50 days, B. S. Graham, unpublished data). All these reasons generate variability in the data, reduce the effect of the latitudinal gradient, but do not challenge its occurrence.

The observed conservation of the δ15N baseline characteristics in these top predators has several implications. First, these species are known to be highly migratory and thus such a gradient in the data is not expected. Indeed, these data suggest that yellowfin tuna and swordfish are relatively resident species at the temporal scale of their tissue isotopic turnover, i.e. 3 months for yellowfin tuna (B. S. Graham, unpublished data). However, this does not preclude large basin-wide movement patterns at the temporal scale of their life time. Furthermore, the coexistence of migrating fish among more resident fish might occur and explain the rather high intra-season variability found in our study. Interestingly, the δ15N predictions along the latitudinal gradient varied two times more for yellowfin tuna than for swordfish (differences of 2.4‰ and of 1.2‰, respectively). This can be interpreted in different ways: (1) yellowfin tuna are more resident than swordfish, (2) swordfish have a slower turnover rate or tissue growth than yellowfin tuna, (3) swordfish do not migrate to highly 15N depleted areas such as the Arabian Sea, which the very low-catch records of this species in this region suggests (Fonteneau 1997). The third hypothesis seems the most plausible given our present knowledge; however, we cannot preclude a mixed influence of the three hypotheses.

Seasonal effect

In the mixed-effects models implemented in this paper for δ15N values, the seasonal effect is random and induced by the grouping of the data. Only intercepts differ between NE and SW monsoon predictions: compared to the SW monsoon, the NE monsoon intercepts are 0.36 and 0.67‰ higher for yellowfin tuna and swordfish, respectively. During the NE monsoon, the waters of the Arabian sea are advected to the South and invade the Somali basin (where part of our data collection was undertaken), potentially increasing the δ15N values of the baseline of this zone compared to SW monsoon (Davis 2005). Conversely, during the SW monsoon, there is a broad equatorward flow of waters from the South Equatorial Current (SEC) along the East African Coast reaching the Somali region (Tomczak and Godfrey 1994; Schott and McCreary 2001; Davis 2005). Further studies involving measurements of the δ15N of the POM over an annual cycle should be investigated to shed some light on the seasonal variations that may occur in the western Indian Ocean.

Interestingly, the seasonal effect is not significant for δ13C values. We believe that seasonal changes in the monsoon regime do not have strong consequences on the carbon isotopes ratios in the sampled areas. In addition, the δ13C ranges we observed in our data were low compared to the intra-individual variability. Our results suggest that muscle δ13C values of fish in these open sea ecosystems of the western Indian Ocean might not be useful to document seasonal changes, to delineate feeding locations or to track fish movement. This is in contrast to studies conducted in the southern Indian Ocean where δ13C has been shown to be a useful tool (Cherel and Hobson 2005, 2007).

Body length effect

Figure 5a indicates changes in the δ15N and δ13C model predictions along a gradient of body length, for fish caught at two different latitudes. In each case, the δ15N values exhibited a stronger body length effect for swordfish than for yellowfin tuna. The isotopic difference between large (220 cm) and small (80 cm) swordfish was about 2.3‰, whereas it was <0.5‰ for yellowfin tuna of 40–160 cm. The same pattern is supported by δ13C values, but isotopic differences are much more reduced (0.8 and 0.4‰ for swordfish and yellowfin tuna, respectively). The lower differences for δ13C are not surprising given that δ15N is known to increase much more with trophic levels than δ13C (DeNiro and Epstein 1981). Body size is indeed known to play a crucial role in predator–prey interactions (Sheldon et al. 1977; Cury et al. 2003). Analyses of stomach contents and nitrogen isotope ratios conducted on fish communities in different marine ecosystems have shown that prey size and trophic level generally increase with increasing predatory body size (Scharf et al. 2000; Jennings et al. 2002; Estrada et al. 2006). In open-sea ecosystems, few studies have yet dealt specifically with size-based predation. Ménard et al. (2006) have shown that the maximum size of the prey consumed by yellowfin tunas tends to increase with tuna body length, but that large yellowfin tunas continue to consume small prey in great proportions. In addition, both adults and juveniles of yellowfin tuna generally show only minor differences in depth distributions (Brill et al. 1999, 2005). Yellowfin tuna spend most of their time in the surface layer, even if some exceptional deep dives have been evidenced by one archival tag (Dagorn et al. 2006). The diet of yellowfin tunas is then mainly composed of organisms present in the upper 100 m (Moteki et al. 2001; Bertrand et al. 2002; Potier et al. 2004, 2007), with no major ontogenetic changes (Ménard et al. 2006). An outstanding diet shift was revealed by Graham et al. (2007) who studied yellowfin tunas collected from nearshore Fish Aggregating Devices around Hawaii. Tunas ranged from 23.5 to 154.0 cm FL and the ontogenetic change concerned small yellowfin tunas between 45 and 50 cm FL. Therefore, we conclude that the body length of yellowfin tuna does not have a strong impact on its δ15N values. On the other hand, large swordfish mainly consume cephalopods (Hernandez-Garcia 1995; Markaida and Hochberg 2005; Young et al. 2006; Potier et al. 2007), while smaller swordfish have a diet focused on mesopelagic fish such as myctophids (Young et al. 2006; Potier et al. 2007). This shift in the dominant prey items has consequences on the δ15N values because mesopelagic fish such as myctophids and paralepidids have shown lower mean δ15N values than cephalopods (Young et al. 2006). In addition, swordfish can catch larger prey specimens as they grow, due to an increase of mouth-gape size, chasing predation and diving capability (Carey and Robinson 1981). Therefore, we fully confirm that body length influences the δ15N values of swordfish, as already shown by Young et al. (2006) with much fewer data. This influence is much more pronounced in swordfish than for yellowfin tuna (although diet shifts for small juvenile yellowfin tuna can occur, as shown by Graham et al. 2007), due to the change in the feeding ecology of swordfish through its ontogeny.

Trophic level differences

Over the body lengths and latitudes common to both species, and for a similar length or latitude, the δ15N values of swordfish were about 0.7–2.8‰ higher than those of yellowfin tuna (Fig. 5a). The greatest δ15N differences between both species were found in large fishes (160 cm) sampled in the south (25°S), while the smallest differences occurred in small fishes (80 cm) sampled in the north (5°N). In a recent study conducted in the same area, Potier et al. (2007) established that (1) the diet composition of swordfish was dominated by mesopelagic cephalopods (Ommastrephidae and to a lesser extent Onychoteuthidae) and by mesopelagic fish (Nomeidae and Diretmidae), while epipelagic prey dominated the diet of yellowfin tuna, (2) swordfish catch larger specimens of the same prey species than yellowfin tuna. This general diet pattern reflects a well-known resource partitioning between both species (Potier et al. 2007). Swordfish undertake large vertical migrations, allowing them to prey actively at great depth, while both adult and juvenile yellowfin tuna spend the vast majority of their time in the surface layer and prey on small organisms (Brill et al. 2005; Ménard et al. 2006; Potier et al. 2007). Consequently, swordfish have access to a larger size range of prey in the environment than yellowfin tuna, and can feed on the predators of yellowfin tuna’s prey. Thus, the observed differences in δ15N values of swordfish and yellowfin tuna, once the body length and latitudinal effects are removed, illustrate different trophic levels between both species due to distinct feeding strategies. Graham et al. (2007) hypothesized that mesopelagic prey might have δ15N values higher than epipelagic species. This assumption could strengthen the δ15N differences between both species, but further investigations should be carried out on the isotopic values of the forage fauna of large pelagics.

Summary and conclusion

This study revealed fish length and latitudinal effects on δ15N and δ13C values of two migratory highly opportunistic predators: yellowfin tuna (T. albacares) and swordfish (X. gladius). However, in these open sea ecosystems of the western Indian Ocean, δ15N was much more useful than δ13C to delineate trophic relationships and to track fish movements. Linear mixed-effects models developed here will allow us to make predictions of δ15N values and to correct for any body length or latitude differences in future food web studies. This study also confirmed that baseline δ15N isotopic variations can be conserved through several trophic levels, and even up to high-trophic levels such as tunas and swordfish. These spatial differences together with differences in the fish length effects according to species illustrated the potential of stable isotopes to investigate complex trophic ecology and feeding strategies, both within and between species, even in the case of highly opportunistic feeding behaviours.

To further investigate these spatial and size variations in the δ15N values of yellowfin tuna and swordfish, isotopic analyses of mesopelagic species together with POM from these regions are needed. Spatial and size-based variation in the δ15N of marine pelagic fish should be considered when using δ15N to detect trophic-level variation in natural communities.