The experiment was carried out at the Beymelek Unit of the Mediterranean Fisheries Research Production and Training Institute, Antalya (Türkiye). From the Central Fisheries Research Institute, Trabzon (Türkiye), juvenile turbots weighing approximately 25 g were moved and stocked into circular tanks with a water capacity of roughly 30 m³. Each tank received 5.4 L/min of brackish water (salinity range: 7.7 to 10.2 g/L) that had been UV (hot cathode lamb with a capacity of 40 mJ/cm², Wedeco Proxima, Aquada UV, Herford, Germany) treated.

After about a 2-month-acclimation, fish were size graded, transferred to 5 m³ tanks and kept there until they weighed roughly 250 g. During the acclimatization of the fish to the conditions before the experiment, each of the 18 tanks, which measured 1.1 × 1.1 × 0.5 m and held 400 L of water, was randomly assigned to house a total of 18 fish.

The fish were then fed a commercial diet (Kılıç Deniz Ürünleri, Muğla, Türkiye) containing 54% protein and 12% lipid for 3 weeks. At the beginning of the experiment, the number of fish in each tank was reduced to 15 and their average weight was 301.6±0.1 g.

Fish were held in an indoor system that allowed natural light to enter and were subjected to a natural photoperiod of 10.5-13 hours light / 11-13.5 hours dark over the study period. Using a hand-held DO meter and pH meter (YSI Model 55 and 63, YSI Inc., Yellowsprings, OH, USA), water parameters including temperature, dissolved oxygen, pH, and salinity were measured every day. The results were 19.28 ± 0.12 °C, 7.17 ± 0.06 mg/L, 7.51 ± 0.02 and 8.30±0.03 g/L, respectively. During the study, fish were collectively weighed at 3-week intervals after a one-day starvation period. The fish were fed their respective diets until apparent satiation by hand at 09:00 and 16:00 h for 10 weeks. At the beginning of the experiment, 10 fish were sampled to determine their initial body composition, while at the end of the experiment, four fish were sampled from each tank to determine their final body composition using an overdose of ethylene glycol monophenyl ether. Another four fishes were sampled to determine organo-somatic indices.

To formulate the experimental diets, a mixture methodology using Scheffe cubic with a D-optimal design was employed because of the limited number of experimental tanks. The center point diet (K) was triplicated, and two additional diets (H and L) were duplicated, while other diets were tested in one tank. The main protein source was an equal blend of two fish meals, Danish LT and Domestic anchovy meal, based on the results of Sevgili et al. (2014a). Wheat flour and fish oil were used as the carbohydrate and lipid sources, respectively. To account for 93.85% of the diets, the minimum and maximum limits of fish meal (FM), wheat flour (W), and fish oil (FO) were set at 46.92-75.08%, 9.38-37.54% and 0.0-9.39%, respectively, as protein, carbohydrate, and lipid sources (Table 1). The mixture design model’s ingredient limitations resulted in variations in protein levels ranging from 45.6% to 63.4%, carbohydrates ranging from 4.9% to 30.5%, and lipids ranging from 5.6% to 17.7% (Table 1). When setting the nutrient boundaries, the feeding habits of turbot (carnivorous) and the results of previous studies were considered. Each diet included an equal amount of wheat gluten meal (5%), vitamin mixture (0.25%), mineral mixture (0.1%), choline chloride (0.2%), and carboxymethyl cellulose (0.5%) (Table 2). The ingredients were ground using a hammer mill with an 800 μm screen (Kocamaz Machine, Model KT-20C, İzmir, Türkiye), weighed at predetermined levels, thoroughly mixed, pelleted using a pelleting machine (with a 9-mm die) without steam, packed in plastic bags, and stored at +4°C until use.

DFI g/kg MBW/ day = (DMI / MBW^0.8) / day

MBW = (geometric mean of IW and FW)^0.8

DGC = [(FW^1/3 – IW^1/3) /day] × 100

FCR = DMI / weight gain

PER = weight gain / protein fed

CF = (average weight / standard length³) × 100

VSI % = (visceral weight / body weight) × 100

HSI % = (liver weight / body weight) × 100

Daily nutrient intake (g kg MBW^0.8 day^-1) = [(N, lipid, energy intake / MBW^0.8) / days.

Daily nutrient gain (g kg MBW^0.8 day^-1) = [(final body weight × final body nutrient) – (initial body weight × initial body nutrient)] / MBW^0.8 / days.

Nutrient retention (%) = 100 × (daily nutrient gain / daily nutrient intake).

N loss (g kg WG^-1) = (N intake – N deposited) / weight gain.

where: DFI = daily feed intake, MBW = metabolic body weight; DMI = dry matter intake; IW = initial weight; FW = final weight; DGC = daily growth coefficient; FCR = feed conversion ratio; PER = protein efficiency ratio; CF = condition factor; VSI = viscera somatic index; HSI = hepato-somatic index.

The fish samples were stored at -20 °C until analysis. Prior to analysis, they were thawed in a refrigerator and homogenized using a kitchen meat chopper (Tefal Le Hachoir 1500, France). Proximate analysis, with the exception of crude lipid, was performed on both the experimental diets and fish using the methods outlined by the AOAC (1990). Dry matter was determined by drying samples at 104 °C until a constant weight was achieved, whereas ash content was determined by incinerating samples in a muffle furnace at 600 °C for 2 hours. Crude protein (N×6.25) was determined by the Kjeldhal method after acid digestion. Finally, the lipid content was determined through ether extraction using an automatic extraction system (ANKOMXT15 Extractor, ANKOM Technology, Macedon, USA).

This study employed principal component analysis (PCA) to identify general relationships between variables and their directions. Data were analyzed using the statistical package JMP v.8.0 for Windows (SAS Institute Inc, Cary, NC, USA) in accordance with the mixture design methodology. The highest order models up to the cubic model using Scheffe polynomials were selected based on the p values of Lack of Fit and ANOVA and R². The main factors (FM, W and FO) were used regardless of the models selected, but effective terms were selected with a stepwise fit model based on the minimum Bayesian Information Criterion.

Optimum dietary macronutrient levels were estimated by examining the prediction profiler at the highest desirability that maximized selected variables, including final weight, DGC, PER, N, and energy retentions and minimized FCR, N, and C losses. Here, the optimization was performed by the highest desirability function of the target goals of the selected responses. Polynomial contrasts up to the cubic level were also used to examine the relationship between DFI and dietary macronutrient concentrations.

Growth and feed utilization performance and organ-body indices are shown in Table 2, whereas whole body compositions and nutrient mass balance values are shown in Tables 3, 4 and 5, respectively. Fish grew from 300 g to a maximum weight of 509 g on experimental diets at the end of the study (Table 2). The center point (diet K) tested in the three tanks yielded an acceptable variation in the mean final weight and FCR, enabling reliable evaluation of the mixture design analysis (Fig. 1). Additionally, the Levene test confirmed the homogeneity of variance among the replicated treatments (diets H, K and L) in terms of final weight (p=0.751) and FCR values (p=0.596). The maximum growth performance was observed in fish on diet N (1.83 %/day of DGC) containing 49.3% protein, 28.9% carbohydrate, and 10.4% lipid, whereas the lowest growth was observed in fish on diet B (DGC of 1.03%/day), which contained 57.1% protein, 23.0% carbohydrate, and 5.7% lipid.

Even though the fish development rate in this study was in line with that of earlier research on large turbot (Regost et al., 2001; Leknes et al., 2012; Sevgili et al., 2012, 2021), a notable inter-treatment variation was observed among the nutrient concentrations considered. FCR values varied between 0.68 in diet F (56.2% protein, 12.8% carbohydrate, and 17.0% lipid) and 1.30 in diet B, consistent with those recorded in turbot in former studies (Regost et al., 2001; Leknes et al., 2012; Sevgili et al., 2012, 2021). The least consumed diet (3.12 g/kg MBW^0.8/day) by turbot was diet J, containing 51.6% protein, 12.3 % lipid, and 30.5% carbohydrate, whereas the highest consumption (5.66 g/kg MBW^0.8/day) was observed in diet D consisting of 61.5% protein, 15.4% lipid, and 13.5% carbohydrate. The DFI values fell within the range of those reported in previous studies (Sevgili et al., 2012, 2021). In the present study, no linear relationship was observed between DFI values and dietary macronutrient compositions, but significant linear, quadratic, and cubic trends (p<0.05) were observed when testing the impact of dietary protein levels using polynomial contrasts, indicating that an interaction between dietary protein and dietary carbohydrate or lipid concentrations was present. However, such a significant trend for DFI was not observed for dietary carbohydrate and lipid levels (p>0.05). Although dietary energy content is known to be the primary factor regulating feed intake in fish, it was not observed in the present experiment, likely because of the narrow range of the boundary. Instead, dietary protein levels appeared to play a more significant role in the voluntary feed intake in turbot. This finding is partly in line with that of Saravanan et al. (2012), who found an opposite relationship between feed intake levels and dietary digestible protein content in Nile tilapia, Oreochromis niloticus.

Dietary protein, carbohydrate, and lipid levels were located on different sides of the biplot by the first two PCAs (Fig. 2). Lipid was separated from other macronutrients by PCA1, whereas dietary protein and carbohydrate were on opposite sides of PCA2. The PCA also revealed that growth performance in turbot was strongly associated with the retentions of C and energy as well as N and energy gains, dietary lipid, and energy levels on the PCA1 axis (Fig. 2). These findings contradict many previous studies on juvenile turbot, which have shown that high dietary lipid levels can lead to reduced growth and nutrient retentions (Caceres-Martinez et al., 1984; Andersen & Alsted, 1993; Sevgili et al., 2014b; Zhang et al., 2015). Turbot are lean fish and have limited ability to tolerate and deposit excessive dietary lipid levels (Regost et al., 2001; Sevgili et al., 2014b; Zhang et al., 2022). However, as the fish grow, they appear to better utilize lipids by promoting the fatty acid β-oxidation (Zhang et al., 2022). The latter researchers showed that larger turbot (80 g) showed comparable expressions of lipolysis-related genes in the liver (pparα and cpt1) when they were fed increased dietary lipid (17.1%), which was not opposed to the findings in smaller fish groups (9 and 50 g). These findings are consistent with those reported for larger turbot (>500 g) by Leknes et al. (2012), who claimed that those fish showed optimum growth with a diet including 43.5% protein and 25.7% lipid. Accordingly, lipid retention values in the present study with larger fish were higher than those in our previous study with juvenile fish (54.4 g) when diets with similar compositions were compared (Sevgili et al., 2014b). Moreover, fish on diets E, F, I and L, including high lipid levels in the present study had remarkably lower energy and C retentions (average 35.8% and 34.6% respectively) (Table 5) than those recorded in the same study (Sevgili et al., 2014b) for diets with 16.0% and 18.6% lipid, which were 38.8% and 44.5%, respectively. These findings suggest that larger turbot are able to store more lipids in their bodies and better benefit from high dietary lipid contents for sparing amino acids as an energy source compared with their smaller counterparts.

Dietary protein level was closely associated with DFI, C, N, and energy intakes, which further supports the abovementioned association between feed intake and dietary protein. However, the dietary protein was moderately related to the losses of C and N (Fig. 2).

Dietary carbohydrate was moderately related to N retention and PER, suggesting a partial protein-sparing effect of carbohydrates in turbot and being consistent with the results of other marine finfish species such as European seabass (Dicentrarchus labrax) (Pérez et al., 1997), Senegalese sole (Solea senegalensis) (Conde-Sieira et al., 2016), gilthead seabream (Sparus aurata) (Basto-Silva et al., 2022) but not with those of European whitefish (Coregonus lavaretus) (Ruohonen et al., 2003; Vielma et al., 2003) and lumpfish (Cyclopterus lumpus) (Hamre et al., 2022). Whole body lipid, C, C/N, and energy levels were closely associated with dietary lipid levels, which is consistent with previous findings on juvenile turbot and shi drum, Umbrina cirrosa (Akpınar et al., 2012; Sevgili et al., 2014b).

In the present study, only significant responses with a p-value of 0.05 were considered in the mixture design analysis, as indicated in Table 6. The models were found to have good R² values (>0.65) and an insignificant lack of fit (p>0.05), which is a prerequisite for a reliable approach. The model coefficients, standard error, and p-values of the terms are given in Table 7. It was observed that not all terms of the regression model for final weight and DGC values were significant (p>0.05) such as FO, FM*FO, W*FO, FM*FO*(FM-FO), and FM*W*FO (shown in Tables 7 and 8). Accordingly, insignificant coefficients can be removed from the model, but the main components should be retained. Thus, the equations for the final weight and DGC can be expressed as follows:

Final weight =586 _FM + 9738 _W – 24855 _FO – 18672 _FM*W + 10404 _FM*W*(FM-W)

DGC = 2.53 _FM + 73.4 _W – 196 _FO – 145 _FM*W + 79.1 _FM*W*(FM-W)

Model terms of the responses, including FCR, PER, N, and energy retentions, N loss, and C loss, were statistically significant (p<0.05) and their model equations are given below.

FCR = –2.58 _FM – 23.0 _W – 150 _FO + 50.9 _FM*W + 223 _FM*FO + 506 _W*FO – 707 _FM*W*FO

PER = 9.20 _FM + 56.5 _W + 360 _FO – 111 _FM*W – 519 _FM*FO – 1155 _W*FO + 1573 _FM*W*FO

N retention (%) = 172 _FM + 1034 _W + 6879 _FO – 2057 _FM*W – 10046 _FM*FO – 22026 _W*FO + 30482 _FM*W*FO

Energy retention (%) = 164 _FM + 895 _W + 6362 _FO – 1871 _FM*W – 9199 _FM*FO – 20187 _W*FO + 28061 _FM*W*FO

N loss (g/ kg weight gain) = –235 _FM – 2080 _W – 13379 _FO + 4450 _FM*W + 19900 _FM*FO + 44724 _W*FO – 63045 _FM*W*FO

C loss (g/ kg weight gain) = –1158 _FM – 9508 _W – 67095 _FO + 20897 _FM*W + 98606 _FM*FO + 218104 _W*FO – 299439 _FM*W*FO

In this study, the optimal formulation for large turbot diets was predicted by compromising the targets of each response (Fig. 3). The optimal formulation was based on the highest desirability (0.87) for all the selected response goals. The goals reached with the highest desirability were 488 g for final weight, 1.67% for DGC, 0.69 for FCR, 2.59 for PER, 46.7 % for N retention, 44.7% for energy retention, 30.3 g/kg weight gain for N loss, and 183 g/kg weight gain for C loss to the environment. The optimal dietary inclusion levels of FM, W, and FO were estimated to be 63.6, 20.8 and 9.4%, respectively (Fig. 3). Figure 4 shows the optimal dietary incorporation of main ingredients and the ternary plot of the selected responses. The resulting optimal dietary macronutrient compositions were calculated using the analysis values given in the footnote of Table 1 and were 49.68% protein, 15.62% lipid, and 14.46% carbohydrate as-is basis. The estimated dietary protein level was higher than what a previous study (Leknes et al., 2012) suggested as optimal (43.5%) for large turbot. Additionally, these researchers recommended increasing dietary lipid content to 25%, which is quite higher than our estimation. However, the maximum limit for dietary lipid inclusion in the current experiment was suggested by the mixture model was 17%t, indicating that a somewhat higher upper limit must have been considered in the present study. The carbohydrate level is more or less consistent with the recommended level of 15% for juvenile turbot by Zeng et al. (2015), who addressed that fish can tolerate higher dietary carbohydrate levels and lower protein levels (about 46%) as water salinity increases. They also claimed that turbot should be reared at a salinity above 12 g/L, which is still higher than the salinity of the present study. Juvenile turbot showed better carbohydrate utilization and less influence from negative effects of its high dietary inclusions (>25%) when diets were supplemented with biotin (Liu et al., 2021; Pan et al., 2022) and taurine (Zhang et al., 2019). Therefore, it is difficult to argue that large turbot could utilize the dietary carbohydrate more effectively than the present study in higher salinity waters or with biotin and taurine supplementations, which clearly warrants further investigation. Further studies are clearly required to fully discover the nutrient requirements of turbot across the aquaculture stage, considering that the market size of turbot can be as high as 4 kg (Ruyet, 2002).

In conclusion, the results of the present study are significant from the perspective of filling the information gap about the optimum dietary macronutrient compositions of diets of large turbot. The results indicate that large turbot (300-500 g) require high dietary protein levels (54% dry matter basis) but can better utilize dietary lipid (>17% on dry matter basis) for energetic use than smaller juveniles. The specific diet formulation for this stage can also include 15.8% dietary carbohydrate on dry matter basis. The optimum nutrient concentrations will allow better diet formulations for this turbot size in terms of growth and nutrient utilization performance and environmental impacts.