Introduction
Gilthead seabream (Sparus aurata) and European seabass (Dicentrarchus labrax) are emblematic species of Mediterranean aquaculture, the second most important aquaculture industry after the Atlantic salmon in the European Union (Llorente et al. 2020). Hatchery stages are critical to the whole production cycle, as they determine the plasticity of body morphology and shape, growth potential and robustness (Kourkouta et al. 2021a, 2021b). Currently, larval quality is assessed through a series of Key Performance Indicators at different stages of development based on the desired traits of fast-growth, robustness of juveniles and adults free of morpho-anatomical defects (Kourkouta et al. 2021b). The aim of the present study was to screen for a set of molecular markers that could be used as quality indices for gilthead seabream and European sea bass larvae, and to develop a time- and cost-efficient tool for real-time batch quality evaluation in the hatchery.
Materials and methods
Gilthead sea bream (SBG) and European seabass (BSS) larvae of four stages [first feeding (FF), flexion (FL), end of larva rearing (ELR) and mid metamorphosis (MM)] were sampled from four and three commercial European hatcheries, respectively. Samples were used for RNA extraction followed by quantitative PCR (qPCR) and whole mount bone staining. For skeletal quality assessment alcian blue /alizarin red whole mount-stained larvae were analysed to establish the incidence of malformations. For both species quality was scored, based on the existence and the severity of deformities (Kourkouta et al. 2021a, 2021b). In the case of BSS, an additional quality score based on axial growth rate was used. Only, the best (BEST) and worst (POOR) performing production batches were used for further analysis. Following total RNA extraction and cDNA synthesis, the expression levels of 22 and 24 genes were measured by qPCR in SBG and BSS, respectively. A Wilcoxon signed-rank test was used to examine significant differences per gene and stage between the two quality extremes scored, BEST and POOR. The multi-collinearity of the variables in the data set, was assessed with a Pearson correlation coefficient. A multivariate approach and permutational analysis of variance (PERMANOVA) evaluated differences in gene expression and the size of the variability explained by the quality score. Random forest was trained as the classification algorithm using the gene profile obtained for the samples of known quality i.e. BEST and POOR. A quality prediction model was built and then used to classify samples for which the quality was masked before analysis.
Results and Discussion
The general strategy applied for both species was as follows: the first step was to screen for candidate genes that presented significant differences between the two quality scores (BEST vs POOR). The genes with no significant differences were excluded from the rest of the analysis. In the second step, correlation analysis was used to identify highly correlated genes (correlation coefficient of > 0.7) in each of the four stages. From each group of highly correlated genes, only the gene with the highest p-value in the Wilcoxon signed-rank test, between samples from the two extremes of quality were used in the next step of the analysis. Based on the criteria outlined above, sets of quality-related genes were selected for each stage. Subsequently, the PERMANOVA results were used to identify the stage/stages in which the differences between the two quality scores (BEST-POOR) were explained by the gene variables. Based on the results, genes were selected that were highly associated with quality at a given stage/stages for SBG and BSS and used for the random forest algorithm. The accuracy along with specificity and sensitivity of the applied random forest model for each stage and each species is presented in Table 1. Finally, the top five genes were identified that, when combined, contributed to generate a model predicting quality with high accuracy. A unique gene set was identified per species, stage, and quality criteria.
The study evaluated a set of genes that participate in key biological functions for their predictive power to estimate quality of commercial batches of SBG and BSS. This addresses a need for tools for real-time monitoring of batch quality during production cycles and may also reflect on the actual juvenile quality. The molecular index generated that consists of the top 5 most informative genes represents a novel, relatively fast and cost-effective tool for global larval quality screening. Finally, the accuracy of the model was ≥ 0.8 in all the cases. The algorithm could be further trained and refined for specific quality characteristics with the input of more data and by further validation with datasets for which quality has been masked.
Acknowledgements
This research was funded by the EU Horizon2020 Research Framework in the context of the project PerformFISH, Grant nº 727610, www.performfish.eu.
References
Kourkouta Ch., Printzi A., Geladakis G., Mitrizakis N., Papandroulakis N., Koumoundouros G., 2021a. Long lasting effects of early temperature exposure on the swimming performance and skeleton development of metamorphosing Gilthead seabream (Sparus aurata L.) larvae. Scientific Reports, 11, Article number: 8787. https://doi.org/10.1038/s41598-021-88306-4
Kourkouta Ch., Tsipourlianos A., Power D.M., Mamuris Z., Moutou K.A., Koumoundouros G. 2021b. Variability of key performance indicators in commercial Gilthead seabream hatcheries. Aquaculture Europe 2021, "Oceans of Opportunity", EAS Conference. October 4-7, 2021, Funchal, Madeira, Portugal. pp. 659-660.
Llorente, I., Fernandez Polanco, J., Baraibar, E., ……….. and Basurco, B., Assessment of the economic performance of the seabream and seabass aquaculture industry in the European Union, MARINE POLICY, ISSN 0308-597X, 117, 2020, p. 103876, JRC117066.