Prediction of fish growth potential from gene expression measurements: insights from gilthead seabream

Abstract

Prediction of growth potential and the effect of genetic, environmental and nutritional effects on growth is a central concern in aquaculture research. On the other hand, objective assessment of growth potential usually relies on mid- to long-term growth trials, which carry very high costs. In this presentation, we will discuss the estimation of long-term growth (t = 3 months) in gilthead seabream (Sparus aurata) based on gene expression measurements performed at an early stage (t = 2 weeks), comparing it with results from a similar previous trial in Atlantic salmon (Salmo salar). For this trial, gilthead seabream juveniles were reared in 21 tanks (1000 L) over the course of 3 months, being fed with a wide range of diets that differed in terms of nutritional composition, type of ingredients and performance/quality. At 2 weeks after the start of the trial, some fish were sampled to obtain liver, muscle and head kidney samples for gene expression analysis. At the end of the 3 month period, all remaining fish were bulk weighed and per-tank performance indicators estimated. Gene expression measurements were performed using reverse transcription and real-time qPCR protocols, with over 30 genes being measured per tissue. These datasets were used to calibrate linear models of growth performance measures using L1-regularized linear regression, which were then further refined by AIC-guided stepwise variable removal. Results show that this approach seems feasible in general, as previously seen with Atlantic salmon, though the prediction quality was not as high as for salmon (R2 = 0.6, rather than R2 = 0.9). The reasons identified for this difference in prediction are threefold: first, the fact that individual measurements of gene expression were performed (as opposed to pooled measurements); second, the wider range of tested dietary formulations (which increases model robustness but decreases its apparent predictive power); third, the relative diversity of ontological categories of genes, which was reduced in this trial. In conclusion, while this study confirms the feasibility of this type of approach, it underlines the challenges of obtaining a robust prediction model for fish growth potential.