Further to my comments above.
It would appear that moderate training actually reorganizes metabolism so as to spare the muscle [against muscle loss] in many teleost's including tilapias (Belal, 2015) with the result that protein growth is promoted over lipid growth in fish that are both fed and continually swum (Christiansen et al., 1989; Lauff and Wood, 1997). The Belal (2015) study indicated reduced body lipid proximate analysis at velocities of 25-35cm/sec. which supports the below hypothesis. In view of a great deal of convincing literature that net protein accretion rates, protein conversion efficiency from the diet, and overall growth rates are all improved if fish are continually swimming (Houlihan and Laurent, 1987; Christiansen et al., 1989; Davison, 1989, 1997), it seems likely that the exogenous fraction of nitrogen excretion will be lower if fish swim aerobically while feeding; that is, amino acids from the free pool will be funneled preferentially toward protein synthesis, rather than toward deamination and oxidation. Because feeding seems to preferentially elevate the concentrations of essential amino acids in the blood plasma (Brown and Cameron, 1991; Espe et al., 1993), whereas swimming preferentially elevates nonessential amino acids (Barton et al., 1995), it may be that the combination is most effective in stimulating protein synthesis. Interestingly, the Specific Dynamic Action (SDA, a measure of protein accretion) effect of feeding continues unabated (Beamish, 1974; Alsop and Wood, 1997) or may even increase (Muir and Niimi, 1972; Blaikie and Kerr, 1996) during sub maximal exercise. Because SDA mainly represents the cost of elevated protein synthesis, this indicates that carbohydrate or lipids are used to a greater extent, not only to power exercise itself, but to power protein synthesis during exercise. This hypothesis then explains the results of the Belal (2015) study.