Flesh color is a complex qualitative trait, where its formation is influenced by several factors, including dietary formulations and absorption of carotenoid pigments, water and lipid content of flesh, as well as reproductive stage (32, 33). Flesh discoloration in Atlantic salmon exposed to prolonged hot summers in Tasmania was recorded in recent years (34, 35). Grunenwald, Adams (18) pointed out that increased dietary astaxanthin, canthaxanthin and vitamin A did not prevent flesh discoloration. In optimal thermal conditions, temperature positively influences plasma Ax level and possibly controls the efficiency of Ax metabolism (36). Atlantic salmon display behaviors that are adaptive to the daily fluctuations in environmental temperature and dissolved oxygen in farming sea cages. Individual fish tend to adjust their depth to stay at the optimal temperature of 12˚C to 18˚C and show a trend of avoiding low dissolved oxygen (37, 38). More recently, Wade, Clark (19) investigated the effects of a summer heatwave on farmed Atlantic salmon from two distinct groups cultured in sea cages. During that summer where the temperature was for 117 days above 18oC, flesh color was reduced initially in the front dorsal and then in the back central region. Furthermore, in autumn, a feeding and coloration recovery did not occur evenly among individuals. In the present study, we sampled fish from a single cage in June 2017. This time point was a few months past the heat stress of summer. From historical observations, while some fish at that site had recovered feeding and flesh color, others could not grow well and improve flesh colour at that time point. Therefore, we expected to find fish with different flesh color phenotypes. We hypothesized that the prolonged elevated temperature would not only affect flesh color, but also induce other changes in response to thermal stress, such as starvation and lipid metabolism. In which, these changes might no longer be recovered in some individuals. The flesh coloration variation in response to thermal stress, lipid metabolism and starvation were caused by molecular mechanisms that we studied at the gene expression level. Using both TruSeq and QuantSeq on the same subsets of HN and LB fish (n = 5 per group), a number of DEGs were detected in both phenotypes.
In the context of iron ion transport, Fertn-M was upregulated in HN fish when compared to LB fish in our study. Iron is an essential trace element playing a key role as a biocatalyst or electron carrier for many biological reactions (39). Because pathogens also require iron for proliferation and production of virulence factors, iron metabolism is closely related to the host innate immune response to prevent pathogens from iron uptake through an iron-withholding strategy (40, 41). This strategy is controlled by the upregulation of positive acute-phase proteins, plasma proteins during acute phase response (39). Ferritin is among these plasma proteins which segregate excess iron into a non-toxic and organic form for iron storage (42). Fertn-M function in Atlantic salmon was well described in previous studies (43). It is competent in both mandatory iron storage stages of iron-oxidation and iron-mineralization (39, 44). Several studies indicated that Fertn-M was activated to respond to many infections and oxidative stress in fish, including Atlantic salmon (39, 45–50). In our previous studies, it was confirmed that aerobic bacteria (Pseudoalteromonadaceae, Vibrionaceae and Enterobacteriaceae families) with varying catabolic pathways that were enriched in low color fish might impact on the host physiology (16, 51). These findings are consistent with our present result and indicate that Fertn-M was downregulated in LB fish since the iron acquisition could be controlled by pathogen and used to alleviate ther stress in LB fish. Thus, we deduced that in the present study in HN fish subjected to thermal stress during hot summer conditions, the regulation of iron storage might have been activated by Fertn-M to limit the iron acquisition by pathogen and relieve thermal stress.
Also, CatB and CatL were upregulated in HN fish in our present study. CatB and CatL are cysteine proteases involved in the self-defense mechanism against fish pathogens and the host immune defense in vertebrates (52, 53). CatB plays essential functions in pathological and physiological processes (54). CatB may be associated with the regulation of follicular apoptosis in zebrafish (55) and affected the expression of CY genes in Japanese spiky sea cucumber (56). They also activate the complex mechanism of fish metabolism and muscle degradation to respond to diverse environmental and biological parameters (57). LB fish downregulated CatL and CatB as compared to HN. This might indicate that LB fish could not activate the self-defense mechanism and respond to environmental conditions as efficiently as HN healthy fish.
Another group of genes (Alb2, Fabp1, ApoCI and ApoAI) related to lipid metabolism were also upregulated in HN fish. Alb2 is one of the most abundant proteins in the plasma, considered as the long-chain fatty acid transporter when it binds together to form complexes that circulate in the plasma (58, 59). A radioactive labelling-based study proposed that Ax is associated with the serum lipoproteins, the primary transporters of esterified fatty acids and serum albumin (60). We have found that Alb is upregulated in the gut of HN fish and propose that this upregulation may increase the Ax absorption and transport from the gut into the blood system and then deposition in the muscle or liver. Due to their hydrophobic property, carotenoids or Ax are described to be closely associated with fatty acids and transported with them through the intestine and blood circulation. Moreover, ApoCI was found to inhibit cholesteryl ester transferase protein activity that leads to increased accumulation of cholesterol in high-density lipoprotein (HDL) and a decrease in low-density lipoprotein (LDL) cholesterol level (61, 62). Finally, upregulation of ApoAI has been linked to increased accumulation of fatty acids and cholesterols into HDL (59, 62, 63). HDL is known to have a major role in carrying fatty acids and carotenoids in Atlantic salmon (64, 65). Thus, upregulation of Alb2, ApoAI and ApoCI in HN fish implies that the uptake, absorption and metabolism of lipid or carotenoids occurs more intensively in HN fish while it might be partially impaired in LB fish.
The CYP450 enzyme family, together with important functions in the oxidation-reduction process, has also been known to induce stress response as a result of a variation in temperature (66). In our study, Cyp2k1, Cyp3a27 and Cyp1b1 were significantly upregulated in HN fish. Studies in birds indicated that CYP450 enzymes influence bird coloration in species such as zebra finches, red siskins and common canaries (67, 68). CYP450 genes were also found to be upregulated in red flesh phenotype Chinook salmon when compared to white flesh phenotype (69). In the latter study, the CYP450 2J19 gene (Cyp2j19), which is considered as a carotenoid ketolase and mediates red coloration in birds, was also upregulated in the pyloric caeca of red flesh phenotype Chinook salmon (70), although in the present study, Cyp2j19 was not a DEG. This finding suggests that Cyp2j19 might not be associated with the red coloration in the Atlantic salmon. However, the three CYP450 genes (Cyp2k1, Cyp3a27 and Cyp1b1) might have a link with flesh pigmentation in our study. Given the rapid evolutionary rate of CYP450s and their multiple roles across many metabolic processes (71, 72), it is plausible to assume that the three CYP450s identified as DEGs in our study neo-functionalized have a role in carotenoid metabolism in Atlantic salmon and more comparative research across salmonids should resolve this.
In our comparison, the neuropeptide gastrin-releasing peptide (Grp) was upregulated in LB fish when compared with HN fish. Aside from functions in the gastrointestinal tract, some neuroendocrine peptides also impact on short-term food satiation and signal the brain to inhibit feeding (73). In vertebrates, the process of food uptake and digestion is regulated and balanced by a series of endocrine events. In the gut-brain axis, neuroendocrine peptides are secreted by the stimulation of food intake available in the gastrointestinal tract. These peptides may act both through endocrine and neural systems. Grp was previously shown to stimulate gastric acid release in the fish gut (73) while inhibiting gastric emptying and mediating satiety in fish (74). In Atlantic cod, it was recorded that in fish fed high rations (1.5% body weight), Grp gut expression was higher when compared to fish fed low rations (0.2% body weight) (74). This indicates that more Grp was released to inhibit food intake and prevent overfeeding. This finding however, does not align with our current results, where Grp was upregulated in LB fish, which did not consume food well, while Grp expression in HN fish was lower despite normal growth and feed intake. It shows that this phenomenon might only occur when fish experienced chronic stress and starvation. Also, in Atlantic cod, Grp expression level in both 30-days starvation and re-feeding periods of starved fish was lower than fed fish (74). This suggests that in LB fish, the control and balance of food intake with the feeding requirement was still dysfunctional likely due to persisting effects of the high thermal stress conditions during summer. Although the stress had been ceased, feeding activities in LB fish had still to recover completely. In addition, Grp was considered as an islet neuropeptide that acts as a stimulator of insulin secretion induced by triggering of nervous system in mice (75). This finding might apply to other vertebrates and be consistent with our result in the upregulation of Grp in LB fish. We propose that Grp could stimulate the secretion of insulin to balance energy metabolism in LB fish due to prolonged starvation. Accordingly, prolonged starvation is also the main reason for flesh discoloration, as fish likely lost a large amount of uptaken carotenoids to compensate for the amount used by physiological activities in the fish body.
When detecting whether there is a missense mutation in carotenoid-related genes, we found Pro21Thr missense mutation on Abcg8, which highly expressed in HN fish. In humans and other mammals, Abcg8 and Abcg5 (ATP-binding cassette, sub-family G (white), members 8 and 5; also known as sterolins) are known to participate in the physiological pathways involving dietary cholesterol and non-cholesterol sterols (76). These sterolins are necessary to secreting cholesterol efficiently into the bile and increase cholesterol levels in plasma and liver when the dietary cholesterol content changes (77, 78). It has been shown that 2% increase in dietary cholesterol improved Ax absorption and deposition as well as Ax levels in the plasma of Atlantic salmon (79). Hence, in light of the aforementioned findings, additional cholesterol in Atlantic salmon diets could potentially result in increased Ax levels in the plasma, which would then be absorbed by the muscle, leading to a more intense flesh coloration. In addition, Abcg8 might have a critical function in increased carotenoid-binding by cholesterols in HN fish. With regard to to ATP-binding cassette, sub family, Zoric (80) reported that a missense mutation in Abcg2-1a, resulting in the substitution of Asparagine with Serine in amino acid position 230 (Asn230Ser), is predicted to be associated with flesh color in Atlantic salmon because it is more prevalent in the pale than in the dark flesh phenotype. The complex Abcg5/g8 was previously associated with carotenoid metabolism and was predicted to function in limiting dietary carotenoid (81). Therefore, in our study, it is possible that Pro21Thr missense mutation on Abcg8 plays an important role together with Abcg5 in the control of secreting carotenoid-binding cholesterols and absorption of dietary carotenoids and mobilization of carotenoids in Atlantic salmon.
In term of SNPs on DEGs identified in TruSeq, the six mutations on Duox2-l and Duoxa1-l that we detected in LB fish in the current study might potentially lead to the mismatched complex, which then affects their function of generating reactive oxygen species (ROS). Dual oxidase 2 (Duox2) is a member of the NADPH oxidase (NOX) family that generates reactive oxygen species (ROS). This enzyme catalyses the electron transfer from NAPDH-FAD to O2, generating superoxide (O2.−) or hydrogen peroxide (H2O2) by reducing molecular oxygen (82). Duox1 and duox2 function solely in regulating specific maturation factors known as duox activators, Duoxa1 and Duoxa2, respectively (83, 84). In a recent study, Duox and Duoxa proteins were found to form stable heterodimers and co-translocate to the plasma membrane (Fig. 8) (85). Opposed to Nox1-5, Duox enzymes have an extended N-terminal extracellular domain known as peroxidase homology (PoxH) domain, followed by an additional transmembrane segment and an intracellular loop containing two calcium-binding EF-hand motifs (Fig. 8). There is a switch in ROS generation from H2O2 to O2, which happens when Duox2 is mismatched with Duoxa1. Following that, the mismatched combination of Duox2 + Duoxa1 releases O2.− together with H2O2, leading to the phenomenon of O2.− leakage (86, 87). In juvenile rainbow trout, dietary Ax supposedly enhanced the antioxidant defense system, which plays a role in the inactivation of ROS (88). Consequently, we hypothesize that in LB fish containing such mutations, the deficiency of Ax or carotenoids can be due to reinforcement of the antioxidant capability.
The qPCR validation confirms that the TruSeq data performance on genes with low expression was much more precise than the QuantSeq data. In addition, although the first ten genes in Fig. 7 (5 DEGs in both libraries and 5 DEGs in QuantSeq only, respectively) had high expression and the last five genes (5 DEGs in TruSeq only) had low expression in both QuantSeq and TruSeq data, the level of expression of the gene did not affect the RT-qPCR result. In RT-qPCR validation, it is evident that genes with high variation within biological replicates can not be considered as differentially expressed (P > 0.05) even though they were picked as DEGs and expressed highly in the QuantSeq and TruSeq data.
A comparison between TruSeq and QuantSeq on two subsets of distinct HN and LB flesh color phenotypes with five replicates revealed many DEGs in both the low and high sequencing depths. As a result, the sequencing analysis were compared to show the benefits and drawbacks of the two methods. In only QuantSeq with a low depth of sequencing, in the comparisons of 39 samples from 4 color phenotypes, there was a high variation in the number of normalized read counts within each group. A recent study of genetic and phenotypic correlations indicated that improvement in growth did not end in any flesh color variation in Atlantic salmon (33). It suggested that flesh color might be affected by multiple genes and the response was not uniform. Accordingly, more replicates resulted in increased variation of gene expression between replicate samples and as a consequence, DEGs were hardly shown in our present study with increased sample size.
In the standard whole transcript method, mRNAs was fragmented before converted to cDNA. Therefore, the longer the transcript, the more fragments were created. On the other hand, 3’ mRNA method creates only one read for each transcript, so the read count reflects the level of gene expression. Bioinformatic analysis is also simplified since exon junction detection, and the normalization of reads to gene length are not required (29, 31). As expected, TruSeq reads mapped transcripts evenly with a minor decline at the 5’ and 3’ end. This result is compatible with the finding from a previous study (28). In contrast, QuantSeq reads covered primarily to the 3’ end. In vertebrates, genes can be alternatively spliced to create many distinct and expressed isoforms (31). We hypothesize that TruSeq method is more suitable to determine differences in gene isoforms than QuantSEq. In addition, the proportion of read numbers mapped using TruSeq should increase with transcript size, while transcript size should not change the proportion of reads mapped to each transcript with QuantSEq. Therefore, only the TruSeq quantification values should be normalized to transcript length. It was therefore expected that the proportion between TruSeq and QuantSeq reads mapped per transcript should increase in favor of TruSeq reads numbers as the size of transcripts increases (29–31).
DEG analysis is one of main approaches of RNA-SEq. Following QuantSeq’s instruction, we loaded 96 QuantSeq libraries onto one lane to reduce sequencing cost and create about 2 million reads for each library. As per QuantSeq’s recommendation, it requires at least 10 million reads per library for QuantSeq transcriptomics in mammals (30). Because the Atlantic salmon genome is smaller than those of mammals, we expected 2 million reads for each QuantSeq library would satisfy the recommendation. With TruSeq method, over 20 million reads for each library were generated to achieve the minimum requirement of DGE analysis. Therefore, TruSeq costs about 6 times more than QuantSeq per sample, however the cost/million reads is the same for either TruSeq or QuantSEq. Some studies have researched the effect of sequencing depth on RNA-Seq, in which a higher power was reached as sequencing depth rises (below 20–30 million reads) (89, 90). Using DESq2 for differential expression analysis, we found that TruSeq detected more DEGs with low expression than QuantSEq. This suggests that the depth of sequencing effects the number of detected DEGs, indicating that future salmon transcriptomics projects might benefit from having more than 2 million reads per library.