Emergence of appetite and circadian rhythmicity in Atlantic salmon brain transcriptome from endogenous to exogenous feeding

The transcriptional profile from normalized data displayed a clear separation between stages, as summarized in Fig 1A. The heatmap shows the expression patterns for all genes identified in the normalized count file for 720 dd and 920 dd. As illustrated by the Venn diagram (Fig 1B), 357 appetite-related genes (Ensembl IDs) were present across all three normalized count files (720 dd, 920 dd, and merged 720 dd and 920 dd), while 90 were absent in any dataset (the complete list of appetite-related genes is provided in S2 Table). Additionally, 15 genes were found to be present in one or two of the three normalized count files (see S2 Table details). Genes associated with appetite and metabolism were categorized by functional class or gene family. Overall, genes that had a higher expression before first feeding (720 dd) compared to after feeding (920 dd) were grouped into calcitonin-related, neuromedins, second-order neuron receptors, adipose-related, hormone receptors (cholecystokinin receptors, leptin receptors), and signaling proteins of cellular metabolism and signaling (transcription factors ampk, akt, creb, mTOR) (Fig 1C). Genes that were higher expressed after the onset of exogenous feeding (920 dd) were grouped in appetite-regulating neuropeptides (npy, cart, pomc), melanocortin receptors, nutrient sensing (e.g., lxr, glut2,glucokinase, cd36), corticotropin-releasing hormone system (including corticotropin-releasing hormone/factor, corticotropin-releasing hormone binding protein, urotensin), second-order neurons (e.g., orexin/hypocretin, galanin, serotonin, somatostatin), thyroid hormone system (thyroid hormones, and thyroid transporters), and genes linked to metabolism (adiponectin, and tumor necrosis factor-alpha) (Fig 1C). Genes grouped in reward-based (hedonic) appetite control (e.g., endocannabinoid receptor and fatty acid amid hydrolase) displayed a mosaic pattern (Fig 1C). Analysis of the differential gene expression comparing 720 dd to 920 dd showed that a higher number of genes associated with appetite were significantly downregulated than those genes that were upregulated at 920 dd (p-value < 0.05, S3 and S4 Tables). However, most appetite-related genes had no significant change in their expression levels between 720 dd and 920 dd.

To determine if genes related to appetite and metabolism displayed a cyclic expression pattern, the normalized count files for 720 dd and 920 dd were analyzed by MetaCycle. Results showed that 16 appetite-related genes among 1090 significant (p-value < 0.05) cycling genes had a cyclic period between 20 h and 28 h at 720 dd (Table 1). Only one gene displayed a significant 24 h periodicity at this stage, the amino acid transporter slc38a2/snat2 (ENSSSAG00000015905), part of the nutrient-sensing group, with an acrophase at 02:00 in the morning. Seven appetite-related genes had a periodicity of 20 h, and eight genes had a 28 h period (Table 1, S2 Fig). The groups containing corticotropin-releasing hormone-related genes (crh-related), genes related to nutrient sensing, and receptors were the dominant groups among the significant cycling genes at 720 dd. At 920 dd, 20 appetite-related genes were among the 1993 significant cycling genes (p-value < 0.05), of which ten had a 24 h period, five had a 20 h period, and the remaining five genes had a 28 h period. Most of these cyclic genes were related to reward-based appetite and transcription factors. The cart3b gene was the only significant cyclic gene at both 720 dd and 920 dd, having a 28 h pattern at both developmental stages. Among other neuropeptides in the melanocortin system, cart4 displayed a 20 h period, while npya2 was cycling with a period of 24 h and had an acrophase at 06:00 in the morning during the dark period, which was 3 h before the first meal of the day (09:00) (Table 1, S3 Fig). Among the 10 appetite-related genes that had a 24 h periodicity at 920dd, six genes peaked during nighttime, and four genes peaked during the daytime (Table 1).

Data is organized according to their function or family, with Ensembl ID, gene name, p-values within corresponding periodicity (20 h, 24 h, and 28 h), and acrophase (time of day the expression peaks for genes having a 24 h periodicity). Abbreviations: Acro; acrophase, dd; day degrees, h; hour. Group: 1: corticotropin-releasing hormone-related genes, 2: glucagon and secretin-related genes, 3: insulin-like growth factor, 4: insulin receptor genes, 5: key melanocortin neuropeptides, 6: nutrient sensing, 7: oxytocin-related genes, 8: receptors. 9: natriuretic: 10: genes related to reward-based appetite, 11: serotonin-related. 12: thyroid system. 13: Transcription factors and signaling pathways. For the expression pattern for individual genes related to appetite, see-. S2 S4 Figs

Differential expression analyses were conducted between timepoints in the circadian sampling series at both 720 dd and 920 dd, and the resulting differentially expressed genes (DEGs) were investigated for the presence of appetite-related genes. At 720 dd, none of the appetite-related genes were among the DEGs. At 920 dd, DEGs were found when comparing the different sampling timepoints against each other (Table 2). Analyses showed that most DEGs related to appetite were found in the morning before feeding (04:00 and 08:00) when the light was still off, and before and after the first meal of the day (08:00 and 12:00) (3 appetite-related genes in each comparison, Table 2).

Each cell shows the total number of differentially expressed appetite-related genes for each comparison, and the number given in parentheses indicates the total number of differentially expressed genes.

Among the DEGs at 920 dd, seven unique genes related to appetite displayed a significantly different expression pattern (Fig 2). Based on their temporal expression pattern, several genes showed rhythmicity consistent with circadian regulation (variation across 12 h) and feeding, while others displayed a rhythmicity more closely associated with feeding-related responses only.

Genes following a circadian rhythm (significant difference over 12 h) included cannabinoid receptor 1 (cnr1, ENSSSAG00000091836), insulin receptor substrates 2b (insrs2b, ENSSSAG00000031099 and ENSSSAG00000065410), and c-type natriuretic peptide 1 (nppc, also referred to as anfc1, ENSSSAG00000074569). The cnr1 was significantly lower expressed at 12:00 than at 24:00 (Fig 2A). This is in line with MetaCycle results showing that cnr1 has a 24 h period rhythm peaking at 24:00 (Table 1). insrs2b (ENSSSAG00000031099) displayed the most pronounced temporal variation, with significantly lower expression during the night, and elevated levels during the daytime (16:00 vs 04:00 and 20:00 vs 08:00) (Fig 2B). This was in line with the MetaCycle results showing a robust period of 24 h (p-value < 0.001) and an acrophase at 18:00 (Table 1). Similarly, the insrs2b paralog (ENSSSAG00000065410) also showed rhythmicity with decreasing expression during the night (Fig 2C). The expression was significantly lower at 08:00 compared to 20:00. The nppc peaked at night, with a significant increase from 16:00–04:00 and then significantly decreased from 04:00–08:00, supporting the acrophase at 04:00 in MetaCycle (Fig 2D, Table 1).

Genes associated with feeding behavior exhibited temporal dynamics aligned with feeding periods. These genes included cocaine- and amphetamine-regulated transcript 4 (cart4, ENSSSAG00000069679), atrial natriuretic peptide receptor 1-like (npp-r1-like, ENSSSAG00000048802), and melanocortin receptor accessory protein (mrap.3, also referred to as mrap2a, ENSSSAG00000103286). cart4 (ENSSSAG00000069679), npp-r1-like (ENSSSAG00000048802), and nppc (ENSSSAG00000074569) were significantly downregulated between 04:00–08:00, a time window preceding the first meal of the day, suggesting a role in pre-feeding regulation (Figs 2D-F). mrap.3 (ENSSSAG00000103286) showed a postprandial response with significantly lower expression at 12:00 compared to 08:00 (Fig 2G). In contrast, both insrs2b paralogs significantly increased after the first meal, indicating sensitivity to feeding events for these genes. To better assess the circadian entrainment of appetite-regulating genes at 920 dd, the acrophase for appetite-related genes was aligned with the acrophase for clock genes as previously reported in Eilertsen et al., [12], (Fig 2H).

Based on characterizations of the parr brain described by Norland et al., [43], ISH was performed to validate the presence of hypothalamic expression of key neuropeptides involved in the melanocortin system in endogenous feeding Atlantic salmon (740 dd) (Fig 3A-E). Expression of npya (npya1/a2) and cart2 (cart2b1/b2) was observed dorsally in the hypothalamus, near the anterior and posterior tuberal nucleus (NAT and NPT, respectively) before the onset of exogenous feeding in Atlantic salmon (Fig 3A-B). Expression of agrp1 hypothalamic-specific was detected in the lateral tuberal nucleus (NLT), whereas pomca (pomca1/a2) was observed in both the NLT and the pituitary (Fig 3C-D). The expression of agrp1 was restricted to only a few neurons in the hypothalamus, as the expression in normalized counts was too low to be included in the RNA sequencing analyses. The overall spatial distribution of the melanocortin neuropeptide mRNA expression in the coronal sections of the hypothalamus is summarized in Fig 3E. These neuropeptides are known to compete for binding to the melanocortin 4 receptor (mc4r). Signaling through the melanocortin receptors can be modulated by the mrap, which influences receptor signaling efficiency (Fig 3F). Transcriptomic analyses revealed significant developmental changes in neuropeptide expression between 720 dd and 920 dd (Fig 3G). Expression levels of cart1b1, 1b2, 4, and npya2 were significantly lower at 920 dd compared to 720 dd, while cart2a, 2b1, 2b2, 3b, npya1, pomca1, and b were significantly higher expressed at 920 dd. The results were in line with DESeq2 analyses showing that these genes were significantly differentially expressed (p-value < 0.05) (S4 Table). To complement the neuropeptide expression profile, we examined the expression levels of mc4r, previously characterized by Kalananthan et al., [44], and the accessory protein mrap. The four mc4r paralogs (mc4ra1, a2, b1, and b2) were detected in the brain transcriptome, with mc4ra1 being significantly higher expressed at 920 dd compared to 720 dd. mrap.1 was lower expressed at 920 dd, while mrap.2 and mrap.3 were significantly higher expressed at 920 dd compared to 720 dd. Their presence confirms that the melanocortin signaling machinery is available in the brain and can respond to changes in neuropeptide input.

Several studies have confirmed that the hypothalamic nucleus lateral tuberis pars ventralis (NLTv) is the putative homolog in teleosts based on the hypothalamic mRNA and protein expression for agrp1, npy, cart, and pomc in different fish species [22,24,30,43,45–57]. In the present study, the mRNA level of these key neuropeptides was expressed in the brain before the onset of exogenous feeding (Fig 3). Recently, we demonstrated that the NLTv region in Atlantic salmon represents a key region in appetite control expressing agrp1, npya, cart2b, and pomca [43]. In addition, the widespread distribution of npya and cart in hypothalamic and extrahypothalamic brain regions, in addition to pomca expression in the pituitary, indicates that several other brain regions are involved in appetite control, either directly or indirectly, in Atlantic salmon [37,38,43], as in mammals [58,59]. This is in line with studies showing whole-brain or region-specific responses for these neuropeptides in Atlantic salmon [37,38,41,44,60,61]. The integration of signals in the hypothalamus is not fully understood in fish, but studies have indicated the presence of a network of hormones, receptors and a nutrient-sensing system that changes the expression of agrp, npy, cart, and pomc in a feeding-related manner (reviewed in [22,30,32]). Therefore, we assume that the expression of these neuropeptides in NLTv and other brain regions has a role in the overall regulation of energy expenditure and metabolism in Atlantic salmon. A model of neuronal regulation during both endogenous and exogenous feeding is presented in Fig 4.

During endogenous feeding (720 dd), the yolk supports Atlantic salmon growth, development, and metabolism as the fish are considered to have an energetically closed system at this stage [1]. However, the precise mechanisms involved in energy homeostasis and yolk sac utilization are not fully understood. During development in Atlantic salmon, leptin and cholecystokinin (cck) are two of the first major metabolic hormones being expressed [62,63]. In adults, leptin and cck proteins suppress food intake in both mammals and several fish species [64–68], and have been suggested to play a role in long and short-term signaling after a meal, respectively [22,32]. Conversely, studies have shown that several hormones related to appetite contradict this function in different fish species [30,69–71]. In this study, the presence of leptin and cck in the brain displayed lower mRNA expression levels at 720 dd compared to 920 dd, while their putative receptors (leptin and cck receptors) showed a higher expression level in the brain at 720 dd compared to 920 dd (Fig 4A, 4B). In mammals, it has been demonstrated that pleiotropic hormones, especially leptin, affect brain development, both in the establishment of hypothalamic circuits, including axon outgrowth, and neuronal plasticity both pre- and postnatally [72–75]. Similarly, there might be a correlation between hormone levels and corresponding receptor levels that can affect neuronal development in Atlantic salmon. In zebrafish, knock-out of leptin has shown that this gene is important in behavior (anxiety, aggression, fear, and circadian rhythm behavior) [76], while knock-out of leptin receptor showed minor differences in morphometric parameters, but did affect appetite-related genes response [77]. In mammals, leptin has an anorexigenic effect, which occurs after several weeks of oral feeding; its early postnatal presence does not seem to affect feeding [74]. During the first feeding period, decreased receptor expression may allow for a more targeted response of feeding-related feedback of leptin in the brain as negative feedback systems can begin to operate, and that hormone levels are now sufficient to activate downstream pathways. However, further studies are needed to investigate the role of leptin in fish development.

Among nutrient sensors, amino acid sensors and the glucose sensor liver X receptor (lxr) displayed a significant cyclic expression pattern at 720 dd (Table 1). This may indicate a sensing mechanism in the brain during endogenous feeding, which in turn may activate appropriate metabolic responses in Atlantic salmon alevins. Several components are involved in glucosensing in fish, including lxr [34,78,79]. While glucose may be unlikely to play a pivotal role in nutrient-sensing pathways in endogenous-feeding, the fish’s energetic budget relies on the catabolism of proteins, lipids, and amino acids stored in the yolk [2], its levels can induce modifications in epigenetic mechanisms, such as histone acetylation, and contribute to altered metabolism [80]. Two lxr paralogs were identified in the Atlantic salmon brain transcriptome, and both were significantly higher at 920 dd compared to 720 dd (S4 Table). Further, we have previously shown that the glycolytic process was upregulated at 920 dd, with several important enzymes involved in the breakdown of glucose to pyruvate among those DEGs [12]. This indicates possible nutritional adaptation to cope with enhanced carbohydrate levels in exogenous feeding compared to endogenous feeding, helping to optimize the energy utilization obtained from food as the brain requires stable glucose levels to support neuronal development and differentiation while protecting the developing brain from potential damage (hypoglycemia) [34]. In general, higher mRNA expression levels of nutrient sensors were observed at 920 dd compared to 720 dd (Fig 4). This indicates adaptive responses where an exogenous-feeding fish can better sense and regulate nutrient levels in the organism, and regulate energy homeostasis accordingly to meet the metabolic demands of a growing fish. This is in line with GO-terms related to lipid and carbohydrate metabolism, and glucose homeostasis being upregulated at 920 dd [12]. Additionally, the digestion of carbohydrates in exogenous feeding fish may help to increase the energy obtained from feed [8].

At 720 dd and 920 dd, different solute carriers (SLC) for amino acid transporters, which mediate the uptake of amino acids, displayed periodicity. SLC transfers nutrients over the membrane and is linked to electrogenic activity that alters the transmembrane potential, which subsequently can stimulate hormone release [81]. During endogenous feeding, Y + L amino acid transporter 1 (slc7a7) and sodium-coupled neutral amino acid symporter 2 (slc38a2a) had a significant cyclic expression pattern, while cationic amino acid transporter 3 (slc7a3) showed a cyclic expression pattern at exogenous feeding. Both slc7 and slc38 are leucine transporters distributed in several brain regions and may function as a transporter of leucine into the cell, sense nutrient levels, or be involved in other cellular processes [82]. The slc38a2a was the only appetite-related gene examined that displayed a 24 h rhythm at 720 dd, while slc7a7 exhibited a 20 h rhythm. Discriminating between functional metabolic oscillations and rhythms associated with developmental progression will be important for understanding the emergence of feeding competence. The present study suggests that SLC transporters may reflect the sensing of amino acid levels, in which early metabolic rhythm can contribute to developmental mechanisms in the transition from endogenous to exogenous feeding. As fish lack a single master clock like mammals [83], further studies can help determine neurodevelopmental dynamics linking how metabolic and biological rhythms become hierarchically organized in fish [84,85].

The hypothalamic nutrient-sensing system integrates metabolic cues through different signaling pathways, including AMP-activated protein kinase (AMPK), mechanistic target of rapamycin (mTOR), and protein kinase B (Akt), which regulate key transcription factors, including brain homeobox transcription factor (BSX), phosphorylated cAMP (CREB), and forkhead box protein 01 (FOX01) and thereby appetite and energy allocation [32,86,87]. The transition from yolk utilization to exogenous feeding marks a shift in which the brain and metabolism must adapt to an external food supply. In this context, higher levels of akt and mtor were observed at 720 dd, and higher ampk expression at 920 dd than at 720 dd (Fig 4). AKT/mTOR are broadly linked to cell growth, cell proliferation, cellular energy metabolism, and neurogenesis [88,89]. High akt/mTOR expression is correlated with rapid embryonic growth and development in carp [90]. Here, the upregulation in alevins may favor fast growth and neurogenesis before the onset of active food intake, while ampk remains relatively low as ATP demands are met by the yolk substrates [87].

As development progresses and exogenous feeding commences, relative lower akt/mTOR levels and elevated ampk is consistent with an adaptive transition to a different brain state, shaped by nutrient availability and food seeking that is aligned with circadian rhythm at 920 dd [12,87]. Ampk may work as an energy sensor where elevated levels of nutrients are inversely correlated with Ampk [32]. The upregulation of brain ampk in exogenous-feeding Atlantic salmon is in line with a recent study demonstrating an upregulation of ampk when longfin yellowtail larvae had an active food-seeking behavior (swirling near the water surface) [5]. In contrast, studies show that akt and mtor are activated in fed conditions in exogenous feeding fish [32,67,91]. Thus, rather than being markers of “fed” or “fasted”, these signaling proteins are better viewed as part of coordinated reprogramming of the brain metabolism and signaling supporting the life-history transition from endogenous to exogenous feeding Atlantic salmon.

At 720 dd, higher levels of the transcription factors creb and bsx and lower levels of fox01 co-occurred with an overall lower expression of agrp1, npya1, cart2a, 2b, 3b, and pomca, and higher expression of npya2, cart1b, 3a, and 4 in Atlantic salmon compared to 920 dd (Fig 4). High expression levels of CREB and BSX are observed together with high levels of Npy and Agrp in mammals, resulting in stimulation of appetite [92,93]. In rainbow trout, both creb and bsx decreased in response to elevated nutrient levels or the presence of satiety signals [67,94]. An unexpected finding in this study was the high expression of creb and bsx, while agrp1 (detected by ISH but not present in the normalized counts for 720 dd, see Fig 3 and S2 Table) and npya1 were lower expressed during endogenous feeding, and the opposite pattern was observed in exogenous feeding Atlantic salmon. These results indicate that creb and bsx may inhibit the expression of agrp1 and npya1 in Atlantic salmon, the presence of paralog-specific interactions between creb/bsx and agrp1/npya1, or that these transcription factors play a role in energy homeostasis and development that is unexplored. Indeed, more studies are necessary to understand the functions of these transcription factors in the transition from endogenous to exogenous feeding.

Spatial distribution of hypothalamic melanocortin neuropeptide expression during endogenous feeding in Atlantic salmon suggests that the melanocortin pathway undergoes developmental reorganization. agrp1 and pomca were restricted to the hypothalamic NLT, whereas npya and cart2b were located more dorsally towards the anterior tuber nucleus (NAT). This organization differs from that observed at the parr stage, in which agrp1, npya, cart2b, and pomca are expressed in neighboring cells of the ventral NLT (NLTv) [43], highlighting the progressive spatial organization as the hypothalamus matures. Comparable developmental plasticity has been described in mammals, where Pomc-positive progenitor cells are widely distributed in the immature hypothalamus, but many of these neurons will gradually adopt a non-POMC fate during hypothalamic maturation [73,74,95]. Ontogenetic analysis of npy and cart expression in Atlantic cod brain demonstrates developmental shifts in expression profile in the hypothalamus, as well as detectable expression in multiple brain regions beyond the hypothalamus, including the telencephalon and preoptic area [96]. Taken together with the persistent expression of hypothalamic and extrahypothalamic expression of these key neuropeptides in the melanocortin system, these findings support that appetite regulation in teleosts emerges through both region-specific differentiation of the NLTv and developmental refinement of whole-brain networks involved in energy homeostasis [36,44,62,96].

The mc4r is considered the primary receptor mediating melanocortin signaling [27] was neither significantly different nor cyclically expressed. Although all four salmon mc4r paralogs were detected [44], only mc4ra1 was higher expressed at 920 dd, suggesting mc4r maintains relatively stable transcriptional levels or may be regulated post-transcriptionally. Interestingly, the differentially expressed melanocortin receptor accessory protein mrap.3 at 920 dd may modulate the melanocortin receptors postprandial availability and signaling strength towards energy homeostasis [97]. While the current study focused on the whole-brain transcriptome, it is important to note that paralog- and region-specific expressions may be masked in the dataset. Future research targeting specific neuronal populations can provide deeper insights into these mechanisms.

At 920 dd, the Atlantic salmon brain is likely to be hunger-promoted, and adapting the fry towards active food-seeking behavior. During this stage, the fish may learn the sensory properties of food and become conditioned to the feeding regime [30,98], consistent with the significant periodicity of several genes related to reward-based (hedonic) appetite (Table 2). Behavioral changes were observed as Atlantic salmon transitioned from a demersal behavior (720 dd) towards an active free-swimming (food-seeking) behavior at 920 dd, indicating neuromuscular readiness along with morphological and physiological adaptations essential for successful feeding [9,63]. Indeed, several reports indicate that marine fish larvae that have started on exogenous feeding may continue to ingest food even though their gut is already full [8,35], in line with the concept that fish larvae are “feeding machines” and are motivated to feed [23]. The neuronal signals generated at 920 dd may allow the salmon to be hungry and motivated to feed, at least at the start of the first-feeding period, although their feeding behavior is inefficient [99].

None of the key neuropeptides of the melanocortin system displayed a detectable significant postprandial transcriptional response in the whole brain of Atlantic salmon at two weeks into first-feeding period (920 dd). In contrast, three weeks into first-feeding period, npya1, pomca1, and pomca2 displayed significant postprandial neuronal anorexigenic responses [36], indicating that several important neurodevelopmental processes and circuit reorganization occur during this period. Additional pathways, including PYY [100], may also influence the early regulation of satiety and energy homeostasis. The postprandial upregulation of two insrs2 and reduced mrap.3 in the salmon brain at 920 dd (Fig 2) further indicates that feeding modulates the sensitivity and signaling efficiency of insulin and melanocortin pathways by intracellular adaptors and accessory proteins, while maintaining transcriptional regulation of ligands and receptors largely unchanged at the whole-brain level. Meal timing also influenced the transcriptional changes, where prolonged feed deprivation (i.e., 13 h overnight) resulted in a significant change in appetite-related genes, as opposed to short feed deprivation (i.e., 6 h), which did not result in a significant change for the genes in our in-house database, consistent with inter-meal interval affecting the food intake [101]. This pattern points to a brain that adjusts metabolic and neuroendocrine responsiveness by tuning signaling capacity rather than altering the transcription of appetite-regulating genes.

Analysis of cyclic appetite-related genes at 720 dd revealed a predominance of transcripts cycling with a periodicity of 20 h or 28 h, whereas at 920 dd the majority had a periodicity of 24 h. These patterns align with Eilertsen et al., [12] showing that only three clock genes were significantly cyclic in the yolk sac alevins, significant cycling of the clock genes and genes of the circadian rhythm pathway in the exogenous feeding fry. Together, these findings indicate that the circadian machinery might not be fully operational in controlling metabolic genes at 720 dd and could represent an intermediate developmental phase where endogenous oscillators are present but are but not fully entrained to the LD cycle [12]. The periodicity of 20 h and 28 h at 720 dd is likely due to lack of coordination or might also indicate data reflecting individual variations. Except for cart3, none of the cyclic appetite-related genes at endogenous feeding were among the genes that cycled in exogenous feeding salmon fry.

The shift in energetics supply from yolk to a pellet diet changed the oscillation of genes and the establishment of a persistent circadian rhythmicity [12], potentially affecting the coordination of rhythmic physiological functions, and clock-controlled gene outputs affecting appetite and feeding behavior. Genes grouped in insulin receptors, melanocortin neuropeptides, nutrient sensing and natriuretic were cyclic at both 720 dd and 920 dd. The unique cyclic expression of corticotropin-related, insulin-like growth factors, oxytocin-related, and thyroid-related genes at 720 dd might be linked to mobilization of energy and yolk-sac utilization, and developmental progression [23,102,103]. Thus, future studies may explore the role of cyclic appetite-related genes at 720 dd and yolk utilization.

Interestingly, cyclic expression of akt with a 24 h period was observed in exogenous-feeding salmon, which peaked at 22:00, after feeding and when the lights were switched off (Table 1). In mammalian model organisms and mammalian cell culture studies, rhythmic activation of AKT can be achieved in response to external signals such as food ingestion, and it also exhibits cell-autonomous circadian oscillations in the absence of external stimuli such as light:dark cycle [104–106]. All appetite-related genes that exhibited a nocturnal peak aligned with a minimum of one clock gene peaking at the same acrophase (Fig 2H). During daytime, acrophase for clock genes occurred earlier in the day compared to appetite-related genes, suggesting a phase delay in the rhythmic regulation of appetite signals relative to the core clock machinery. The photoperiod is considered the most important synchronizer of biological rhythm, but periodic feeding can also work as an entrainment cue in fish, leading to physiological and behavioral patterns involved in food anticipatory behavior [107–115]. Multiple internal rhythms can also coexist and be entrained by different cues [85]. Our results suggest that there might be an interaction between LD regime and feeding that influences the overall expression of cyclic genes, while underlying mechanisms remain to be solved. Future research should aim to manipulate light-dark cycles and uncouple the feeding time from the LD regime, combined with tissue-specific analyses.

As the fish did not undergo handling except euthanasia, special approval for the experiment was not required according to Norwegian National legislation via the Norwegian Animal Welfare Act (LOV-2015-06-09-16-65) and Regulations on the Use of Animals in Experiments (FOR-2017-04-05-451), given by the EU (Directive 2010/63/EU) for animal experiments. All the fish were euthanized on-site with an overdose of metacaine (MS-222^TM; MSD Animal Health, Netherlands) before being handled further. The trial was conducted at the Bergen High Technology Center (University of Bergen, Bergen, Norway), in an animal facility approved by the Norwegian Food Safety Authority (VSID2135). All personnel involved in the experiment had undergone training (FELASA-C) approved by the Norwegian Food Safety Authority, which is mandatory for running experiments involving animals included in the Animal Welfare Act.

The experimental design, RNA sequencing, and analyses are described in Eilertsen et al., [12] and include normalized count files and differential gene expression analyses generated in DESeq2 [116], and lists of cyclic genes determined by MetaCycle [117]. In summary, two sibling groups of Atlantic salmon eggs (obtained from Mowi, Tveitevågen, Askøy, Norway) were fertilized and incubated under light:dark (LD) period of 14:10 h at 5.7 ± 0.5˚C. Just before first feeding, the fish were transferred to feeding tanks and reared under a 14:10 h light:dark period at 11.05 ± 0.3˚C. The age of the developing alevins and fry was determined using day degree (dd) as a product of days and temperature (day × °C) from the day of fertilization until the last sampling point. To ensure precise age estimation, daily temperature values were computed as the mean of measurements recorded in 10-minute intervals. Food was introduced from 762 dd, and the fish were fed a commercial diet twice a day (09:00–09:30 and 16:00–20:00). Two 48 h circadian sampling series were carried out at 720 dd and 920 dd (i.e., before and after the onset of the exogenous feeding period) with 4 randomly sampled fish every 4 h starting and ending at noon. In total, 104 fish were collected. The brains were dissected from the skull, the total RNA was extracted. Whole brains were dissected for RNA sequencing to maximize the specificity of neuronal tissue. This approach was preferred over head-region or brain-region dissections which could introduce more variability while we acknowledge that whole-brain RNAseq may dilute low-abundant or region-specific transcripts. All RNA sequencing data have been deposited in the European Nucleotide Archive (ENA) under accession number PRJEB74272. Accessible via https://www.ebi.ac.uk/ena/browser/view/PRJEB74272↗.

The raw sequencing results were trimmed by Trimmomatic v0.38 [118] and subsequently alignment to the published Atlantic salmon genome (http://ftp.ensembl.org/pub/release-106/gtf/salmo_salar/↗) using STAR v2.7.0 [119]. The output files from the aligner were processed by Samtools 1.6 [120], and counts were generated using HTSeq v0.11.2 [121]. Using DESeq2 1.26.0 [116], the three normalized count arrays were generated corresponding to (i) 720 dd, (ii) 920dd, and (iii) 720dd and 920 dd together. The normalized count files were filtered by excluding genes with a coefficient of variance greater than 100, and genes with an average count of less than 5 within each sample. DESeq2 was applied to do differential expression analyses using Wald statistical model where genes with counts less than 10 per comparison were not included in the comparisons, and an adjusted p-value threshold was set to < 0.05.

To identify genes exhibiting daily rhythms for the two 48 h circadian sampling series of 720 dd and 920 dd, the normalized count arrays were analyzed by MetaCycle v1.2.0 [117] incorporating JTK_CYCLE algorithm [122]. Period detection parameters were set to default settings in JTK_CYCLE, with a minimum of 20 h and a maximum of 28 h, using Bonferroni correction for combined p-values. Rhythmic genes were identified at multiple adjusted p-value thresholds (p < 0.1, p < 0.05, p < 0.01 and p < 0.001) with a minimum meta2d_Base of 10 counts.

Exploration of the three normalized count arrays was done using Venny v2.1 (RRID:SCR_016561) to visualize shared and unique EnsemblID across the normalized count arrays and in-house database of genes involved in appetite and metabolism. Up to four gene lists were compared and intersections were identified and extracted directly from the interactive diagram. The transcriptome data was further explored in RStudio R.4.2.2 [123], with the packages pheatmap [124] and ggplot2 [125] used to create figures. Heatmap for the whole transcriptome was visualized by calculating the mean value of each gene at each sampling time point, while the heatmap for appetite-related genes was made by calculating the mean value for each “category” at each sampling time point. Expression values were scaled by row to highlight relative differences of gene or gene group. Hierarchical clustering of rows was performed using correlation-based distance, while column order was preserved to reflect sampling sequence. The dendrogram height parameter was adjusted to display row clustering (treeheight_row = 20) and the column dendrogram was suppressed. The 48 h circadian sampling series were further analysed using CircaCompare [126] to generate cosinusoidal expression curves for appetite-related genes identified as rhythmic by MetaCycle, enabling comparison between the two developmental stages.

To identify candidate genes involved in appetite and metabolism, an in-house database of 464 genes putatively involved in these processes was generated independently of the RNA-seq data (S1 Table). The in-house database included (i) key receptors and neurotransmitters of the melanocortin system [20,127] including existing functionally characterized key neuropeptides and melanocortin 4 receptors in the Atlantic salmon melanocortin system [37,38,41,44,60,61] updated to the current annotated salmon genome (Ssal_v3.1), (ii) transcription factors and signaling pathways related to the melanocortin system in the fish hypothalamus [32], (iii) genes related to the adipocytokine signaling pathway (KEGG entry map04920, https://www.genome.jp/kegg/pathway.html↗), (iv) genes related to nutrient sensing and reward system linked to appetite control in teleosts [30,32,34,78,82,91,128,129], and (v) in-house database for appetite-controlling genes in Atlantic cod Gadus morha [130]. To generate the Atlantic salmon in-house database, annotated genes in Atlantic cod, human, zebrafish, and rainbow trout (Oncorhynchus mykiss) genomes were blasted (BLASTp, https://blast.ncbi.nlm.nih.gov/Blast.cgi?PROGRAM=blastp&PAGE_TYPE=BlastSearch&LINK_LOC=blasthome↗) against the Atlantic salmon genome (Ssal_v3.1). BLASTn and multiple sequence alignments (VectorNTI) were used to identify Atlantic salmon paralogs.

Atlantic salmon were sampled at 740 dd for chromogenic ISH. The fish were euthanized (200 mg/L of MS-222), and heads were cut off and fixated in 4% paraformaldehyde in phosphate-buffered saline pH 7.4. Atlantic salmon heads were rinsed in 1 × phosphate buffered saline and immersed in 25% sucrose/25% OCT (CellPath, UK) for 24 h as described in Eilertsen et al., [131]. The heads were embedded in 100% OCT and coronal parallel cryo-sectioned across the hypothalamus at 10 µm (Leica CM 3050s, Leica Biosystems, Germany) and dried at 65 °C for 30 min. The topographical distribution of agrp1, npya, cart2b, and pomca in the hypothalamus was investigated based on their hypothalamic expression in the Atlantic salmon parr brain [43]. Synthesis of antisense digoxigenin (DIG)-labeled riboprobes and ISH using 4 M urea was carried out as described in Norland et al., [43]. Sections were imaged using a Leica DM600B microscope (Leica Microsystems, Germany) with a digital Leica DFC 320 (Leica Microsystems) camera.

The data presented in this publication have been deposited in the European Nucleotide Archive (ENA) under accession number PRJEB74272. Accessible via https://www.ebi.ac.uk/ena/browser/view/PRJEB74272↗. All relevant data are within the manuscript and its Supporting Information files.