The AMPK system of salmonid fishes was expanded through genome duplication and is regulated by growth and immune status in muscle

5′adenosine monophosphate-activated protein kinase (AMPK) is a master regulator of energy homeostasis in eukaryotes. This study identified expansions in the AMPK-α, -β and -γ families of salmonid fishes due to a history of genome duplication events, including five novel salmonid-specific AMPK subunit gene paralogue pairs. We tested the hypothesis that the expanded AMPK gene system of salmonids is transcriptionally regulated by growth and immunological status. As a model, we studied immune-stimulated coho salmon (Oncorhynchus kisutch) from three experiment groups sharing the same genetic background, but showing highly-divergent growth rates and nutritional status. Specifically, we compared wild-type and GH-transgenic fish, the latter achieving either enhanced or wild-type growth rate via ration manipulation. Transcript levels for the fifteen unique salmonid AMPK subunit genes were quantified in skeletal muscle after stimulation with bacterial or viral mimics to alter immune status. These analyses revealed a constitutive up-regulation of several AMPK-α and -γ subunit-encoding genes in GH-transgenic fish achieving accelerated growth. Further, immune stimulation caused a decrease in the expression of several AMPK subunit-encoding genes in GH-transgenic fish specifically. The dynamic expression responses observed suggest a role for the AMPK system in balancing energetic investment into muscle growth according to immunological status in salmonid fishes.


Results
Phylogenetic analysis of AMPK subunits and nomenclature system. Three separate maximum likelihood (ML) phylogenetic analyses of amino acid sequence were performed to reconstruct the evolutionary history of the AMPK-α, -β and -γ subunit families (Figs 1-3, respectively). By including sequences gathered from multiple teleost lineages, including salmonids, these analyses accommodated the presence of paralogues retained from the tsWGD and ssWGD events. For the included salmonid species, the databases searched contain RefSeq genes predicted within reference genomes assembly for each species, namely: Atlantic salmon (Salmo salar) 20 , rainbow trout (Oncorhynchus mykiss) (unpublished; NCBI accession: GCA_002163495), Chinook salmon (O. tshawytscha) 34 , coho salmon (O. kitsuch) (unpublished; NCBI accession: GCA_002021735.1) and Arctic charr (Salvelinus alpinus) 35 . For all AMPK genes, we employed a standard nomenclature system advocated elsewhere 36 with putative paralogues retained from the teleost WGD named "gene A" and "B". In cases where there was no evidence for the retention of teleost paralogues, we used "gene A". We gave salmonid-specific paralogues a sub-annotation of "1" or "2" (e.g. "gene A1" and "A2") 36 .
Novel salmonid paralogues of AMPK-α. The AMPK-α tree split into two clades (AMPK-α1 and -α2), represented by all included vertebrate species, as reported elsewhere 6,8 (Fig. 1). In both clades, the branching of represented ray-finned fishes follows expected species relationships, assuming a lack of retention of teleost-specific paralogues, with spotted gar (Lepisosteus oculatus) being the sister clade to teleosts 37 . However, in both the AMPK-α1 and -α2 clades, the represented salmonid sequences are split across distinct clades represented by different salmonid species (Fig. 1). Within the AMPK-α2 clade, the salmonid sequences split into two monophyletic clades (Fig. 1). However, in the AMPK-α1 clade, the salmonid sequences did not split cleanly into two monophyletic clades (Fig. 1). As the AMPK-α amino acid sequences for salmonids are highly conserved, we performed ML phylogenetic analyses at the nucleotide-level, which provides greater phylogenetic signal 19 . Note, this was done for all amino acid trees with any evidence for salmonid-specific paralogues. In the nucleotide level analyses, the salmonid sequences split into two clades representing each of the included species for both AMPK-α1 and AMPK-α2 (Fig. S1). As salmonid sequences branch as the sister clade to northern pike (Esox lucius), which did not undergo the ssWGD, we conclude that the salmonid clades are products of (2019) 9:9819 | https://doi.org/10.1038/s41598-019-46129-4 www.nature.com/scientificreports www.nature.com/scientificreports/ ssWGD. We hereafter name these paralogue pairs AMPK-α1A1 and AMPK-α1A2, along with AMPK-α2A1 and AMPK-α2A2. Adding further weight to these inferences, the chromosomal location of the same genes in Atlantic salmon 20 , specifically AMPK-α1A1 (~53 Mbp along Ssa11) vs. AMPK-α1A2 (~139Mbp along Ssa01) and AMPK-α2A1 (~29 Mbp along Ssa10) vs. AMPK-α2A2 (~22Mbp along Ssa23) matches coordinates of large blocks of duplicate genes that have maintained collinearity since the ssWGD 20 .
Novel salmonid paralogues of AMPK-β. The AMPK-β subunit tree split into two clades (AMPK-β1 and -β2), represented by all included vertebrate species, as reported elsewhere 6,8 (Fig. 2). The AMPK-β2 clade followed the same pattern as the AMPK-α clades, with no evidence for retention of teleost-specific paralogues. Again, the salmonid AMPK-β2 sequences are split across different clades representative of different salmonid species (Fig. 2). However, the salmonid AMPK-β2 sequences did not split into two clades (Fig. 2), as to be expected if they originated during the ssWGD event. This was likely an artefact, as the nucleotide level ML analysis, split the salmonid AMPK-β1 sequences into two monophyletic clades, each representing the five included species (Fig. S3). Hence, two AMPK-β2 paralogues were evidently retained from ssWGD, named AMPK-β2A1 and AMPK-β2A2. The chromosomal location of AMPK-β2A1 (~44 Mbp along Ssa03) and AMPK-β2A2 (~43 Mbp along Ssa14) are also consistent with an origin from ssWGD 20 .
For the AMPK-β1 clade, there is evidence for two teleost paralogues which are likely the result of the tsWGD. (Fig. 2). Spotted gar is the sister to two teleost clades represented by different lineages, including salmonids (Fig. 2). As a result, the salmonid sequences are named AMPK-β1A and AMPK-β1B, but there is no evidence for salmonid-specific paralogues in either teleost clade (Fig. 2).
GH-transgenesis alters constitutive expression of AMPK subunit genes in skeletal muscle. Constitutive differences in transcript level among the three experimental groups were quantified in control (PBS-injected) skeletal muscle for each of the fifteen unique AMPK subunit genes; done using qPCR with primers specific to salmonid-specific paralogues (Table 1). A statistically significant group effect was observed for nine of the genes ( Table 1). Several of the subunit genes were significantly upregulated in the GH-transgenic groups when compared to wild-type (WT), with differing effects for transgenic full ration (TF) and transgenic restricted ration (TR) fishes. AMPK-α2A1 (1.7-fold), AMPK-β1B (2.1-fold), AMPK-γ2B1 (4.4-fold) and AMPK -γ2B2 (5.0-fold) were upregulated in TF vs. WT (Table 1). AMPK-α2A2 (1.9-fold), AMPK-β1A (1.9-fold), AMPK-β1B (1.9-fold) and AMPK-γ1A1 (1.8-fold) were significantly upregulated in TR vs. WT (Table 1). AMPK-γ2B1 was significantly upregulated in TF vs. TR (4.0-fold), while AMPK-β1A (1.5-fold) and AMPK-γ3A (1.8-fold) were both downregulated in TF vs. TR (Table 1). www.nature.com/scientificreports www.nature.com/scientificreports/  www.nature.com/scientificreports www.nature.com/scientificreports/ Immune status alters AMPK subunit expression in GH-transgenic skeletal muscle. Challenge with two immune stimulants, peptidoglycan ( Table 2) and polyinosinic:polycytidylic acid (Poly I:C) (Table 3), intended to generate a pro-inflammatory and antiviral response, respectively, caused marked changes in the expression of several AMPK subunit genes. We observed systematic responses between the GH-transgenic and WT groups, recaptured by significant strain:treatment interaction effects (Tables 2 and 3). In the GH-transgenic groups, immune stimulation almost invariably caused downregulation of AMPK subunit expression, when many of the same genes were upregulated in WT (Tables 2 and 3). Several AMPK subunit genes were affected by immunostimulation in either but not both TF and TR, indicating an effect depending on nutritional status.  Table 2. AMPK subunit gene expression responses of TF, TR and WT coho salmon groups to peptidoglycan stimulation Transcript levels (mean ± s.d., n = 5) are given for the three experiment groups (peptidoglycaninjected) and are quantitatively comparable across groups/genes. '*' Indicates a significant change in expression due to treatment; for these genes, fold change values are shown, which were calculated by dividing the mean peptidoglycan treatment by the mean control transcript levels (  Table 3. AMPK subunit gene expression responses of TF, TR and WT coho salmon groups to Poly I:C stimulation Transcript levels (mean ± s.d., n = 5) are given for the three experiment groups (Poly I:C-injected) and are quantitatively comparable across groups/genes. '*' Indicates a significant change in expression due to treatment; for these genes, fold change values are shown, which were calculated by dividing the mean Poly I:C treatment by the mean control transcript levels ( www.nature.com/scientificreports www.nature.com/scientificreports/ Interestingly, the same AMPK subunit genes typically showed highly coupled responses to separate peptidoglycan and Poly I:C challenges (compared Tables 2 and 3).

Discussion
The first objective of this study was achieved by establishing the complete set of AMPK subunit genes retained in the salmonid lineage from the tsWGD and ssWGD events. Phylogenetic analyses revealed genetic expansions of all three AMPK subunits as a product of the tsWGD event, confirming past findings 6,8 . Furthermore, we identified additional salmonid-specific paralogues due to the ssWGD, an event that has been associated with an high overall paralogue retention rate 20,21 , which nonetheless shows variability across different gene families (e.g. 38 ). Evidently, salmonids retain at least fifteen transcriptionally active and presumably functional genes encoding AMPK complex proteins, four for AMPK-α, four for AMPK-β and seven for AMPK-γ. This represents a significant expansion compared to humans (seven genes) and other teleosts such as zebrafish (ten genes). Our phylogenetic analyses will be useful for researchers wishing to further explore the salmonid AMPK system, providing a reference to robustly interpret signals of functional divergence and conservation among paralogues.
The second objective of the study was achieved by determining the extent to which AMPK genes are transcriptionally regulated by growth and immune status in coho salmon skeletal muscle, testing the hypothesis that AMPK plays a role in the regulation of resource allocation across different physiological systems. We confirmed that AMPK is extensively transcriptionally regulated by GH-transgenesis, with some variation between TF and TR, and by immune stimulation from both bacterial and viral mimics, with striking differences between GH-transgenics and WT. This is consistent with our past work on the same coho salmon muscle samples, where GH-transgenesis was shown to alter the expression responses of both the innate immune system, as well as the GH and IGF pathways, in comparison to WT 27 . However, it is important to emphasize that our study provides no information on changes in AMPK regulation at the level of total protein abundance, or phosphorylation status, both which may fail to correlate with transcript levels.
Changes in baseline mRNA expression for AMPK subunit genes usually involved upregulation in TF and/ or TR compared to WT, including for AMPK-α2A1 (TF only), AMPK-γ2B1A (TR only) and AMPK-γ2B2 (TF only). The AMPK-α subunit contains the main phosphorylation site leading to AMPK activity making it a strong candidate for mediating the effects of GH-transgenesis. Indeed, enhancements in AMPK-γ2 expression lead to a concomitant AMPK activity escalation in mice 39 . Our results are thus contrary to the assumption that increased AMPK activity would lead to the scaling down of anabolic pathways and increases in catabolic pathways, considering the increased growth phenotype of TF. However, in mouse, overexpression of AMPK-α and γ subunits can lead to the accumulation of glycogen 40,41 , which may support faster growth rate. Perhaps related to this observation, our previous proteomic study highlighted pervasive changes in skeletal muscle carbohydrate metabolism in the same GH-transgenic animals when compared to WT, with GH-transgenics showing an increased abundance of proteins involved in glycolysis, glycogenolysis and gluconeogenesis 42 . Differing constitutive expression of both AMPK-γ2B paralogues between TF and TR may be caused by differences in growth rate, as induced by elevated endocrine IGF-I in TF salmon 30 compared to both the TR and WT groups. The additional increase in expression of AMPK-β1A in TR compared to WT, suggests a direct modification due to GH, independent of IGF-I mediated effects, given that IGF-I levels are comparable in TR and WT 30 . Contrary to these increases, AMPK-γ3A was significantly lower in TF compared to TR. In mammals, AMPK-γ3 is predominantly expressed in skeletal muscle 43 , where it is important for glycogen metabolism, indicating that decreased expression of AMPK-γ3 leads to glycogen accumulation [44][45][46] , with potential implications for other glycogen driven anabolic pathways. These findings highlight disruption of WT AMPK signalling due to GH-transgenesis across all three AMPK subunit families.
The most established function of AMPK is sensing cellular energy status, leading to modifications of metabolic pathways, but AMPK also has known roles in regulating immune responses [47][48][49][50] . The transcriptional response of innate immune markers (previously measured for the same set of samples used in this study) highlighted a strong impact of GH-transgenesis on immune function in coho salmon skeletal muscle 27 . After exposure to Poly I:C, no appreciable antiviral response was present in TF, whereas robust upregulation of gene markers for type-I interferon signalling occurred in WT fish 27 . Treatment with peptidoglycan led to an attenuated upregulation of inflammatory cytokines in TF fish compared to WT fish 27 . Interestingly, in TR the same immune marker genes showed an intermediate response to immune stimulation, suggesting that accelerated growth rate was the primary cause of attenuated muscle immune function in TF 27 . Moreover, both TF and TR showed complex alterations in www.nature.com/scientificreports www.nature.com/scientificreports/ the expression of GH and IGF pathway genes relative to WT in response to immune stimulation 27 . Such alterations in the relationship between growth and immune systems in skeletal muscle suggest an accompanying requirement for a shift in AMPK function as the main cellular energy sensor. Our data provides support for this concept through complex differences in the expression of AMPK subunit genes as a function of growth and immune status.
A common set of AMPK-α subunit genes experienced downregulation of mRNA expression in both TF and TR vs. WT in response to peptidoglycan and Poly I:C. Interestingly, many of the same genes were upregulated in WT, which may serve to activate catabolic AMPK pathways needed for energetic reallocation towards a robust immune response 48,51 . Whatever the underlying cause for upregulation in WT fish, the reciprocal downregulation observed in the GH-transgenic strain suggests a disruption of the normal AMPK system transcriptional response to immunological challenge. The AMPK-α subunit contains the main site for phosphorylation at Thr172, leading to increases in AMPK activity by more than 100-fold 52 . The observed reduced expression of AMPK-α subunit genes during immune stimulation could lead to disruption of signalling pathways in the cytosol, transcription and gene expression in the nucleus and a reduced ability to uptake glucose into skeletal muscle 46,53 . Additionally, AMPK-α1 has been shown to be crucial for AMPK activation due to glucose deprivation, including from cytotoxic T lymphocytes 51 . Alterations to such functions could potentially impact the ability of the AMPK complex to sense cellular energy status with impacts on the allocation of energetic investment into growth and immune functions. Both immune stimulants caused reduced expression of AMPK-β2A2 in TF and TR, while AMPK-β1B also showed significantly lower expression for TF compared to WT in response to peptidoglycan. This decline in AMPK-β subunit gene expression may lead to overall reduction in AMPK activity due to the importance of the AMPK-β subunit in AMP binding and subsequent activation of AMPK via phosphorylation 54 . Glycogen serves an important role inhibiting AMPK activation through binding of the β subunit 55 , indicating that the observed decrease in AMPK-β expression in TF could lead to an impaired ability for AMPK to respond properly to excess glycogen, thus potentially reducing the ability of the skeletal muscle tissue to produce a coordinated shift into a required catabolic state for a proper immune response. The trend of decreased AMPK subunit expression in GH-transgenic salmon continued with AMPK-γ, highlighted by reductions of AMPK-γ2B1 and −2B2 in response to both peptidoglycan and Poly I:C for TF compared to WT. Similar to AMPK-α, decreased expression of AMPK-γ2B1 and −2B2 in TF exposed to peptidoglycan and Poly I:C is compounded by the heightened expression of the WT group in response to either immune stimuli. However, the importance of AMPK-γ to enzyme activity cannot be understated, as it is the primary binding site for AMP and ATP used by the AMPK complex to determine cellular energy status 56,57 . A reduced ability to sense dynamic cellular energetic status could contribute to an attenuation of cross-talk between the growth and immune systems linked to excess GH, causing an attenuated immune response in GH-transgenic fish.
Given the increased number of AMPK subunits present in the salmonid genome, it is important to contrast expression responses among salmonid-specific paralogues. For example, paralogue pairs for both AMPK-α1A and AMPK-γ2B showed similar expression changes in response to peptidoglycan and Poly I:C in all three strains, suggesting maintenance of functional elements controlling their transcriptional regulation. However, for AMPK-α2A and AMPK-β2A, single genes in each paralogue pair (AMPK-α2A2 and AMPK-β2A2, respectively) were significantly altered by the same immune challenges, suggesting loss of immune-responsive elements in one gene duplicate, perhaps by sub-functionalization. In addition, in several cases the constitutive mRNA expression levels of AMPK subunit genes was notably different among paralogue pairs, including for AMPK-β1A, AMPK-β2A, AMPK-γ1A and AMPK-γ2B. The overall importance of divergent or conserved expression profiles between many salmonid-specific paralogues remains unclear. However, as no two salmonid-specific AMPK subunit paralogues act in a completely antagonistic fashion, the most likely scenario is that paralogues are acting in a cooperative and/or synergistic fashion. These findings further emphasise an ongoing need to properly define the functional physiological importance of paralogue expression divergence following ssWGD, which represents an important ongoing study area 18 .
In conclusion, salmonid fishes retain an expanded set of genes encoding the AMPK-α, β and γ subunits, owing to a history of whole genome duplication events. Using a coho salmon model that allowed us to contrast the impact of immune stimulation on GH-transgenic vs. wild-type fish, we demonstrate that genes from each AMPK subunit family, which together are required for full AMPK activity 58 , show coordinated transcriptional responses to changes in growth and immune status in skeletal muscle. While additional work is needed to understand how such complex mRNA responses correspond to AMPK function via changes in the translation (and post-translational status) of individual AMPK subunit paralogues, our findings support a role for the AMPK system in balancing investment between growth and immunity in fish.

Materials and Methods
Phylogenetic analyses of AMPK subunits. Protein-level phylogenetic analyses were performed separately for the AMPK-α, -β and -γ subunits, which represent different gene families with unique evolutionary histories (e.g. 6,8 ). As a start point, the complete set of AMPK-α, -β and -γ family members from human (Homo sapiens) (AMPK-α1: NP_006242.5; AMPK-α2: NP_006243.2; AMPK-β1: NP_006244.2; AMPK-β2: NP_005390.1; AMPK-γ1: NP_002724.1; AMPK-γ2: NP_057287.2; AMPK-γ3: NP_059127.2) were used as queries in BLASTp 59 searches against the NCBI non-redundant protein database to extract putative orthologues from a range of vertebrates, including salmonids with available reference genome sequences. The collected protein sequences for AMPK-α, (n = 38), AMPK-β (n = 42) and AMPK-γ (n = 65) were separately aligned using MAFFT v7 60 with default settings followed by quality filtering (default settings) using the GUIDANCE2 algorithm 61 through the GUIDANCE server 62 . This led to the following number of aligned amino acid positions: 506, 231 and 327 for the AMPK-α, -β and -γ datasets, respectively. The IQ-TREE maximum likelihood (ML) approach 63 and server 64 were used to determine the best-fitting amino acid substitution model (JTT + G4 in each case) separately for the three individual subunit alignments and build consensus ML trees with the same model, using 1,000 ultrafast bootstrap replicates 65 . www.nature.com/scientificreports www.nature.com/scientificreports/ In addition, we performed six separate nucleotide-level phylogenetic analyses, one per each putative pair of salmonid-specific paralogues identified in the amino acid ML trees. In these analyses, all salmonid-specific paralogues from each included salmonid species were aligned along with the relevant northern pike (Esox lucius) orthologue (closest outgroup species to the ssWGD event) 25 using MAFFT v7, as described above. Subsequently, ML trees were built in IQ-TREE as described above except incorporating the best-fitting model of nucleotide substitution and without alignment filtering. Consensus ML trees were visualized and edited using FigTree v1.4.3 (http://tree.bio.ed.ac.uk/software/figtree/).

Primer design and quantitative PCR.
Details of all primer pairs used for quantitative PCR (qPCR) are provided in Table S1. New primers matching to fifteen unique AMPK subunit genes identified in salmonids (see Results section) were designed for coho salmon (Table S1). We designed primers in regions that were as conserved as possible across salmonid species and distinguishing among any identified pairs of salmonid-specific paralogues (noted in Table S1). Design of new primers was aided by the use of NetPrimer (PREMIER Biosoft), which predicted no self-or cross-dimers. Primers were also predicted to either span an exon-exon boundary or be positioned in different exons, based on cross-referencing with the reference coho salmon genome (NCBI accession; GCA_002021735.1).
The cDNA samples used as the template for all qPCR analysis reported in this study were prepared during a past study 27 (full methods described therein). Briefly, three coho salmon groups were used at matched body sizes, which required sampling at different ages due to differences in growth rate: i) satiation-fed wild-type (WT) fish aged 19 months; ii) satiation-fed GH-transgenic fish aged 6 months (transgenic full ration: 'TF'); iii) GH transgenic fish aged 17 months fed a restricted wild-type ration (transgenic restricted ration: 'TR'). Fish were sampled either 30 h after Poly I:C injection to mimic a viral infection (n = 5 cDNA biological replicates per TF, TR and WT), 30 h after peptidoglycan injection to mimic a bacterial infection (n = 5 cDNA samples per TF, TR and WT) or 30 h after PBS injection as a control (n = 5 cDNA samples per TF, TR and WT).
An Mx3005P qPCR System (Agilent Technologies) was used to measure transcript levels of all target genes. Each reaction contained 5 µl of 1:100 cDNA (corresponding to 2.5 ng of reverse-transcribed total RNA), 500 nM sense/antisense primers and 7.5 µl Brilliant III Ultra-Fast SYBR Green (Agilent Technologies), totalling 15 µl per reaction. Conditions for thermal cycling were 1x cycle of 95 °C for 3 min, followed by 40x cycles of 95 °C for 20 seconds and 64°C for 20 seconds. The presence of a single product for all assays was confirmed with a dissociation analysis (thermal gradient from 55°C to 95°C). Technical duplicates were performed for each assay, and each 96-well plate contained all samples for each target gene. Duplicate no-template controls (cDNA replaced with water) were used in each plate. LinRegPCR was used to determine the efficiency of each qPCR assay following published recommendations 66 . Two reference genes (RpL13 and ACTB) selected for data normalization were shown to be the most suitable among a panel of five candidates in a previous study using the same samples (primer sequences are provided therein) 27 . The program GenEx (MultiD Analyses AB) was used for efficiency correction and normalization to arbitrary transcript levels on a quantitatively comparable scale for all AMPK subunit genes measured.
Statistical analyses. R-studio v1.0.136 (Rstudio, Boston, MA) interfacing with R v3.3.2 ("Sincere Pumpkin Patch") was used for statistics and graphical functions, using the normalized qPCR transcript values. Two genes had non-detectable Ct values (AMPK-γ2A1 and AMPK-γ2A2 for 2 and 7 out of 45 samples, respectively); these missing values were imputed 67 to increase statistical power, using missForest 68 . A linear model was fit to each AMPK subunit gene to determine the overall effect of strain (WT vs. TF vs. TR: fixed factors), treatment (PBS vs. Poly I:C or PBS vs. peptidoglycan: fixed factors) and, where relevant, the strain:treatment interaction. Where statistically significant overall effects were observed (P < 0.05), pair-wise comparisons using Tukey's method were completed to determine differences among the different levels in the model. The model residuals were tested for normality (Anderson-Darling test) and homogeneity in variance (Levene's test) as well as through visual assessment. When the data failed to meet these assumptions, Box-Cox transformations were completed using the 'car' package 69 , which recovered normality and homogeneity in variance for all comparisons, except the strain comparison for AMPK-γ3A, which required a Kruskal-Wallis test.