At Least Two Genes Encode Many Variants of Irak3 in Rainbow Trout, but Neither the Full-Length Factor Nor Its Variants Interfere Directly With the TLR-Mediated Stimulation of Inflammation

The interleukin-1-receptor-associated kinase 3 (IRAK3) is known in mammals as a negative feedback regulator of NF-κB-mediated innate-immune mechanisms. Our RNA-seq experiments revealed a prototypic 1920-nt sequence encoding irak3 from rainbow trout (Oncorhynchus mykiss), as well as 20 variants that vary in length and nucleotide composition. Based on the DNA-sequence information from two closely related irak3 genes from rainbow trout and an irak3-sequence fragment from Atlantic salmon retrieved from public databases, we elucidated the underlying genetic causes for this striking irak3 diversity. Infecting rainbow trout with a lethal dose of Aeromonas salmonicida enhanced the expression of all variants in the liver, head kidney, and peripheral blood leucocytes. We analyzed the functional impact of the full-length factor and selected structural variants by overexpressing them in mammalian HEK-293 cells. The full-length factor enhanced the basal activity of NF-κB, but did not dampen the TLR2-signaling-induced levels of NF-κB activation. Increasing the basal NF-κB-activity through Irak3 apparently does not involve its C-terminal domain. However, more severely truncated factors had only a minor impact on the activity of NF-κB. The TLR2-mediated stimulation did not alter the spatial distribution of Irak3 inside the cells. In salmonid CHSE-214 cells, we observed that the Irak3-splice variant that prominently expresses the C-terminal domain significantly quenched the stimulation-dependent production of interleukin-1β and interleukin-8, but not the production of other immune regulators. We conclude that the different gene and splice variants of Irak3 from trout play distinct roles in the activation of immune-regulatory mechanisms.


INTRODUCTION
The immune system of vertebrates enables rapid, very efficient, and sustainable defense against microorganisms and viruses. Shortly after the detection of pathogens, amplifier mechanisms boost various destructive activities, which may also be directed against the host itself. To avoid such pathophysiological conditions, the immune system is tightly regulated by a vast number of inhibitory factors (1). One of the factors that counteracts pro-inflammatory processes is the interleukin-1receptor-associated kinase 3 (IRAK3) alias IRAK-M. In general, all four IRAK-family members (IRAK1-4) are composed of four defined regions: an N-terminal death domain, the ProST region, which is rich in proline, serine, and threonine residues, and a kinase domain adjacent to the C-terminal region (2)(3)(4)(5). In contrast to IRAK3, the IRAK-family members−1,−2, and−4 positively regulate inflammatory processes. While mammalian species express all four IRAK paralogs, genomes of several non-mammalian vertebrates feature only one or two IRAK paralogs (2). An IRAK3 ortholog has, for instance, been lost in sauropsidia, amphibians (2,6), and several fish species.
Two components of the putative myddosome of rainbow trout (Oncorhynchus mykiss) have already been characterized by our group, the TLR adapter myd88 and its binding partner irak4 (21)(22)(23). So far, irak1 has not been investigated in trout, but has in other fish species (24,25), and irak2 has been lost in fish (26). The first mention of irak3 as a promising marker for the early immune response of zebrafish is found in a report by Stockhammer et al. (27), but it took another 9 years until the first Irak3 factor was structurally und functionally characterized in a (serranid) fish species (28).
Meanwhile, we also began analyzing the structure and function of Irak3 from rainbow trout, but we encountered several problems. First and above all, we consistently obtained irak3 amplificates with various irak3-specific PCR primers, which all showed atypical shoulder peaks in melting-curve analysis, indicating the presence of various isoforms. We wondered (i) how many Irak3 variants are expressed in trout, (ii) how would they differ structurally, and (iii) what physiological effects the different variants might cause during the immune response of trout. The present report characterizes more than 20 irak3-mRNA isotypes including their splice variants from rainbow trout. We expressed four structurally very different irak3 transcripts in two cell models and quantified their potential to interfere with immune-activating mechanisms.

Identification of Irak3-Encoding Sequences in RNA-Seq Datasets From Rainbow Trout
At the outset of our investigations of an IRAK3 ortholog in salmonid fish, no full-length cDNA sequence for any teleostean ortholog was known. Therefore, we performed a BLAST search against the next-generation RNA-sequence reads of rainbow trout O. mykiss, which had been generated in our institute using the Illumina RNA-seq technology. The procedure is only briefly presented below, since a more detailed presentation of the experiment has already been published elsewhere (29). Total RNA was extracted from six tissues (gills, head kidney, heart, liver, spleen, white muscle) of rainbow trout. Trout were 10-11 months in age with an average weight of 333.0 ± 54.9 g. Library constructions followed the TruSeq RNA Sample Preparation v2 Guide supplied with the TruSeq RNA Sample Prep Kit v2 (Illumina) with minor modifications. Three micrograms of total RNA were purified, fragmented, and then used as template for cDNA synthesis. Subsequently, Illumina paired-end adapters were ligated to the 3 ′ -adenylated cDNA ends. After determining their DNA concentrations, the libraries of each tissue from eight individuals per strain, marked with distinct adapters, were loaded into separate Illumina-flow cells. Thirty-six sequencing cycles were conducted on an Illumina Genome Analyzer using the Multiplex Paired-End-Sequencing kits (Illumina). The resulting short-sequence reads were filtered for quality by our in-house Linux program FILTRIX (unpublished). The filter steps included the removal of adapter sequences to trim away low-quality regions and merge overlapping reads. Based on this primary-data analysis comprising 332 million high-quality sequencing reads of the trout transcriptome, we used the bioinformatics software Unipro UGENE v1. 16 (30) to search for irak3 reads.
In addition, we performed 3 ′ -RACE experiments to elucidate the 3 ′ -end sequence of the truncated irak3 variants. To this end, we used the sense RACE-primer f1 5 ′ -AAGCTGTCTGAT TTTGGGACGG-3 ′ derived from the 5 ′ -region of exon 9 (5 ′ upstream of the site where the coding region of the truncated irak3 variant terminates) and a nested RACE-primer f2 5 ′ -TGC CGGAGGAATACATACGATG-3 ′ located further downstream on the same exon. An oligo(d)T oligonucleotide served as antisense primer. All PCR amplicons were cloned into pGEM-T Easy vector (Promega) and sequenced using the MegaBACE capillary sequencer (GE Healthcare). Each nucleotide position was sequenced at least four times.

In-silico Structural Characterization of Irak3 Variants
Global sequence alignments were performed using the EMBOSS/Stretcher tool (https://www.ebi.ac.uk/Tools/psa/ emboss_stretcher/). Phylogenetic analyses were conducted using the Molecular Evolutionary Genetics Analysis package MEGA7 (31). We first clipped all irak3 sequence files at position 994 in exon 9 (relative to the first coding nucleotide of variant a), which represents the common fragment of the truncated and full-length Irak3-encoding cDNAs and then subjected these to multiple alignment using the ClustalW algorithm. The dendrogram was reconstructed using the neighbor-joining method based on the maximum-likelihood method. The tree with the highest log likelihood (−1338.9) was chosen and optimized manually. The online tool Microsatellite Repeats Finder (http://insilico.ehu. es/mini_tools/microsatellites/) was used to identify repetitive elements in the genomic DNA. Functional motifs and domains of conceptually translated (Expert Protein Analysis System proteomics server, Swiss Institute of Bioinformatics; http://www. expasy.ch/) amino acid (aa) sequences were identified using the NCBI (National Center for Biotechnology Information; http:// blast.ncbi.nlm.nih.gov/) algorithm in accordance with results from the search with SMART (Simple Modular Architecture Research Tool; http://smart.embl-heidelberg.de/). The threedimensional (3D) structures of particular Irak3 variants were predicted by the Protein Homology/analogY Recognition Engine V 2.0 (Phyre 2 ) (32) and visualized using OpenRasMol software (33).

Establishment of Irak3-Expression Plasmids
We constructed the plasmids that express the full-length irak3 factor (m) and its variants (a, g, h, k, l, and m ′′ ) from rainbow trout to analyze their ability to interfere with TLR-dependent NF-κB activation in in-vitro models. In the first step, we designed variant-specific oligonucleotide primers featuring restriction sites for HindIII and EcoRI at their 5 ′ and 3 ′ ends, respectively. Subsequently, a standard-PCR reaction using cDNA from rainbow trout as template was conducted using the common sense primer 5 ′ -ACCCAAGCTTGATATGGACTCGTCTATG TACCTGTA-3 ′ and the antisense primer 5 ′ -TCAGAATTCCTCC TGGGTGGTCATGGAGA-3 ′ specific for variants m and m ′′ or the antisense primer 5 ′ -TCAGAATTCAAAGCATCGTATGTAT TCCTCCG-3 ′ specific for variants a and l (restriction sites of all primers are underlined). The resulting amplicons were inserted into pGEM-T Easy Vector (Promega). This subclone was digested with HindIII and EcoRI at 37 • C overnight. The retrieved fragments were finally inserted "in frame" into our expression vector 280, a modified (34) CMV-driven pKS(+) Bluescript plasmid (Promega).
To analyze the intracellular spatial distribution, we tagged the full-length Irak3 factor m and its variants a and m ′′ with fluorescent labeling. Then, we digested the irak3-a, -m and -m ′′ subclones with HindIII and EcoRI at 37 • C overnight and inserted the retrieved fragments into the respective cloning sites of plasmid pAM505, which expresses the green fluorescent protein (GFP; AF140578) under the control of the CMV promoter.
All plasmids were prepared with endotoxin-free reagents (EndoFree Plasmid Maxi Kit; Qiagen) and sequenced to confirm their appropriate construction.

Cell Culture, Transfection, and Luciferase Assay
The human embryonic kidney-cell line HEK-293 (ATTC) and the embryonic cell line CHSE-214 from Chinook salmon Oncorhynchus tshawytscha (Sigma-Aldrich) were cultured in EMEM medium (Sigma-Aldrich), supplemented with 10% fetal calf serum (PAN), 1% non-essential aa (Biochrom), and 2 mM L-glutamine (Biochrom). The two cell lines were cultured in humidified atmosphere containing 5% CO 2 , at either 37 • C (HEK-293) or 20 • C (CHSE-214). Prior to and during the stimulation experiments, both cell lines were proven to be free of mycoplasma contamination using the Lookout Mycoplasma PCR detection kit (Sigma-Aldrich).
Endotoxin-free preparations of the expression constructs were transfected into HEK-293 cells using Lipofectamin 2000 (ThermoFisherScientific). For these co-transfection assays, we used 50 ng of the ELAM-driven NF-κB-reporter vector and increasing concentrations (50, 500, 2,000 ng) of the respective irak3-expression constructs. Since the HEK-293 cells do not express most TLRs (35), 200 ng of the plasmid that expresses bovine TLR2 (34) were co-transfected. The total DNA concentration was adjusted to 4,100 ng/assay by adding the empty cloning vector. After transfection, the cells were distributed into six wells of a 24-well plate. Three wells were left as unstimulated controls. The other three wells were stimulated for 24 h with the TLR2 ligands Pam 2 CSK 4 (10 ng/ml, Invivogen) and FSL-1 (100 ng/ml, Invivogen) or with heat-killed bacteria Escherichia coli, strain 1303 (30 µg/ml). The luciferase activity of HEK-293-cell lysates was measured with the Dual-Luciferase Reporter Assay System (Promega) with the Lumat LB 9501 luminometer (Berthold). Values were normalized against the protein concentration as determined using the NanoDrop 2000 photometer (ThermoFisherScientific). Every transfection was assayed in triplicate; each transfection experiment was performed at least twice.
CHSE-214 cells were transfected with 2,000 ng of the plasmid that expresses trout irak3 variants a, l, m, and m ′′ using X-tremeGENE HP DNA Transfection Reagent (Roche/Sigma-Aldrich). Since CHSE-214 cells endogenously express Tlr3 and Tlr5 (36), we stimulated them for 3 or 6 h with a mixture of the TLR3 ligand poly(I:C) of low molecular weight (10 µg/ml, Invivogen) and the TLR5 ligand flagellin from Salmonella typhimurium (100 ng/ml, Invivogen). These incubation times and doses were determined in pre-tests. After the stimulation period, cells were harvested in ice-cold PBS for RNA isolation and the subsequent profiling of immune-gene expression.

Confocal Microscopy and Determination of Cell Viability
HEK-293 cells were transfected with 1 µg of plasmid expressing the GFP-tagged factors. Optionally, transfected cells were stimulated for 30 min with 10 ng/ml Pam 2 CSK 4 prior to observation with fluorescence microscopy using the confocal microscope LSM 780 (Carl Zeiss Microscopy), equipped with a 63 × oil-immersion DIC objective and an incubation chamber to maintain ∼37 • C and 5% CO 2 during the microscopic investigations. Nuclei were stained by adding Hoechst 33342 dye (1 µg/ml, Sigma-Aldrich) to the medium for 30 min, immediately before the start of the observation. Quantification of the live/dead ratio of transfected cells was performed with the Nucleocounter3000 (Chemometec). Dead cells were stained using propidium iodide (PI, 1 µg/ml). The fluorescence microscopic observation of the HEK-293 cells overexpressing the full-length (m), truncated (a), or splice (m ′′ ) irak3 variants were repeated at least twice with different plasmid preparations.

Quantitative Real-Time Expression Analyses of in vivo and in vitro Samples
The correlation between IRAK3-transcript and protein levels has been proven (16). For expression profiling, we designed a panel of transcripts that encode inflammatory and inhibitory factors from trout, including the primers that discriminate between different irak3 transcripts ( Table 1). To determine the abundance of the entirety of irak3 transcripts, we designed oligonucleotide primers specific to exon 1 of the irak3 gene (covering the region between positions 2 and 95 on the coding sequence of all unspliced variants) sharing perfect sequence identity across the multiple variants. A second pair was placed on exons 10 and 11 (positions 1,136 to 1,298 on the coding sequence of factors m and n), which are absent in all truncated variants, to quantify the abundance of the full-length transcripts.
To profile the infection-modulated irak3 expression, we followed the principles of the 3Rs (37) for ethical use of animals and utilized samples from a previous experiment. Here, rainbow trout were infected with 1 × 10 7 Aeromonas salmonicida ssp. salmonicida (approved animal experiment, LALLF M-V/TSD/7221.3-2.5-008/10.). The peritoneal injection has been previously described (22). Tissue samples from peripheral blood leucocytes (PBLs), head kidney and liver were used from four individuals at 0, 6, 12, and 24 h post-infection. The RNA was isolated from these samples in separate tubes using TRIzol (Invitrogen) and subsequently purified with the ISOLATE II RNA Micro Kit (Bioline). After cDNA synthesis, the concentrations of the individual cDNA aliquots were adjusted at 10 ng/5 µl. Samples were than measured using the Biomark/Fluidigm. To this end, cDNAs were individually preamplified in 10 cycles using the PreAmp Master Mix (Fluidigm) and subsequently treated with exonuclease I (New England BioLabs) to degrade single-stranded oligonucleotide primers. Finally, multiplex RT-qPCR was conducted using the 48.48 Fluidigm Gene Expression biochips, which were first primed in the MX IFC Controller (Fluidigm). Then they were loaded with the pre-amplified cDNA samples and eventually analyzed using the Biomark HD system with the manufacturer's thermal protocol "GE Fast 48 × 48 PCR+Melt v2.pcl." EvaGreen fluorescence dyes (Bio-Rad) served as DNA-binding reporter molecules to allow for the quantification of the amplified target fragments.
All melting curves were analyzed individually to validate the absence of unspecific products. Amplicons were visualized on agarose gels in order to assess product size and quality. Crossing points (CP) between 5 and 35 cycles were considered detectable. The obtained real-time data were all normalized with a factor based on the geometric mean values from three potential reference genes. The qBase+ software (Biogazelle, Ghent University, Belgium) validated the suitability of the quantified reference genes, eukaryotic translation elongation factor eef1a1 (38), ribosomal protein rps5, and 18S ribosomal RNA (39), for data normalization. Individual copy numbers were calculated based on external gene-specific standard curves (10 7 -10 3 copies per 5 µl; R 2 > 0.999).

RESULTS
From our 332 million RNA-sequencing reads from rainbow trout, we derived an ORF that encodes a 1920-nt comprising irak3-cDNA sequence. This trout sequence resembles reasonably well the human IRAK3-encoding sequence (∼52% nucleotidesequence identity) and the IRAK3 ortholog in orange-spotted grouper Epinephelus coioides (∼63%), which represents the first published example of an IRAK3 ortholog from teleost fish. The aa sequence derived from the conceptually translated cDNA shares >96% similarity with those from segments of the factor encoded by salmon and on average approximately 60% similarity with the ortholog factors from a wide range of fish species. The similarity of the Irak3-aa sequences is only about 40% in comparison to mammalian species.

Two irak3 Genes in Rainbow Trout Share Structural Similarities
Inquiring into the gene structure, we used our RNA-seq-derived irak3-prototype sequence to search for IRAK3 orthologs in the genome assemblies of rainbow trout (40)(41)(42) published by other groups and consortia. We identified two relevant entries for genomic trout sequences, (i) irak3 (Gene ID: 101268966) on an unplaced scaffold (MSJN01015042, MSJN01016613, MSJN01023158) and (ii) LOC110499741/irak3-like (Gene ID: 110499741) on chromosome 21 (MSJN01004319). Comparing our cDNA sequence with these two different genomic deposits revealed that the latter, not contiguous sequence file consecutively comprises 13 exons that encode Irak3. Their lengths vary between 63 and 472 nt (excluding UTRs; Figure 1, upper scheme). The alignment of these exon sequences with the formerly mentioned three other genomic irak3-sequence files showed that they altogether harbor exons 1, 2, 5-7, and 9-13. The thus proven segmentation of the irak3-encoding gene from trout into 13 exons complies with a recent report on a homologous irak3-gene structure in grouper (28). We compared the genomic sequences of both irak3 genes from rainbow trout in order to understand if they stem from two different genes. The exon sequences are highly similar (>98% sequence identity). Due to large sequence gaps in both of the irak3 genes, we conducted local alignments of comparable intronic fragments, which yielded identities between 90 and 99%. While this high degree of sequence similarity indicates their Frontiers in Immunology | www.frontiersin.org close relatedness, clear differences exist, and validate that they do belong to two different genes. For instance, the positions of the partially degenerated micro-or minisatellites within the intronic sequences appear to be conserved in both genes, albeit slightly altered ( Table 2). Remarkably, we found repetitions of coding sequences in the introns of both irak3 genes. For instance, the 5 ′ -most 15 to 17 nt of exon 5 are repeated four times in intron 4, directly upstream of exon 5. Intron 7 of the irak3 gene from trout features three tandem repetitions of exon 7 in a row, separated by a longer (>700 nt) and a shorter (314 nt) non-coding intronic sequence, each flanked by the canonical 5 ′ -GT donor and 3 ′ -AG acceptor dinucleotide splice sites. Similarly, a short fragment of exon 9 resides in intron 10, again flanked by core splice sites. Nonetheless, both deposits of genomic trout sequences for irak3 and LOC110499741/irak3-like gene still contain gaps of unknown lengths in introns 1, 2, 3, 4, 11, and 12.

More Than 20 irak3-Transcript Variants of Different Lengths Were Cloned From Rainbow Trout
We detected as many as 45 single nucleotide changes (SNCs) between nt positions 106 and 1,913 (numbering the first nucleotide of the translational start codon ATG as position 1) in our RNA-seq-derived irak3-consensus sequence (Figures 2A,B). To validate the variable nt positions in the irak3 sequences of trout, we derived irak3-specific oligonucleotides (Figure 2A) from our prototype-irak3 sequence for subsequent PCRs. Amplicons were cloned and more than 100 clones were sequenced. The cloned sequences validated 23 of the 45 SNCs ( Figure 2B). Surprisingly, we found that the lengths of the encoded ORFs eventually differed significantly, ranging from 276 to 1,920 nt (Figure 1). We named those varying irak3 sequences in alphabetical order (a-m) according to their frequency in our clone collection and submitted them to the "European Nucleotide Archive" (ENA, project ID: PRJEB29555). The most abundant irak3-transcript variants encoded truncated versions of the prototypic 1,920-nt irak3 ORF (Figure 1, table). They contained ORFs comprising 1,044 or 1,005 nt, subsequently termed as variants a (NCBI accession code: , and k (LR031303). The variants m (LR031305) and n (LR031307) represent full-length irak3 transcripts complying with two trout irak3 sequences, which have been submitted to the ENA database (GBTD01132160 and GFIN01027674). Our two full-length sequences and those two derived from the database share an overall identity of >98%, differing in 20 positions due to those SNCs and a triplet insertion at the 3 ′ end of file GFIN01027674. In addition, the above-mentioned genome-derived irak3-like sequence LOC110499741 from file XM_021576800 features an ORF comprising 2,208 nt and is hence 288 nt longer than our cDNA-derived full-length irak3 sequences. This additional 288-nt sequence stems from a microsatellite [(TACCGGCTCAAACCCCTAGTCTATAACATCACAAGG) 8 ; c.f. Table 2] that had been incorporated into the cDNA sequence, since it does not interrupt the ORF in conceptual translation. A bootstrap test of our phylogenetic analysis ( Figure 2C) only confirmed the strong similarity of variants a and c (98%) as well as of j and l (86%). The cDNA sequence derived from the irak3-like gene LOC110499741 was clearly separated from the other 14 variants (a-n), as well as from the sequences GBTD01132160 (termed here "1") and GFIN01027674 ("2") because of the insertion of the 288-nt microsatellite.

Some Truncated Transcripts Are Caused by Alternative Splicing
We found 10 shorter irak3 transcript variants (Figure 1)

Some Truncated Transcripts Contain a Nonameric Sequence Absent in the Full-Length cDNA Sequence
The structural cause for the truncation of the irak3 variants comprising the ORFs of 1,044 and 1,005 nt in length could not be explained based solely on the available DNA-sequence information. These truncated cDNA sequences feature a 9-ntsequence motif at the 3 ′ -end of their ORFs that is not found in the copies of the full-length cDNAs, or in the available genomic DNA sequences from trout. To elucidate the origin of the truncated transcript variants, we performed 3 ′ -RACE experiments with the sense primer RACE-f1 annealing to the 5 ′ -region of exon 9 and the nested primer RACE-f2 annealing further downstream on the same exon (Figure 2A). The resulting 3 ′ -RACE sequence (LR213462, Figure 3A) did not match either the full-length irak3-cDNA sequences of trout or the substantial amount of intronic sequence information from the irak3 and the irak3-like genes. It shares instead a high identity with an irak3-like sequence from Atlantic salmon (LOC106609858, XM_014208880). The trout 3 ′ -RACE sequence matches from position 16 (located in exon 9) to the 3 ′ end of the salmon irak3 UTR (positions 2,192-2,628) with >86% similarity (Figure 3B).
Remarkably, comparing both sequences revealed that a point mutation introduced a translational stop codon in the truncated irak3 variants from trout ( Figure 3B, right side). The TGG triplet in the salmon sequence at positions 2,189-2,191 is converted into a TAG stop codon at positions 1,042-1,044 in the truncated irak3 variants (a, b, b ′ , c, c ′ , c ′′ , f, g, h, i, j, k, and l) from trout. Another noteworthy feature is a 275-nt long sequence residing in the 3 ′ UTR of both salmonid sequences ( Figure 3C). Conceivably, this sequence stems from a retroposon element, since it is also present in many other genes from trout and salmon including such evolutionary unrelated genes as specc1, orai1, eef1e1, anapc1, and morn3. This mobile genetic element must have been inserted into the ancestral salmonid irak3 gene, before the separation of the trout and salmon species.
Irak3 Transcripts From Rainbow Trout Contain 45 Single Nucleotide Exchanges, but Only Three Different Alleles Were Found to Be Expressed in Individuals The 45 identified SNCs in the irak3-cDNA sequences from rainbow trout are not randomly distributed across the coding sequence (Figures 2A,B). Rather, no exchanges were found in exons 1 and 6, whose coding sequences are strictly conserved. Moreover, exon 8 features only synonymous exchanges. The occurrence of the SNCs in the other exons does not follow an unambiguous pattern that would indicate any closer relatedness between the different irak3 variants, as proven by our phylogenetic analysis ( Figure 2C). We tried to identify the minimal number of alleles from which the plethora of variants might possibly be encoded. To this end, we separately examined the irak3 reads obtained from different tissues from each individual that was included in the RNA-seq analysis. We examined potential linkage groups formed by neighboring SNCs on the short 20-70-nt long reads and identified no more than three different SNC-linkage groups per individual. These individual linkage groups were expressed in all tissues analyzed. Hence, no tissue-specific linkage group was found, but SNC combinations did vary among the different individuals. Based on the assumption that one of these SNC combinations comes from the paternal allele and the other from the maternal allele, we conclude that the different alleles originated from (at least) two irak3 genes.
Taken together with the structural analysis of the two highly segmented irak3-encoding genes, which revealed the presence of duplicated/multiple exon copies and/or exon fragments, it seems likely that the various irak3 variants represent differentially spliced transcripts derived from only two irak3 genes.

Several Irak3 Variants From Trout Lack Function-Relevant Domains and Amino-Acid Residues
The full-length variants of irak3 in rainbow trout encode 639aa comprising proteins (m, VDB38424; n, VDB38426), which contain the following domains in accordance with the structural characteristics of IRAK-family members  444-639). Twentynine of the 45 SNCs mentioned above (almost two-thirds of the SNCs) represent non-synonymous mutations that contribute to a striking diversity of Irak3 variants in rainbow trout. Remarkably, the relative abundance of non-synonymous exchanges is very low in the pseudokinase domain at only 0.6%, while it is significantly higher at in the ProST region (2.9%), followed by the C-terminal domain (2.0%) and the death domain (1.4%). Even more diversity of Irak3 factors is caused through the length reduction of several variants. The Irak3 variants a to k are shorter by at least 292 aa residues compared with the full-length factors m and n. The shortenings involve the deletion of about one third of the pseudokinase domain and the C-terminal domain (Figure 2A). In contrast, variants m ′ and m ′′ contain the complete C-terminal domain and a part of the death domain, but lack the entire pseudokinase domain.
The death domain is encoded by the first two coding exons of the irak3 and the irak3-like gene. Strikingly, the exon-1-encoded sequence is completely conserved in all irak3 variants, except the splice variants m ′ and m ′′ (Figure 1). The extremely short ORF of variant l (276 nt; Figure 1) only encodes an isolated death domain. The death domain of most other Irak3 variants is followed by the ProST region. This region contains 11 proline (16%), 15 serine (22%), and 4 threonine residues (6%) in a stretch of 69 aa residues. The relatively high abundance of these three aa residues corresponds approximately to those determined for all vertebrate IRAK proteins (43). Nevertheless, two of the vital ProST residues (Ser-119, Ser-127) are affected by base exchanges in some irak3 variants (a, c, e, h, i, j, k, n; Figure 1A).
The pseudokinase domain covers approximately the central half of the full-length Irak3 aa sequence from trout. This domain harbors the IRAK3-characteristic residue Asn-292, which is present in all full-length and truncated Irak3 variants, but absent in most splice variants (m ′ , m ′′ , b ′ , c ′′ , f, f ′ , j, and l). The Cterminal domain is encoded by exons 12 and 13 and is therefore only present in the full-length factors m and n and their splice variants m ′ and m ′′ .

Infection With A. salmonicida Modulates the Expression of irak3 and Other Immunoregulatory Genes
In healthy rainbow trout, the proportion of full-length transcripts within the entirety of all irak3-transcript variants varied in different tissues and cells: 29.1% (liver), 33.7% (head kidney), and 40.2% (PBLs) (Figure 4A). Hence, the transcripts that encode the full-length factor contribute much less than half of all irak3 transcripts.
We used samples from a previous infection trial with live A. salmonicida pathogens (44) to determine the infection-related regulation of the expression of inflammation inhibitors. The copy numbers of all irak3 variants were significantly elevated in the liver at 6 hpi and in the head kidney at 24 hpi ( Figure 4A). The infection-related modulation of the irak3-transcript levels showed a significant ∼2.7-fold increase in the liver and PBLs as early as 6 hpi and an approximate doubling of the transcript number was recorded in the head kidney, but only as late as 24 hpi. The ratio of full-length vs. truncated irak3 variants was only slightly modulated during infection (<10% variation in each tissue/cell type).
The modulations of the expression levels of other inhibitorencoding genes were also tissue-specific ( Figure 4B). We detected in the head kidney very high transcript concentrations of numerous inhibitory factors (il1rl1, tollip, inpp5d, tnfaip3, cfh, cd59, a2m, il4/13, tgfb-2, gpr84) mostly at 24 hpi. In contrast, the expression of only a few regulators (nfkbia, socs1, and gpr84) increased in PBLs, especially at 12 hpi. In the liver, many genes were down-regulated, including complement-system regulators, some of which were not detectable in other cell/tissue types.
Certain genes were clearly upregulated across all cells/tissues and most time points. Tnfaip3/a20 and the three nfkbia variants all significantly increased in the liver and head kidney at almost all times after infection, and considerable copy numbers were found in the PBLs as well. Socs1 and gpr84 were also strongly upregulated in all three cell/tissue types, but only at 12 hpi. Therefore, these data reveal that the A.-salmonicida infection triggered a broad variety of factors and mechanisms that counteract inflammation, validating that this infection trial was successful.

Overexpression of irak3-a in Mammalian Cells Reduces Cell Viability
Previous investigations revealed that IRAK3 is equally distributed in the cytoplasm and nucleus of human cells and shifts toward the cytoplasm upon immune stimulation (45,46). We wanted to see, whether this also applies to the ortholog from trout and whether the Irak3 variants might alter this pathogen-stimulated spatial redistribution of the factor. To this end, we transfected the full-length irak3 or selected variants tagged with GFP into HEK-293 cells together with a plasmid expressing bovine TLR2 (Figure 5A). Twenty-four hours after transfection, we observed that the full-length factor Irak3-m and its severely crippled splice variant m ′′ were almost uniformly distributed in the cytoplasm of the cells. The truncated Irak3-a, featuring the N-terminal half of the full-length factor and notably lacking the entire C-terminal domain, was visible both in the cytoplasm and in the nucleus (Figure 5B). This was seen in multiple transfections performed with different plasmid preparations. A stimulation of the cells with 10 ng/ml Pam 2 CSK 4 did not change the localization (data not shown).
Strikingly, we found a higher proportion of rounded cells in those cultures that were transfected with the truncated Irak3a variant (c.f. Figure 5A). Since the rounding up of otherwise fibroblast-like HEK-293 cells indicates cell death, we quantified the dead cells in cultures that were transfected with one of the three variants m, a, and m ′′ . The fraction of dead cells doubled to >12% in cells transfected with irak3-a compared to non-transfected cells or cells transfected with irak3-m andm ′′ (Figure 5C). The in vivo most abundantly expressed irak3transcript variant appears to have a somewhat cytotoxic effect, at least if massively overexpressed.  1.0). The quantified transcripts are listed as gene symbols on the left and categorized according to their affiliation to immune processes; orthologs whose expression is IRAK3-dependent in mammals according to Zhou et al. (5) are marked in red. Significant copy-number changes (p < 0.05) are underlined. All expression values were normalized against the geometric means of the reference genes eef1a1 and rps5 as evaluated with the qBase+ software.

Full-Length Irak3-m and Truncated Variant a Increase Basal NF-κB Levels, but Do Not Reduce Stimulation-Induced NF-κB Levels in Mammalian Cells
In the next step, we investigated the influence of the Irak3 variants m, a, m ′′ , and l (without GFP tag) on TLR2-induced NF-κB activation in mammalian HEK-293 cells. We transfected increasing amounts of constructs that express these factors into those cells. The co-transfection of an NF-κB-responsive luciferase-reporter gene allowed monitoring the NF-κB activity.
Considering the effect of overexpressed Irak3 factors on the basal NF-κB activity without stimulating TLR2 signaling, we found that overexpressing the full-length factor m and the truncated factor a substantially increased the basal level of NF-κB activity in a dose-dependent fashion (Figure 6A), eventually to 5.4 ± 0.4 or 2.5 ± 0.2-fold (factors m and a, respectively) if the highest dose (2 µg of the plasmids) was used. In contrast, overexpression of the more severely shortened splice variants l and m ′′ induced only minor levels of active NF-κB. Stimulating TLR2 signaling with the PAMPs Pam 2 CSK 4 and FSL-1 in non-irak3-transfected HEK-293 cells increased the level of active NF-κB by 20.2 ± 0.4-fold compared with unstimulated non-irak3-transfected cells. Overexpressing the full-length factor m and the truncated variant a lowered the TLR2-signalinginduced degree of NF-κB activation down to 4.0 ± 0.2 or 2.8 ± 0.3-fold, respectively, compared to identically transfected, unstimulated controls. Splice variant l was basically ineffective (Figure 6B). Taking into account the stimulatory effect of the factors m and a on the basal NF-κB activity in the cells, it becomes clear that none of the transfected Irak3 variants substantially altered the ligand-dependent stimulation of NF-κB ( Figure 6C). Altogether, the effect of the Irak3 variants on dampening the stimulated NF-κB activity was rather modest, not dose-dependent, and most likely independent from the properties of the transfected constructs.
We obtained almost congruent results when using heat-killed E. coli bacteria instead of PAMPs to stimulate HEK cells (data not shown). In addition, we also examined the effect of isoforms of the truncated irak3 variants (g, h, k), but we did not observe any differences in modulating the levels of active NF-κB compared with Irak3 variant a (data not shown).

Splice Variant irak3-m ′′ , but Neither the Full-Length nor the Truncated Variants, Reduces the Expression of Pro-Inflammatory Interleukins in Salmonid Cells
We examined the potentially different physiological significance of the prototypic Irak3 and its shorter variants. To this end, we overexpressed the full-length irak3, the truncated variant a and the splice variants m ′′ and l in the salmonid cell line CHSE-214 and quantified their influence on the transcript levels of selected immune genes (Figure 7). Transfected cells and non-transfected controls were stimulated with a mixture of the TLR3 ligand poly (I:C) and the TLR5 ligand flagellin for 6 h. qPCR analyses of irak3-transcript levels revealed that overexpression of the fulllength irak3 factor increased the basal irak3-transcript level in unstimulated cells by approximately 2,000-fold above the level of ∼25 copies per ng of total RNA measured in non-transfected CHSE-214 cells.
Stimulating the non-transfected control cells for 6 h significantly increased the expression of the pro-inflammatory cytokines il1b and cxcl8 by ∼11-fold as well as the expression of the inflammation dampener nfkbia-1-3 (3.9-to 5.5-fold), tnfaip3 (2.6-fold), and grp84 (2.4-fold) (Figure 7). Surprisingly, overexpression of neither the full-length irak3 factor m, nor the truncated factors a and splice variant l caused any noteworthy changes in the expression of the selected indicator genes for monitoring inflammation. In stark contrast, splice variant m ′′ clearly and significantly decreased the transcript levels of il1b and cxcl8 by a >5-fold, accompanied by a slight increase in inpp5d copies (∼2-fold) over the levels found in stimulated but non-transfected cells. The 3-h stimulation yielded similar effects, albeit to a lesser extent (Supplementary Figure 1).

DISCUSSION
The interleukin-1-receptor-associated kinase 3 (IRAK3) was identified as a negative feedback regulator of inflammatory events (47,48) and hence as a relevant factor for properly calibrating the immune response against infections. Its counterpart in the economically important salmonids had not yet been examined in more detail (1).

At Least Two and Possibly More Genes Encode Irak3 in Rainbow Trout
At the outset of our study, we derived an irak3-consensus sequence from our sequence data obtained previously in an RNAseq analysis of rainbow-trout-tissue samples. An alignment of the resulting consensus sequence with database entries of genomic DNA sequences suggested the existence of two different, yet closely related irak3 genes that are encoded in rainbow-trout genome. Thirteen exons encode the full-length Irak3 factor in this salmonid species.
The detection of severely truncated cDNA sequences featuring a nonameric sequence, which is absent in both the full-length irak3-cDNA sequences and the two different genomic irak3sequences, might suggest that these truncated variants are encoded by a third gene copy. However, we identified a highly similar counterpart to this nonamer in the 3 ′ -UTR of an incomplete irak3 transcript from Atlantic salmon. It might be plausible that the truncated transcripts from trout result from splicing a section of the 3 ′ UTR to a degenerated copy of exon 9. This hypothetic copy may reside in an intronic region of the irak3 gene similar to the fragments of exon 5, exon 7, and exon 9 within particular introns of the irak3 genes. This could explain why the same nonameric sequence motif is found in two different transcripts featuring truncated exon-9 sequences of either 151-nt or 110-nt length. That assumption is supported by the fact that the respective nonameric sequence contains the motif TGCTT, which has previously been identified in an analysis of >20,000 human mRNAs as one of the 10 most frequent "accessory splicing signals" (49,50).
In mammals, at least half of all full-length transcripts appear together with a splice variant (51). Splice variants also exist from IRAK3 transcripts. For example, splice variant IRAK3 (2) is known in humans and IRAK3 (9−11) in mice (13). We also found spliced transcript variants that are clearly derived from the transcripts encoding the full-length or the truncated factor. In some cases, exons are spliced-out using the canonical splice-donor and acceptor dinucleotides flanking the enclosed intron. In other cases, the RNA splicing machinery apparently uses alternative more 5 ′ -located splicedonor dinucleotides. This produces crippled exon sequences, which are incorporated into the resulting transcripts (b ′ , c ′′ , f, m ′ , m ′′ ). Intriguingly, splicing at alternative, non-canonical splice-acceptor or -donor sites apparently produces an even larger proportion of differently shortened transcripts. We recently described a similar observation for the acute-phase gene tcbp coding for the trout C-polysaccharide binding protein (39). The usage of non-canonical splice sites may thus be a quite common phenomenon in teleost fish and salmonids in particular. Apparently, this species complex is still struggling with the consequences of the recent whole-genome duplication (41,52) to sort out the most efficient gene copies from an excessive number.

High Selection Pressure Rests on the Pseudokinase Domain of Irak3 From Rainbow Trout
Within the 1,920 nt ORF of the irak3 consensus sequence, we detected 45 single-nucleotide changes. Hence, about every 40th base was found to be variable, potentially yielding more than 2 45 = 3 × 10 13 combinations. A recent report indicates that on average every 64th nucleotide position may be polymorphic in the pseudotetraploid genome of rainbow trout (42). Our study thus revealed an even more frequent occurrence of nucleotide exchanges in the irak3 sequences of rainbow trout. The frequency of SNCs appears to be gene-specific, since the trout irak4 paralog of irak3, for instance, revealed only six SNCs in a 1,425 nt sequence [ Supplementary Figure 2; (22)].
Mapping the 29 non-synonymous SNCs on the full-length irak3-consensus sequence from rainbow trout illustrates that the pseudokinase domain is affected by the fewest aa-residue exchanges compared to the other three Irak3 domains. This pseudokinase domain, which is obviously under a higher FIGURE 7 | Expression profiling of stimulated Irak3-overexpressing CHSE-214 cells. The HeatMap illustrates the averaged fold-change values (according to the legend on the right) of the mRNA concentrations measured in CHSE cells 6 h after stimulation with 10 µg/ml poly (I:C) and 100 ng/ml flagellin, relative to unstimulated control cells (set as 1.0). The quantified transcripts are listed as gene symbols on the left; orthologs whose expression is IRAK3-dependent in mammals according to Zhou et al. (5) are highlighted in red. Significant copy-number changes (p < 0.05) of stimulated vs. non-stimulated untransfected cells are marked with asterisks; significant copy-number changes of stimulated non-transfected vs. transfected cells are underlined. All expression values were normalized against the geometric means of the reference genes eef1a1 and rps5. selection pressure, discriminates the inflammation-supporting IRAK members−1 and−4 from the negatively regulating IRAK3 (3,12). To our knowledge, there is no evidence yet that the pseudokinase domain of IRAK3 orthologs lacks phosphorylation activity. In general, the pseudokinase domain has so far received only little scientific attention. Therefore, we do not know much about the biological significance of exchanged aa residues within this domain.
The domain-mapping-of-disease-mutations database currently lists 15 non-synonymous SNPs in the <2 kb ORF of the human IRAK3 gene. Several studies have proven significant associations between a particular SNP in one of the resulting IRAK3 variants and unbalanced immune reactions, including (i) asthma (53,54), (ii) sepsis (55) and sepsis-induced acute lung injury (56), as well as (iii) a positive response to treatment in rheumatoid arthritis (57), and (iv) protection against Bordetella pertussis after vaccination (58). Additionally, certain SNPs of the human IRAK3 gene are associated with the risk of cancer (59)(60)(61). Clearly, our study was not designed to analyze the physiological effects of SNPs/SNCs. It remains to be seen if any of these many Irak3 variants detected in our study are associated with a disease trait in rainbow trout.

Irak3 Variants Differentially Activate Basal NF-κB Levels, but Not Their Final Level After Stimulation
We selected four major structural variants of Irak3 from rainbow trout to study and compare their functions in vitro in the human HEK-293-cell system. This cell line is a well-established "workhorse" for the heterologous expression of immune factors from fish (23,36,39,(62)(63)(64), especially for the reconstitution of the TLR-signaling cascade. The chosen Irak3 factors included the prototypic full-length Irak3-m factor (639 aa), the truncated Irak3-a variant lacking a part of the pseudokinase and the complete C-terminal domain (347 aa), and the splice variant Irak3-m ′′ lacking the ProST and the pseudokinase domain (199 aa). We were also curious to see, whether IRAK3-l, which consists only of a death domain (91 aa), would be functional. The HEK-293 cells were transfected with increasing amounts of plasmids that express either one out of the four selected Irak3 variants together with a plasmid expressing the bovine TLR2 and an NF-κB-responsive reporter vector. This experimental setup accounted for the well-explored fact that TLR signaling activates the NF-κB-transcription-factor complex (10) and should allow IRAK3 to counteract NF-κB-induced proinflammatory processes.
The overexpression of the full-length factor m and the truncated factor a from rainbow trout significantly increased basal NF-κB activity, while the shorter splice variants m ′′ and l were basically ineffective. Modulation of basal NF-κB activity was not necessarily expected, since overexpressing other regulatory factors of the TLR-signaling cascade, such as Il1rl1/St2 (36) or Irak4 from rainbow trout (22), left the level of basal NF-κB activity unchanged. The different modes of action of the full-length Irak3 factor m and its variants a, m ′′ , and l also became evident after stimulating TLR2transfected HEK-293 cells with relevant PAMPs. Noteworthy, the summation of basal and induced NF-κB-induction folds yielded, in effect, similar values for the final NF-κB-activity levels after stimulation. It was rather unexpected that the overexpression of full-length irak3-m did not alter the final level of NF-κB activity after stimulation with TLR2 ligands, given that Irak3 has often been proven to be a dampener of TLR-dependent inflammation (12,47,48,65). However, a survey of the pertinent literature showed that NF-κB levels after pathogen-dependent stimulation has not often been used as a read-out system. The first analysis on human IRAK3 (alias IRAK-M) 20 years ago demonstrated that overexpressed IRAK3 increased basal NF-κBactivity levels without stimulation (3). This is generally in line with subsequent studies (4,28) and also with our findings. Many other reports on IRAK3 function did not examine its direct impact upon the TLR-mediated activation of NF-κB activity, but instead used different read-out systems to validate the inflammation-dampening effects of IRAK3, such as the level of cytokine-gene expression (13,66,67). We show here that Irak3 from rainbow trout does not directly target NF-κB to inhibit inflammatory processes.
Notwithstanding, our overexpression experiment in HEK-293 cells proved that the truncated Irak3 variant a increases the basal level of activated NF-κB almost as efficient as the fulllength variant, although it lacks the entire C-terminal domain. This domain is apparently not involved in the mechanism that regulates the observed effect. Parallel microscopic observations detected a, perhaps unrelated to NF-κB activation, difference between full-length and the truncated Irak3 variants. We found that the truncated variant Irak3-a increased cell death, in contrast to the full-length variant. Although the death domain is a wellexplored structural requirement for the execution of apoptotic processes (68), there is no evidence, to our knowledge, that IRAK3 directly affects apoptosis in cells. For this reason, we can only speculate whether the physiological expression of the C-terminally truncated Irak3-a variant has a toxic effect on cells in vivo.

The Irak3 Splice Variant m ′′ May Act as a Scavenger Molecule of TLR Signaling
Our studies conducted in the salmonid CHSE-214 model cells focused on the Irak3-dependent modulation of immune-gene expression after stimulating the TLR axis. We found that the stimulation of immune-gene expression is largely uninfluenced by the overexpression of either the full-length factor irak3-m or its variants a or l. We note in this regard that previous studies on the function of the mammalian IRAK3 have been performed primarily on macrophages and lung epithelial cells (13,71). An embryonic cell such as CHSE-214 may not provide all the factors required for the correct functioning of Irak3. Moreover, we found a low endogenous expression level of irak3 in CSHE-214 cells, but it is remarkable that the additional, eventually 2,000-fold increased concentration of the transcripts encoding the Irak3 factors m, a, and l did not influence at all the induced immunegene expression. Only overexpression of the splice variant m ′′ significantly lowered the expression of il1b and cxcl8. The stimulated induction of these two proinflammatory cytokines is known to heavily depend on the activation of the TLR-signaling cascade. The assumption that Irak3 splice variant m ′′ hinders TLR signaling is supported by the fact that its overexpression also led to decreased nfkbia-3 expression. It is known from mammals that activated TLR signaling concomitantly induces the expression of NF-κB-inhibitory factors such as NFKBIA (alias IκBα). This negative feedback loop conceivably prevents excessive NF-κB activation (69,70). The Irak3 splice variant m ′′ consists almost exclusively of the C-terminal domain. It lacks parts of the death domain and the entire ProST region which is required in the mammalian ortholog for interacting with IRAK4 and activating NF-κB (3)(4)(5). This particular architecture of m ′′ might prevent its integration (via TIR-TIR interactions) into the TLR-associated myddosome and could explain why only high concentrations of m ′′ reduce the NF-κB activity levels (c.f. section Irak3 Variants Differentially Activate Basal NF-κB Levels, But not Their Final Level After Stimulation). Our 3D-model analysis shows that m ′′ exposes the C-terminal domain more prominently than the full-length factor Irak3-m. This domain contains the motif Pro-X-Glu-X-X(aromatic/acidic residue), which is crucial for the interaction with the downstream signaling factor TRAF6 (72) and is well-conserved in the fulllength Irak3 factor and its splice variant m ′′ . Hence, Irak3m ′′ might perform its inhibitory function on the expression of il1b and cxcl8 by scavenging TRAF6 in a highly efficient way. This could, in turn, reduce the intensity of TLR-mediated NF-κB activation.
It is poorly understood in the mammalian ortholog, how the shortening of the full-length factor impacts the function of IRAK3, as encoded by the respective splice variants. IRAK2 factors might perhaps serve as a paradigm. There are four isoforms of IRAK2 in mouse and two alternatively spliced factors act as inhibitory factors (73), while the other two support the TLR-signal transfer in a positive manner (74). This example illustrates how the splicing of a particular mRNA transcript may reverse the basic function of the original protein and extend its functional spectrum with either additional or even antagonizing aspects.

Expression Profiling Provides no Indications That any of the irak3 Variants From Trout Induces the Expression of Other Immune Regulators
In parallel to irak3, we profiled the expression of a broader set of immune inhibitors in infected rainbow trout with focus on those genes whose expressions are considered IRAK3-dependent. Previous publications on mammalian models reported that IRAK3 induces the expression of genes coding for SOCS, INPP5D (SHIP1), TNFAIP3 (A20), NFKBIA (IκBα), SOD2, and GPR84 (5,15,17). In fact, we observed an early upregulation of irak3 transcripts concomitant with increased levels of inpp5d, tnfaip3, nfkbia-2 and -3, sod2, and gpr84 in the head kidney and/or liver. However, our overexpression studies on the salmonid cell line CHSE-214 did not provide any evidence for the involvement of Irak3 factors in stimulating the expression of specific immune inhibitors. Only the overexpression of variant m ′′ led to an increase in the expression of inpp5d.
Using discriminating primers, we determined the abundances of all irak3 transcripts and the full-length irak3 transcripts during the infection of trout with A. salmonicida. Within 24 hpi, irak3 transcript levels increased significantly both in the liver and head kidney, but not in PBLs. These results complement our previous observation that irak3-transcript levels were elevated at 72 hpi with A. salmonicida in the gills of rainbow trout (75). The infection-related increase in the expression of irak3 has previously been reported in embryonic zebrafish (27).
In summary, our in-vivo analyses of selected Irak3 factors from rainbow trout revealed that both, the full-length and the truncated Irak3 factor contain the structural requirements that allow modulating the activation status of NF-κB. The splice variant Irak3-m ′′ from trout limits the expression of inflammatory cytokines. However, we found no evidence that Irak3 from trout actually induces the expression of other immune regulators in contrast to its mammalian counterpart.

CONCLUSIONS
The mammalian IRAK3 factor controls not only the various TLRdependent signaling cascades but also the signal transduction downstream of the interleukin-1 receptor (IL1R1). While only one full-length and one splice variant of human IRAK3 have been reported, we found more than 20 irak3 cDNA variants in the salmonid fish rainbow trout, which probably stem from two duplicated genes. It is possible that the presence of these several irak3-transcript variants is only a collateral and functionally insignificant consequence of the polyploidization history of salmonids (41,52). Nevertheless, these multiple Irak3 variants may represent not only sheer abundance, but could cover different and/or complementing tasks. The individual irak3 variants from trout could, for instance, integrate specifically into the different cascades mediated by IL1R1 and more than a dozen TLRs (44,(76)(77)(78)(79) present in pseudo-tetraploid salmonids. Although our study suggests that the trout Irak3 factors do not interfere with the stimulation-dependent NF-κB activation, we found that only, and perhaps significantly, a specific Irak3splice variant eventually downregulates the expression of certain cytokine genes. This demonstrates that the great structural diversity of the irak3 factors from trout may also translate into some functional diversity.

DATA AVAILABILITY STATEMENT
The novel datasets generated for this study are included in the manuscript/Supplementary Files. All irak3-cDNA sequences have been submitted to the "European Nucleotide Archive" (project ID: PRJEB29555).