Two alternative pathways for docosahexaenoic acid (DHA, 22:6n-3) biosynthesis are widespread among teleost fish

Docosahexaenoic acid (DHA) plays important physiological roles in vertebrates. Studies in rats and rainbow trout confirmed that DHA biosynthesis proceeds through the so-called “Sprecher pathway”, a biosynthetic process requiring a Δ6 desaturation of 24:5n−3 to 24:6n−3. Alternatively, some teleosts possess fatty acyl desaturases 2 (Fads2) that enable them to biosynthesis DHA through a more direct route termed the “Δ4 pathway”. In order to elucidate the prevalence of both pathways among teleosts, we investigated the Δ6 ability towards C24 substrates of Fads2 from fish with different evolutionary and ecological backgrounds. Subsequently, we retrieved public databases to identify Fads2 containing the YXXN domain responsible for the Δ4 desaturase function, and consequently enabling these species to operate the Δ4 pathway. We demonstrated that, with the exception of Δ4 desaturases, fish Fads2 have the ability to operate as Δ6 desaturases towards C24 PUFA enabling them to synthesise DHA through the Sprecher pathway. Nevertheless, the Δ4 pathway represents an alternative route in some teleosts and we identified the presence of putative Δ4 Fads2 in a further 11 species and confirmed the function as Δ4 desaturases of Fads2 from medaka and Nile tilapia. Our results demonstrated that two alternative pathways for DHA biosynthesis exist in teleosts.

latter being β-oxidised to DHA in peroxisomes 17 . This pathway, known as the "Sprecher pathway", was subsequently confirmed to be operative in rainbow trout Oncorhynchus mykiss 18,19 . The first question that arose after the demonstration of this pathway was whether the same or different ∆6 Fads catalysed the reactions with C 18 and C 24 substrates 15 . It was demonstrated that the same ∆6 Fads carried out the conversions of 18:3n−3 to 18:4n−3 and 24:5n−3 to 24:6n−3 in humans 20 and rat 21,22 . In fish species, it is still unclear whether the same Fads catalyses the two ∆6 desaturation reactions or if two ∆6 Fads (isoenzymes) are involved 12,23,24 . Studies using yeast as a heterologous expression system confirmed that the bifunctional ∆6∆5 Fads from zebrafish (Danio rerio) had ability to desaturate both C 18 and C 24 substrates at the ∆6 position 24 . However, the Nibe croaker (Nibea mitsukurii) ∆6 Fads catalysed the desaturation of C 18 but not C 24 substrates 25 . These findings suggested that the DHA biosynthetic capability varied among teleost fish and, interestingly, recent findings have demonstrated that, unlike other vertebrates, teleost fish have acquired alternative pathways for DHA biosynthesis during evolution 6 .
The "∆4 pathway", first described in the marine protist Thraustochytrium sp. 26 , is a more direct pathway involving one single elongation of EPA to docosapentaenoic acid (DPA, 22:5n−3), which is subsequently desaturated at the ∆4 position to produce DHA. Although for many years Δ4 desaturases had not been found in any vertebrate species, a Fads2 with Δ4 desaturase activity was first discovered in rabbitfish (Siganus canaliculatus) 27 . Since then, Fads with ∆4 desaturases have been found in several teleost species such as Senegalese sole (Solea senegalensis) 28 , pike silverside (Chirostoma estor) 29 and striped snakehead (Channa striata) 30 . Recently, human cells expressing the baboon FADS2 had the ability for direct ∆4 desaturation of 22:5n−3 to 22:6n−3 31 . Thus, the existence of the ∆4 pathway among teleosts appeared to be more widespread than initially believed.
It is interesting to note that, unlike other vertebrates, current evidence suggests that all fads-like genes found in teleost fish are Fads2 orthologues 32 . Thus the functional diversity among fish Fads2 described above has been hypothesised to be dependent upon various factors including the phylogenetic position of species, in combination with environmental and ecological factors 6 . In the present study, we aimed to elucidate the pathways for DHA biosynthesis existing in species representing major lineages along the tree of life of teleost fish 33 . In particular, we have investigated the prevalence of the Sprecher pathway among teleost fish by determining the Δ6 activity towards C 24 substrates (24:5n − 3 and 24:4n−6) of desaturases with different substrate specificities (Δ6, Δ5 and Δ4), and derived from fish species with different evolutionary and ecological backgrounds. Furthermore, we have taken advantage of the now known key amino acid (aa) residues determining Δ4 desaturase ability of Fads 34 to identify teleost taxa, with publically available genomic or transcriptomic databases, in which their desaturase repertoire enables them to biosynthesise DHA through the more direct Δ4 pathway.
Putative Δ4 desaturase collection and phylogenetics. The phylogenetic tree comparing the deduced aa sequence of the fish Fads with those of human and rat is shown in Fig. 3. All Fads1 clustered together and were separate from all Fads2 in the tree. All teleost Fads2 studied in the present study strongly clustered within the teleost group (99% bootstraps), with desaturases from early divergent teleost species (e.g. A. gigas, A. japonica, C. gariepinus, S. salar and O. mykiss) clustering separately from desaturases from species belonging to more recent Figure 3. Phylogenetic tree comparing the amino acid sequences of teleost Fads2 with non-teleost vertebrate Fads-like from the cartilaginous fish and mammals (human and mouse). The numbers represent the frequencies (%) with which the tree topology presented was replicated after 1,000 iterations. The functionally characterised Fads were shown with their corresponding regioselectivity (Δ6, Δ5, Δ6Δ5 and Δ4). Asterisks ("*") indicate Fads2 that have been subjected to further functional analyses in the present study, including the newly cloned Δ4 Fads2 from medaka (Oryzias latipes) and Nile tilapia (Oreochromis niloticus) highlighted in bold. Crosses (" †") indicate Fads2 that possess the YXXN amino acid residues determining Δ4 desaturase activity 32 . Branches including Teleostei and Acanthopterygii Fads2 sequences are indicated.
lineages (95% bootstraps) (Fig. 3). Among the latter, one can find all the sequences with YXXN residues determining Δ4 activity 34 including the previously studied Δ4 desaturases from S. canaliculatus, S. senegalensis and C. estor and the herein characterised Fads2 from medaka (Oryzias latipes) and Nile tilapia (O. niloticus). Clearly, all Fads2-like proteins from Nile tilapia and other cichlids formed a monophyletic clade (99% bootstraps), itself comprising a subgroup with Fads2 sequences possessing the abovementioned distinctive YXXN motif for Δ4 desaturases and another group that includes the Δ6Δ5 Fads2 from Nile tilapia (Fig. 3).  Table 2), demonstrating that they were both Δ4 desaturases as predicted. Interestingly, both enzymes showed some activity as Δ5 desaturases, since they were able to convert 20:4n−3 and 20:3n−6 into 20:5n−3 (EPA) and 20:4n−6 (ARA), respectively (Table 2). Generally, the Nile tilapia Δ4 Fads2 showed higher conversion activities compared to that of medaka and both proteins were more efficient towards n−3 substrates compared to n−6 substrates.

Discussion
It has been largely accepted that DHA biosynthesis in vertebrates proceeds through the Sprecher pathway 6, 9 . While most earlier investigations focussed on mammals, studies in O. mykiss confirmed that the Sprecher pathway also operated in fish 18,19 . It was subsequently demonstrated that the same ∆6 Fads-like enzyme that acts on C 18 PUFA precursors at the initiation of the LC-PUFA biosynthesis ( Fig. 1) was also responsible for the desaturation of 24:5n−3 required in the Sprecher pathway 20,22 . Despite the plethora of studies reporting on the functions of fish Fads 6 , the capability of fish Fads to operate towards 24:5n−3, and therefore to contribute to DHA biosynthesis through the Sprecher pathway, had not been fully established. For that purpose, we herein conducted a retrospective study investigating the ability to operate as ∆6 desaturases towards 24:5n−3 and 24:4n−6 of a range of previously characterised Fads2 from fish belonging to lineages distributed along the phylogenetic tree of teleosts 33 . Using a newly developed method involving yeast, we were able to establish that, with the exception of the Nibe croaker Fads2, all teleost non-∆4 desaturases tested in this study had the ability to efficiently convert 24:5n−3 and 24:4n−6 into 24:6n−3 and 24:5n−6, respectively, confirming their ability for ∆6 desaturation of C 24 PUFA substrates. Such ability was observed in Fads2 from species spread across the evolutionary history of teleosts from basal (e.g. A. gigas and A. japonica) and recent (e.g. S. canaliculatus and S. aurata) lineages, and with different regioselectivities including ∆6 desaturases (e.g. O. mykiss and S. aurata) and bifunctional ∆6∆5 desaturases (e.g. C. estor and S. canaliculatus). Since all these Fads2 also showed ∆6 desaturase activity towards C 18 PUFA (18:3n−3 and 18:2n−6), the present results confirmed that the same Fads2 can function as ∆6 desaturases at both steps of the LC-PUFA biosynthetic pathway as described above for mammals 20,22 . This is in agreement with studies on zebrafish ∆6∆5 Fads2, which showed the ability to operate as ∆6 desaturase on C 18 35 and C 24 PUFA substrates, using for the latter a yeast supplemented with 24:5n−3 24 . Interestingly, we could also confirm that the previously characterised ∆5 Fads2 from Atlantic salmon S. salar 36 , also showed ∆6 activity on C 24 PUFA (24:4n−6 and 24:5n−3) and 18:3n−3, suggesting that this enzyme is indeed a bifunctional ∆6∆5 desaturase. Bifunctionality appears a relatively widespread feature among fish Fads2 as a consequence of sub-(acquisition of additional substrate specificities) and neo-functionalisation (substitution and/or acquisition of new substrate specificities) events that have occurred in teleost Fads2 6,29 . More specifically, bifunctional ∆6∆5 Fads2 have been found in D. rerio 35 , S. canaliculatus 27 , O. niloticus 37 , C. estor 29 , C. gariepinus 38 and C. striata 39 . Moreover, all the ∆4 Fads2 found so far in fish also exhibited some ∆5 desaturase activity [27][28][29][30] , although none of them had ∆6 activity, which is consistent with the lack of ∆6 desaturase activity towards C 24 PUFA substrates observed in all the ∆4 Fads2 assayed in the present study. Interestingly, our results show that the two pathways of DHA biosynthesis, namely the Sprecher and ∆4 pathways, co-exist within some species such as S. canaliculatus and C. estor since, in addition to the role of their ∆6∆5 Fads2 in the Sprecher pathway uncovered in the present study, the existence of ∆4 desaturases in their genomes potentially enables them to further operate via the ∆4 pathway 27,29 .
The Δ4 pathway was first reported in the rabbitfish S. canaliculatus 27 , with further Δ4 desaturases subsequently described in S. senegalensis, C. estor and C. striata [28][29][30] . In the present study, we have expanded the number of fish lineages and species in which putative Δ4 desaturases exist. In particular, putative Δ4 desaturases were identified in 11 species belonging to Cichliformes (O. niloticus, Maylandia zebra and Haplochromis burtoni), Beloniformes (O. latipes), Blenniiformes (Tomicodon sp., Acyrtus sp. and Enneanectes sp.), Cyprinodontiformes (Poecilia reticulata, Fundulus heteroclitus and Austrofundulus limnaeus) and Centrarchiformes (Micropterus salmoides). It is very likely that the number of species with Δ4 Fads2 will expand when further genomic and/or transcriptomic data become available. This is particularly true for species within groups such as Cichliformes and Beloniformes, in which we found putative Δ4 Fads2 in all species studied in each group. While the actual functional activity of the retrieved desaturases should be assessed in each individual species, the characterisation of two of the newly identified putative Δ4 desaturases, namely the O. latipes and O. niloticus Fads2, as functional Δ4 enzymes supported the effectiveness of our in silico search strategy using a conserved aa region containing the YXXN motif responsible for Δ4 activity to identify functional Δ4 desaturases 34 . Overall, these results clearly showed that the presence of Δ4 Fads2 among teleosts was far more common than initially believed when the first vertebrate Δ4 desaturase was discovered in S. canaliculatus 27 . However, the presence of Δ4 Fads2 appears to be restricted to teleost species within groups regarded herein as "recent lineages", indicating that the acquisition of the Δ4 pathway occurred later during the evolution of teleosts 6,29 .
In more basal teleost lineages, namely Osteoglossiformes (e.g. A. gigas), Anguilliformes (e.g. A. japonica), Cypriniformes (e.g. D. rerio), Siluriformes (e.g. C. gariepinus) and Salmoniformes (e.g. S. salar and O. mykiss), the Sprecher pathway appears to be the only possible route available for DHA biosynthesis. This is supported by, not only the apparent absence of Δ4 Fads2 in their genomes, but also the relatively higher capacity for Δ6 desaturase towards 24:5n−3 of their Fads2, as denoted by normalising the Δ6 conversions of 24:5n−3 (Δ 24:5n−3 ) with that towards a control substrate (Δ control ). Thus, Fads2 from early divergent teleosts, along with the cartilaginous fish S. canicula, had relatively high capacity for ∆6 desaturation towards 24:5n−3, with Δ 24:5n−3 /Δ control ≥ 0.82. In contrast, Fads2 from other species (S. aurata, C. estor and S. canaliculatus) had lower Δ 24:5n−3 /Δ control ≤ 0.43, indicating lower activity of the Sprecher pathway. While exceptions to this pattern are likely to exist given the functional diversity among teleost Fads2 6, 29 , the apparent lower contribution of the Sprecher pathway to DHA biosynthesis in late-diverging teleosts coincided with the occurrence of Δ4 Fads2 enabling certain species an alternative route for DHA biosynthesis. The limited activity of the Sprecher pathway among these teleost species might be not only restricted to their lower desaturation capability on 24:5n−3 stated above, but also to the absence of key elongase enzymes such as Elovl2, responsible for the production of the Δ6 desaturase substrate 24:5n−3 [40][41][42] . Although Elovl4 can partly compensate such an absence in certain tissues [42][43][44][45] , loss of Elovl2 in the genomes of Acanthopterygii, a group that includes all the late-diverging species considered in this study 46 , can notably compromise the efficient production of 24:5n−3 as precursor for DHA biosynthesis via the Sprecher pathway. Lack of key enzymatic capabilities in LC-PUFA biosynthetic pathways has been speculated to be a consequence of species having readily available essential LC-PUFA in their diets 24,41 . This is the case of marine teleosts, particularly higher trophic species, in which no selection pressure to retain complete and active LC-PUFA biosynthetic pathways has been exerted. For example, extreme cases of marine teleosts with loss of enzymatic activities include the pufferfish (e.g. Tetraodon nigroviridis and Takifugu rubripes), which lack Fads2 in their genomes 46 . In the present study, we observed that the marine carnivore Nibe croaker N. mitsukurii possess a Fads2 that was the only non-Δ4 Fads2 studied that showed no detectable activity towards 24:5n−3. These results were consistent with the inability of N. mitsukurii Fads2 to desaturate 24:5n−3 to 24:6n−3 in yeast 25

Conclusions
The present study demonstrated that, with the notable exception of Δ4 desaturases, fish Fads2 have the ability to operate as Δ6 desaturases towards C 24 PUFA enabling them to synthesise DHA through the Sprecher pathway. However, the so-called "Δ4 pathway" represents an alternative route in some species. Through in silico searches, the present study revealed that the presence of Δ4 Fads was more common than initially believed, and reported three new orders and 11 species in which putative Δ4 desaturases were identified. Importantly, two putative Δ4 Fads2 desaturases retrieved from medaka and Nile tilapia in silico were confirmed as functional Δ4 enzymes. Interestingly, functional characterisation of the S. salar Fads2 previously characterised as a Δ5 desaturase confirmed this enzyme has also Δ6 desaturase activity and should be therefore regarded as a bifunctional Δ6Δ5 desaturase. Overall our results demonstrate that two alternative routes for DHA biosynthesis can exist in teleost fish. Whereas the Sprecher pathway appeared to be widely spread across the entire clade, a more scattered distribution was observed for the Δ4 pathway.  N. mitsukurii). The desaturase sequences from fish species listed above were used for phylogenetic analysis and selected sequences were subjected to functional characterisation as described.

Methods
Determination of Δ6 desaturase activity of fish Fads2 towards C 24 PUFA in co-transformant Saccharomyces cerevisiae. We first investigated the ability for Δ6 desaturase activity towards C 24 PUFA substrates, i.e. 24:4n−6 and 24:5n−3, the latter being an intermediate in the Sprecher pathway for DHA biosynthesis. Such activities were tested in a total of 15 Fads sequences belonging to 12 species of fish (Table 3), through a newly developed yeast-based assayed as follows. Yeast competent cells InvSc1 (Invitrogen) were co-transformed with two different plasmid constructs prepared as described below. First, the D. rerio elovl2 open reading frame  (ORF) 48 was ligated into the yeast expression vector p415TEF (a centromeric plasmid with a LEU2 selectable marker) to produce the construct p415TEF-elovl2, in which the expression of the D. rerio elovl2 was controlled under the yeast TEF1 promoter (constitutive expression). Second, the ORF of the corresponding fish Fads (Table 3) was cloned into the episomal yeast vector pYES2, in which the Fads expression was under the control of the GAL1 promoter (inducible expression). Selection of transformant yeast containing both constructs was performed by growing the co-transformed yeast on S. cerevisiae minimal medium minus uracil minus leucine (SCMM −ura−leu ) plates. One single colony was grown in SCMM −ura−leu broth for 24 h at 30 °C, and subsequently subcultured in individual Erlenmeyer flasks at 0.1 OD 600 (t 0 ) and supplemented with either 0.75 mM Na salts of 22:4n−6 (docosatetraenoic acid, DTA) or 22:5n−3 (DPA) (0.75 mM). Co-transformed yeast were then grown for 24 h (t 0 + 24 h) allowing the D. rerio Elovl2 to convert the exogenously added C 22 substrates (DTA or DPA) into their corresponding C 24 elongation products 24:4n−6 and 24:5n−3, respectively. In order to test the ability of the fish desaturases to introduce Δ6 double bonds into the newly synthesised 24:4n−6 and 24:5n−3 in yeast, the fads expression was then induced (t 0 + 24 h) by addition of 2% galactose, after which the recombinant yeast were further grown for 48 h (t 0 + 72 h) before collection. As positive controls, a subculture aliquot of the same colony used for the above described assay was supplemented with an n−3 PUFA substrate for which the corresponding assayed Fads had previously shown activity (Table 3) and galactose (2%) at t 0 . More specifically, co-transformant yeasts were grown in the presence of 18:3n−3 as controls for Δ6 (e.g. AgΔ6Fads2) or Δ6Δ5 (e.g. DrΔ6Δ5Fads2) desaturases, 20:4n−3 for Δ5 desaturases (e.g. SsΔ5Fads2) and 22:5n−3 for Δ4 desaturases (e.g. CeΔ4Fads2).
The yeast co-transformed with empty p415TEF and pYES2 vectors were also prepared as negative controls.
In silico retrieval of putative Δ4 desaturases. For retrieval of putative Δ4 desaturase sequences from databases, an alignment of the four functionally characterised Δ4 desaturases from rabbitfish (ADJ29913), Senegalese sole (AEQ92868), pike silverside (AHX39206) and striped snakehead (ACD70298) was performed using the Clustal Omega Multiple Sequence Alignment tool (http://www.ebi.ac.uk/Tools/msa/clustalo/). The conserved aa sequence PPLLIPVFYNFNIMXTMISR, which included the four key aa residues (underlined) accounting for Δ4 regioselectivity 34 , was used as a query for blast searches. The majority of the putative Δ4 desaturase sequences were obtained from the NCBI Non-redundant protein sequences (nr) database using the blastp algorithm. We further explored the Expressed Sequence Tags (EST) and Transcriptome Shotgun Assembly (TSA) databases using the tblastn algorithm. In addition, the Fish-T1K website (http://www.fisht1k.org) was also used for the tblastn search. Among the retrieved sequences, we selected only those that contained "Y" and "N" in positions 1 and +4, respectively, within the four aa domain YXXN, as these have been reported previously to be crucial for Δ4 function 34 .

Phylogenetic analysis of Fads desaturases.
A phylogenetic tree was built to compare the deduced aa sequences of the fish Fads considered in the present study. The neighbour-joining method 49 , with the CLC Main Workbench 7 (CLC bio, Aarhus, Denmark), was used to construct the phylogenetic tree, with confidence in the resulting tree branch topology measured by bootstrapping through 1,000 iterations. The alignment of Fads aa sequences used for constructing the phylogenetic tree was performed with MAFFT using the L-INS-i method 50 . Non-teleost fish sequences from S. canicula and mammalian (human and mouse) Fads2 sequences were also included in the analysis. obtained were purified, digested with the restriction enzymes, and ligated into similarly digested pYES2 yeast expression vector (Invitrogen) producing the constructs pYES2-Olfads2 and pYES2-Onfads2, respectively. Yeast competent cells InvSc1 (Invitrogen) were transformed with the plasmid constructs using the S.c. EasyComp TM Transformation Kit (Invitrogen). Selection of yeast containing the pYES2 constructs was performed on S. cerevisiae minimal medium minus uracil (SCMM −ura ) plates. One single yeast colony of each transformation (pYES2-Olfads2 or pYES2-Onfads2) was grown in SCMM −ura broth for 2 days at 30 °C, and subsequently subcultured in individual Erlenmeyer flasks until optical density measured at a wavelength of 600 nm (OD 600 ) reached 1, after which galactose (2%, w/v) and a PUFA substrate were added as sodium (Na) salts. The exogenous supplemented PUFA included Δ6 ( Fatty acid analysis of yeast. Total lipids extracted from yeast samples 52 were used to prepare fatty acid methyl esters (FAME). FAME extraction, purification and analysis were performed as described by Li et al. 27 . For functional characterisation of the desaturases from O. latipes and O. niloticus, substrate FA conversions were calculated as the proportion of exogenously added FA substrate desaturated as [product area/(product area + substrate area)] × 100 51 . Substrate FA conversions for the Δ6 desaturase activity towards C 24 substrates were calculated using the same formula as above considering the areas of 24:5n−3 and 24:4n−6 produced endogenously by the D. rerio Elovl2 as substrates for calculations. When necessary, GC-MS was used to confirm the identity of the products 27 .