Metabolic Engineering of ω3-Very Long Chain Polyunsaturated Fatty Acid Production by an Exclusively Acyl-CoA-dependent Pathway*

ω3-Very long chain polyunsaturated fatty acids (VLCPUFA) are essential for human development and brain function and, thus, are indispensable components of the human diet. The current main source of VLCPUFAs is represented by ocean fish stocks, which are in severe decline, and the development of alternative, sustainable sources of VLCPUFAs is urgently required. Our research aims at exploiting the powerful infrastructure available for the large scale culture of oilseed crops, such as rapeseed, to produce VLCPUFAs such as eicosapentaenoic acid in transgenic plants. VLCPUFA biosynthesis requires repeated desaturation and repeated elongation of long chain fatty acid substrates. In previous experiments the production of eicosapentaenoic acid in transgenic plants was found to be limited by an unexpected bottleneck represented by the acyl exchange between the site of desaturation, endoplasmic reticulum-associated phospholipids, and the site of elongation, the cytosolic acyl-CoA pool. Here we report on the establishment of a coordinated, exclusively acyl-CoA-dependent pathway, which avoids the rate-limiting transesterification steps between the acyl lipids and the acyl-CoA pool during VLCPUFA biosynthesis. The pathway is defined by previously uncharacterized enzymes, encoded by cDNAs isolated from the microalga Mantoniella squamata. The conceptual enzymatic pathway was established and characterized first in yeast to provide proof-of-concept data for its feasibility and subsequently in seeds of Arabidopsis thaliana. The comparison of the acyl-CoA-dependent pathway with the known lipid-linked pathway for VLCPUFA biosynthesis showed that the acyl-CoA-dependent pathway circumvents the bottleneck of switching the Δ6-desaturated fatty acids between lipids and acyl-CoA in Arabidopsis seeds.

3-Very long chain polyunsaturated fatty acids (VLCPUFA) are essential for human development and brain function and, thus, are indispensable components of the human diet. The current main source of VLCPUFAs is represented by ocean fish stocks, which are in severe decline, and the development of alternative, sustainable sources of VLCPUFAs is urgently required. Our research aims at exploiting the powerful infrastructure available for the large scale culture of oilseed crops, such as rapeseed, to produce VLCPUFAs such as eicosapentaenoic acid in transgenic plants. VLCPUFA biosynthesis requires repeated desaturation and repeated elongation of long chain fatty acid substrates. In previous experiments the production of eicosapentaenoic acid in transgenic plants was found to be limited by an unexpected bottleneck represented by the acyl exchange between the site of desaturation, endoplasmic reticulum-associated phospholipids, and the site of elongation, the cytosolic acyl-CoA pool. Here we report on the establishment of a coordinated, exclusively acyl-CoA-dependent pathway, which avoids the rate-limiting transesterification steps between the acyl lipids and the acyl-CoA pool during VLCPUFA biosynthesis. The pathway is defined by previously uncharacterized enzymes, encoded by cDNAs isolated from the microalga Mantoniella squamata. The conceptual enzymatic pathway was established and characterized first in yeast to provide proof-ofconcept data for its feasibility and subsequently in seeds of Arabidopsis thaliana. The comparison of the acyl-CoA-dependent pathway with the known lipid-linked pathway for VLCPUFA biosynthesis showed that the acyl-CoA-dependent pathway circumvents the bottleneck of switching the ⌬6-desaturated fatty acids between lipids and acyl-CoA in Arabidopsis seeds.
Human development and health depend in many respects on the availability of long chain multiply unsaturated fatty acids of 20 or 22 carbons in length that contain up to 6 methyleneflanked cis-double bonds. These fatty acids are classified under the designation very long chain polyunsaturated fatty acids (VLCPUFAs). 2 Nutritionally important VLCPUFAs include arachidonic acid (ARA, 20:4 ⌬5, 8,11,14 ), an 6-fatty acid, and the 3-fatty acids eicosapentaenoic acid (EPA, 20:5 ⌬5, 8,11,14,17 ) and docosahexaenoic acid (DHA, 22:6 ⌬4,7,10,13,16,19 ). 3-VLCPUFAs are of particular interest from a nutritional standpoint since the uptake of these fatty acids is considered to be low in Western diets (1). 3-VLCPUFAs have long been investigated for their importance during human fetal development and the formation and function of the central nervous system, brain, and retina. In addition to their structural functions in membranes, several medical studies indicate that 3-fatty acids have antiinflammatory properties and, therefore, might be useful in the management of inflammatory and autoimmune diseases such as cardiovascular disease, major depression, arthritis, inflammatory bowel disease, asthma, and psoriasis (2).
Whereas mammals, including humans, can convert the essential precursors 6 -18:2 ⌬9,12 or 3-18:3 ⌬9,12,15 (␣-linolenic acid) to VLCPUFAs, a considerable proportion of VLCPUFA has to be taken up directly as components of the diet (3). Naturally occurring producers and sources of 3-VLCPU-FAs are microorganisms, including marine bacteria and microalgae. These organisms represent the starting point of the aquatic food chain by which VLCPUFAs ultimately accumulate in fish oils. At the moment the main source of 3-VLCPUFAs for human consumption are fatty ocean fish, such as salmon, mackerel, tuna, or herring. Unfortunately, the increased demand for fish and fish oils has led to a depletion of fish stocks in vast ocean areas worldwide. Thus, as an exclusive source of VLCPUFAs, fish cannot cover the needs of a growing world population (4). Altogether, the increasing interest to find alternative sources of 3-VLCPUFAs has led to various attempts to produce 3-VLCPUFAs via the biotechnological introduction of new biosynthetic pathways in plants, ultimately aiming for VLCPUFA-production in annual oilseeds, e.g. in rapeseed or linseed.
VLCPUFA synthesis in transgenic plants established so far starts from the plant endogenous fatty acids 6 -18:2 ⌬9,12 or 3-18:3 ⌬9,12,15 and requires two distinct catalytic activities, desaturases and elongases. Both activities work together in an alternating manner to establish the final product. Genes for the biosynthesis of VLCPUFAs have been isolated from organisms of different kingdoms (algae, amoeba, fungi, moss, and flowering plants) and have been characterized by heterologous expression in various model systems regarding the properties of the encoded enzymes.
In recent years a range of tests was done to establish the biosynthesis of VLCPUFAs in plants by introducing a variety of different front-end desaturases and elongases (5,6). Two previous approaches are of particular relevance for the study presented here. First, the so called alternative pathway for the biosynthesis of VLCPUFAs was successfully tested in Arabidopsis leaves (7). This pathway is based on the sequential action of a ⌬9 elongase (from Isochrysis galbana) (8), a ⌬8 desaturase (from Euglena gracilis) (9), and a ⌬5 desaturase (from the fungus Mortierella alpina) (10). The ⌬9 elongase converts CoA-bound 18:2 ⌬9,12 and/or ␣-18:3 into 20:2 ⌬11:14 and/or 20:3 ⌬11,14,17 , which could be further converted to ARA and EPA via lipidbound ⌬8 and ⌬5 desaturation, respectively. In this fashion a rate-limiting ⌬6-elongation step was avoided, and EPA accumulation of up to 3% of the leaf total lipid content of Arabidopsis thaliana was achieved. It must be noted, however, that leaf tissue is not oleogenic and that the fatty acids are directly incorporated into membrane lipids rather than stored as oils as is the case in seeds and as would be desirable for VLCPUFA production. Moreover, a crucial problem was observed. The ⌬9-elongation products (20:2 ⌬11,14 and 20:3 ⌬11,14,17 ) accumulated to very high levels in the acyl-CoA pool of the transgenic Arabidopsis plants (11), indicating an inefficient transfer of these non-native fatty acids out of the acyl-CoA pool.
In the second approach the coexpression of ⌬6 and ⌬5 desaturases of the diatom Phaeodactylum tricornutum with a ⌬6 elongase from the moss Physcomitrella patens in linseed showed that the enzymes carried out ⌬5 and ⌬6 desaturation on lipid-bound substrates in the plant with a positional specificity for the sn-2 position of phosphatidylcholine (PC), whereas the ⌬6 elongase preferred acyl-CoA-species as substrates (12,13). Comprehensive acyl-CoA and lipid analysis of the EPA-producing transgenic linseed plants demonstrated that production of VLCPUFAs in plants requires not only the interplay of desaturases and elongases but also the transfer of acyl groups from the PC pool into the CoA pool and vice versa. This transfer is very likely catalyzed by the enzymatic activity of an acyl-CoA:lysophosphatidylcholine acyltransferase (LPCAT). The LPCAT endogenous to linseed does not accept the newly formed ⌬6-desaturated fatty acids, and thus, the ⌬6-desaturated fatty acids accumulated to high levels. In brief, the back-and-forth trans-acylation of fatty acids between the PC and CoA pools by the action of LPCAT represents a rate-limiting factor for VLCPUFA production by this transgenic approach in linseed (12).
In the research mentioned as well as in several other studies, e.g. Refs. 14 and 15, desaturases from various organisms were analyzed for their feasibility to produce VLCPUFAs. The accu-mulated evidence indicates that the front-end desaturases characterized from algae, fungi, and moss are lipid-dependent desaturases and accept glycerolipid-linked substrates. In contrast to these lipid-dependent desaturases, a number of mammalian front-end desaturases accept acyl-CoA species as substrates (16 -18). Recently, one acyl-CoA-dependent ⌬6 desaturase from the microalga Ostreococcus tauri was identified (19). From fatty acid analyses performed on microalgae like O. tauri, however, it could be concluded that further desaturases with a preference for acyl-CoA substrates can be isolated from those organisms. Such plant-like acyl-CoA-dependent enzymes with the correct substrate specificities may allow bypassing the rate-limiting transport and exchange of intermediates of VLCPUFA synthesis between lipid-bound desaturation in the PC pool and elongation steps in the acyl-CoA pool. Many vascular plants synthesize 18:2 ⌬9,12 and 18:3 ⌬9,12,15 in their seed oils but do not produce or incorporate VLCPUFAs with chain lengths above 18 carbon atoms or more than three double bonds into triacylglycerol (TAG). To produce the 3-fatty acid, 3-20:5 ⌬5, 8,11,14,17 (EPA), in seed oils, it is necessary to introduce one additional elongation and two desaturation steps. To avoid the accumulation of 6 byproducts of desaturation, the enzymes selected should be specific for 3 substrates. A second known constraint to circumvent is the rate-limiting shuttling of fatty acids between PC and CoA pools observed by Abbadi et al. (12).
To contribute to a solution for these problems, we report on the isolation and characterization of acyl-CoA-dependent desaturases from the microalgae Mantoniella squamata and O. tauri. Using the new enzymes, an entirely acyl-CoA-dependent 3-VLCPUFA biosynthetic pathway consisting of ⌬6and ⌬5 desaturase enzymes was successfully established in Saccharomyces cerevisiae and in seeds of A. thaliana plants. This modified pathway may allow a more efficient flux during VLCPUFA biosynthesis and avoids the bottleneck after ⌬6 desaturation described by Abbadi et al. (12).

EXPERIMENTAL PROCEDURES
Materials-Restriction enzymes and DNA-modifying enzymes were obtained from MBI Fermentas. Standards of fatty acids as well as all other chemicals were from Sigma; methanol, n-hexane, and isopropanol (all high performance liquid chromatography grade) were from Baker. Fatty acids and acyl-CoAs were either obtained from Cayman Chemicals or Larodan. Basic molecular biological and biochemical techniques were performed as described (20).
Algae Material and Growth Conditions-M. squamata SAG 65.90 (21) was obtained from the algae culture collection Göttingen (SAG, Germany). O. tauri strain OTTH0595-genome was obtained from the Roscoff Culture Collection. Non-axenic cultures were grown in batch cultures under long daylight (14 h) conditions with 45 mol of photons m Ϫ2 s Ϫ1 in 200 ml of Brackish Water Medium (1/2 SWES) complemented with soil extract at 20°C.
cDNA Library Construction and Random Sequencing of the cDNA Library-Total RNA from 7-day-old M. squamata cultures were isolated using the RNAeasy kit (Qiagen) per the manufacturer's instructions. Poly Aϩ mRNAs were prepared using oligo-dT cellulose (22) (see above) and reverse-transcribed via a reverse transcription kit (Promega) and then used for the construction of the cDNA library with a Lambda ZAP Gold library construction kit (Stratagene). After in vivo mass excision of the cDNA library, plasmid recovery and transformation of Escherichia coli (Stratagene), plasmid DNA was prepared on a Qiagen DNA preparation robot (Qiagen) according to the manufacturer's instructions and submitted to random sequencing by the chain termination method using the ABI PRISM Big Dye Termination Cycle Sequencing Ready Reaction kit (PerkinElmer Life Sciences). Analyses and annotations of the EST sequences between 100 and 500 base pairs resulted in a non-redundant EST data base. Data base screening of the cDNA library yielded two sequences that were annotated as putative desaturases.
Isolation and Cloning of Desaturase Sequences-To obtain full-length cDNA sequences, the rapid amplification of cDNA ends (RACE) technique was used. Therefore, 5 l of total RNA isolated from M. squamata were reverse-transcribed and ligated to adaptor-ligated double-stranded cDNA by using the Marathon cDNA amplification kit (BD Bioscience). Adaptorligated double-stranded cDNA was used as template for 5Ј-and 3Ј-RACE PCRs to obtain the missing 5Ј-prime and 3Ј-prime ends of the coding sequences for several desaturases. For RACE reactions, gene-specific primers were designed from EST sequence information. Primers used in 5Ј-and 3Ј-RACE were as follows: for Ms⌬6 (MsI) as 5Ј-RACE primer, 5Ј-CATC-CGGGCGGCAGCGTCATCTTCTAC-3Ј, and as 3Ј-RACE primer, 5Ј-GGAGAAGAGGTGGTGGATGACCTGG-3Ј; for Ms⌬5 (MsII) as 5Ј-RACE primer, 5Ј-CCGAGTGAGGGGAG-TACGTGGCGGG-3Ј, and as 3Ј-RACE primer, 5Ј-CACTCTC-CGGCGGGCTCAACTACC-3Ј. A 50-l standard reaction contained 1ϫ Advantage2 DNA polymerase buffer, 1 l of Advantage 2 DNA Polymerase (BD Bioscience), 0.2 mM concentrations of each dNTP, 0.5 M 5Ј or 3Ј primer, 0.5 M adaptor primer 1 (AP1), and 5 l of adaptor-ligated double-stranded cDNA. RACE-PCR amplification was performed as follows: 30 s at 94°C; 5 cycles of 5 s at 94°C, 4 min at 72°C; 5 cycles of 5 s at 94°C, 4 min at 70°C; 20 cycles of 5 s at 94°C, 4 min at 68°C. Amplified products were isolated from agarose gels, purified using a gel extraction kit (GE Healthcare), and subsequently cloned into pGEM-T (Promega). Cloned inserts were sequenced using the ABI PRISM ® 3100 Genetic Analyzer (Applied Biosystems).
Cloning cDNA for Desaturases into Yeast Expression Vectors-From the 5Ј-and/or 3Ј-cDNA sequence data obtained by RACE-PCR, the putative translation initiation codons and stop codons were identified, and this sequence information was used to obtain full-length cDNA clones of the putative desaturases from M. squamata. Gene-specific primers were designed to the 5Ј-and the 3Ј-ends of the coding regions of the corresponding nucleotide sequences, introducing restriction sites for cloning into the yeast expression vectors and the yeast consensus sequence for enhanced translation in front of the start codons (23). The open reading frames of the Ms⌬6 (MsI) and Ms⌬5 (MsII) and of Ot⌬5 (OtII) were amplified and modified by PCR. The following pairs of primers were used (restrictions sites are in bold, translation initiation sequences are in italics, and start or stop codons are underlined): for Ms⌬6 as forward primer, 5Ј-ATGCGCGGCCGCACATAATGTGTCCTCC-CAAGGAAT-3Ј, and as reverse primer 5Ј-ATGCAGATCTC-TAGTGAGCGTGCGCCTTC-3Ј; for Ms⌬5 as forward primer, 5Ј-ATGCGCGGCCGCACATAATGCCCCCGCGCGA-GACCA-3Ј, and as reverse primer, 5Ј-ATGCAGATCTTCAC-CCGATGGTTTGAAGG-3Ј; and for Ot⌬5 as forward primer, 5Ј-ATGCGCGGCCGCACATAATGGGGACGACCGCGCG-CGAC-3Ј, and as reverse primer, 5Ј-ATGCAGATCTTCATC-CGACGGTTTGGAGGGACGG-3Ј. The amplified cDNAs were cloned into the pGEM-T vector (Promega) before being released and cloned into the yeast expression vector (pESC-LEU or pESC-TRP, respectively, Stratagene) using the restriction sites inserted by PCR, yielding pESC-LEU-Ms⌬6 and pESC-TRP-Ms⌬5.
Cloning and Vector Construction for Production of Transgenic Arabidopsis Plants-Plasmids for plant transformation were constructed based on the vector pUC19. For the generation of the constructs, a triple cassette containing three seedspecific USP promoters (26), three OCS terminators, and three different polylinkers between each promoter and terminator was first introduced to the vector pUC19 (Pharmacia Corp.), yielding the USP123OCS plasmid. The open reading frames of the different desaturases and elongases were modified by PCR to create appropriate restriction sites adjacent to the start and stop codons, cloned into the pGEM-T vector (Promega), and sequenced to confirm their accuracy. The primers used were (restrictions sites are in bold, and start or stop codons are underlined): Ms⌬6, forward, 5Ј-ATGCGCGGCCGCACATA-ATGTGTCCTCCCAAGGAAT-3Ј, reverse, 5ЈATGCTCTAG-ACTAGTGAGCGTGCGCCTTC-3Ј; Ms⌬5, forward, 5Ј-ATG-CCCATGGACATAATGCCCCCGCGCGAGACCACCAC-3Ј, reverse, 5Ј-ATGCACCGGTTCACCCGATGGTTTGAA-GGC-3Ј; Ot⌬6, forward, 5Ј-ATGCGCGGCCGCACATAATG-TGCGTGGAGACGGAAAAT-3Ј, reverse, 5Ј-ATGCTCTAG- The open reading frames were then released using the restriction sites created by PCR and successively inserted into the same sites of the polylinkers of the USP123OCS plasmid. The resulting cassette, containing the three genes each under the control of the USP promoter, was released by digesting the USP123OCS plasmid with SbfI or SacI and cloned into the corresponding sites of the binary vector pCAMBIA3300 yielding the constructs triple-Ms, triple-Ot, and triple-Pt. The binary vector pCAMBIA3300 (CAMBIA) uses the bar gene with the cauliflower mosaic virus 35 S promoter as a selectable marker in plants. The binary plasmid constructs were transformed into chemically competent Agrobacterium tumefaciens cells (strain EH105).
Expression in S. cerevisiae-S. cerevisiae cells strain INVSc1 (Invitrogen) were transformed as described (27). For induction, expression cultures were grown for 48 h to 72 h at 22°C in the presence of 2% (w/v) galactose supplemented with 150 -350 M concentrations of appropriate fatty acid substrate and in the presence of 1% Igepal CA 630 (Nonidet P-40) from Sigma-Aldrich. Cells were harvested by centrifugation at 1200 ϫ g for 5 min, and the pellets were washed twice with H 2 O before being used for further analysis. The host strain transformed with the empty vector(s) was used as negative control in all experiments.
Arabidopsis Transformation-A. thaliana ecotype Columbia (Col-0) plants were transformed by floral dipping (37). T2 seeds were collected from individual T1 plants resistant to ammonium glufosinate and analyzed individually by GC.
Fatty Acid Analysis-Fatty acid methyl esters (FAMEs) were obtained by methylation of microalgae or yeast cell sediments with 0.5 M sulfuric acid in methanol containing 2% (v/v) dimethoxypropane at 80°C for 1 h. FAMEs were extracted in 2 ml of n-hexane, dried under N 2 , and analyzed by gas chromatography (GC). FAMEs of single or pooled Arabidopsis seeds were prepared by transesterification with trimethylsulfonium hydroxide (28). FAMEs of TLC-separated individual lipids were obtained by transmethylation with 333 l of toluol/methanol (1:2 v/v) and 167 l of 0.5 M NaOCH 3 at room temperature for 20 min. FAMEs were extracted in 500 l of NaCl, 50 l of HCl (37%), and 2 ml of n-hexane, dried under N 2 , and analyzed by GC. The GC analysis was performed with an Agilent GC 6890 system coupled with a flame ionization detector equipped with a capillary 122-2332 DB-23 column (30 m ϫ 0.32 mm; 0.5 m coat-ing thickness; Agilent). Helium was used as carrier gas (1 ml min Ϫ1 ). Samples were injected at 220°C. The temperature gradient was 150°C for 1 min, 150 to 200°C at 15°C min Ϫ1 , 200 to 250°C at 2°C min Ϫ1 , and 250°C for 10 min. Data were processed using the HP ChemStation Rev. A09.03. FAMEs were identified by comparison with appropriate reference substances.
Lipid Analysis-For lipids analysis expression has been carried out with 50-ml cultures. Harvested cell pellets were homogenized in 5 ml of chloroform/methanol 1:2 (v/v), and lipids were extracted on a shaker for 4 h and then for 20 h with 5 ml of chloroform/methanol 2:1 (v/v) at 4°C. The resulting organic phases were combined and dried under N 2 . The remaining lipids were dissolved in 1 ml of chloroform. Separation of lipid classes (neutral lipids and phospholipids) was achieved using a silica column (Bond Elut SI, 100 mg/ml; Varian). Lipid extracts were loaded on the silica column pre-equilibrated with chloroform and then fractionated into the lipid classes by elution as follows; neutral lipids with chloroform and phospholipids with methanol/glacial acetic acid (9:1. v/v). Isolation of individual components of the phospholipid class was achieved by thin layer chromatography using appropriate standards and with methanol/chloroform/glacial acetic acid (25:65:8, v/v) as developing solvent.
Lipid Analysis of Arabidopsis Seeds-For lipid analysis, 10 mg of seeds were homogenized in 4 ml of chloroform/methanol/ glacial acetic acid (2:1:0.1 v/v/v) and incubated for 24 h at 4°C. Seed residues were pelleted (2 min, 3000 ϫ g). The supernatant was collected, and the pelleted seed residues were incubated with 2 ml of n-hexane for 30 min at room temperature. The resulting organic phases were combined and dried under N 2 . The dried lipids were dissolved in 200 l of chloroform. Separation of lipid classes (TAG and different phospholipids) was achieved by thin layer chromatography with methanol/chloroform/glacial acetic acid (25:65:8, v/v) as a developing solvent. Lipids (TAG, PC, phosphatidylinositol (PI)/phosphatidylserine (PS), and phosphatidylethanolamine (PE)) were identified according to authentic standards (Avanti), scraped out from the thin layer chromatography plates, and reextracted with developing solvent for subsequent fatty acid analysis.
Acyl-CoA Species, Synthesis, Extraction, and Analysis-Authentic standards for saturated and monounsaturated acyl-CoA esters with acyl-chain lengths from C12 to C18 were obtained from Sigma. Standards for polyunsaturated acyl-CoAs For acyl-CoA analysis of yeast cells, 20 ml of liquid cultures were harvested at an A 600 of 1.5-2.0, and acyl-CoA-species were extracted as described (19). The conversion of acyl-CoA esters to their etheno derivatives and acyl-CoA analysis was performed as described in Larson and Graham (29).

RESULTS AND DISCUSSION
The Marine Microalga, M. squamata, Accumulates High Levels of VLCPUFAs-The biosynthesis of VLCPUFAs in various plant species has previously been achieved by the introduction of genes from different organisms that encode enzymes responsible for fatty acid desaturation, elongation, and transesterification (7,12,14,15). In this study it was the aim to discover enzymes that exhibit greater specificity for 3 substrates and could be used in an exclusively acyl-CoA-dependent pathway for VLCPUFA biosynthesis, which abolishes a need for ratelimiting acyl shuttling.
To test whether M. squamata was suitable as a donor organism for the isolation of genes encoding acyl-CoA-dependent ⌬6 and ⌬5 desaturases, the fatty acid composition of the alga was analyzed. M. squamata was found to accumulate particularly 3-VLCPUFAs up to 20 mol% of the total fatty acids, which were predominantly associated with TAG but were also found in phospholipids and the acyl-CoA pool. These observations prompted us to pursue further attempts to isolate candidate genes encoding enzymes with the desired characteristics.
Isolation of Putative Acyl-CoA-dependent Front-end Desaturases from M. squamata and O. tauri-Using an EST data base established from M. squamata cDNA, two partial sequences, MsI and MsII, were annotated as putative desaturases. Based on the EST information, full-length clones were isolated by a RACE-PCR approach (supplemental Fig. 1). The amino acid sequence deduced from the MsI cDNA shared 66.5% amino acid identity with the sequence from an acyl-CoA-dependent ⌬6 desaturase from the related microalga, O. tauri (Ot⌬6, Gen-Bank TM accession number AY746357), 64% identity with an unknown partial sequence of a putative desaturase sequence from Ostreococcus lucimarinus CCE9901 (GenBank TM accession number XM_ 001421036), and 42% identity with the ⌬5 desaturase from Pavlova salina (GenBank TM accession number DQ995517). The deduced amino acid sequence of the second clone, MsII, shared 64% identity with an uncharacterized putative ⌬5 desaturase from O. lucimarinus CCE9901 (Gen-Bank TM accession number XM_001420818), 66% identity with a putative ⌬5 desaturase from O. tauri (GenBank TM accession number CR954212), 51% identity with the known ⌬5 desaturase from P. patens (GenBank TM accession number DS54492), and 32% identity with the previously described ⌬5 desaturase from Marchantia polymorpha (GenBank TM accession number AY583465).
In addition, the putative ⌬5 desaturase from O. tauri (Gen-Bank TM accession number CR954212) was cloned by a PCRbased approach, taking advantage of available genomic information, and named OtII. A comparison of the putative desaturase-like sequences from the microalgae revealed prop-erties characteristic for front-end desaturases, including the presence of an N-terminal HPGG motif (cytochrome b 5 binding domain) (30,31), three conserved histidine boxes most likely involved in the coordination of the diiron center of the active site (32), and the presence of a typical His to Gln substitution in the more variable third histidine box.
Biosynthetic Capacity of M. squamata Enzymes Heterologously Expressed in Yeast in Comparison to Known Lipid-dependent Desaturases-To confirm catalytic activity and substrate specificities of the putative desaturases encoded by MsI, MsII, and OtI, the sequences were individually expressed in the S. cerevisiae strain INVSc1 in the presence of potential fatty acid substrates for ⌬6or ⌬5-fatty acid desaturases (Table 1). With expression of empty vector controls, only yeast-endogenous fatty acids and the added substrates were detected. With expression of MsI, a new fatty acid product was observed only when 18:3 ⌬9,12,15 (␣-linolenic acid) was supplied ( Table 1). The product fatty acid was identified as 18:4 ⌬6,9,12,15 (stearidonic acid), indicating ⌬6-desaturation by MsI. Therefore, MsI is referred to as Ms⌬6 from this point on. Fatty acids other than ␣-linolenic acid, including yeast-endogenous fatty acids, were not accepted by Ms⌬6 (Table 1), indicating that only a narrow range of substrates are accepted by Ms⌬6. This interesting substrate preference differs from that of the previously identified Ot⌬6. Whereas Ot⌬6 accepts both 18:2 ⌬9,12 and 18:3 ⌬9,12,15 (␣-linolenic acid) as substrates, Ms⌬6 is more selective for the 3-substrate, ␣-linolenic acid ( Table 1). The pronounced preference of Ms⌬6 for 3 substrates is a desirable trait for an 3-VLCPUFA pathway in plants, because both 6 and 3 substrates may be competitively available for conversion. A preference for 3 substrates has previously been described for enzymes from Primula luteola (34) and Echium species (35); however, the reported ratio of 5.8 of converted 3 to 6 fatty acids for the Primula enzyme (34) was much lower than that of 114 observed here for Ms⌬6.
When MsII was expressed, new fatty acid products were detected in the presence of 20:3 ⌬8,11,14 (di-homo-␥-linolenic acid), 20:3 ⌬11,14,17 or 20:4 ⌬8,11,14,17 substrates. The products were identified as ARA and EPA, respectively. The conversion of 20:3 ⌬11,14,17 into the corresponding ⌬5 desaturation product was observed only in minor amounts. Therefore, MsII is referred to as Ms⌬5 from this point. Other fatty acids supplied to the cultures were not converted (Table 1). Thus, in contrast to Ms⌬6, Ms⌬5 exhibited a slightly broader range of accepted substrates, which is comparable with those of known lipid-dependent ⌬5 desaturases, which show roughly equal activities against 3 and 6 substrates (Table 1). When OtII was expressed in yeast, trace amounts of ⌬5-desaturated fatty acid products were detected in the presence of 20:3 ⌬8,11,14 or 20:4 ⌬8,11,14,17 , which were identified as ARA and EPA, respectively (data not shown), and OtII was consequently termed Ot⌬5. Because of the low activity of Ot⌬5 with expression in yeast, experiments described in the following paragraphs were performed with Ms⌬6 and Ms⌬5 only.
The desaturation efficiencies of the newly discovered enzymes were compared with those of the known acyl-CoA-dependent ⌬6 desaturase from O. tauri, Ot⌬6 (19), and the lipiddependent enzymes from P. tricornutum, Pt⌬6 and Pt⌬5, (33). To avoid the influence of different expression levels of trans-genes in yeast, at least six individual expression cultures were set up in parallel and individually analyzed, and results were averaged. This experimental design minimizes the possibility that differences in fatty acid product accumulation originate solely from differences in transgene expression. Fig. 1A shows the efficiency of Ms⌬6 conversion of 18:3 ⌬9,12,15 to 18:4 ⌬6, 9,12,15 in comparison to those of the known acyl-CoA-dependent ⌬6 desaturase from O. tauri and the lipid-dependent ⌬6 desaturase from P. tricornutum. As shown in Fig. 1A, Ms⌬6 was equally efficient as Ot⌬6 under identical conditions, and about 34% of the available 3-18:3 substrate was desaturated by both enzymes. In contrast, the lipid-dependent Pt⌬6 converted only 15% of the supplied ␣-linolenic acid to stearidonic acid, resulting in a substantially lower efficiency of conversion. An important characteristic of Ms⌬6 is the high desaturation efficiency, which exceeded that of the O. tauri acyl-CoA-dependent ⌬6 desaturase.
The low desaturation efficiency of Ms⌬5 upon expression in yeast (Fig. 1, B and C) could neither be increased by changes of expression conditions nor by optimization of codon usage of the algal enzyme (data not shown). A very similar problem arose for Ot⌬5, which exhibited only trace activity upon expression in yeast.

Uniform Distribution of Ms⌬6 and Ms⌬5 Desaturation Products in Different Yeast Lipid Classes Indicates Acyl-CoA
Dependence-Because the distribution of product fatty acids in different lipid classes can serve as an indication for desaturation in the acyl-CoA pool, yeast cultures expressing Ms⌬6 and Ms⌬5 were supplied with exogenous 18:3 ⌬9,12,15 or 20:3 ⌬8,11,14 , respectively, and the distribution of desaturation products in individual lipid classes was analyzed. Analogous experiments were performed in parallel with the acyl-CoAdependent Ot⌬6 and the lipid-dependent Pt⌬6 and Pt⌬5.
When ⌬5 desaturases were tested for the distribution of the desaturation product, 20:4 ⌬5, 8,11,14 , in various lipid classes, the product was detected in the neutral lipid fraction and in all phospholipids analyzed (Fig. 2B). The expression of Ms⌬5, 20:4 ⌬5, 8,11,14 , resulted in the accumulation of about equal proportions in PC, PI/PS, PE, CL, and in the neutral lipid fraction. In contrast, 20:4 ⌬5, 8,11,14 formed with expression of Pt⌬5 accumulated predominantly in PC and was detected only in low amounts in PI/PS, PE, or CL.
In contrast to the results for lipid-specific desaturases (here for comparison, Pt⌬5, Fig. 2B), where desaturation products are mostly enriched in PC, as shown by Domergue et al. (13), the data indicate that fatty acids produced by Ms⌬6 and Ms⌬5 do not associate predominantly with PC and can be detected to roughly equal amounts in all lipid species analyzed. These results suggest that, similar to the acyl-CoA-specific ⌬6 desaturase from O. tauri, Ms⌬6 and Ms⌬5 may use acyl-CoA esters as substrates rather than lipid-linked acyl groups.

Desaturation Products of Ms⌬6 and Ms⌬5 Are Detected More Rapidly in the CoA Pool Than Products of Lipid-dependent
Desaturases-For direct verification of acyl-CoA dependence of Ms⌬6 and Ms⌬5, acyl-CoA profiles of the respective yeast expression cultures were determined. Control cultures expressing Ot⌬6, Pt⌬6, or Pt⌬5 were tested in parallel. Data presented were obtained by the method described by Domergue et al. (19). Exogenous substrates (18:3 ⌬9,12,15 for ⌬6 desaturases or 20:3 ⌬8,11,14 for ⌬5 desaturases) were added to induced yeast cultures at an A 600 of 1.5-2.0. In cultures not supplied with exogenous fatty acid substrates, the patterns of total fatty acids and of fatty acids bound to CoA reflected only yeast-endogenous fatty acids (data not shown). Within 5 min of adding exogenous 18:3 ⌬9,12,15 substrate to cultures expressing Ms⌬6, Ot⌬6, or Pt⌬6, 18:3 ⌬9,12,15 was detected as a new peak in the total fatty acid pools (Fig. 3, left) and in the acyl-CoA pools as a shoulder of 16:1 ⌬9 -CoA in the acyl-CoA chromatogram (see supplemental Fig. 2). In addition, a new peak representing the 18:4 ⌬6,9,12,15 product of ⌬6 desaturation appeared simultaneously in the acyl-CoA pool of yeast expressing Ms⌬6 and Ot⌬6 but not in that of yeast expressing Pt⌬6 (Fig. 3, right). The 18:4 ⌬6,9,12,15 product was not detectable in the total fatty acid pools of any culture at this early time point, and 18:4 ⌬6, 9,12,15 appeared in the total fatty acid profiles of cultures expressing Ms⌬6, Ot⌬6, and Pt⌬6 only at time points exceeding 1 h after application of 18:3 ⌬9,12,15 . After 1 h, trace amounts of 18:4 ⌬6, 9,12,15 appeared also in the acyl-CoA pool of yeast-expressing Pt⌬6. The appearance of the ⌬6 desaturase product in the acyl-CoA pool before its appearance in the total fatty acid pool suggests that Ms⌬6 desaturates 18:3 ⌬9,12,15 to 18:4 ⌬6, 9,12,15 in an acyl-CoA-dependent manner, consistent with data on Ot⌬6 presented before (19). Reciprocally, the delayed appear-ance of 18:4 ⌬6,9,12,15 -CoA product in the acyl-CoA pool with expression of Pt⌬6 indicates that desaturation by Pt⌬6 occurred on fatty acids linked to phospholipids rather than on those associated with CoA.
Changes in total fatty acids and the acyl-CoA pool with expression of the ⌬5 desaturases Ms⌬5 and Pt⌬5 were analyzed before the addition of exogenous 20:3 ⌬8,11,14 substrate and afterward at defined times of 4, 8, and 24 h because these desaturases were not as efficient as the ⌬6 desaturases, and ⌬5-desaturation products were only detectable at those later time points. Within 1 h of adding exogenous 20:3 ⌬8,11,14 substrate to cultures expressing Ms⌬5 or Pt⌬5, 20:3 ⌬8,11,14 was detected as a new peak in the total fatty acids and in the acyl-CoA pools of all cultures. After 1 h, a new peak representing the 20:4 ⌬5,8,11,14 -CoA product of ⌬5 desaturation appeared in the acyl-CoA pool of yeast expressing Ms⌬5 or Pt⌬5. Simultaneously, the desaturation product 20:4 ⌬5, 8,11,14 appeared in the total fatty acid pool of yeast expressing Pt⌬5 but not in that of yeast expressing Ms⌬5. In Ms⌬5 cultures only at times exceeding 24 h after substrate addition, the 20:4 ⌬5, 8,11,14 product was detectable in the fatty acid pool in trace amounts (data not shown).

Desaturation Products of Ms⌬6 and Ms⌬5 Are More Efficiently Elongated Than Those of Lipid-dependent Desaturases-
To provide further evidence that Ms⌬6 and Ms⌬5 may be acyl-CoA-dependent enzymes, the desaturases were individually coexpressed with acyl-CoA elongases. The rationale of this experiment is that fatty acid elongation taking place in the acyl-CoA pool (13) should be more efficient with fatty acids desaturated within the acyl-CoA pool than with those desaturated by lipid-dependent enzymes, which require additional acyltransferase activities. The efficiencies of the combined desaturations/elongations were compared with those resulting from expression of the known desaturases Ot⌬6, Pt⌬6, or Pt⌬5, in the same context.
Side-by-side Assessment of Acyl-CoA-dependent Versus Lipiddependent EPA Biosynthetic Pathways Reconstituted in Yeast-In the next step we tested whether the coexpression of the CoA-dependent M. squamata desaturases would lead to more effective biosynthesis of EPA than with lipid-dependent desaturases. To test this hypothesis, we compared the capacity to accumulate EPA from yeast expressing Ms⌬6 together with PSE1 and Ms⌬5 with that of yeast coexpressing Pt⌬6 with the PSE1 and Pt⌬5. All cultures were supplemented with ␣-linolenic acid. Fatty acid products formed in all cultures were identified as 3-18:4 ⌬6, 9,12,15 , 20:3 ⌬11,14,17 , 3-20:4 ⌬8,11,14,17 , and 3-20:5 ⌬5, 8,11,14,17 (EPA, Fig. 4C). Most importantly, in both expression combinations, the biosynthesis of EPA was established, and EPA accumulated to 0.24 and 0.7% that of the total fatty acids, respectively. EPA accumulation through the CoAdependent pathway was, therefore, to 2-3 times higher than that with expression of the lipid-dependent enzymes (Fig. 4C) even though the Ms⌬5 desaturase was not as efficient as the Pt⌬5 desaturase (for comparison see Fig. 1, B and C). A possible explanation is that flux from the substrate, 18:3 ⌬9,12,15 to EPA was more effective in the CoA-dependent pathway than in the lipid-dependent pathway, where the ⌬6 desaturation product accumulated to high levels and was not available for further conversion.
Thus, whereas the use of lipid-dependent desaturation is limiting here in yeast, as described previously for linseed by Abbadi et al. (12), a more efficient interplay of the acyl-CoA-dependent enzymes may avoid the LPCAT bottleneck and may push the intermediates toward the end product, EPA. However, EPA accumulation in yeast was overall lower than expected. A possible explanation is that yeast cofactors may be different from those present in algae naturally harboring the enzymes used and that the functionality of the enzymes may, thus, be restricted in the heterologous system. The existence of 20:3 ⌬11,14,17 resulted from the direct elongation of the substrate 18:3 ⌬9,12,15 by PSE1 as mentioned above. This byproduct accumulated to higher amounts in the lipid-dependent expression cultures than observed with the CoA-dependent EPA biosyn- Yeast cultures coexpressing either Ms⌬6, Ot⌬6, or Pt⌬6 with the elongase PSE1 or Ms⌬5 or Pt⌬5 with the elongase Otelo5 were analyzed for their fatty acid composition. A, elongation efficiencies (product ϫ 100/(educt ϩ product)) of the ⌬6 elongase provided with its ⌬6-desaturated substrate (18:3 ⌬9,12,15 ) by the respective ⌬6 desaturase enzymes. B, elongation efficiencies of the ⌬5 elongase provided with its ⌬5-desaturated substrate (20:4 ⌬8,11,14,17 ) by the respective ⌬5 desaturase enzymes. As controls the corresponding empty vectors and the elongases were expressed separately. In the absence of the respective desaturation products, only elongation of the feeding substrate was detected (data not shown). C, establishment of an exclusively acyl-CoA-dependent pathway for EPA biosynthesis in yeast. EPA accumulation in yeast expressing Ms⌬6 together with PSE1 and Ms⌬5 were compared with yeast expressing Pt⌬6, PSE1, and Pt⌬5. As control, the corresponding empty vector combination was also expressed (data not shown). The yeast cultures were grown in the presence of 250 M 18:3 ⌬9, 12,15 . Data shown are representative for six independent experiments. thesis pathway. In addition, the fatty acid 20:4 ⌬5, 11,14,17 was detected in both expression combinations, which is in agreement with the substrate specificities determined before (see Table 1).
The establishment of an acyl-CoA-dependent pathway for EPA production in yeast (Fig. 4C) provided proof of concept data for the feasibility of the chosen approach. Because results obtained with the yeast model cannot be readily transferred to the situation in plants, the next set of experiments aimed at establishing acyl-CoA-dependent EPA production in Arabidopsis seeds.
Establishment of Acyl-CoA-dependent EPA Biosynthesis in A. thaliana-To test whether acyl-CoA-dependent EPA biosynthesis could be established in plants, Ms⌬6 and Ms⌬5 were coexpressed with PSE1 in plants, subsequently referred to as triple-Ms plants. For direct comparison of the efficiency of EPA production by different enzymes, a construct containing Ot⌬6 and Ot⌬5 combined with PSE1 was also tested, which will be referred to as triple-Ot. To provide a side-by-side control against the efficiency of lipid-dependent EPA biosynthesis, Pt⌬6 and Pt⌬5 were coexpressed with PSE1 in plants, which reflect the approach previously reported (2, 36) and which will be referred to as triple-Pt plants. The plant transformation constructs encoded each gene under the control of the seed-specific USP promoter (26). The fatty acid analysis of individual T2 seeds of triple-Ms, triple-Ot, and triple-Pt plants showed several new fatty acids that were identified as 3-18:4 ⌬6,9,12,15 , 3-20:3 ⌬11,14,17 , 3-20:4 ⌬8, 11,14,17 , and EPA (Fig. 5, B-D) and for triple-Ot and triple-Pt additionally as 6 -18:3 ⌬6,9,12 , 6 -20:3 ⌬8,11,14 , and ARA (Fig. 6, B-D), indicating that in triple-Ms plants the 3 pathway and in triple-Ot and triple-Pt plants both the 6 and the 3 pathways had been successfully established. The fatty acid content of the endogenous fatty acids serving as substrates for the VLCPUFA pathway, 18:3 ⌬9,12,15 and 18:2 ⌬9,12 , respectively, are more or less unaltered in the transgenic Arabidopsis seeds ( Fig. 5A and 6A, respectively). Among the new fatty acids produced in the triple-Ms and the triple-Ot seeds, 3-20:4 ⌬8,11,14,17 is most abundant. In triple-Pt seeds, the most abundant fatty acids were 6 -18:3 ⌬6,9,12 , 3-20:3 ⌬11, 14,17 , and at least the first desaturation product of the 3-pathway, 18:4 ⌬6, 9,12,15 . The accumulation of both ⌬6-desaturated fatty acids in the triple-Pt plants is consistent with results obtained with this construct in tobacco, rapeseed, or linseed (2,36) and confirms the presence of the bottleneck in lipid-dependent VLCPUFA biosynthesis. In contrast, in the triple-Ms and triple-Ot no or only minor accumulation of the ⌬6 desaturation product 18:4 ⌬6,9,12,15 was observed, and nearly all 18:4 ⌬6, 9,12,15 was elongated to the respective 3-20:4 ⌬8,11,14,17 product with very high efficiency (97% conversion) by the ⌬6 elongase in case of the triple-Ms plants. These results indicate that the acyl-CoA and PC pool bottleneck has been successful avoided by the use of strictly CoA-dependent desaturation (triple-Ms and triple-Ot). The low proportions of 3-20:5 ⌬5, 8,11,14,17 (EPA) in the triple-Ms or triple-Ot plants indicate that the ⌬5 desaturase activity of Ms⌬5 and, even more that of Ot⌬5, still limit the accumulation of EPA in these seeds. This result is consistent with the proof of concept data from yeast expression studies, where Ms⌬5 also showed only low desaturation efficiency, and Ot⌬5 was nearly inactive.
The expression of the lipid-dependent enzymes in Arabidopsis seeds showed that LPCAT activity with non-native ⌬6-polyunsaturated fatty acids was limiting in Arabidopsis, as previously observed in linseed (12). This observation led to the hypothesis that the selectivity of the LPCATs might be a ubiquitous problem in vascular plants. By avoiding the LPCAT activity in an exclusively acyl-CoA-dependent pathway for VLCPUFA production, no ⌬6 intermediates were accumulating. The accumulation of EPA was nonetheless rather low both in yeast and in Arabidopsis seeds. The low accumulation even with acyl-CoA-dependent enzymes is likely a consequence of inefficient ⌬5-desaturation after successful elimination of the LPCAT bottleneck (compare Fig. 1B with Figs. 5 and 6). Future studies will be directed toward the isolation of more efficient acyl-CoA-dependent ⌬5 desaturases, which currently have not been identified.
Although the changes in the total fatty acids of seeds of triple-Ms plants show that the accumulation of ⌬6 desaturation intermediates has been successfully avoided using the enzymes from M. squamata, another important concern for VLCPUFA production is the incorporation into TAG. To test whether the newly formed fatty acids accumulated in TAG or in membrane phospholipids, the distribution of the various fatty acid products of the newly established biosynthetic pathways was analyzed, and the percentages of each product associated with different lipids were calculated. An initial analysis of the lipid composition of different transgenic seeds indicated that all transgenic lines accumulated similar proportions of TAG ( Table 2). The comparison of plants expressing the triple-MS  and triple Ot-constructs indicates a preferential incorporation of CoA-bound 3 fatty acids over 6 fatty acids in TAG ( Table  2). The distribution of ARA and EPA in different lipid classes of Arabidopsis seeds showed that the VLCPUFA produced by acyl-CoA-dependent pathways accumulated overall to a higher degree in TAG than those produced by the lipid-linked pathway (Table 2). These observations invite some speculation on the routes of TAG biosynthesis in Arabidopsis seeds. Based on the fact that, in contrast to lipid-bound VLCPUFAs, CoAlinked VLCPUFAs were preferentially incorporated into TAG rather than into phospholipids, we suggest that TAG is rather produced by an acyl-CoA:diacylglycerol acyltransferase than by a phospholipid:diacylglycerol acyltransferase enzyme in Arabidopsis seeds. It is interesting to note that incorporation of 3 products into TAG was significantly higher than that of 6 products (compare in Table 2). A possible explanation for the higher incorporation of 3 products into TAG would be a preference of the acyl-CoA:diacylglycerol acyltransferase and/or phospholipid:diacylglycerol acyltransferase for the conversion of 3-substrates.
From plant transformation studies it is widely known that the insertion sites of transgenes in the plant genome play an important role for biotechnological efficiency. In particular, multigene constructs with more than one promoter require certain distances between different promoter-gene-terminator packages. The low efficiency of EPA accumulation in the transgenic Arabidopsis plants reported here may, thus, in part be a consequence of less than optimal transgene insertion and positional effects. Another possible explanation for low product accumulation are so-far unknown cofactors or other accessory proteins required by the algal enzymes for optimal functionality that are missing in the plants, as was already mentioned for yeast.
In summary, VLCPUFA can be produced in plants using an exclusively acyl-CoA-dependent pathway that avoids rate-limiting shuttling of fatty acid intermediates between lipids and coenzyme A. Although technical constraints, such as low catalytic activities of the ⌬5 desaturases or low expression levels of transgenes in plants still limit VLCPUFA accumulation in the experiments reported here, the data show that in principle there is no requirement for acyl-shuttling into lipids. VLCPUFA produced by an acyl-CoA-dependent fashion are in fact more efficiently incorporated into TAGs than those produced from the lipid-bound fatty acid substrates, suggesting that acyl-CoA-dependent VLCPUFA biosynthesis may offer advantages over other approaches to improve the composition of plant oils.