Domain Swapping Localizes the Structural Determinants of Regioselectivity in Membrane-bound Fatty Acid Desaturases of Caenorhabditis elegans*

Most fatty acid desaturases are members of a large superfamily of integral membrane, O2-dependent, ironcontaining enzymes that catalyze a variety of oxidative modifications to lipids. Sharing a similar primary structure and membrane topology, these enzymes are broadly categorized according to their positional specificity or regioselectivity, which designates the preferred position for substrate modification. To investigate the structural basis of regioselectivity in membrane-bound desaturases, the Caenorhabditis elegans ω-3 (FAT-1) and “Δ12” (FAT-2) desaturases were used as a model system. With the use of unnatural substrates, the regioselectivity of C. elegans FAT-2 was clearly defined as ν+3, i.e. it “measures” three carbons from an existing double bond. The structural basis for ν+3 and ω-3 regioselectivities was examined through construction and expression of chimeric DNA sequences based on FAT-1 and FAT-2. Each sequence was divided into seven domains, and chimeras were constructed in which specific domains were replaced with sequence from the other desaturase. When tested by expression in yeast using exogenously supplied substrates, chimeric sequences were found in which domain swapping resulted in a change of regioselectivity from ν+3to ω-3 and vice versa. In this way, the structural determinants of regioselectivity in FAT-1 and FAT-2 have been localized to two interdependent regions: a relatively hydrophobic region between the first two histidine boxes and the carboxyl-terminal region.

Fatty acid desaturases are part of multicomponent systems that catalyze the oxygen-and nicotinamide adenine dinucleotidedependent syn-dehydrogenation of unactivated aliphatic regions of their fatty ester (acyl-lipid) or thioester (acyl-acyl carrier protein or acyl-CoA) substrates (1)(2)(3)(4)(5)(6). In addition to the family of soluble fatty acid desaturases, of which the plant stearoyl-acyl carrier protein desaturase is well characterized, there is a large group of structurally distinct integral membrane desaturases. Membrane desaturases of widely varying substrate specificity and regioselectivity are scattered among a range of taxa. The yeast Saccharomyces cerevisiae has a single stearoyl-CoA ⌬9 desaturase, whereas many bacteria lack such desaturases entirely. At the other extreme, the biosynthesis of (4Z,7Z,10Z,13Z,16Z,19Z)-docosahexaenoic acid in some marine fungi likely requires the successive action of six structurally similar membrane desaturases, each varying in substrate specificity and regioselectivity (7,8).
Based on structural and catalytic similarities, membrane desaturases are included in a superfamily of oxidative enzymes along with alkane hydroxylase, xylene monooxygenase, carotene ketolase, and sterol methyloxidase (9,10). These are thought to contain a histidine-coordinated diiron center at the active site. While essentially no three-dimensional structural information is available for these difficult to purify enzymes, primary structure similarity, especially the conservation of three histidine-rich motifs, and similarity of hydrophobicity patterns support their structural similarity. On this basis, a model for the membrane topology of this superfamily has been proposed as depicted in Fig. 1 (9,11,12), although an alternative topology has been suggested (13).
In addition to their oxidative prowess, fatty acid desaturases are remarkable in their individual positional specificity (3). Separate desaturases have the ability to introduce a double bond at positions ranging from three carbons from the carboxyl to three carbons from the methyl terminus of fatty acyl substrates. This represents as much as 20 Å in the span of regioselectivity for these enzymes for flexible substrates with few distinguishing functional group landmarks. In fact, the substrate structural references that determine membrane desaturase regioselectivity are relatively complex (6). There appear to be three modes of regioselectivity. The ⌬x desaturases introduce a double bond between Cx and C(xϩ1) in the fatty acid moiety of the substrate. The -x desaturases reference the methyl end of the substrate introducing a double bond between the -x and -(x-1) positions. There is a third mode of regioselectivity called ϩx in which the double bond is introduced relative to an existing double bond. An example of this is the so-called extraplastidial "⌬12" 1 desaturase of plants. The natural substrate of the enzyme is oleoyl phosphatidylcholine, which it converts to linoleoylphosphatidylcholine. However, the enzyme will also desaturate 19:1(10) 2 to 19:2(10,13) suggesting a regioselectivity that yields a double bond that is three carbon atoms distal to an existing double bond within a limited range relative to the carboxyl terminus (6,14). The recently discovered ⌬5/⌬6 desaturase of zebrafish apparently represents a -3 desaturase (15) in that it references double bonds at the ⌬8 and ⌬9 positions.
Given the nutritional and commercial importance of unsaturated fatty acids of various types (16), it is essential that we develop an understanding of the structure-function relationships of integral membrane fatty acid desaturases. Integral membrane proteins are notoriously difficult to study in vitro. Given this, we have undertaken a study of the structural determinants of desaturase regioselectivity through the use of domain swapping experiments coupled with heterologous expression in yeast. The "⌬12" (FAT-2) and -3 (FAT-1) fatty acid desaturases from the nematode Caenorhabditis elegans were used as a model system. These enzymes are involved in the biosynthesis of polyunsaturated fatty acids in C. elegans that range from 18:2(9,12) to 20:5 (5,8,11,14,17) (17)(18)(19)(20). The two enzymes share 51% amino acid sequence identity. As such, they are two of the most similar desaturases, which differ distinctly in regioselectivity and especially regioselective mode (3), the manner in which the position of the incipient double bond is determined. In fact, the precise regioselective mode of the so-called "⌬12" desaturase of C. elegans (FAT-2) has not been determined previously. Here we report characterization of the regioselectivity of FAT-2 and the use of domain swapping experiments in an attempt to localize the structural determinants of regioselectivity in membrane-bound fatty acid desaturases.
Preparation of fat-1 and fat-2 Constructs-A construct containing the C. elegans fat-1 gene was obtained by PCR-amplifying the sequence from plasmid pDM015 (21) using oligonucleotide primers fat-1start and fat-1end (Table I). The resulting fragment was gel-purified and cloned into vector pYES2.1/V5-His-TOPO (Invitrogen) to yield plasmid pSAS001 suitable for the galactose-inducible expression of the native FAT-1 enzyme in S. cerevisiae.
The open reading frame of fat-2 was cloned into the vector pYES2.1/ V5-His-TOPO to yield plasmid pSAS050 as described previously (22). To enhance translation in yeast, a T4A mutation has been introduced into the sequence (22).
Construction of Chimeric Enzymes-The templates for chimera construction were the plasmids pSAS001 and pSAS050 containing fat-1 and fat-2 coding regions, respectively. DNA sequences from both "parental" genes were recombined (fused) together using the megaprimer PCR method (23). Sequences of the fusion and outlying primers used in the construction of chimeras are given in Table I. The fusion primers combine nucleotide sequence from both desaturase parents and were used to generate PCR fragments corresponding to specific regions of one of the desaturases. After purification from agarose gels (using GenElute Minus EtBr spin columns, Sigma), these fragments served as "megaprimers" in a subsequent PCR. The complementary sequence at the 5Ј-end of the megaprimers allowed annealing to the other desaturase gene, and extension of the overlap using Vent polymerase (New England Biolabs) yielded a recombinant molecule, which was amplified using the outlying primers (fat1start or fat2start and fat1end or fat2end). The temperature profile for a typical megaprimer fusion PCR was 94°C for 2 min one time; 94°C for 45 s, 55°C for 1 min, 72°C for 5 s ϩ 5 s/100 bp of product 30 times; 72°C for 3.5 min one time.
The temperature profile for a typical extension PCR was 94°C for 3 min one time; 94°C for 1 min, 55°C for 1 min, 72°C for 1 min, 30 s 30 times; 72°C for 5 min one time. PCR products were purified from a 1.2% agarose gel and TA cloned into the yeast expression vector pYES2.1/ V5-His-TOPO behind the galactose-inducible GAL1 promoter. The accuracy of chimeric constructs was confirmed by DNA sequencing.
Gel Electrophoresis and Western Blotting-Yeast cells were disrupted using a French press. Proteins from disrupted cells were solubilized in SDS sample buffer (containing 125 mM Tris-HCl, pH 6.8, 4% SDS, 20% glycerol, 0.2 M dithiothreitol, 0.001% bromphenol blue), incubated at 90°C for 5 min, and analyzed by SDS-PAGE on 4 -15% polyacrylamide gradient gels. Protein bands were visualized by staining with GelCode Blue stain reagent (Pierce) or electrophoretically transferred to polyvinylidene difluoride membranes (Amersham Biosciences) for Western blot analysis. Blots were blocked with 1% skim milk and 0.1% Tween 20 in phosphate-buffered saline and then incubated with primary antibody. For this, antiserum from a rabbit immunized with a synthetic oligopeptide (NH 2 -FVRGQTQTIDR-COOH, corresponding to amino acid sequence shared by FAT-1 (amino acids 297-307) and FAT-2 (amino acids 268 -278)) conjugated to keyhole limpet hemocyanin was used. After the primary antibody incubation, the membrane was washed with phosphate-buffered saline containing 0.1% Tween 20 and then incubated with peroxidase-conjugated goat anti-rabbit IgG (Am-  (fat-1) and "⌬12" (fat-2) fatty acid desaturases All primers are listed in 5Ј to 3Ј directions. Sequence in bold indicates complementarity to the sequence of the desaturase specified in the primer name. Underlined sequence is complementary to the other desaturase. Italicized sequence in the outlying primers indicates the position of XbaI (in fat1start and fat2start) and SacI (in fat2end and fat2end) restriction sites.
ersham Biosciences) and visualized by enhanced chemiluminescence (ECL, Amersham Biosciences) using x-ray film. Fatty Acid Analysis-GC-flame ionization detection analysis of fatty acid methyl esters was performed as described previously (22). For GC/MS analysis, fatty acid methyl esters of interest were partially purified from the total fatty acid methyl ester fraction by HPLC using an Agilent 1100 Series HPLC system with the fraction collector collecting 0.5-ml fractions from 2 ϫ 12.5-cm Whatman Partisphere C 18 columns connected in series and using a linear solvent gradient starting at 90% acetonitrile, 10% water with increasing acetone to 30% in 20 ml.
The fractions containing the fatty acids of interest were saponified (21) and derivatized to diethylamides according to Nilsson and Liljenberg (25). The diethylamide reaction products were further purified by applying the reaction mixture to a short silica gel column (ϳ 0.5 ml), washing with 10 column volumes of dichloromethane, and eluting the diethylamides with 10 column volumes of acetone. The acetone was removed under a nitrogen stream, and the samples were dissolved in 0.5 ml of dichloromethane for GC/MS analysis.
GC/MS analysis of the diethylamides was accomplished using an Agilent 5973 mass selective detector coupled to an Agilent 6890N gas chromatograph using G1701DA MSD Chemstation software (for instrument control and data analysis) and equipped with a 30-m ϫ 0.25-mm DB-23 column with 0.25-m film thickness (J&W Scientific). The chromatograph conditions included a split injection (20:1) onto the column using a helium flow of 0.4 ml/min, an initial temperature of 160°C for 1 min, and a subsequent temperature ramp of 4°C/min to 240°C. The mass selective detector was run under standard electron impact conditions (70 eV), scanning an effective m/z range of 10 -450 at 3.32 scans/s.

RESULTS
Regioselectivity of the C. elegans "⌬12" Desaturase-To further characterize the regioselective mode of FAT-2, INVSc1 yeast expressing pSAS050 were grown in culture with commercially available ⌬10 monoenoic fatty acids. The use of 17:1(10) and 19:1(10) as substrates and analysis of the corresponding FAT-2 products allows the discrimination between the ⌬12, -6, and ϩ3 types of regioselectivity. The fatty acid composition of the final culture was assessed by GC analysis and compared with yeast cultures expressing an empty plasmid grown in parallel. HPLC fractionation and GC analysis of fatty acids from the pSAS050 (FAT-2)-expressing strain with exogenously added 17:1(10) or 19:1(10) revealed peaks not present in the empty vector control strain. In the case of cultures supplied with 17:1(10), the major new GC peak coeluted with 18:0 and was revealed upon fractionation by HPLC. GC/MS analysis of the HPLC-purified peaks identified them as the ϩ3 desaturation products 17:2(10,13) and 19:2(10, 13). the 19:1(10) desaturation product is also consistent with double bonds at the ⌬10 and ⌬13 positions (data not shown). The above double bond assignments are supported by MS peak intensity patterns that are consistent with 18:2(9,12) and 20:2 (11,14) spectra obtained under the same conditions (data not shown). Also the product of 19:1(10) desaturation was found to be insensitive to the dienophile 4-methyl-1,2,4-triazoline-3,5-dione under conditions for which conjugated fatty acids react (data not shown). The latter observation confirms that desaturation by FAT-2 did not occur at the ⌬12 position for 19:1 (10).
Chimera Construction and Heterologous Expression-All of the integral membrane fatty acid desaturases are thought to share the same membrane topology and overall three-dimensional structure (but see Ref. 13). This is particularly true of C. elegans FAT-1 and FAT-2 since they share 51% amino acid sequence identity. Furthermore, in replacing amino acid sequence domains of one of these enzymes with the corresponding domains from the other enzyme, we would expect, at least in some cases, to maintain this general topology and structure as well as (possibly altered) desaturase activity. With this in mind, we set out to investigate the structural elements that determine regioselectivity in the two enzymes through domain swapping experiments.
Transition points for chimera construction were chosen roughly on a structural basis as follows. Using the computer program TopPred II (26) to predict the location of the transmembrane helices in FAT-1 and FAT-2, a membrane topological model was constructed based on earlier proposals (9, 11, 12) (Fig.  1). For chimeric sequence construction, four divisions were made in the locations where the sequence was predicted to cross the membrane interface on the cytosolic side. Other divisions were placed after the second conserved histidine cluster, and another was placed after the third histidine cluster. These transition points divide the desaturases into seven broad structural domains: A, the amino-terminal region (amino acids 1-78, FAT-1 numbering); B, the first set of putative transmembrane helices (amino acids 79 -122); C, the conserved His boxes H1 and H2 and an intervening relatively hydrophobic segment (amino acids 123-174); D, a second relatively hydrophobic segment (amino acids 175-232); E, the second set of putative transmembrane helices (amino acids 233-280); F, a relatively hydrophilic domain containing His box H3 (amino acids 281-328), and G, the carboxylterminal region (amino acids 329 -403). Fig. 3 schematically depicts chimeras made from introducing single (Fig. 3A) and multiple (Fig. 3B) domains from the opposing desaturase. Chimeras are named according to the identity of the parental sequence and the letter designator of the domain introduced from the other parent. For example, -3:D denotes a chimera consisting of FAT-1 sequence with domain D replaced by domain D from FAT-2.
Expression levels of the various wild type and chimeric fatty acid desaturases were determined by immunoblotting as indicated in Fig. 4. For the most part, within a given group of chimeras, desaturase protein levels in yeast cells were similar or somewhat reduced compared with the corresponding wild type enzymes. Thus for single domain chimeras with an -3 background, it is estimated from Fig. 4A that expression levels were 30 -100% of the wild type FAT-1, and -3:G is no exception. Chimeras with a "⌬12" background showed a slightly wider range of expression with ⌬12:B, ⌬12:C, and ⌬12:D levels being reduced relative to wild type FAT-2 and the other chimeras. Again the chimera with the G domain swap was expressed at levels comparable to the corresponding wild type enzyme. The double domain swap enzymes were expressed at levels that were comparable or slightly reduced compared with the wild type levels (Fig 4, C and D). Given the aforementioned expression patterns, it was possible to compare semiquantitatively and qualitatively the activities of the various enzymes in yeast cells.
Functional Characterization of Single Domain Swap Chimeras-Yeast is an ideal host for heterologous expression in the study of membrane-bound desaturases and related lipid-modifying enzymes (27). Because the polyunsaturated products of the -3 and "⌬12" desaturases are not metabolized further, the final levels of product in the total lipid extract may be used as a semiquantitative indicator of enzyme function (21,28). For functional characterization, the wild type and chimeric enzymes were expressed in the S. cerevisiae strain INVSc1 supplied with various fatty acid substrates, and the recombinant yeast cells were analyzed for fatty acid content. As a negative control, yeast was also transformed with the empty plasmid pYES2.1. Plasmids were constructed so as to give rise to native amino acid sequences with the exception of a T4A mutation in FAT-2 that allows better conformity with Kozak consensus sequences (22,29).
The fatty acid desaturase activities of the single domain swap chimeras are shown in Fig. 5. All chimeras containing a single substituted domain were catalytically active and retained the regioselectivity of their parent desaturase. None of the -3 chimeras showed ϩ3 activity with 16:1(9) or 18:1(9) (data not shown). For wild type FAT-1, as well as that of related -3 chimeras, the -3 desaturase activity on the substrates 18:2(9,12) and 18:3(6,9,12) (yielding 18:3(9,12,15) and 18: 4(6,9,12,15), respectively) showed the same general trends (Fig. 5A). However, there are some notable differences in relative activities for the two substrates. Yeast cells expressing wild type FAT-1 accumulated more product from 18:2(9,12) than from 18:3 (6,9,12). Yeast expressing most chimeras gave roughly equal amounts from each substrate with the notable exception of -3:D, which appears to have a preference for 18:3 (6,9,12). In comparing overall activity, -3:E and -3:F showed product accumulations that were equal to or higher than the wild type -3 enzyme. The other -3 chimeras with single domains derived from FAT-2 sequence showed reduced accumulations relative to wild type enzyme. The -3:G chimera was most severely affected, giving accumulations of less than 20% of wild type levels with both substrates.
Functional Characterization of Double Domain Swap Chimeras-The results of the single domain replacement chimeras suggest that the structure conferred by any single domain cannot be solely responsible for the determination of regioselectivity. However, these results give a clue to what could be involved. It is clear, of course, that the G domains from each desaturase support enzyme activity at least in the wild type sequences. Consequently we reasoned that the explanation for low activity in both -3:G and ⌬12:G chimeras could be a disruption of essential structural or functional interactions present in the wild type enzyme between domain G and other domains. If such G-specific interactions could be reconstituted, wild type activity levels might be restored along with possible effects on regioselectivity.
On this basis, a set of chimeras was constructed that contained both domain G and one of the other six domains from the opposing desaturase (Fig. 3B). The majority of these chimeras were inactive on a variety of substrates including 18:1(9) and 18:2 (9,12), and fatty acid profiles of total lipid extracts of yeast expressing these chimeras were indistinguishable from the plasmid-only control (data not shown). However, two chimeric sequences were discovered in which domain swapping resulted in a change of regioselectivity from ϩ3 to -3 and vice versa. When domains D and G were simultaneously replaced with sequence from the other desaturase, the resulting chimera displayed regioselectivity consistent with the parent that contributed those domains. Product accumulation resulting from the expression of chimera ⌬12:DϩG is compared with wild type FAT-1 in Fig. 6A. Yeast expressing the ⌬12:DϩG chimera was found to convert the fatty acid substrates 18:2(9,12) and 18: 3 (6,9,12) into -3 desaturation products at levels roughly comparable to that of yeast expressing wild type FAT-1. Similarly expression of the -3:DϩG chimera in yeast resulted in the accumulation of 18:2(9,12) from endogenous 18:1(9) (see Fig.  6B). Thus, while the structure of the D and G domains alone does not determine positional specificity, the combination of the two domains from one type of desaturase is able to determine the regioselectivity of the chimeric enzymes.

DISCUSSION
There is very little known about structure-function relationships for the superfamily of integral membrane proteins, which includes most fatty acid desaturases. Sequence comparisons, site-directed mutagenesis, and spectroscopic results are consistent with histidine-rich coordination of a diiron center at the active site of these enzymes (10,30). Eight highly conserved histidine residues have been shown to be required for function with the exception of one that is replaced by glutamine in some "front-end" desaturases, which introduce double bonds near C-1 in the substrate (3,31). Shanklin and colleagues (32) have identified a number of individual amino acids that are important for the chemoselectivity of the FAD2-like enzymes, which tend to have both desaturase and hydroxylase activities to varying degrees. A domain swapping study was undertaken by Napier and colleagues (33) to investigate the regioselectivity and substrate specificity of two front-end desaturases. This work hinted at the importance of the carboxyl terminus of desaturases in determining specific regiochemical and/or substrate characteristics of these enzymes. Additionally the inactivity of a carboxyl-terminal deletion mutant of the Bacillus subtilis ⌬5 desaturase supports the importance of domain G in desaturase function (13).
In this work, two regioselectively distinguishable desaturases were chosen that have a high degree of sequence similarity. While the substrate classes of FAT-1 and FAT-2 are not known precisely, it has been suggested that they are both acyl-CoA desaturases (18). Their natural substrates are similar in length and have double bonds at C(x-3) where x is the position of the incipient double bond, i.e. the "⌬12" desaturase acts on a ⌬9 substrate, and the -3 desaturase acts on -6 substrates. However, the enzymes differ in both regioselectiv- ity and regioselective mode; FAT-2 "measures" from a double bond (as indicated by its activity on 17:1(10) and 19:1(10)), and FAT-1 measures from the methyl end of the substrate. The ϩ3 regioselectivity of FAT-2 determined by this study is very similar to that found for the plant oleate desaturase (FAD2) (14).
Our initial approach to investigating the structural determinants of desaturase regioselectivity was to construct single domain chimeric enzymes in which one domain in a wild type enzyme was replaced with the domain sequence of the other enzyme. It is useful to consider the possible activity of a given example of such a chimeric enzyme. It may be inactive. It may be active and retain the regioselectivity of the "parent" desaturase, suggesting that the domain is functionally interchangeable but not solely responsible for regioselectivity. The regioselectivity could correspond to that of the introduced domain, suggesting that it is the major determinant of positional specificity. Other possibilities are that the chimeric enzyme could have both regioselectivities or a novel regioselectivity such as -4, for example.
In the case presented for the C. elegans FAT-1 and FAT-2, single domain chimeras of both enzymes show activity with retention of parental positional specificity. The conclusion to be drawn from this is that no single domain is either necessary or sufficient to confer regioselectivity on the enzyme. This leaves the possibility that two or more domains act in concert to determine regioselectivity. In the interest of avoiding a complete search of two-domain chimera "space," we attempted to rationalize the single domain chimera results that indicated that domain G was somehow important. Replacing both G and a second domain might result in restoration of native interactions between the two domains. Most of the two-domain chimeras were inactive. Clearly there are limits to the interchangeability of most domains. The chimeras in which D and G were simultaneously swapped had regioselectivities corresponding to the identity of the D and G domains, i.e. if D and G domains corresponded to FAT-1, the chimera had -3 activity. Therefore, these two domains combined appear to be responsible in large part to the determination of regioselectivity in the -3 and "⌬12" desaturases of C. elegans. The importance of domain G in C. elegans desaturase activity is consistent with previous studies indicating that removal or replacement of sequence at the carboxyl terminus of desaturases can abolish activity (13). The results from our work on C. elegans desaturases extend this observation to show that the carboxyl terminus is actually an important determinant of regioselectivity.
Domain D is important for both regioselectivity and substrate specificity. Alteration of this relatively hydrophobic region near the second histidine box in the FAT-1 background changes the relative activity on linoleate versus ␥-linolenate (Fig. 5A). This and its importance in regioselectivity suggest that domain D may be intimately involved in binding and positioning of the substrate relative to the active site.
It is notable that none of the putative transmembrane domains are implicated in the determination of regioselectivity and that both D and G domains are thought to be on the same side of the endoplasmic reticulum. While neither domain contains the putative active site histidine boxes, domain D is immediately adjacent to the second His box, and domain G is immediately adjacent to the third His box.
Inspection of alignment of domains D and G for FAT-1 and FAT-2 reveals a total of 63 amino acid differences (Fig. 7). Further work is in progress to define more specifically the amino acid residues that are important for regioselectivity.