The Multifunctional Protein in Peroxisomal β-Oxidation

Plant fatty acids can be completely degraded within the peroxisomes. Fatty acid degradation plays a role in several plant processes including plant hormone synthesis and seed germination. Two multifunctional peroxisomal isozymes, MFP2 and AIM1, both with 2-trans-enoyl-CoA hydratase and l-3-hydroxyacyl-CoA dehydrogenase activities, function in mouse ear cress (Arabidopsis thaliana) peroxisomal β-oxidation, where fatty acids are degraded by the sequential removal of two carbon units. A deficiency in either of the two isozymes gives rise to a different phenotype; the biochemical and molecular background for these differences is not known. Structure determination of Arabidopsis MFP2 revealed that plant peroxisomal MFPs can be grouped into two families, as defined by a specific pattern of amino acid residues in the flexible loop of the acyl-binding pocket of the 2-trans-enoyl-CoA hydratase domain. This could explain the differences in substrate preferences and specific biological functions of the two isozymes. The in vitro substrate preference profiles illustrate that the Arabidopsis AIM1 hydratase has a preference for short chain acyl-CoAs compared with the Arabidopsis MFP2 hydratase. Remarkably, neither of the two was able to catabolize enoyl-CoA substrates longer than 14 carbon atoms efficiently, suggesting the existence of an uncharacterized long chain enoyl-CoA hydratase in Arabidopsis peroxisomes.

Plant fatty acids can be completely degraded within the peroxisomes. Fatty acid degradation plays a role in several plant processes including plant hormone synthesis and seed germination. Two multifunctional peroxisomal isozymes, MFP2 and AIM1, both with 2-trans-enoyl-CoA hydratase and L-3-hydroxyacyl-CoA dehydrogenase activities, function in mouse ear cress (Arabidopsis thaliana) peroxisomal ␤-oxidation, where fatty acids are degraded by the sequential removal of two carbon units. A deficiency in either of the two isozymes gives rise to a different phenotype; the biochemical and molecular background for these differences is not known. Structure determination of Arabidopsis MFP2 revealed that plant peroxisomal MFPs can be grouped into two families, as defined by a specific pattern of amino acid residues in the flexible loop of the acylbinding pocket of the 2-trans-enoyl-CoA hydratase domain. This could explain the differences in substrate preferences and specific biological functions of the two isozymes. The in vitro substrate preference profiles illustrate that the Arabidopsis AIM1 hydratase has a preference for short chain acyl-CoAs compared with the Arabidopsis MFP2 hydratase. Remarkably, neither of the two was able to catabolize enoyl-CoA substrates longer than 14 carbon atoms efficiently, suggesting the existence of an uncharacterized long chain enoyl-CoA hydratase in Arabidopsis peroxisomes.
Fatty acids are degraded by the sequential removal of two carbon units in a process known as ␤-oxidation (Fig. 1). This process is ubiquitous and takes its name from the oxidation taking place at the carbon atom ␤ to the carboxyl group. The discovery of a peroxisomal ␤-oxidation system was made in plants (1), and whereas peroxisomal ␤-oxidation in animals appears to be merely a fatty acid chain-shortening machine feeding mitochondrial ␤-oxidation, plant and fungal ␤-oxida-tion takes place almost entirely in the peroxisomes. The products of the reaction are H 2 O 2 , acetyl-CoA, reducing equivalents, and a variety of chain length-shortened acyl-containing molecules like the plant hormones jasmonic acid (2) and indole-3-acetic acid (3,4).
The conversion of storage triacylglycerols by ␤-oxidation provides metabolic energy and carbon skeletons for germination and early post-germinative seedling growth in oil seed plants (5) and metabolic energy and carbon for the production of hydrolytic enzymes in cereals (6). It is a salvage pathway for fatty acids during foliar senescence (7), supplies respiratory substrates to carbohydrate-deprived tissue (8), and at a lower level, is a constitutive property of all plant tissues most likely involved with membrane lipid turnover (7).
AtMFP2 is induced during germination and expressed primarily during post-germinative growth (10), whereas AtAIM1 is expressed predominantly in siliques, flowers, and seedlings older than 8 days (3). Mutations in the AtMFP2 gene give rise to a sucrose-dependent seedling phenotype (11), whereas AtAIM1 disruption causes abnormal inflorescence development resulting in low fertility (3). The different phenotypes reveal the diverse physiological roles of the two genes. In vivo and in vitro characterization of the chain length specificity of AtMFP2 suggests a long chain hydratase coupled to a short, medium, and long chain-specific dehydrogenase (11). AtAIM1 seems to have higher affinity for short chain acyl-CoA, although aim1 plants have elevated levels of C18:1 and C18:2 unsaturated fatty acids (3). The aim1 plants cannot convert 4-(2,4-dichlorophenoxy)butyric acid efficiently to the toxic indole-3-acetic acid analogue, 2,4-dichlorophenoxyacetic acid (3), and AIM1 activity is essential for woundinduced jasmonate production (2).
The molecular organization of the ␤-oxidation system differs between organisms and organelles. In the yeast lipolytica, the five ACX isoforms present form a heteropentameric complex and are imported into the peroxisomes as such (12). Bacterial and mammalian mitochondrial multifunctional complexes include ECH, HACD, and KAT activities (13,14). Monofunctional enoyl-CoA isomerases/hydratases, HACD, and KAT, with a preference for shorter chain length substrates, are present in mammalian mitochondria in addition to the MFPs. In mammalian and plant peroxisomes, ACX and KAT activities reside on soluble monofunctional isozymes with different substrate specificities, whereas the ECH and HACD activities are catalyzed by MFPs (15,16). In plants, MFPs are targeted for the peroxisomes only; no MFPs have been observed in mitochondria or as having mitochondrial targeting signals.
Channeling has been shown to occur between the active sites in bacterial and mammalian MFPs (17)(18)(19), but no tight binding protein-protein complexes of ␤-oxidation enzymes have been isolated from peroxisomes to date. By channeling, the intermediates need not equilibrate with the bulk solvent but are immediately directed from one active site to the next. In plants, an indication of specific isozyme functions by reduced jasmonate formation in the AtACX1/AtACX5 double mutant (20) and other evidence for specific involvement of AtACX1 and AtKAT2 in jasmonate synthesis (21)(22)(23)(24) suggest that weak or transient protein-protein interactions and channeling could be mechanisms by which plants control the flow of metabolites through ␤-oxidation.
The structure of the full-length AtMFP2 reported here is the first structure to be determined for a full-length peroxisomal MFP. AtMFP2 is a 79-kDa protein composed of three domains and a linker helix. It forms a stable monomeric structure and resembles the mammalian peroxisomal MFE-1 in domain organization and size (25). The substrate specificity profiles published here elucidate significant differences between the hydratase domains. In AtAIM1 the hydratase substrate specificity peaks at crotonoyl-CoA extending up to 2-trans-tetradecenoyl-CoA, whereas AtMFP2 hosts a short to medium chain-length hydratase peaking around 2-trans-octenoyl-CoA and 2-trans-dodecenoyl-CoA. Surprisingly, the substrate specificity profiles of AtMFP2 and AtAIM1 reveal that neither of the two is capable of efficiently catabolizing straight chain substrates longer than 2-trans-tetradecenoyl-CoA. The structure of AtMFP2 is compared with mono-and multifunctional enzymes, providing insight into the catalytic mechanism and substrate-binding sites as well as the molecular basis for isozyme specificity.

EXPERIMENTAL PROCEDURES
Expression and Protein Purification-The AtMFP2 was recloned from the Riken database clone pda05150 (26) into a pET24b expression vector (NdeI and NotI restriction sites, Novagen) after PCR amplification using forward primer 5Ј-GCGGCGATTAATGGATTCACGAACCAAGGGGAAG-ACG-3Ј, reverse primer 5Ј-ATAGTTTAGCGGCCGCTT-AGTGATGGTGATGGTGATGCAACCGTGAGCTGGC-3Ј (AseI and NotI restriction sites are underlined; coding regions are in bold), and blunt-end ligation in pSTblue-1 (Novagen). Escherichia coli BL21(DE3) cells were transformed with the resultant vector, which codes for the full-length protein and a C-terminal His 6 tag. The integrity of the recloned AtMFP2 was confirmed by DNA sequencing (Agowa). Gene expression was achieved by incubating the cells in shaking flask in 2ϫYT medium, 50 mg/ml kanamycin at 35°C until a cell density (A 600 ) of 0.6 was reached. The induction of gene expression was achieved by adding isopropyl 1-thio-␤-D-galactopyranoside to a final concentration of 0.4 mM and persisted overnight. Cells (ϳ10 g of cell paste/liter of medium) were harvested by centrifugation. The pET24b-AtMFP2 plasmid was additionally transformed into the methionine auxotroph E. coli B834(DE3) strain (Novagen) for production of selenomethionine-substituted protein for phase determination. In this case the starter cultures were grown in Luria-Bertani-Broth at 35°C over night. Cells from the starter cultures were harvested by centrifugation and resuspended in Milli-Q water prior to transfer to prewarmed SelenoMet Medium Base with SeleonoMet Nutrient Mix and 40 mg/liter selonomethionine added (Athena Enzyme Systems, Baltimore) plus 100 mg/liter kanamycin. Gene expression was induced by adding isopropyl 1-thio-␤-D-galactopyranoside to a final concentration of 0.4 mM after the cell density had reached an A 600 of 0.5. The cultures were allowed to produce protein overnight at 35°C after which they were harvested by centrifugation. This procedure gave a yield of ϳ5 g of cell paste/liter of medium. Native and selenomethionine-substituted protein batches were purified following the same purification scheme. Approximately 10 g of wet cell paste was resuspended in 30 ml of buffer A 1 (20 mM imidazole, 500 mM NaCl, 30 mM phosphate buffer, pH 7.4), and cell membranes and DNA were disrupted by the addition of 180 mg of lysozyme (Sigma-Aldrich), 3 ml of 10ϫ BugBuster (Novagen), and 1500 units of Benzonase (Sigma-Aldrich). The cell extract was cleared by centrifugation for 40 min at 18,000 ϫ g and 4°C, filtered through a 0.45-m syringe filter (Millipore), and applied to a 5-ml HisTrap column (GE Healthcare) preequilibrated with buffer A 1 . The column was washed with a step gradient going from 20 mM to 35 mM and finally to 70 mM imidazole in 500 mM NaCl, 30 mM phosphate buffer, pH 7.4. Each gradient step lasted for 5 column volumes with a flow rate of 1 ml/min. Finally, the bound AtMFP2 was eluted with 500 mM imidazole, 500 mM NaCl, and 30 mM phosphate buffer, pH 7.4. The AtMFP2-containing fractions were pooled and concentrated in a Vivaspin20 30,000 MWCO filter device (Sartorius). The concentrated sample (2 ml) was filtered through a 0.22-m syringe filter (Millipore) and applied to a Superdex 200 prep grade (26/60) gel filtration column (GE Healthcare) equilibrated and run with 20 mM Hepes, 150 mM NaCl, pH 7.5. The sample was collected at an elution volume of ϳ235 ml. The fractionated AtMFP2 was pooled and concentrated to 10 mg/ml in a Vivaspin20 filter device before dialysis against 0.5 liters of 20 mM EDTA, 20 mM Hepes, pH 7.5, in a 10,000 MWCO dialysis tube (SpectraPor) followed by exhaustive dialysis against 20 mM Hepes, pH 7.5. The sample was mixed with 100% glycerol in a 1:1 volume ratio before storage at Ϫ20°C.
Arabidopsis peroxisomal acyl-CoA oxidases, AtACX, were produced and purified for substrate preparation. AtACX1 and AtACX4 were prepared as described by Pedersen and Henriksen (27). N-terminally His 6 -tagged AtACX3 was expressed from a pET24b vector construct prepared from the pda08625 clone (Riken) in RosettaGami2 (DE3) cells (Novagen). The cells were cultured, and gene expression was induced as described for AtMFP2, except induction was done for 18 h at 14°C. The cells were harvested by centrifugation and resuspended in buffer A 2 (300 mM NaCl, 10 M FAD, 30 mM phosphate buffer, pH 7.8), to which was added 2% buffer B 2 (buffer A 2 ϩ 500 mM imidazole), 5% (w/w) lysozyme (Sigma-Aldrich), 100 units/g Benzonase (Sigma-Aldrich), and BugBuster (Novagen) (30 ml of buffer A 2 /10 g of wet cell paste). The cells were lyzed at room temperature for 20 min on a shaking platform. Cell debris and the insoluble fraction were removed by centrifugation, and the supernatant was fractionated on a 5-ml HisTrap column (GE Healthcare) equilibrated with 2% buffer B 2 at 1 ml/min. The column was washed with 15% buffer B 2 prior to elution of the bound AtACX3 with 50% buffer B 2 . The 280 peak fractions were assayed by SDS-PAGE analysis; the AtACX3-containing fractions were pooled, and the buffer was exchanged for the buffer used for size exclusion chromatography (30 mM MOPS, 300 mM NaCl, 10 M FAD, pH 7.0) in a Vivaspin20 filter device. The concentrated sample was fractionated on a Superdex 75 16/60 prep grade column (GE Healthcare) run at a flow rate of 0.2 ml/min. The peak fractions were pooled and glycerol added to a final concentration of 5% (v/v) to stabilize the protein and sampled by SDS-PAGE. Pure AtACX3 fractions were pooled and concentrated to 2 mg/ml in a Vivaspin20 filter device.
AtAIM1 was expressed in RosettaGami2 cells harboring a pET46 Ek/LIC (Novagen) construct encoding full-length N-terminally His 6 -tagged AtAIM1 prepared from clone pda02497 (Riken). The cells were treated as described for AtACX3, except that the induction temperature was 18°C. The supernatant was fractionated like AtACX3, except buffer A 2 was supplemented with 500 mM NaCl, pH 7.4, and it contained no FAD. The column was washed with 10% and eluted with 25% buffer B 2 . The buffer was exchanged for MonoS buffer (30 mM MES, 200 mM NaCl, pH 6.0) in a Vivaspin20 filter device, and the sample was applied to a MonoS 10/100 GL column (GE Healthcare) running at 5 ml/min. After the column was washed with MonoS buffer, fractionation was achieved using a 0 -50% isocratic gradient over 30 column volumes. The buffer for the gradient was 30 mM MES, 1.8 M NaCl, pH 6.0. AtAIM1 eluted at 350 mM NaCl; it was concentrated to 10 mg/ml over the Vivaspin20 filter device and stored at Ϫ20°C in MonoS buffer supplemented with 5% (v/v) glycerol.
Crystallization and Data Collection-AtMFP2 was dialyzed thoroughly against 20 mM Hepes, pH 7.5 (10,000 MWCO), to remove any glycerol prior to crystallization. NAD ϩ and acetoacetyl-CoA were added to a final concentration of 1 mM. Large, well diffracting crystals were obtained when the protein sample was mixed with mother liquor (4.2 M NaCHO 2 , 2% (v/v) glycerol) and 40% (v/v) polypropylene glycol in 2:2:0.4 l microseeded drops at 20°C. Crystals grew to a maximal size of 0.3 ϫ 0.02 ϫ 0.02 mm 3 within 3 days after which they were transferred to a cryoprotective solution consisting of 4 M NaCHO 2 , 10% (v/v) glycerol and stored in N 2(l) . In the case of the selenomethionine-substituted protein, the mother liquor consisted of 3.9 M NaCHO 2 , 4% (v/v) glycerol, and the drop was made at a 2:2:0.5 ratio. Data were integrated with Mosflm (28) and scaled with Scala (29) ( Table 1). Truncate (29) was used to generate amplitudes. Both data sets suffered from anisotropy; a diffraction anisotropy server (30) was employed to correct for the anisotropy at the model building stage but did not improve the map quality, so anisotropy correction was not incorporated in the refinement.
Phase Determination and Refinement-Nineteen selenium sites of the possible 22 were identified with the AutoSol procedure from PHENIX (31) with a partial model of the Pseudomonas fragi (Pf) MFP (individual domains from the poly(Ala) model, Protein Data Bank code 1WDK) (13) from a previous molecular replacement run in PHASER (32) as input for AutoSol SAD selenium-site identification. Phase extension using Resolve (31) produced a partially interpretable electron density map, which was subjected to automated model building with the phenix.autobuild module (33) using keyworded scripts. This resulted in a 74% complete structural model (R fac / R free ϭ 34/37%), which was further refined and optimized using phenix.refine and Coot (34). Structure validation was done using Coot and MolProbity (35). Side chain atoms with poor real space correlation have been omitted from the model. As a consequence of the high overall B-factor, only a very limited number (25 solvent molecules) of solvent molecules could be identified. Neither co-factor nor substrate-like molecules were identified in the electron density despite the presence of NAD ϩ and acetoacetyl-CoA in the crystallization conditions. The final AtMFP2 model covers amino acid residues 7-719, but three loops had poor or very disordered density and were omitted from the model.  Table 1.
Enzyme Activity-AtMFP2 and AtAIM1 2-trans-enoyl-CoA hydratase activity was determined using in situ synthesized substrates from C4-to C18-CoA. Aliquots of 50 M saturated acyl-CoA esters were oxidized with a mixture consisting of AtACX1, AtACX3, and AtACX4 at 50 nM each in 175 mM Tris-HCl, pH 8.5, 2.5% (w/v) polyethylene glycol 400, and 40 pM catalase from bovine liver (Sigma-Aldrich). AtACX2 was not included because the AtACX1 substrate profile covers the C16and C18-CoAs (Ref. 36 and this study). The reactions were incubated in a FluorSTAR Optima plate reader (BMG Labtech GmbH, Offenburg, Germany) at 27°C, and the oxidation was monitored by recording the increase in absorbance at 260 nm due to the resonance between the introduced ␤-double bond and the CoA-thioester bond (37). Upon maximum conversion the absorbance had increased ϳ40% as expected from the ratio between the ⑀ 260 of saturated acyl-CoA and crotonyl-CoA (the ⑀ 260 of saturated and 3-hydroxy-acyl-CoA used was 16,400 and 22,600/M cm, respectively, for the 2-trans-enoyl-CoA substrates (38)). The hydratase assay was started by dispensing equal volumes of buffer, AtAIM1, or AtMFP2 with a multichannel pipette diluting the acyl-CoA substrates present by 10%. The final concentration of AtAIM1 or AtMFP2 in the reactions was 0.5 nM, and hydration was monitored by recording the decline in absorbance at 260 nm in the FluoSTAR in kinetics mode. To ensure maximal conversion for the dehydrogenase assay, the MFP concentration was adjusted to 5.5 nM. After full conversion, the dehydrogenase reaction was started by the addition of saturating amounts of NAD ϩ to a final concentration of 1 mM. The plate was read at 340 nm in kinetics mode in a SpetraMAX 340PC 384 plate reader (Molecular Devices) at 27°C. Experiments with 25 M substrate were conducted under conditions similar to those described above. To examine the effect of Tween 20 and delipidated bovine serum albumin (BSA; fraction V, 96% purity, Sigma), dehydrogenase assays were set up with C16-CoA and AtMFP2 as described above but titrated with 0 -4.9% (w/v) Tween 20 or 0 -22 M delipidated BSA. The concentration of MFP in the Tween 20 assay was 1 nM MFP, and it was 0.5 nM MFP in the BSA assay adjusted to 75 nM MFP in the C16-and C18-CoA dehydrogenase reactions. BSA was delipidated by incubation in 90% methanol at 4°C overnight. The protein was sedimented by centrifugation at 20,000 ϫ g and washed twice before residual methanol was removed under vacuum.
Structure Analysis-The Coot program (34) was used for structure evaluation. PyMOL was used to prepare Figs. 2, 3, and 7. Domain interfaces and packing was evaluated using the PISA (39), Protorp (40), Prism (41), and Hotsprint (Protein Data Bank code 1WDK) (42) servers. Secondary structure matching (SSM) superpositions were carried out in Coot. Sequence alignment including structural information was done in STRAP (43). Sequence identities were calculated by the FFAS (fold and function assignment system) server (44), consensus motifs were analyzed using WebLogo (45), and alignments and phylogenetic tree analysis were calculated by ClustalW (46) and depicted by TreeView (47). Protein surface topography was calculated using MOE (Chemical Computing Group).

RESULTS
Overall Structure-AtMFP2 is a 725-amino acid residue enzyme organized in two compact structural entities separated by an ␣-helical linker. The two AtMFP2 entities correspond to the two regions in the sequence to which ECH and HACD activities have been mapped (9)    supplemental Fig. S1 online). The interaction area between the N-terminal ECH domain (not including the linker region) and C-terminal HACD entity is only 152 Å 2 , whereas the helical linker has a 420 Å 2 interface with the ECH domain and an 817 Å 2 interface with the HACD entity, showing that the helical linker is absolutely essential for the spatial arrangement of the multifunctional protein complex (Fig. 2, orange). The HACD entity can be divided further into two domains (HACD N and HACD C ) with substantial interdomain interaction (730 Å 2 ). The HACD domains interface is dominated by hydrophobic contacts (55% nonpolar residues in the interface) but has a relatively loose interface fit (gap volume index, 3.9) (48).
ECH Domain-The ECH domain is a crotonase-like fold with a crotonase fold core of four turns of ␤-␤-␣ structure extended by a large, mostly ␣-helical C terminus. Ten ␤-strands form two mixed ␤-sheets lying almost perpendicular to one another. On one side, the larger sheet is packed against a single ␣-helix, whereas it is stacked with a three-layered arrangement of ␣-helices on the other side of the sheet. The crotonase-like family also includes the ECH domain of the PfMFP complex (13), the monofunctional hexameric or trimeric enoyl-CoA hydratases (49) and enoyl-CoA isomerases (50,51), the 6-oxo camphor hydrolases (52), the monofunctional hexameric 4chlorobenzoyl-CoA dehalogenases (53), and the monofunc- tional monomeric methylmalonyl CoA decarboxylase from E. coli (54). They all share the four turns of ␤-␤-␣ structure plus the following ␣-␤-␣-␤ structural motif called the T1 domain. It was originally characterized as a motif involved in trimerization of monofunctional proteins. A final common ␣-helical extension, HE, connects the T1 motif with a trimerization module in monofunctional proteins and represents the ␣-helical domain linker in AtMFP2 and PfMFP.
Crotonase-like enzymes catalyze diverse reactions but share the use of a CoA-thioester substrate and the formation of a carbanion species as part of their reaction scheme. The enzymes have a low overall sequence identity but have been shown to contain a common sequence pattern, the enoyl-CoA hydratase/ isomerase signature, 106  up the third ␤-␤-␣ section and is part of the interface binding the pantetheine, thioester, and acyl groups of the substrate but not the nucleotide part of CoA. Along with the helical linker, the small ␤-sheet embodies the CoA-binding site in MFP-ECHs (Fig. 2), whereas in monofunctional enzymes the last ␣-helix making up the CoA-binding site originates from a neighboring subunit.
The SSM superposition (55) of AtMFP2-ECH/monofunctional mitochondrial rat, Rattus norvegicus, RnECH (Protein Data Bank code 1MJ3), and AtMFP2-ECH/PfMFP-ECH gives core r.m.s.d. values of 1.5 and 1.7 Å, respectively, including AtMFP2 amino acid residues 8 -183 in the former case and 8 -244 in the latter (amino acid residues numbers will refer to AtMFP2 in the following discussion unless otherwise specified).
HACD Domains-The two HACD domains form a globular entity with the supposed active site residing at the bottom of a . The 3-hydroxyacyl-CoA dehydrogenase signature is indicated by the magenta color of the backbone trace. The proposed reaction mechanism of AtMFP2 is based on the present crystal structure and the reaction mechanism and structures of human HACD NAD ϩ and its substrate analogue complexes (56,70). cleft ( Fig. 3 and supplemental Fig. S1 online). The N-terminal domain (residues 308 -496) is 6-phosphogluconate dehydrogenase-like with eight strands in the central ␤-sheet flanked by ␣-helices. It has a Rossmann fold NAD binding motif, but no co-factor is bound in the crystal structure. The structure of the N-terminal AtMFP2-HACD domain is similar to the structures of the HACD domains found in PfMFP, peroxisomal RnMFE-1 (15), and human mitochondrial HACD (56) with SSM superposition r.m.s.d. values of 1.1, 1.6, and 1.1 Å, respectively, by superposition of residues 308 -496. The C-terminal HACD domain (residues 497-719) is a dimer of 5-helix motifs belonging to the hydroxyacyl-CoA dehydrogenase-like protein family. The first helical motif (residues 497-575) is connected to the other (residues 624 -719) by a long ␣-helix. Also the C-terminal HACD domains of MFPs are very similar (e.g. the SSM r.m.s.d. to PfMFP is 1.6 Å). Monofunctional mitochondrial dehydrogenases contain only one 5-helix motif, but the functional dimer of these enzymes creates an entity very similar to the C-terminal HACD domain (e.g. the SSM r.m.s.d. to human heart HACD (56) is 2.2 Å when including 177 of 222 amino acid residues). In AtMFP2, the first ␣-helices of each 5-helix motif run antiparallel and create an interface corresponding to the dimerization interface in monofunctional HACDs.  (57)), is located at the interface between the N-and C-terminal HACD domains and in between the two halves of the C-terminal domain (Fig. 3A). It carries the conserved active site residue Asn499. One highly conserved amino acid residue, Asp 523 , is engaged in hydrogen bonds important for the structural integrity of the C-terminal domain. The Asp 523 carboxylate accepts hydrogen bonds tying the second and third ␣-helix together. The side chain of another highly conserved residue, Asp 540 , is disordered in the current structure.
Plant MFPs-There is still not a vast amount of GenBank TM data available for plant MFPs, but two subfamilies of peroxisomal MFPs are emerging (supplemental Fig. S2 online). Although monocots seem to have more MFP isozymes than dicots, the isozymes appear to fall within either an AtAIM1-or AtMFP2-like subgroup. A single barley isozyme (GenBank TM GI: 151418576) identified in expressed sequence tags from inflorescence, developing caryopsis, apex, and germinating seeds could represent a third subfamily of plant MFPs, but the physiological significance of this gene product is unknown. The MFP2-like subfamily is characterized by a large proportion of Gly residues (ϳ15%) in the ECH flexible loop and only few conserved residues (Fig. 2C). The AtAIM1-like subfamily is characterized by an ECH-flexible loop having only one Gly residue and high sequence conservation (Fig. 2C).
Substrate Specificity-The substrate specificities of AtMFP2 and AtAIM1 were determined in vitro using recombinant acyl-CoA oxidases to synthesize the 2-trans-enoyl-CoA substrates immediately before sampling with AtMFP2 and AtAIM1. The oxidase reaction was monitored by UV absorbance spectroscopy at 260 nm confirming equal amounts of freshly prepared enoyl-CoA substrates to be available in the assays (Fig. 4A). Both AtMFP2 and AtAIM1 have short to long chain (C4 -C14) acyl-CoA ECH activity, but the AtAIM1 substrate profile is shifted toward the shorter substrates peaking at C4-CoA compared with AtMFP2 peaking around C8-and C12-CoA (Fig.  4B). The dehydrogenase substrate profile of AtMFP2 is slightly skewed toward the longer chain lengths compared with its hydratase profile and is about 10 times slower (Fig. 4C). The AtAIM1 dehydrogenase substrate profile lacks its C4 hydratase peak and peaks at C8-CoA instead of the C12-CoA of AtMFP2. The total production of NADH with C16-CoA after 1 h of incubation at 27°C and 10-fold more MFP remains indistinguishable from zero (Fig. 4D). Incubation with 25 instead of 50 M C16-or C18-CoA did not change this (data not shown). In the presence of Tween 20, some dehydrogenase activity is detected with C16-CoA peaking at around 0.1% (w/v) Tween 20 (Fig.  5A). However, the activity of the MFPs with C14-CoA is equally affected, maintaining the picture of negligible activity with substrates longer than C14-CoA of either AtAIM1 or AtMFP2. Delipidated BSA also affects the activity of AtMFP2 with C16-CoA up to around 6 M above which the positive effect declines (Fig. 6A). With the preparation of substrates unaffected (Fig.  6B), the presence of 6 M BSA in the reactions adversely affected the hydratase reaction of AtAIM1 (compared with Fig. 4B), whereas the activity with C16-CoA was below the increased noise level of the assay due to the presence of BSA absorbing in the UV range. With 150 times more MFP present in the C16-and C18-CoA wells than in the C14-CoA wells, AtMFP2 dehydrogenase activity with C18-CoA substrate is detected (Fig. 6D). Under these conditions, the activity of 0.
Two active site glutamic acids have been proposed to act in catalysis (62,63), and an oxyanion hole has been proposed to polarize the thioester carbonyl group. The N-terminal Glu activates a water molecule for nucleophilic attack at the C3 position (hereafter called Glu N (Glu 119 in AtMFP2)). The C-terminal Glu acts as a proton or, alternatively, as a hydrogen bond acceptor (hereafter called Glu C (Glu 139 in AtMFP2); Fig. 2 and supplemental Fig. S3 online). An active site water molecule hydrogen bonded to Glu C and Gly 147 is found at a position from where it could execute a nucleophilic attack on the enoyl-CoA substrate. The hydrogen bonding network used to deduce a reaction mechanism for RnECH is only partly conserved in AtMFP-ECH (Fig. 2A). The Gln 162 (rat) residue observed to donate hydrogen bonds to the active site Glu C in the rat enzyme is replaced by a hydrophobic residue in many crotonase-like enzymes including both AtMFP2 and PfMFP. The hydrogen bonds from Gln to Glu C provide an argument for a negatively charged Glu C in the reaction mechanism (59). The occurrence of a more hydrophobic Glu C environment would be expected to raise the pK a of the active site Glu C and, consequently, favor the concerted reaction mechanism over a stepwise mechanism, but it could also indicate an adaption to plant peroxisomal pH.
The development of a partial positive charge on the thioester carbonyl during the reaction could be stabilized by the oxyanion hole created by hydrogen bonds donated by the amide nitrogen atoms of Phe 68 and Gly 116 . Gly 116 is positioned like the backbone nitrogen atom with a similar function in monofunctional RnECH, methylmelanyl-CoA decarboxylase (54), and 4-chlorobenzoyl-CoA dehalogenase (53), whereas Phe 68 is somewhat misplaced due to the disorder of the 71-82 loop and high temperature factors of the 68 -87 region.
The acyl-binding pocket appears wide open in AtMFP2 because of the disorder of the 71-82 loop ( Fig. 2B and supplemental Fig. S3A online). The corresponding loop is observed to be disordered or flexible in all other crystal structures of crotonase-fold enzymes as well. The flexibility could reflect an ability to adapt the acyl-binding pocket structure to the substrate in accordance with the relatively broad substrate specificity profiles of the enzymes with regards to acyl chain length and the presence of conjugated systems. The part of the acyl-binding pocket that is included in the structure of AtMFP2-ECH is dominated by hydrophobic residues (Phe 33 , Phe 68  An MFP channeling mechanism, with the nucleotide being the pivot shaft for transfer of the acyl chain from ECH to the HACD active site, has been suggested in bacteria (13). If channeling is indeed a common phenomenon among the MFPs, the proposed nucleotide environment of the MFPs is expected to be conserved and would likely include an extended conformation of the acyl-CoA in order to bridge the 40 Å span from the ECH to the HACD active site. The linker and additionally the AtMFP2-ECH motif, -Asn 25 -Pro 26 -Pro 27 - (Fig. 2B), conserved among peroxisomal MFPs and some bacterial MFPs, would interfere with the binding of acyl-CoA in the bent conformation observed in monofunctional ECHs (Fig. 2B). Pro 26 -Pro 27 carries one of the two cis-prolines found in the AtMFP2 structure. There is no crystal structure of an acyl-CoA⅐MFP-ECH complex. PfMFP was crystallized in the presence of acetoacetyl-CoA and pentaethylene glycol n-octyl ether (C 8 E 5 ) (13), and the C 8 E 5 molecules were bound in the ECH/HACD domains. The adenine-binding pocket as defined by the PfMFP-HACD⅐acyl-CoA complex (Protein Data Bank code 1WDM; acetoacetyl-CoA B factor, 160 Å 2 ) (64) consists of residues that are not strictly conserved between AtMFP2 and PfMFP. This pocket resides at the ECH/HACD interface. The three-dimensional shape of this interface is highly conserved however, and the hydrophobicity patches and potential for hydrogen bonds to the adenine moiety persists. A highly conserved MFP pattern of Arg/Lys amino acid residues in the helical linker ( 290 FFSQRG-TAKVP 300 ; bold residues are highly conserved) and the ␣helix from the HACD domain facing the domain interface ( 481 KKIKK 485 ) could support the existence of a common ECH/ HACD adenine-binding site, with positively charged amino acid residues supporting the relocation of the pyrophosphate in the region proposed by Ishikawa et al. (13) as the center of rotation, but supportive mutagenesis, structural, or biophysical evidence is not available yet.
Dehydrogenase Activity-AtMFP2-HACD catalyzes the oxidation of the L-3-hydroxyacyl-CoA hydroxyl group to a keto group while reducing NAD ϩ to NADH. For mitochondrial short chain HACDs, the reaction takes place by hydride transfer at the si face of the nicotinamide ring (65). The NAD ϩbinding site, defined by the presence of a 319 Gly-X-Gly 321 -X-X-G 324 phosphate binding motif (66), is located in the first ␤-␣ transition in the N-terminal AtMFP2-HACD domain adjacent to the active site Ser 428 , His 449 , Glu 461 , and Asn 499 logo (Fig.  3A). Amino acid residues observed to hydrogen bond to the cofactor in monofunctional HACDs (67,68) and PfMFP are conserved and have similar conformations in AtMFP2-HACD (Glu 342 , Glu 401 , Lys 406 , and Asn 624 ), but no significant electron density is observed in the AtMFP2 co-factor-binding site. A similar absence of positive electron density for the co-factor is observed in the crystal structure of the truncated recombinant RnMFE-1-HACD (15), whereas NAD ϩ is present in the active site of the PfMFP ␤-oxidation complex but with an average B-factor of 98 Å 2 (13). As evidenced by the enzyme activity characterization included in this study, the AtMFP2 used for crystallization experiments is an active enzyme transferring electrons to NAD ϩ . Interestingly, at position 347 in AtMFP2 and AtAIM1, a relatively well conserved Gly is replaced by a bulky Phe colliding with the expected position of a ribose hydroxyl group of NAD ϩ and therefore likely impeding NAD ϩ binding and lowering the affinity of AtAIM1 and AtMFP2 for NAD ϩ . The NAD ϩ -binding site, as defined by the structure of PfMFP, is included in Fig. 3

by superposition with
AtMFP2. Also included is the superimposed acetyl-CoA from the same PfMFP structure. The conserved residues His 449 and Glu 461 have been identified as being important for dehydrogenase activity by sitedirected mutagenesis of E. coli MFP, and a reaction mechanism has been proposed and later modified (56,69,70). The reaction mechanism implies that the hydrogen bond between Glu 461 and His 449 would orientate His 449 for optimal proton subtraction from the substrates hydroxyl group. Ser 428 and Asn 499 side chains could stabilize the developing negative charge on O3 of the substrate (Fig. 3B). A conserved cis-peptide bond (Ser 452 -Pro 453 ) adjacent to the active site His 449 appears essential for orienting the substrate with backbone hydrogen bonds to O1 and the mercaptoethylamino group.
The relative orientation of the N-and C-terminal domains of AtMFP2-HACD results in the formation of a relatively open Y-shaped NAD ϩ and acyl binding cleft between the three domains, thereby exhibiting a relatively solvent-exposed binding site not designed for substrate discrimination based on acyl chain length or conjugation of double bonds. The N-and C-terminal domains of human monofunctional HACD have been shown to undergo a small rearrangement upon the binding of substrate and a larger rearrangement upon the formation of the ternary complex with NAD ϩ . The large rearrangement sequesters the active site from solvent and brings the active site residues into optimal position for catalysis to occur (56). The conserved active site residues of AtMFP2-HACD seem too far apart to facilitate the electron/hydride transfer depicted in Fig. 3B. The His 449 N ⑀ to Asn 499 N ␦2 distance is 4.3 versus 3.7 Å in the ternary human HACD complex. The structures of the bacterial PfMFP represent an ensemble of slightly different domain packing (64), suggesting that the HACD subdomains are likely to move upon substrate binding. The domain movements inferred to occur upon substrate/co-factor binding in PfMFP affects only the ECH/HACD N interface and not the HACD N / HACD C interface. This is not consistent with the HACD N / FIGURE 6. BSA effect on MFP activity and substrate profile. A, AtMFP2 was titrated with BSA, and dehydrogenase activity was determined. B, substrate was prepared in the presence of 6 M delipidated BSA, and hydratase activity was subsequently determined, C. D, dehydrogenase activity was determined with 150 times more MFP in the C16-and C18-CoA wells compared with 0.5 nM in the C14-CoA well. The data are not normalized to M enzyme to show residual long chain dehydrogenase activity at high enzyme concentrations. Conditions otherwise are as described for Fig. 4. HACD C domain closure upon NAD ϩ /substrate binding observed in monofunctional human HACD and suggests that either the monofunctional enzymes and MFP have different mechanisms of action or that not all functional states of PfMFP are covered by the current crystal structures.
Substrate Specificity-Our data clearly characterize AtAIM1 as a short to medium chain length-specific hydratase coupled to a broad chain length-specific dehydrogenase, whereas AtMFP2 is characterized as a medium chain length-specific hydratase coupled to a broad chain length-specific dehydrogenase (Fig. 4,  B-D). Because of the difficulties in obtaining 2-trans-enoyl-CoA substrates, few studies have addressed the actual substrate specificities of AtMFP2 and AtAIM1. However, our data showing the in vitro substrate profiles of AtMFP2 and AtAIM1 for short chain substrates are in accord with the comments on substrate preference made by Richmond and Bleecker (3) and the observation that MFP2 mutants have unaltered crotonyl-CoA hydratase activity levels (11). However, at the other end of the profile, neither AtMFP2 nor AtAIM1 has significant activity with enoyl-CoA substrates longer than C14 (tested with 2-trans-enoyl-C16-and -C18-CoA). Considering the abundance of C18 -C22 fatty acids in Arabidopsis seeds, with eicosenoic acid (C20:1) being the predominant seed-specific fatty acid (73), this is remarkable.
To address potential solubility or micelle formation issues as the reason for the lack of activity with longer chain length acyl-CoA substrates, dehydrogenase activity was assayed at 25 M substrate or in the presence of either Tween 20 or delipidated BSA. Changing the substrate concentration alone did not result in detectable activity, whereas both Tween 20 and delipidated BSA affected the activities of AtMFP2 in a complex manner with a concentration threshold for a positive effect to occur followed by a peak and a gradual decline of effect (Figs. 5A and 6A). The activity enhancing effect may be due to solubilization of the substrates by the detergent or lipid-binding BSA. It is, however, noticeable that the effect was the same for long and medium chain substrates. It is also possible that it represents a direct effect on the dynamics of the enzymes. The decline of the positive effect in the case of the detergent is likely because of denaturation of the enzymes, whereas the declining effect of BSA is possibly because of competitive binding of the acyl-CoA substrates. In the presence of either 0.1% (w/v) Tween 20 or 6 M BSA, corresponding to the concentrations at maximum activity enhancement with C16-CoA, the activity with C16-CoA and C18-CoA remains negligible relative to the activity with C14-CoA (Figs. 5B and 6C). The addition of excess MFP to the C16-and C18-CoA reactions clearly illustrates that AtMFP2 activity with these substrates is significantly (about 116 and 560 times) lower than with C14-CoA (Fig. 6D). It thus seems unlikely that the lack of activity with substrates longer than C14-CoA is due to solubility issues.
MFPs limited to short to medium chain hydratase activities are known to exist in cucumber peroxisomes and rat liver mitochondria, but they are usually supplemented with isozymes harboring long chain activity as well (14,71). The existence of additional hydratases has been identified and inferred in Arabidopsis peroxisomes (11,72), but none of these has been shown or suggested to be specific for long chain substrates. Our data suggest that an additional long chain hydratase is active in Arabidopsis peroxisomes.
Relying on the endogenous ACXs present in the seedlings, Rylott et al. (11) addressed the substrate specificity of AtMFP2 in vivo by feeding crude seedling homogenates with substrates of various chain lengths and detecting the resultant acyl-CoA products by mass spectrometry. Interestingly, they conclude that AtMFP2 is a long chain-specific hydratase coupled to a broad range dehydrogenase. However, when feeding with C18-CoA, 3-hydroxy-octadecanoyl-CoA accumulates at wild-type concentrations in the MFP2 mutant, suggesting that wild-type 2-trans-octadecenoyl-CoA hydratase activity remains in the MFP2 mutant. In a complementary experiment, the pools of long chain fatty acids in Arabidopsis seedlings were quantified. The data show an initial increase and consecutive decrease in virtually all fatty acid pools identified from C16 to C20. The wild-type and mutant seedlings exhibited the same tendencies, but the tendencies were stronger in the wild type, increasing more and ending with lower concentrations of eicosanoic acid (C20:0) and eicosenoic acid (C20:1) after 5 days, relative to the mutants. That the decrease also appears in the mutant insinuates the presence of long chain hydratase activity in the MFP2 mutant seedlings. The weaker tendencies observed in the MFP2 mutants could be attributed to the observed 35% reduction in overall fatty acid levels in the mutants, indicating that the state of the seedlings is affected in a complicated manner. The data illustrate the inherent difficulties encountered when interpreting in vivo data in a complex background such as ␤-oxidation with its numerous isozymes and involvement in hormone metabolism.
In vitro experiments employing recombinant enzymes may correspond to an incomplete biological system. However, because all experiments were done at concentrations lower than the critical micelle concentration of C16-CoA (202 Ϯ 5 M (74)), all chain lengths of 2-trans-enoyl-CoA were synthesized equally well by the ACX mixtures employed (Fig. 4A), and the lower substrate concentration or added Tween 20 or BSA did not alter the ratio between conversion of C14-CoA and the longer chain lengths, it is difficult to envisage how the observed substrate profiles could be artificial. Complexes between the MFPs and thiolases could potentially influence the tertiary structure and substrate binding of the MFPs similar to the situation in PfMFP, but it has not been possible to replicate this situation in vitro.
The dehydrogenase substrate profiles of AtMFP2 and AtAIM1 are skewed toward the longer chain lengths compared with their hydratase profiles and are about 10 times slower (Fig.  4C). The AtAIM1 dehydrogenase substrate profile lacks its C4 hydratase peak and peaks at C8-CoA instead. The hydratase reaction being an order of magnitude faster than the dehydrogenase activity is common in multifunctional enzymes (18,71). The dehydrogenase assay was of course impeded by the fact that hydroxyacyl-CoA substrates longer than C14-CoA were inefficiently synthesized, but even upon incubation with 75 nM MFP for 1 h at 27°C, when full hydration had occurred (not shown), AtMFP2 dehydrogenation of C16-CoA proceeded about 120 times slower than with C14-CoA (Fig. 6D). It is thus safe to conclude that both AtAIM1 and AtMFP2 are broad