Structure of RhlG, an Essential β-Ketoacyl Reductase in the Rhamnolipid Biosynthetic Pathway of Pseudomonas aeruginosa*

Rhamnolipids are extracellular biosurfactants and virulence factors secreted by the opportunistic human pathogen Pseudomonas aeruginosa that are required for swarming motility. The rhlG gene is essential for rhamnolipid formation, and the RhlG enzyme is thought to divert fatty acid synthesis intermediates into the rhamnolipid biosynthetic pathway based on its similarity to FabG, the β-ketoacyl-acyl carrier protein (ACP) reductase of type II fatty acid synthesis. Crystallographic analysis reveals that the overall structures of the RhlG·NADP+ and FabG·NADP+ complexes are indeed similar, but there are key differences related to function. RhlG does not undergo the conformational changes upon NADP(H) binding at the active site that in FabG are the structural basis of negative allostery. Also, the acyl chain-binding pocket of RhlG is narrow and rigid compared with the larger, flexible substrate-binding subdomain in FabG. Finally, RhlG lacks a positively charged/hydrophobic surface feature adjacent to the active site that is found on enzymes like FabG that recognize the ACP of fatty acid synthesis. RhlG catalyzed the NADPH-dependent reduction of β-ketodecanoyl-ACP to β-d-hydroxydecanoyl-ACP. However, the enzyme was 2000-fold less active than FabG in carrying out the same reaction. These structural and biochemical studies establish RhlG as a NADPH-dependent β-ketoacyl reductase of the SDR protein superfamily and further suggest that the ACP of fatty acid synthesis does not carry the substrates for RhlG.

Pseudomonas aeruginosa is a Gram-negative bacterium that is capable of existing in multiple environmental niches and is an opportunistic pathogen that infects immunocompromised and cystic fibrosis patients (1,2). It is a particularly serious pathogen in hospitals and is relatively resistant to many commonly used antibiotics. Key pathogenic features of the organism are that it produces an extracellular rhamnolipid surfactant that facilitates the acquisition of hydrophobic carbon sources (3), the development of biofilms (4,5), and the dissemination of hydrophobic signaling molecules (6); it acts as a heat-stable hemolysin (7); and it is required for swarming motility (8,9). P. aeruginosa swarming is a multicellular form of organized movement allowing the rapid colonization of surfaces (8,10). Rhamnolipids are virulence factors that are found in the sputa of cystic fibrosis patients (11) and inflict damage on eukaryotic cells (12)(13)(14)(15). Rhamnolipids also have potential industrial applications because of their ability to reduce the surface tension of water (16).
Rhamnolipids are composed of molecules having either one or two rhamnose sugars linked to a dimer of ␤-hydroxyacids (primarily ␤-hydroxydecanoate) (17). The terminal stages of the rhamnolipid biosynthetic pathway are carried out by the dTDP-rhamnose transferases RhlB and RhlC that produce lipids with one or two rhamnose moieties, respectively (18,19). RhlA is essential for rhamnolipid synthesis, and although its biochemical activity has not been demonstrated, it is proposed to be involved in forming the ␤-hydroxydecanoyl-␤-hydroxydecanoate precursor (9,20). RhlG is an essential component in the rhamnolipid biosynthetic pathway because inactivation of the rhlG gene leads to the loss of rhamnolipid production (21). Based on the similarity of RhlG to FabG, 2 the NADPH-dependent ␤-ketoacyl-ACP reductase of type II fatty acid synthesis, RhlG is proposed to perform the same reaction as FabG and divert intermediates in fatty acid synthesis into the rhamnolipid pathway (21). A transacylase may be involved downstream of RhlG, and the mechanism for the transport of rhamnolipids out of the cell is unknown. This work reports the x-ray structure of RhlG and compares its structure and biochemical activities to the FabG of type II fatty acid biosynthesis.

EXPERIMENTAL PROCEDURES
Expression and Purification of RhlG-The P. aeruginosa rhlG gene was amplified by PCR and cloned into pET15b vector within NdeI/ BamHI restriction sites. The recombinant RhlG protein with a N-terminal His-tag was expressed in Escherichia coli strain BL21(DE3) and purified by nickel-affinity chromatography as described previously (22). The SeMet-labeled RhlG was expressed in an E. coli methionine auxotroph strain B834 grown in M9 minimal medium supplemented with glucose and amino acid mix containing SeMet (23).
The Oligomerization State of RhlG-The Stoke's radius of RhlG was evaluated using gel filtration chromatography on a Superdex TM 200 10/300 GL column (Amersham Biosciences) equilibrated with 20 mM Tris-HCl, pH 8, containing 200 mm NaCl. Protein molecular weight markers (Amersham Biosciences) were ferritin (440 kDa), catalase (232 kDa), bovine serum albumin (66 kDa), ovalbumin (43 kDa), chymotrypsinogen A (25 kDa), and RNaseA (13.7 kDa). Sedimentation velocity experiments were conducted at 50,000 rpm and 4°C using a ProteomeLab XL-I analytical ultracentrifuge, equipped with both absorbance and interference optical detection systems, an eight-hole Beckman An-50 Ti rotor, and cells containing sapphire windows and charcoal-filled Epon double-sector centerpieces. The sedimentation velocity profiles of the protein (50 g/ml, 400 l) were collected at 1-min intervals with the Rayleigh interference optical system. The SEDNTERP program (24) was used to estimate the molecular weight and the partial specific volume. Data were modeled as a superposition of Lamm equation solutions, c(s), with the software SEDFIT 9.2 (25,26). The sedimentation coefficient distribution c(s) was calculated with maximum entropy regularization, the molecular weight distribution, c(M), was calculated using the value determined from the c(s) distribution, and the f/f 0 value was determined from the c(s) analysis (26).
Crystallization and Structure Determination-RhlG and SeMet-RhlG were crystallized at 18°C by the vapor diffusion hanging drop method. Upon setup, the drop contained 50 mM HEPES, pH 7, 5 mm Tris, pH 8.0, 0.15 M ammonium sulfate, 6% polyethylene glycol 4000, 0.1 m NaCl, 0.5 mM EDTA, 0.5 mm dithiothreitol, and 2.5 mg/ml protein. The reservoir contained 1 M HEPES, pH 7, 0.3 m ammonium sulfate, and 12% polyethylene glycol 4000. Orthorhombic crystals in space group C222 1 appeared in ϳ3 days. RhlG⅐NADP ϩ co-crystals, also in space group C222 1 , were formed in the presence of 1 mM NADPH. Crystals were flash-frozen in mother liquor with 20% glycerol for cryoprotection. A 2.9 Å SeMet RhlG⅐NADP ϩ single wavelength anomalous dispersion data set was measured at the SER-CAT 22ID beamline of the Advanced Photon Source, the 2.3 Å native RhlG⅐NADP ϩ data set was measured at SER-CAT 22BM, and the 2.7 Å native RhlG data set was measured at SER-CAT 22ID. The data were integrated using MOSFLM (27) and scaled using SCALA (28). A selenium substructure solution comprising 12 of the possible 20 selenium sites was found using CNS-1.1 (29). These were fed into Resolve (30) for 2-fold NCS detection, density modification, and model building. Roughly 60% of the asymmetric unit was traced automatically. The homologous FabG⅐NADP ϩ structure (PDB ID 1Q7B) was then used as a guide to manually complete the model using the O program (31). Of the two molecules in the asymmetric unit, molecule A was more complete, and it was therefore used to model poorly defined regions of molecule B. The model was refined using REFMAC5 (32) and subjected to another round of building and refinement. The native data were processed similarly, but the R merge with the RhlG⅐NADP ϩ data were 30% despite less than a 1% change in the unit cell parameters. Therefore, the apo structure was solved by molecular replacement using AMoRE (33) and the RhlG⅐NADP ϩ structure as the search model. Addition of waters, B-factor refinement, and energy minimization were all performed using CNS-1.1. Refer to Table  1 for data measurement and refinement statistics.
ACP-dependent Gel Reconstitution Assay of RhlG/FabG Activity-RhlG/FabG activity was tested in an ACP-dependent assay by analyzing the formation of ␤-hydroxyacyl-ACP with a conformational-sensitive gel electrophoresis method. The assay contained 100 M ACP, 1 mm ␤-mercaptoethanol, 100 M [2-14 C]malonyl-CoA (specific activity, 55 Ci/mol), 45 M octanoyl-CoA, 100 M NADPH, 1 g of purified FabD, 1 g of purified FabH in 0.1 M sodium phosphate buffer, pH 7.0, and indicated amount of FabG (1 ng to 0.256 g) or RhlG (2.20 -140 g) in a final volume of 40 l. The ACP, ␤-mercaptoethanol, and buffer were preincubated at 37°C for 30 min to ensure the complete reduction of ACP. The substrate for the RhlG/FabG reaction was generated using FabD to transfer the malonyl group from CoA to ACP to produce malonyl-ACP, and Mycobacterium tuberculosis FabH to condense octanoyl-CoA and malonyl-ACP to form ␤-ketodecanoyl-ACP. The reaction was initiated by the addition of RhlG/FabG. After incubation at 37°C for 30 min, the reaction was stopped by adding 8 l of 6ϫ sample buffer to the reaction mix and placing the reaction tubes on ice. Then 40 l of each of the assay mixtures was loaded onto conformational sensitive gels that contained 13% acrylamide and 2.5 M urea. The electrophoresis was performed at 25°C under constant current of 32 mA/gel. The gels were stained in Coomassie Blue, destained, and dried with a vacuum gel drier at 80°C. The gels were exposed against PhosphorImager screens that were scanned by a Typhoon 9200. The product of FabG, ␤-hydroxy-[1-14 C]decanoyl-ACP, was quantitated with ImageQuant 5.2.
ACP-independent Spectrophotometric Assay of RhlG/FabG Activity-This assay measured the disappearance of NADPH, the cofactor for the RhlG/FabG reaction, spectrophotometrically at 340 nm. The reaction mixture contained 0.5 mM acetoacetyl-CoA, 0.2 mm NADPH, 8 g of FabG or up to 100 g of RhlG protein, 0.1 M sodium phosphate buffer, pH 7.4, in a final volume of 300 l. The reaction was initiated by the addition of acetoacetyl-CoA. Decrease in the absorbance at 340 nm was recorded for up to 5 min. The initial rate was used to calculate the enzymatic activity.

RESULTS
Purification and Oligomeric State of RhlG-Purification of RhlG by metal chelate affinity chromatography followed by gel filtration yielded pure protein with an apparent mass of 29 kDa as shown by SDS gel electrophoresis (Fig. 1A). RhlG possessed a Stoke's radius determined by gel filtration chromatography consistent with a molecular mass of a 50.4 kDa globular protein (Fig. 1A), indicating that RhlG is a dimer in solution. The analysis of purified RhlG by analytical ultracentrifugation verified that the protein exists as a dimer with a molecular mass of 58.4 kDa at protein concentrations Ͻ0.8 mg/ml (Fig. 1B). At higher protein concentrations, RhlG tetramers (116 kDa) were also observed (not shown).
The Structure of RhlG-Crystals of RhlG in space group C222 1 were obtained in the absence and presence of NADP ϩ , but the RhlG⅐NADP ϩ co-crystals grew more reproducibly and diffracted to higher resolution. These co-crystals were selected for ab initio structure determination using single wavelength anomalous diffraction phasing with SeMet-RhlG. A data set was collected at the "peak," and this was sufficient to identify the selenium sites and to obtain initial phases to 2.9 Å. The final high resolution model was refined against 2.3 Å data collected from a native RhlG⅐NADP ϩ co-crystal. The asymmetric unit comprised an AB dimer, but NADP ϩ was only bound to monomer A and was absent from monomer B (Fig. 2). The N-terminal His-tag was not visible in either monomer and was absent from the final model. In addition, residues 97-104 in monomer B were not visible in the electron density. Analysis of the structure using PROCHECK (34) revealed a high quality structure, and this is consistent with the final R work and R free values of 21.2 and 26.4%, respectively. The RhlG crystals without NADP ϩ diffracted to 2.7 Å, and the initial structure was determined from the RhlG⅐NADP ϩ model using rigid body refinement. In the final model, residues 93-104  and 94 -104 were missing from monomers A and B, respectively, and the N-terminal His-tag residues were also absent. The final R work and R free values were 22.4 and 28.6%, respectively. Pertinent data collection and refinement statistics for all structures are presented in Table 1. Each monomer of RhlG adopts a Rossmann fold structure that contains seven ␤-strands and eight ␣-helices placing the enzyme in the SDR superfamily of reductases (Fig. 2) (35). As indicated by the similarities between the RhlG and FabG primary structures (Fig. 3), the overall fold (Fig. 2) is very similar to that of FabG (23,36,37), and the description will therefore be brief. The fold contains two right-handed ␤␣␤␣␤ motifs (␤1-␣1-␤2-␣2-␤3 and ␤4-␣4-␤5-␣5-␤6) connected by helix ␣3. Helices ␣6 -␣7 form a separate helix-turn-helix substructure, and ␣8 -␤7 conclude the fold with strand ␤7 completing the seven-stranded parallel ␤-sheet. The organization of the dimer is identical to that of FabG, with monomer-monomer interfaces involving ␣4 and ␣5.
Detailed Comparison of RhlG to FabG-FabG undergoes a series of complex conformational changes in response to NADP ϩ binding that mediate negative allosteric kinetic behavior (23,37). Most notably, the three active site residues Ser 138 , Tyr 151 , and Lys 155 (E. coli FabG) are not in their catalytic conformation unless NADP ϩ is present. The same is not true for RhlG (Fig. 4). The three FabG⅐NADP ϩ active site residues are superimposable on RhlG residues Ser 148 , Tyr 162 , and Lys 166 , thereby confirming their identical roles in catalysis. In contrast, the adenineribose-phosphate half of the cofactor binds more tightly to RhlG than to FabG as judged from the interactions with the protein. In FabG, the binding pocket includes one arginine that interacts with the phosphate and a number of hydrogen bonding interactions. In RhlG (Fig. 5), arginines 19 and 41 interact with the phosphate, and the guanidinium group of Arg 41 stacks onto the adenine rings to fully enclose this end of the cofactor. Also, the backbone amides of Arg 41 , Asp 42 , and Leu 66 as well as the hydroxyl of Ser 18 mediate hydrogen bonding interactions. Finally, it may be significant that only one active site in the asymmetric unit has a bound cofactor because both active site clefts are accessible in the crystal packing arrangement. One obvious difference between monomers A and B is the ordering of residues within the ␤4-␣4 loop (residues 93-104), which is clearly correlated with cofactor binding. This loop, which is adjacent to but not directly a part of the nicotinamide-binding pocket, is only ordered in the presence of NADP ϩ (monomer A). Residues 94 -96 are in van der Waals contact with Met 199 when NADP ϩ is present, thereby sequestering the cofactor from solvent.
An important feature of the FabG catalytic mechanism, and perhaps of many other SDR superfamily members, is the presence of a "proton relay" that shuttles protons from the solvent into the substrate (23, 37).  In E. coli FabG, Tyr 151 forms a hydrogen bond with the carbonyl oxygen at the C-3 of the substrate, thereby positioning the substrate for catalysis. Catalysis occurs by Tyr 151 donating a proton to the carbonyl oxygen and NADPH donating a hydride ion to the C-3 carbon. The side chain of Lys 155 binds the ribose hydroxyls of NADPH, and these groups interact with a series of water molecules to ensure that the Tyr 151 hydroxyl hydrogen is replenished following catalysis. In FabG, the proton relay absolutely depends on the presence of NADP ϩ , both to provide crucial hydrogen bonding elements and to effect necessary conformational changes in the active site residues and the backbone that allow the proton relay to form. A characteristic kink in helix ␣4 at a conserved asparagine (Asn 110 ) provides a key hydrogen bond acceptor to the proton relay by releasing the backbone carbonyl oxygen from its helical hydrogen bonding interactions (37). All of these structural requirements for the proton relay are present in RhlG, including the conserved Asn 116 , and two of the predicted four water molecules are clearly present with marginal electron density for a third water molecule at the expected location (Fig. 6). Unlike FabG, the structural requirements for the proton relay are present in the absence of cofactor, and the RhlG active site is always primed for catalysis.
Although very similar, RhlG⅐NADP ϩ and FabG⅐NADP ϩ have significant differences in four regions of their structures that have functional implications. These regions in the RhlG dimer are highlighted in Fig. 7A, and they generally surround the active site cleft containing the bound NADP ϩ . Region one encompasses the ␤4-␣4 loop, and these residues are flexible in the absence of cofactor. The difference may be related to conformational flexibility; however, it is one of the most conserved regions in the FabG protein family, and this conservation does not extend to RhlG. The second region involves the ␣4-␤5 loop that approaches the active site across the dimer interface from monomer B. RhlG contains a four-residue extension to this loop compared with FabG and lacks the cluster of arginines and lysines that are found in FabG sequences. Region three is within the ␤5-␣5 loop that is adjacent to the ␣4-␤5 loop. The fourth region is in the ␣6-␣7 helix-turn-helix substructure. Although both substructures are similar in the two enzymes, they achieve this structure in different ways. In FabG, the helix-turn-helix is maintained by a hydrophobic interaction on the inner surface (Phe 183 and Ile 200 in E. coli), a salt bridge on the outer surface (Glu 185 and Arg 197 ) and a central leucine residue (Leu 209 ). In RhlG, Trp 222 corresponds to Leu 209 in FabG, and a salt bridge between Arg 194 and Asp 213 replaces the Phe 183 -Ile 200 interaction on the inner surface. These changes considerably reduce the interior volume and hydrophobicity of the inner cavity of the RhlG motif. Also, there are two additional proline residues in the RhlG motif (Pro 196 and Pro 207 ) compared with FabG that reduce the flexibility of the feature. One residue that is invariant in both RhlG and FabG is a methionine at the N terminus of helix ␣6 (Met 199 in RhlG and Met 188 in E. coli FabG). In RhlG, this side chain extends into a small pocket flanked by the pyrophosphate of NADP ϩ and the backbone of residues 94 -96. In FabG, helix ␣6 is in an open conformation, and the methionine side chain makes no interactions with the cofactor or the protein. Finally, it is apparent that the surfaces of FabG and RhlG have different electropotential characteristics (Fig. 7, B and C). Most notably, the positive potential associated with ␣4-␤5 loop in FabG is absent in RhlG, and the hydrophobic inner surface of the ␣6-␣7 substructure is less extensive in RhlG. These differences give rise to a different substrate docking interface and a more constricted active site tunnel in RhlG.
Enzymatic Activity-The high degree of sequence similarity between RhlG and FabG (Fig. 3) led to the suggestion that RhlG is a ␤-ketoacyl reductase (21). However, our RhlG structural analysis suggests that the ACP of fatty acid synthesis may not carry RhlG substrates because of the lack of the electropositive/hydrophobic ACP binding surface or docking site adjacent to the active site entrance that is characteristic of proteins that bind ACP (38). We assessed this point by measuring the activity of  RhlG as a reductase using ␤-ketodecanoyl-ACP as substrate (Fig. 8, A  and B). The in vitro assay detects formation of ␤-hydroxy-[1-14 C]decanoyl-ACP, the product of FabG, by conformational sensitive gel electrophoresis. The specific activity of RhlG was 0.0245 pmol/min/g, ϳ2000-fold less active than that of FabG (49.7 pmol/min/g). RhlG activity was not detected when NADH was substituted for NADPH in the assay (not shown). The stereochemistry of the product was inferred as ␤-D-hydroxydecanoyl-ACP based on the ability of purified E. coli FabZ of type II fatty acid synthesis to form 2-trans-decenoyl-ACP (not shown). The potential for CoA thioesters to also be RhlG substrates was examined using a spectrophotometric assay and ␤-ketobutyryl-CoA as the substrate (Fig. 8C). Unlike FabG, NADPH oxidation was not detected at any level of RhlG suggesting that CoA thioesters are not better substrates than ACP thioesters.

DISCUSSION
Our crystallographic analyses of RhlG have confirmed that the structure of the protein is highly homologous to that of FabG. In addition, the proposal that RhlG and FabG are both stereospecific NADPH-dependent ␤-ketoacyl reductases that catalyze the equivalent reduction of a carbonyl group to an enoyl is fully supported by their identical active sites and the way in which they bind the cofactor NADPH. The identical spacing of the active site lysine and tyrosine is particularly important because it is dictated by the carbonyl moiety where the reduction is taking place. For example, in the structurally homologous FabI enzyme that reduces a double bond substrate rather than a carbonyl group, the two residues are spaced further apart. Also, like FabG, the nicotinamide ring of the bound NADP ϩ adopts the syn conformation consistent with FIGURE 7. Comparison of the RhlG dimer to the FabG dimer in the cofactor bound state. A, the RhlG⅐NADP ϩ dimer shown as ribbons. The four regions that show significant differences from the FabG⅐NADP ϩ structure are indicated in cyan. Arg 194 and Glu 157 line opposite sides of the active site entrance, and form a hydrogen bond in the presence of NADP ϩ . A salt bridge between Arg 194 and Asp 213 helps to maintain the ␣6/␣7 substrate binding cleft. B and C, the electrostatic surface potential of the RhlG dimer (B) and the FabG dimer (C). NADP ϩ was removed prior to generation of the molecular surface. The views are identical and correspond to A looking into the active site tunnel. A yellow arrow marks the tunnel entrance. Note that the FabG ACP-binding surface centered on arginines 129 and 172 is missing in RhlG. Also, the hydrophobic substrate-binding surface of FabG centered on Ile 200 is much smaller in RhlG. Finally, Met 199 in RhlG engages the active site and closes the substrate-binding pocket, whereas its counterpart in FabG, Met 188 , is exposed, and the pocket is open. The electrostatic potential surface was generated using GRASP (43). Negative charges are indicated by red, positive charges by blue, and neutral charges by white. The scale ranges from Ϫ12 to 12 k b T (k b ϭ Boltzmann constant, T ϭ temperature). the catalysis of pro-S hydrogen transfer from the cofactor to the acyl chain substrate.
However, there are key differences between the two proteins. These differences not only provide important clues as to the role of RhlG in the rhamnolipid biosynthesis, but they also provide insights into the catalytic mechanism of FabG. The first major difference is that the RhlG active site is fully competent for catalysis in the absence of NADPH. In contrast, the catalytic competency of FabG depends on the presence of cofactor that induces significant conformational rearrangements of the three active site residues upon binding. In addition, cofactor-dependent changes in the FabG backbone conformation are required to set up the proton relay, but this is not the case for RhlG. We have shown that FabG is a negative allosteric enzyme, and the conformational changes appear to mediate this kinetic behavior. Consistent with the rigidity of the active site, RhlG does not appear to be an allosteric enzyme, although we have no obvious explanation for why only one of the active sites contains cofactor in our RhlG⅐NADP ϩ structure.
The second difference concerns the structure and nature of the FabG and RhlG acyl chain-binding pockets. Although we and others have not been able to visualize a bound acyl chain in FabG or RhlG, structural work on the homologous FabI enzyme strongly suggests that the pocket lies on the inner surface of the ␣6-␣7 substructure that is directly adjacent to the active site. In FabG, this surface is very hydrophobic (see Fig. 7C), and the helix-turn-helix motif adopts variable locations, suggesting that it can switch from an open to a closed conformation when cofactor and substrate are both present. In contrast, RhlG has a rigid ␣6-␣7 substructure with a smaller hydrophobic surface, and the motif adopts a closed conformation.
In the presence of NADP ϩ , residues 198 -213 within ␣6-␣7 associate with Glu 157 and Gln 158 from the ␤4-␣4 loop via a hydrogen bond (2.6 Å) between Arg 194 and Glu 157 . In FabG⅐NADP ϩ , the active site tunnel is 6.1 Å between nearest neighbors Gln 203 and Asn 145 . This 3.5-Å wider cavity is consistent with the ability of FabG to accept acyl chain lengths between 4 and 18 carbons, including cis-unsaturated substrates of 12 carbons and longer, whereas RhlG is predicted to have a far stricter acyl chain specificity in overall size and length. Given the predominance of ␤-hydroxydecanoate in rhamnolipids, RhlG has no selective pressure to recognize substrates that differ significantly from ␤-ketodecanoyl-linked molecules. The reduced flexibility of the RhlG motif appears to result from two additional proline residues compared with FabG, and the interaction of Met 199 with the phosphates of the cofactor on one side and the ␤4-␣4 loop on the other. Met 199 is exposed and fully conserved in the FabG protein family, and its role was previously unknown. The structure of RhlG suggests that it helps to lock the ␣6-␣7 substructure in the closed position when both substrates are present in the active site. Consistent with this, the ␤4-␣4 loop is highly conserved in FabG and is flexible in the absence of cofactor in both FabG and RhlG.
The third and final difference relates to the ACP docking site that appears to be missing from RhlG. In a series of studies, we have shown that an electronegative/hydrophobic "recognition helix" of ACP (helix ␣2) docks to an electropositive/hydrophobic patch that is adjacent to the active sites of the enzymes of type II fatty acid synthesis (38,39). In FabG, the patch is centered on conserved arginines 129 and 172 on the adjacent monomer in the region of the ␣4-␤5 loop, and ACP presumably supplies the substrate across the monomer-monomer interface (40) (Fig. 7C). Inspection of the equivalent electrostatic surface adjacent to the active site entrance of RhlG does not indicate the presence of an ACP interaction domain (Fig. 7B). Importantly, RhlG lacks the two conserved arginines. In addition, the ␣4-␤5 loop where ACP appears to bind in FabG has a significantly different structure in RhlG where it is extended by four residues. These structural features strongly suggest that ACP does not interact with RhlG and are FIGURE 8. Enzymatic activity of RhlG. A, the substrate ␤-ketodecanoyl-ACP was generated in situ using M. tuberculosis FabH, malonyl-ACP, and octanoyl-CoA, and the formation of ␤-hydroxydecanoyl-ACP as a function of RhlG concentration was measured following separation of the products by conformationally sensitive gel electrophoresis as described under "Experimental Procedures." The inset in A shows the activity of FabG assessed under the same experimental conditions. B, a representative gel image of the ACP-dependent reductase activity of FabG and RhlG. RhlG exhibited significantly less activity than FabG although 50 g of RhlG and only 1 g of FabG were used. C, activity of FabG (f) and RhlG (Ⅺ) toward ␤-ketobutyryl-CoA as a function of protein determined by monitoring the continuous oxidation of NADPH in the spectrophotometric assay described under "Experimental Procedures." consistent with the poor activity of RhlG toward the ACP thioesters that are intermediates in type II fatty acid synthesis.
Our kinetic results support an essential role for the NADPH-dependent RhlG ␤-ketoacyl reductase in rhamnolipid formation, but they are not consistent with its proposed function as an enzyme that diverts acyl chains from fatty acid biosynthesis as proposed by Campos-Garcia et al. (21). To fulfill this role, RhlG must be able to effectively compete with FabG for ␤-ketoacyl-ACPs. For example, LpxA is an enzyme that diverts the ␤-hydroxytetradecanoyl-ACP from fatty acid synthesis to lipid A formation and is able to perform this role by effectively competing with the FabZ dehydratase of fatty acid elongation for ␤-hydroxytetradecanoyl-ACP (41). However, our biochemical assays clearly demonstrate that RhlG is three orders of magnitude less active with ACP thioesters compared with FabG, and therefore, would not be able to effectively compete with FabG for substrate. Another problem with this proposal is that the product of both FabG and RhlG would be ␤-D-hydroxydecanoyl-ACP; thus RhlG cannot divert acyl chains for fatty acid synthesis and would not be essential to rhamnolipid formation.
Taken together, our structural and biochemical data suggest that RhlG does not represent an exit point from lipid biosynthesis to rhamnolipid biosynthesis but rather that the two pathways are quite distinct. Although our current data do not clarify the role of RhlG in the rhamnolipid biosynthetic pathway, they do suggest that there are missing enzymes upstream of RhlG that specifically supply substrates to this enzyme that are critical to ␤-hydroxydecanoate formation. RhlG has presumably evolved from FabG to perform this specialized role in rhamnolipid biosynthesis, and it therefore makes sense that it has lost the necessity for allosteric control and substrate malleability, and become a "hard-wired," single substrate enzyme. An effective way to separate the pathways would be a specialized ACP that only recognizes RhlG and the other component enzymes, and the very different ACP-binding sites on FabG and RhlG support this idea. P. aeruginosa has two ACP-like carriers in its genome in addition to the ACP of fatty acid synthesis (42), and we are exploring the possibility that one of these putative carrier proteins is devoted to the rhamnolipid biosynthetic pathway.