Insights into Acinetobacter baumannii fatty acid synthesis 3-oxoacyl-ACP reductases

Treatments for ‘superbug’ infections are the focus for innovative research, as drug resistance threatens human health and medical practices globally. In particular, Acinetobacter baumannii (Ab) infections are repeatedly reported as difficult to treat due to increasing antibiotic resistance. Therefore, there is increasing need to identify novel targets in the development of different antimicrobials. Of particular interest is fatty acid synthesis, vital for the formation of phospholipids, lipopolysaccharides/lipooligosaccharides, and lipoproteins of Gram-negative envelopes. The bacterial type II fatty acid synthesis (FASII) pathway is an attractive target for the development of inhibitors and is particularly favourable due to the differences from mammalian type I fatty acid synthesis. Discrete enzymes in this pathway include two reductase enzymes: 3-oxoacyl-acyl carrier protein (ACP) reductase (FabG) and enoyl-ACP reductase (FabI). Here, we investigate annotated FabG homologs, finding a low-molecular weight 3-oxoacyl-ACP reductase, as the most likely FASII FabG candidate, and high-molecular weight 3-oxoacyl-ACP reductase (HMwFabG), showing differences in structure and coenzyme preference. To date, this is the second bacterial high-molecular weight FabG structurally characterized, following FabG4 from Mycobacterium. We show that ΔAbHMwfabG is impaired for growth in nutrient rich media and pellicle formation. We also modelled a third 3-oxoacyl-ACP reductase, which we annotated as AbSDR. Despite containing residues for catalysis and the ACP coordinating motif, biochemical analyses showed limited activity against an acetoacetyl-CoA substrate in vitro. Inhibitors designed to target FabG proteins and thus prevent fatty acid synthesis may provide a platform for use against multidrug-resistant pathogens including A. baumannii.

Antibacterial resistance is a large burden to modern healthcare, and is predicted to result in 10 million deaths annually if no action is taken by 2050 49 . In particular, Acinetobacter baumannii, a Gram-negative opportunistic pathogen, is highlighted as a major concern to human health. A. baumannii is implicated in both nosocomial and community acquired infections 24 , and its pathogenic success can be largely attributed to its highly plastic genome, resulting in the acquisition of numerous virulence and resistance determinants enabling it to thrive in hospital environments 31,50,69 . Specifically, its ability to form biofilms on medical equipment contributes to severe nosocomial infections 31 . Increasing drug resistance has resulted in the use of last resort antibiotics, including tigecycline and colistin, and pan-drug resistant strains may become common within the next two decades 70 . Carbapenem-resistant A. baumannii was recently listed as an urgent threat by the Centre for Disease Control (CDC) and as the top priority for drug development by the World Health Organisation (WHO) 12,62 . For these reasons, research into novel drug development is essential for successful treatment of multi-drug resistant pathogens, including A. baumannii.
Fatty acids can be produced via two distinct pathways: type I and type II Fatty Acid Synthesis (FASI and FASII, respectively). In mammals, FASI is performed by an enzyme with a large multidomain and multifunctional structure 6 . In contrast, bacteria utilise the FASII system which involves a series of discrete and functionally specific enzymes that produce fatty acids primarily destined for incorporation into membrane lipids. Acyl-carrier protein (ACP) is attached to the acyl chain via a thioester linkage and acts to deliver the substrate between enzymes of the FASII pathway. In brief, this pathway begins with condensation of acetyl-CoA and malonyl-ACP

Methods and materials
Genome mining and phylogenetic analysis. Identification of FabG-like protein sequences across a panel of A. baumannii genomes from different clonal groups (A. baumannii AB5075_UW, AYE, WM99c, ACICU, ATCC 17978, D1279979 and SDF) were performed using V5VHN7 and A0A0E1FTA3 (UniProtKB) as queries for BlastP searches against the NCBI non-redundant protein sequence database 4 . Duplicate sequences obtained from independent BlastP searches were removed. All FASTA sequences (n = 175) were used as input data and a multiple sequence alignment performed using ClustalW as an integrated part of GenomeNet 36 with default settings applied. Phylogenetic reconstructions from the multiple sequence alignment were performed using the "build" function of ETE3 v3.1.1 in GenomeNet. Protein phylogeny was examined using the midpoint rooted Maximum-Likelihood method (100 bootstraps) using RAxML v8.1.20 with model PROTGAMMAJTT and default parameters applied 60 . The resulting tree was annotated in FigTree v1.4.4 (http:// tree. bio. ed. ac. uk/ softw are/ figtr ee/). All A. baumannii genomes included in the phylogenetic analyses were aligned using Mauve 17 to confirm the presence/absence of identified SDRs and their relative genetic positioning within the examined genomes. Identification of conserved SDR fingerprint motifs from FASTA sequences were performed manually. To determine ACP interacting residues from the identified A. baumannii SDRs, representative protein sequences ([ST]x 12 Yx 3 K) from clades with fully conserved active site motifs (n = 28) were aligned against known FabG sequences using methods as described above with gap open penalties set at 12.
Expression and purification. Genes encoding the putative enzymes were codon optimized for Escherichia coli expression and cloned into the pMCSG21 vector at the SspI site, producing a 6-His fusion protein with TEV cleavage site for tag removal (Genscript, Piscataway, NJ). Plasmids were transformed into competent E. coli BL21(DE3) pLysS cells (Merck-Millipore) and selected on spectinomycin (100 µg/mL) Luria-Bertani (LB) agar plates. A starter culture was used to inoculate 1 L of auto-induction media 61 , incubated at room temperature for 36 h until an OD 600 > 3.0 was reached. Cells were harvested by centrifugation and resuspended in 'Buffer A' containing 20 mM imidazole, 300 mM NaCl, 50 mM phosphate buffer pH 8.0. Cell membranes were lysed via two freeze-thaw cycles followed by the addition of 2 mg/mL lysozyme and 0.025 mg/mL DNAse. The cell lysate was clarified using centrifugation and filtration (using a 0.45 μm PVDF syringe filter) prior to purification over a 5 mL Nickel-Sepharose HisTrap HP column (GE Healthcare). A 10-column volume wash with Buffer A removed unbound contaminants. Following this, the sample was eluted by a gradient of elution 'Buffer B' (containing 50 mM phosphate buffer pH 8.0, 300 mM NaCl, and 500 mM imidazole) over 5-column volumes. The resulting eluate was treated with TEV protease overnight to cleave the affinity tag, then further purified using a Superdex 200 26/60 column (GE Healthcare) in tris-buffered saline ( www.nature.com/scientificreports/ peak from gel filtration was collected and concentrated using a 10 kDa MW centrifugal filter (Amicon/Millipore) and all samples were assessed for purity by SDS-PAGE.
Crystallization, data collection, and structure modelling. Purified  Pellicle and biofilm formation analyses. ATCC 17978 and ΔAbHMwfabG were grown ON in LB broth, sub-cultured to OD 600 of 0.05 in LB broth (5 mL) in polypropylene tubes and grown in the dark for 72 h at 37 °C under static conditions. Quantitative measurement of pellicles was assessed by the addition of 1 mL of methanol, the floating pellicle was subsequently removed and resuspended in 1 mL of PBS. Resuspended pellicles and planktonic growing bacteria (200 µL) were transferred to 96-well microtitre plates (GreinerBio-one) and OD 600 values determined using a Spectrostar nano (BMG Labtech). Following removal of the remaining culture media, biofilm was quantified by crystal violet staining, thorough washing and quantitation by measuring the optical density at 590 nm using previously described methods 21 .
Motility. Motility analyses of strains ATCC 17978 and ΔAbHMwfabG were performed by inoculating colony material in the centre of a semi-solid LB agar plate (0.35% Eiken Agar) as per previous studies 3 . Bacterial migration was quantified (in centimetres) 10 h post inoculation at 37 °C.
Minimal inhibitory concentration analyses. The minimal inhibitory concentration (MIC) of strains ATCC 17978 and ∆AbHMwfabG was determined using a previously described method 68 . In brief, overnight cultures were diluted in cation-adjusted Mueller-Hinton media. Cells were then transferred to a 96-well microtiter tray containing a two-fold dilution series of colistin, chloramphenicol, gentamicin or streptomycin, with the final volume being 100 μL. Plates were incubated in a humidified chamber at 37 °C for 18 h. Growth was determined by visual examination. www.nature.com/scientificreports/ Two isolates from the international clone (IC) I lineage (AB5075_UW and AYE) and two from the IC II lineage (ACICU and WM99c) were included. The remaining three isolates examined are not categorized within the described IC lineages. Two of these are the widely studied strains ATCC 17978 and SDF, where the latter has undergone significant insertion sequence (IS) mediated genome reduction and is defined as avirulent 27 . The A. baumannii strain D1279779 was isolated from an outpatient in Northern Australia and represents a communityacquired A. baumannii isolate 24 .

Results and discussion
A total of 175 protein sequences were identified, where phylogenetic analysis revealed clustering of these sequences into 36 distinct clades (Fig. 1). The length of proteins identified from 35 of the 36 clades ranged between 241 and 303 amino acids, indicative of members belonging to the classical family of SDRs 38 . In contrast, all members of the clade represented by ABO12488.2 were 463 amino acids in length.
The majority of SDRs were present in two or more strains, with two exceptions identified (ABO13863.1 and CAM84732.1). These members were found to be encoded on endogenous plasmids present in ATCC 17978 and AYE, respectively, and thus suggest a foreign origin. The non-IC lineage strains ATCC 17978 and D1279979 encoded the greatest number of distinct SDR proteins (n = 29) from the examined A. baumannii genomes. The higher abundance can be directly attributed to unique SDR sequences being encoded within 'regions of genomic plasticity' unique to these strains, as previously defined by Farrugia and colleagues 24 . Clades represented by KGP65003.1 and KGP65050.1 were found to be specific for the IC I lineage, as they were only present in AYE and AB5075_UW strains whilst the clade represented by ACC57255.1 was the only IC II specific SDR identified. The avirulent A. baumannii strain SDF was found to encode 15 distinct SDRs where two sequences encoded C-terminal truncations as a result of ISAba7 mediated insertions. It could be hypothesised that sequences conserved within SDF catalyze essential functions whilst homologs absent from this strain may indicate redundancy or catalyze functions that enable metabolic versatility and/or virulence. Our analyses indicate significant diversity of SDRs present in A. baumannii, where the presence of specific members/proteins are influenced by the evolution of distinct global lineages and the impressive genetic plasticity afforded by this bacterium.
Comparative analysis of conserved SDR domains. All identified sequences were found to harbour the N-terminal coenzyme binding fingerprint motif most similar to the consensus of the classical SDR family, TG X3 [AG][FILV]G 38 . Members of five distinct clades were found to encode amino acid substitutions at important conserved residues within the motif, potentially indicating reduced or no functional activity.
Twenty-nine of the 36 clades harboured the classical SDR catalytic triad active site consensus sequence, [ST] x 12 Yx 3 K 38 . No residues constituting the active site sequence could be identified for the two clade members as represented by ABO12830.2 and thus are likely non-functional. The remainder of sequences that differed from the consensus were due to deviation from the consensus sequence pattern or absence of the serine/threonine residue critical for substrate stability.

Identification of A. baumannii
FabG homologs with putative 3-oxoacyl ACP functionality. Similar to many FASII enzymes, FabG homologs encode a conserved ACP-binding signature surface which facilitates ACP docking and subsequent donation of the acyl chain from the ACP prosthetic group 76 . It has been shown that Arg 129 and Arg 172 , residues located at the entrance to the active site tunnel play a significant role in ACP docking in E. coli 76 . FabG is deemed essential in most bacteria as it is the only isozyme capable of reducing β-ketoacyl-ACP substrates required for endogenous fatty acid production, however, some unique exceptions have been reported 25,35,45,73 . To identify SDRs with putative 3-oxoacyl ACP activity in A. baumannii, representative protein sequences from distinct clades encoding the typical catalytic motif ([ST]x 12 Yx 3 K) were aligned against sequences with known reductive capacity towards 3-oxoacyl ACP substrates (Supplementary Figure 1). The multiple sequence alignment revealed that sequences from clades represented by ABO11256.2 and ABO10977.2 encoded the conserved hydrophobic residues adjacent to the active site which support ACP interaction. ABO11256.2 is conserved across all examined genomes and is encoded in an operon with other enzymes with predicted roles in fatty acid biosynthesis. Further, large scale mutagenesis studies have defined ABO11256.2 to be the only SDR that is essential for A. baumannii viability, and thus is likely to be the primary FASII FabG of the bacterium 28,65 . Interestingly, ABO10977.2 is only present in four of the seven genomes, where it was absent from the IC I isolates AYE and AB5075_UW as well as SDF (Fig. 1). These phylogenetic analyses have deduced clades with ACP interacting residues to be genetically similar, and form sister clades with ABO12488.2, a conserved clade where all members are approximately double in length to that of ABO11256.2 and ABO10977.2.
To further explore the unique SDR, the representative ABO12488.2 sequence was used as a query for conserved domain searches which revealed that members of this clade harbor an N-terminal flavodoxin-type domain and a typical ketoreductase domain at its C-terminus, indicative of HMwFabG proteins such as MtFabG4 19 . HMwFabG represent a genetically distinct group of β-oxoacyl reductases with homologs often restricted to species from Actinobacteria and Proteobacteria phyla and catalyze the reduction of β-oxoacyl-ACP using NADH as a coenzyme 19 . Similar to low molecular weight FabG proteins, hydrophobic residues are also responsible for interactions with ACP substrates albeit in a different position (Arg 111 and Lys 150 ) 20 . A multiple sequence alignment of MtFabG4 against ABO12488.2 clade members revealed the conservation of Arg 111 and Lys 150 (Arg127 and Lys166 in ABO12488.2) residues, inferring the HMwFabG member of A. baumannii can also catalyze the reduction of β-oxoacyl-ACP substrates.
By using comparative analyses, we have deduced A. baumannii SDR members with putative 3-oxoacyl ACP activity.  28 . We therefore pursued techniques in X-ray crystallography to elucidate the three-dimensional structure of this important enzyme. The structure of apo FabG was resolved at 1.9 Å in the space group P 4 3 2 1 2 using molecular replacement with starting model PDB: 4WJZ (58% sequence identity, Hou et al. 34 ). The structural model displayed good stereochemistry, with a final overall R work = 0.19 and R free = 0.21 (complete statistics in Table 1). The model showed two protomers in the crystal asymmetric unit with each protomer consisting of 244 residues arranged as an archetypal Rossmann fold ( Fig. 2A). The secondary structure displayed a central β-sheet (β3-β2-β1-β4-β5-β6-β7) skirted by 9 α-helices (Fig. 2B). An α7 turn α8 motif forms a capping region, or flexible 'lid' to the active site cavity. This capping region is observed in most SDRs, and its flexibility is reflected with poorer density and higher B factors in this region. Overall, from generation of crystal symmetry mates, the tertiary structure is a homotetramer arranged as two anti-parallel dimers, consistent with conventional FabG enzymes, and the elution profile on a size exclusion column (≈ 100 kDa, or 4 × 26 kDa protomers). Interactions between chains were confirmed by PDBsum (Supplementary Table 3) 42 . A/B interactions involve an average total interface area of 1502 Å 2 , and 12 hydrogen bonds at the β7 strand of each opposing protomer. The interaction between protomers B and C is made through the α4 to α4′, α5′ to α5, and α5-β5 loop regions of each chain with an average total interface area of 1492.5 Å 2 , 6 salt bridges, and 22 hydrogen bonds. No interactions were observed between diagonal protomers, A/D and B/C. In terms of catalytic mechanism, the reduction reaction is well understood and occurs when a hydride from the coenzyme nicotinamide ring and a proton from an active site tyrosine is donated to the C3 carbon and C3 oxygen of the substrate β-ketoacyl-ACP, respectively 26,52 . For SDR proteins, the catalytic residues are highly conserved. In FabG this triad is well positioned to accept the hydride ion and includes Ser 138 , Tyr 151 located on the loop region between β5 and α7, and Lys 155 located on the α5 helix. Following the reaction, it is known that a proton relay system replenishes the tyrosine via a residue (commonly Asn, Gly, or Ser; position Asn 110 in E. coli) that forms a kink in the α4 helix near the active site, crucial for positioning a backbone carbonyl for participation in a water network 26,52 . Whilst an asparagine residue is commonly observed at this site, our FabG shows a His 110 residue in this location, however, it is the kink and the helical backbone that is essential rather than the residue sidechain, and this kink is observed in α4 of our structure.
Specific residues that determine coenzyme preference in SDR proteins has been elucidated and coenzyme binding families assigned 38,63 . For FabG, the presence of a basic residue in the Gly motif (TGASRGIG 18 ) suggested preference for NADPH and categorization into the cP1 subfamily 38 . To observe binding mechanisms, we www.nature.com/scientificreports/ www.nature.com/scientificreports/ performed co-crystallization with NADPH. Crystals diffracted on the MX1 beamline at the Australian Synchrotron to a resolution of 1.85 Å in the spacegroup P 4 3 2 1 2 and the final model displayed good stereochemistry and an overall R work and R free of 0.16 and 0.19 respectively ( Table 1). The crystallographic asymmetric unit revealed two protomers, and strong positive density allowing NADPH to be modelled within each protomer. The NADPH is bound through an extensive binding pocket, whereby the nicotinamide moiety is bonded with Lys 155 and Tyr 151 , and the adenine ribose is bonded through Ser 14 , Arg 15 , Thr 37 , Asp 59 , Val 60 , and Asn 86 (Fig. 2C). Interestingly, contrary to previous reports of large conformation changes in FabG upon coenzyme binding (such as that in E. coli), our structure reveals little change at the active site between apo and holo proteins (chain A comparison reveals r.m.s.d = 0.107 Å) 51,52 . An 'active' arrangement of catalytic residues for apo proteins is also observed in Plasmodium falciparum PDB: 2C07 67 , Staphylococcus aureus PDB: 3OSU (unpublished), and Bacillus anthracis PDB: 2UVD 74 . Superimposition of A. baumannii FabG ± coenzyme, E. coli FabG ± coenzyme, and P. falciparum apo FabG highlight the positioning of catalytic residues at the active site in these models (Fig. 2D). Enzyme assays complemented clear preference for NADPH by FabG (Fig. 2E), confirming biological relevance for the NADPH bound structure. Although enzyme assays were conducted in the presence of a substrate mimic, acetoacetyl-CoA, we believe FabG has an ability to bind ACP due to the presence of two conserved ACP-binding residues, Arg 129 and Arg 172 (Supplementary Figure 1), as these residues have been shown to interact with ACP in other FabG models 76 . Overall, analysis of the FabG sequence, structure, and enzymatic assays, suggests FabG (ABO11256.2) is indeed a functional, and is likely the primary 3-oxoacyl-ACP reductase of A. baumannii.

Structural analysis of AbSDR; a short chain dehydrogenase/reductase. Another putative A.
baumannii gene annotated as a putative 3-oxoacyl-ACP reductase was AbSDR (ABO10977.2), however, unlike FabG, it was not conserved across all species examined. AbSDR is a classical SDR enzyme, where our bioinformatics analysis revealed this protein to harbor a conserved domain belonging to the FabG_rel super family (cl36988, E-value; 3.56e-127) rather than the BKR_SDR_c subgroup of the NADB_Rossman super family that includes other FabG proteins, including A. baumannii FabG (cd05333, E-value; 8.57e-121). We therefore solved the crystal structure to determine if features of this protein were consistent with those for a typical 3-oxoacyl-ACP reductase. The best quality diffraction data for AbSDR was collected on the Australian Synchrotron MX2 beamline to a resolution of 1.9 Å. The crystal was indexed in P 3 1 (Table 1). Each protomer displays a Rossmann fold (Fig. 3A), and a schematic view of structural topology (Fig. 3B) highlights a central β-sheet displaying a β3-β2-β1-β4-β5-β6-β7 pattern. This β-sheet is flanked by two groups of helices (α3, α4, α5, α6 and α2, α1, α8). Interestingly, AbSDR contains only one alpha helix (α7) in the flexible capping subdomain, which is conserved in SDR proteins. Alternatively, our FabG protein displays two helices α7 and α8 in the same region. This subdomain is responsible for dimerization (generating a dimer of dimers [tetramer]) which is involved in the formation of a tunnel to the active site 51 .
Overall, four protomers were present in the crystallographic asymmetric unit, each displaying high structural homology (greatest r.m.s.d of 0.158 Å, Fig. 3A). The tetramer is very similar to A. baumannii FabG and the biological unit is confirmed by gel filtration elution volume (tetramer ≈ 100 kDa) and SDS-PAGE (protomer ≈ 26 kDa). To further confirm that the homotetramer observed in the asymmetric unit was indeed representative of the biological unit, we performed PDBsum analysis (Supplementary Table 4) 42 . The quaternary structure showed a homotetramer with two interfaces: the α8 and β7 of opposing antiparallel protomers at one interface (A/B interface area 1363.5 Å, 1 salt bridge, 12 hydrogen bonds, and 184 non-bonded contacts) and the α4 and α6 helices of opposing antiparallel protomers at the other (A/C interface area 1642.5 Å, 4 salt bridges, and 22 hydrogen bonds, and 212 non-bonded contacts). A weak interaction is observed between the diagonal B/C interface, with 5 non-bonded contacts.
In terms of catalytic ability, the active site residues are conserved, and closely resemble that of our FabG structure. At the active site, three SDR catalytic residues, Ser 140 -Tyr 153 -Lys 157 are present, and Ser 111 creates a kink in the α4 helix. Superimposition shows that similar to FabG, the active site architecture is in an 'active' conformation, even without the presence of coenzyme suggesting conformational change may not be necessary for coenzyme binding (Fig. 3C). By analogy, and due to the conserved nature of the active site residues, we assume a similar catalytic reaction to that already determined for SDR proteins 26 . AbSDR has two basic residues Arg 12 and Arg 34 that suggests specificity towards a NADPH coenzyme rather than NADH, and classification in the cP3 coenzyme-binding subfamily 38,63 . We investigated the binding site to assess ability of AbSDR to bind NADPH, by superimposing AbSDR with our NADPH bound FabG model, revealing the two NADPH determining basic residues, Arg 12 and Arg 34 , alongside other conserved residues involved in coenzyme binding (Fig. 3C). Functional assays did not show clear activity, unlike that for FabG where all available NADPH was converted to NADH in the reaction (Fig. 3D). In terms of ability to bind ACP, AbSDR residues are conserved at the ACP-coordination site, including Lys 131 and Arg 174 (Supplementary Figure 1).

Comparative analyses of A. baumannii FabG and
AbSDR. Since our enzyme activity analysis revealed limited NADPH conversion by AbSDR, we searched for closely related structures. Comparison of our structure to others in the PDB was conducted using the DALI server 33 and a protomer from each is superimposed in (Fig. 3E) 1.6, unpublished). While the most closely related structures are annotated as FabG enzymes, majority are unpublished and without supporting biochemical assays, and therefore, FASII functions remain putative. Previous analyses of other FabG-like SDR enzymes have shown no activity in vitro in A. baumannii 16 and roles in steroid metabolism in M. tuberculosis 71 . As SDR proteins maintain conserved structural features, whilst performing an array of functions, it can be difficult to assign biological function based on their structure or sequence. FabG2 from Xanthomonas campestris (a plant-associated bacterial pathogen) is unable to catalyze the reduction of short-chain (3-oxoacyl-ACP) substrates in an initial FabG reaction, however, can participate when longer substrate chains (≥ C 8 ) are available 35 . FabG3 from X. campestris, whilst functional against acyl chains of various lengths, showed lowered reductive activity compared to functional FabG1 and was not essential for fatty acid synthesis, rather likely involved in xanthomonadin synthesis 73 . FabG1 and FabG3 of X. campestris contain the necessary hallmarks of a FabG-catalytic triad, N-terminal coenzyme binding motif, and an ability to bind ACP 35,73 . X. campestris belong to the same class of gamma-proteobacteria as A. baumannii, and as such there may be potential that AbSDR performs a similar ancillary function within fatty acid biosynthesis. We suggest the annotation for AbSDR remains as a putative 3-oxoacyl-ACP reductase.

A. baumannii HMwFabG: A conserved high molecular weight 3-oxoacyl-ACP reductase.
A much larger protein (463 aa) annotated as a 3-oxoacyl-ACP reductase was revealed during initial A. baumannii sequence analysis. High molecular weight 3-oxoacyl-ACP reductases are noted in Proteobacteria and Actinobacteria, however, only one crystal structure for a HMwFabG has been characterized to date: MtFabG4 PDB: 3M1L 19 . We therefore investigated the structure of this protein, to better understand this class of enzymes and its potential role in A. baumannii. After successful expression and purification of the recombinant protein, crystallography trials yielded two crystal forms with different space groups. These structures were modelled using molecular replacement with starting model PDB: 5VP5 (48% sequence identity, unpublished). The first structure was resolved to a resolution of 1.8 Å, with P1 symmetry, and a R work = 0.18 and R free = 0.21 (designated as crystal form 1 or CF1). The structure from the second crystal form was resolved to a resolution of 1.65 Å, modelled in the P 1 2 1 1 space group, had overall R work = 0.2 and R free = 0.24, (and was designated crystal form 2 or CF2). All data collection and refinement statistics for both structures are listed in Table 1. The CF1 model showed poor electron density for residues 1-25 in chain A and 1-22 in chain B and for this reason, these regions could not be modelled. For CF2, N-terminal residues could be built in chain A and show a bent helix structure (residues 4-22), unlike MtFabG4 where this region was truncated (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16) to assist with protein solubility during expression and purification 19 . Chain B of CF2 had regions of flexibility and poor electron density between residues 1-22 and 405-413 where it could not be built. As the models were structurally very similar, (r.m.s.d between protomers of the two models was 0.179 Å), CF2 (PDB: 6UUT) was used for the remaining analysis.
The crystallographic asymmetric unit contains two protomers (Fig. 4A). Each protomer has two domains; an N-terminal domain: β2-β3-β1-β4-β5-β6-β7 and C-terminal domain: β14-β13-β12-β11-β8-β9-β10 (Fig. 4B). The N-terminal domain displays a flavodoxin-like fold important for dimeric chain interaction and orientation as previously described 19 . The C-terminal domain resembles a typical FabG enzyme Rossmann fold with a β-α-β cross-over region, fingerprint coenzyme binding residues and catalytic residues. In terms of quaternary structure, both models complemented elution traces from gel filtration suggesting HMwFabG is a dimer in solution with protomers an approximate size of ≈ 48 kDa, also confirmed by SDS-PAGE. This was further confirmed through calculation of interface statistics using PDBsum (Supplementary Table 5) 42 . The average total interface area was 3057 Å 2 , and had 4 salt bridges, 32 hydrogen bonds, and 333 non-bonded contacts. These interactions create an antiparallel homodimer formed between helices at the N-terminal domain of one protomer with the C-terminal domain of the opposing protomer. That is, α4 and α6 of the N-terminal domain interact with α11 and α13 on the C-terminal domain opposing protomer. Superimposition of the two structures also reveals an extra β2-α3-β3 motif following the α2 helix of our structure, where a loop region (residues 73-80 not modelled) is seen in MtFabG4 20 . Overall, the general structures of HMwFabG proteins are conserved, and the catalytic C-terminal domain closely resembles the machinery used by classical low molecular weight FabG enzymes. Within the C-terminal domain of HMwFabG is the Ser 357 -Tyr 370 -Lys 374 catalytic triad and Asn 329 creates a kink in the α10 helix important for proton replenishment and this domain closely resembles the structure of a classic 3-oxoacyl-ACP reductase, including our FabG.
Preference for NADH coenzyme is determined by Asp 256 which is suggested to decrease available volume of the 2′phosphate accommodating region, thereby providing an unfavourable binding environment for the larger NADP(H) coenzyme 20 . Our enzyme assays showed clear preference for NADH as the catalytic coenzyme (Fig. 4C). Attempts to co-crystallize HMwFabG and NADH were unsuccessful. Since our structure was homologous with the MtFabG4, which also prefers NADH, we compared the two models to analyse the ability of HMwFabG to accommodate coenzyme and substrate 20 (Fig. 4D). Superimposition with MtFabG4 (49% sequence identity) NAD + and HXC ligands reveals our structure possesses an active site that can accommodate coenzyme and substrate, and highlights key residues involved with interactions between coenzyme and substrate (Fig. 4D) 20 . These residues are conserved in HMwFabG, with the exception of Thr 313 and Lys 169 . Lys 169 , shows significant sidechain movement, however, conformational rearrangement at this site has been noted previously 20 .
Essential ACP binding residues are identified for low molecular weight FabG 76 and HMwFabG 20 . In our HMwFabG, these residues translate to Arg 127 and Lys 166 which contribute to a positively charged cluster of residues at the entrance to the active site that enables ACP binding (Supplementary Figure 1). Both our structural and functional data suggest that HMwFabG has the capacity to participate as a reductase in the FASII pathway of A. baumannii.   Given NADH is a lower energy molecule relative to that of NADPH and is largely associated with catabolic pathways, a role for HMwFabG enzymes under nutrient limited environments has been proposed 18 . To examine the role of AbHMwFabG, the HMwFabG from the well-studied laboratory reference strain ATCC 17978 (A1S_2061) was deleted by allelic replacement, generating ∆AbHMwfabG. The ability of ∆AbHMwfabG to grow in M9 minimal media (nutrient poor) or LB media (nutrient rich) were assessed. Interestingly, unlike that of MtFabG4, no significant differences were identified for ∆AbHMwfabG compared to WT in nutrient poor media (Fig. 5A), whilst ∆AbHMwfabG was significantly impaired in its ability to grow in nutrient rich conditions (Fig. 5B). The HMwFabG encoded in P. aeruginosa PAO1 (PA4786) has been shown to be involved in the production of a quorum sensing signalling molecule 30 , which can impact lifestyle changes such as switching to a sessile growth state. To examine the impact HMwFabG may have on A. baumannii lifestyles; motility, biofilm and pellicle formation were examined. No significant differences were observed in biofilm formation (Fig. 5C), whilst ΔAbHMwfabG was significantly impaired in its ability to form a pellicle (Fig. 5D). Pellicles are communities of bacteria that form at the air-liquid interface and provide a favourable environment for strict aerobes such For all panels, results are the mean ± standard error of the mean (SEM) from at least biological quadruplicates. Statistical analyses were performed using an unpaired, two-tailed Student's t-test; * p < 0.05, *** p < 0.001, ns = not significant. www.nature.com/scientificreports/ as A. baumannii as they support efficient oxygen uptake from the air and nutrient acquisition from the media below 29,46 . Similar to biofilms at the solid-liquid interface, bacteria in pellicles are encased in an extracellular polymeric matrix comprised of various molecules including exopolysaccharides, proteins, extracellular DNA and lipids. The matrix is known to enhance motility, horizontal gene transfer and promote cell-to-cell signalling 44 . Accordingly, motility of the ΔAbHMwfabG mutants was significantly lower compared to that of the parental strain (Fig. 5E). A proteomic analysis revealed AbHMwFabG to be up-regulated in 4-day pellicles compared to cells grown in the planktonic state 40 , which is in support of our findings for a role of AbHMwFabG in pellicle biogenesis. AbHMwFabG may aid in the synthesis of fatty acids that constitute the pellicle matrix or be involved in the production of quorum sensing molecules that A. baumannii produces, such as N-hydroxydodecanoyl-Lhomoserine lactone, which facilitates efficient cell-to-cell communication in this sessile growth mode. To confirm that the ΔAbHMwfabG mutant was responsible for pellicle formation, we sequenced the complete genome of the ΔAbHMwfabG mutant and did not identify any mutations (in addition to the target gene) that are likely to impact A. baumannii. The only notable mutation we identified included a premature stop codon in a hypothetical gene (ACX60_RS15975). However, upon further examination of other A. baumannii genomes, this gene was found to readily accumulate premature stop codons (e.g. in A. baumannii strain AB0057). Hence, a contribution to the phenotypes reported is highly unlikely. Overall, the biological advantage of A. baumannii possessing a secondary HMwFabG enzyme is currently unknown. In Mycobacterium it is hypothesized that this enzyme is required for survival in unfavourable conditions 18 . Upregulation of MtFabG4 was observed when cells were treated with the aminoglycoside, streptomycin 57 . Interestingly, RNA sequencing of colistin treated A. baumannii revealed expression of HMwFabG was upregulated by 5.28-and 4.0-fold (log 2 ) after 15 min and 1 h shock treatments, respectively 32 . Furthermore, transcription of the low molecular weight fabG was downregulated 3.73-and 3.03-fold after 15 min and 1 h colistin shock treatments, respectively 32 . These findings suggest that A. baumannii has the capacity to switch between FabG homologs, which is dependent on environmental factors. However, drug resistance analyses of ATCC 17978 and the ∆AbHMwfabG mutant by defining the minimal inhibitory concentration to colistin, chloramphenicol, gentamicin or streptomycin did not reveal changes greater than two-fold (Supplementary Table 6).
Recent advances in the development of inhibitors against MtFabG4 are reviewed comprehensively by Dutta 18 . Successful inhibitors targeting MtFabG4 NADH binding sites include triazole linked polyphenol or polyphenolaminobenzene hybrids 7,8 . Dual inhibitors of MtFabG4 and HtdX, another enzyme that belongs to the same operon in M. tuberculosis and is presumed to play a role in fatty acid metabolism, bind at catalytic loops 7,8 . A specifically designed synthetic inhibitor of MtFabG4, S-S006-830, found two putative binding regions; the first near the substrate/coenzyme binding sites in the C-terminal domain and the other at a smaller pocket on the N-terminal domain 58 . Given the structural similarity between MtFabG4 and HMwFabG of A. baumannii, these inhibitors may also provide a new opportunity to treat drug-resistant A. baumannii and should be examined. The type of inhibitors designed would likely determine whether these act as narrow or broad-spectrum therapeutics. Inhibitors that act within conserved binding site regions of FabG would likely inhibit all FabG isoforms, however, may also affect other bacterium including normal flora, an off-target effect similar to most antibiotics in current use. Inhibitors that specifically target one FabG isoform (for example a HMwFabG N-terminal domain-targeting drug) would likely be narrow spectrum but may allow A. baumannii to switch preference to another FabG. In this case, these inhibitors may be useful under specific environmental conditions, though the conditions for lifestyle dependent FabG switching remains largely unknown. Moreover, the use of dual cocktail inhibitors targeting individual FabG isoforms may also be a potential therapeutic option.

Conclusion
In this study, we investigated the FASII pathway of A. baumannii, providing insights into potential 3-oxoacyl-ACP reductase (FabG) homologs. Our bioinformatics analysis identified potential SDR candidates that harboured necessary catalytic residues and binding motifs and thus may function as FabG. We provide high quality crystal structures for three enzymes: a functional low molecular weight FabG, a putative FabG protein, AbSDR, and a HMwFabG. We have provided evidence that the HMwFabG of A. baumannii is required for optimal growth in nutrient rich conditions, motility and pellicle biogenesis. The exact function and pathways involved in HMw-FabG activity remains elusive but may contribute to its survival and success as a pathogen. A. baumannii is a pathogen of extreme concern in healthcare institutions where nosocomial infections are increasingly difficult to treat due to antimicrobial resistance. Specific targeting of the FabG or HMwFabG enzymes in Acinetobacter may be a promising new focus for antimicrobial design.