The genome of a Bacteroidetes inhabitant of the human gut encodes a structurally distinct enoyl-acyl carrier protein reductase (FabI)

Enoyl-acyl carrier protein reductase (FabI) catalyzes a rate-controlling step in bacterial fatty-acid synthesis and is a target for antibacterial drug development. A phylogenetic analysis shows that FabIs fall into four divergent clades. Members of clades 1–3 have been structurally and biochemically characterized, but the fourth clade, found in members of phylum Bacteroidetes, is uncharacterized. Here, we identified the unique structure and conformational changes that distinguish clade 4 FabIs. Alistipes finegoldii is a prototypical Bacteroidetes inhabitant of the gut microbiome. We found that A. finegoldii FabI (AfFabI) displays cooperative kinetics and uses NADH as a cofactor, and its crystal structure at 1.72 Å resolution showed that it adopts a Rossmann fold as do other characterized FabIs. It also disclosed a carboxyl-terminal extension that forms a helix–helix interaction that links the protomers as a unique feature of AfFabI. An AfFabI·NADH crystal structure at 1.86 Å resolution revealed that this feature undergoes a large conformational change to participate in covering the NADH-binding pocket and establishing the water channels that connect the active site to the central water well. Progressive deletion of these interactions led to catalytically compromised proteins that fail to bind NADH. This unique conformational change imparted a distinct shape to the AfFabI active site that renders it refractory to a FabI drug that targets clade 1 and 3 pathogens. We conclude that the clade 4 FabI, found in the Bacteroidetes inhabitants of the gut, have several structural features and conformational transitions that distinguish them from other bacterial FabIs.

Bacterial fatty acid biosynthesis (FASII 3 ) consists of a conserved collection of individual proteins that produce the fatty acids required for membrane phospholipid biogenesis (1). The FASII substrates are linked to acyl carrier protein (ACP) 3 and undergo successive rounds of condensation, reduction, dehydration, and reduction to extend the acyl chain by two carbons with each elongation cycle (Fig. 1A). The final reduction step of each elongation cycle is catalyzed by enoyl-ACP reductase (ENR) (Fig. 1A). There are two widely distributed, distinct ENR protein families, FabI and FabK (2)(3)(4)(5). FabI proteins use NAD(P)H as the reductant and belong to the short-chain dehydrogenase superfamily of enzymes that use a Rossmann fold to bind the nucleotide cofactor (6,7). Most short-chain dehydrogenases have an active site dyad consisting of Tyr-Xaa 3 -Lys, but the bacterial FabI subfamily is distinguished by having a differently spaced catalytic dyad consisting of Tyr-Xaa 6 -Lys (8,9). FabK is a flavoprotein (FMN) that belongs to the NAD(P)H-dependent flavin oxidoreductase family (4). FabK adopts an overall TIM barrel fold (4,5) and requires NAD(P)H as a reductant (2), but how the NADP(P)H and FMN cofactors collaborate in catalysis is not clear (3,5). There are also two FabI-like ENRs (FabL and FabV) with a limited distribution in bacteria (10 -14). These two reductases also belong to the short-chain dehydrogenase superfamily but are distantly related to FabI. The function of ENR in the FASII pathway is to pull each cycle of elongation to completion and is the rate-determining step of fatty acid elongation (15).
The role of ENR as a pacemaker of the essential FASII pathway has made it the subject of intense investigation as a target for antibiotic drug discovery (16,17). AFN-1252 (afabicin (Debio 1452)) is the most clinically advanced of these compounds and is designed to specifically target staphylococcal FabI (18 -21). Most commonly used, broad-spectrum antibiotics not only attack the pathogen but also devastate the commensal gut microbiome, leading to multiple complications (22,23). As a pathogen-specific antibiotic (16), AFN-1252 has the desirable property of having no effect on the size or composition of the gut microbiome during therapy (24). Firmicutes and Bacteroidetes are the two most abundant bacterial phyla in the gut microbiome. Most Firmicutes (85%) are in Class Clostridia that encode fabK and not fabI (25), therefore these organisms are expected to be refractory to growth inhibition by FabI-targeted drugs (2,3). The Bacteroidetes occupants of the gut microbiome are more diverse. Some contain only a fabI homo-log, some contain only a fabK homolog, and some contain both a fabI and fabK (Fig. 1B). This bioinformatic analysis suggests that some commensal Bacteroidetes may be susceptible to FabI-targeted therapeutics. We constructed a phylogenetic tree to understand the evolutionary relationships between the FabIs expressed in bacteria (Fig. 1C). This analysis shows that there are four distinct FabI clades. Representative FabIs from clades 1-3 have been biochemically and structurally characterized.
Bacteroidetes/Chlorobi phyla are in the fourth FabI clade, suggesting that Bacteroidetes FabI may have unique features not found in structures from the other clades.
The goal of this study was to structurally and functionally characterize the FabI from Alistipes finegoldii (AfFabI), a human Bacteroidetes commensal anaerobe containing a clade 4 FabI. Like other FabIs, AfFabI is a tetramer and adopts a Rossmann fold to bind the nucleotide cofactor. The unique feature of AfFabI is the carboxyl-terminal ␣9 helix that forms a coiledcoil structure with the ␣9 helix of an adjacent protomer to form a protomer-protomer contact that is absent in the other three FabI clades. Upon NADH binding, the intertwined ␣9 -␣9 helices unravel to form a structured loop that is involved in sealing the lid over the active site of the opposite protomer. Deletion of the unique carboxyl-terminal domain results in a folded, but inactive enzyme because of severely compromised NADH affinity. The AfFabI active site uses the same Tyr-Xaa 6 -Lys catalytic dyad as other FabIs, but the surrounding residues create a unique active site environment that renders AfFabI refractory to a FabI therapeutic (AFN-1252) that effectively targets the clade 1 and 3 FabIs.

Bioinformatics
The distribution of species in Bacteroidetes taxa encoding FabI, FabK, or both, was determined by counting the high homology FabI and FabK TBLASTN hits for each unique species entry in the RefSeq Representative Genomes database (12/22/ 2019) using the Alistipes FabI and FabK sequences and an e-value cut-off of 1e-100 for each hit. Environmental Bacteroidetes species encode for FabI only, FabK only, or both (not shown). Likewise, the individual species in the five major human-associated Bacteroidetes genera (Alistipes, Bacteroides, Parabacteroides, Porphyromonas, and Prevotella) encode for one of the three possible combinations of FabI and FabK (Fig.  1B). Prevotella, a prominent genus of the gut microbiome, had the highest number of species that encode only a FabI, but all three possible combinations were found in the individual species from all the commensal Bacteroidetes genera. A phylogenetic analysis was conducted to understand the evolutionary relationships between the FabIs (Fig. 1C). Clade 1 FabIs are encoded in the Proteobacteria and Firmicute phyla and are represented by the prototypical Escherichia coli (26) and Staphylococcus aureus (27) crystal structures. Clade 2 are the mycobacterial enzymes typified by the structure of InhA of Mycobacterium tuberculosis (28). Clade 3 FabIs consist of the plastid FabIs and include the structurally characterized FabIs of Chlamydia trachomatis (29) and Plasmodium falciparum (30). The FabI sequences from the closely related Bacteroidetes and Chlorobi phyla form a distinct fourth clade (Fig. 1C). The phylogenetic relationship between the various Bacteroidetes/Chlorobi FabI sequences is in overall agreement with the phylogenetic relationships between the organisms based on their 16S RNA sequences (31). These data show that Bacteroidetes/Chlorobi clade 4 FabIs have a divergent evolutionary history from the other clades, suggesting that there may be structural and/or functional differences between clade 4 FabIs and the other three characterized clades.

A. finegoldii FabI and FabK
We selected A. finegoldii as a prototypical gut commensal bacterium to examine the properties of the clade 4 FabIs. A. finegoldii is predicted to have both a FabI and FabK. A. finegoldii FabI (AfFabI) (UniProt ID: A0A174E195) has 32% identity with Neisseria gonorrheae FabI. The A. finegoldii FabK (AfFabK) (UniProt ID: I3YI65) has 42% identity with Strepto-coccus pneumoniae FabK and contains the signature FMNbinding motif. The ENR functions of AfFabI and AfFabK were verified by determining whether plasmids directing their synthesis would complement E. coli strain JP1111 (fabI(Ts)) ( Fig. 2,  A and B). The positive controls for the experiment were strain JP1111 carrying plasmids expressing E. coli FabI (EcFabI), C. trachomatis FabI (CtFabI), or Clostridium acetobutylicum FabK (CaFabK), and the negative control was the expression vector lacking a gene insert. All strains grew at the permissive temperature (30°C) (Fig. 2A). Strain JP1111 containing the empty vector failed to grow at the nonpermissive temperature (42°C), whereas the plasmids expressing AfFabI, AfFabK, and all the positive controls grew at 42°C (Fig. 2B). These data show that A. finegoldii encodes both a FabI and FabK ENR as predicted from the bioinformatic analyses.
AFN-1252 is a FabI inhibitor that is known to inhibit clade 1 and clade 3 FabIs (29,32,33). The sensitivity of AfFabI (clade 4) to AFN-1252 was compared with other FabIs using E. coli strain ANS1 (⌬tolC) expressing a series of FabIs as described previously (33). Strain ANS1 was transformed with a series of plasmids expressing different FabI and FabK enzymes and the AFN-1252 minimal inhibitory concentrations were determined using a microbroth dilution assay (Fig. 2C). Growth of strains ANS1/pEcFabI (clade 1) and ANS1/pCtFabI (clade 3) were both inhibited by AFN-1252 whereas all strains expressing a FabK were refractory to AFN-1252 growth inhibition. A key result was that strain ANS1/pAfFabI was refractory to AFN-1252 inhibition indicating that clade 4 FabIs may have an active site that is distinct from the clade 1 and clade 3 proteins.

Biochemical properties of AfFabI
An amino-terminal His-tagged version of AfFabI was expressed in E. coli and purified by affinity and gel filtration chromatography to obtain a homogeneous 34-kDa protein based on SDS gel electrophoresis (Fig. 3A, inset). AfFabI eluted as a single species on the calibrated XBridge BEH SEC column (Fig. 3A). Its Stokes radius was consistent with a molecular weight of 145 kDa (Fig. 3A, inset), indicating that like other characterized FabI proteins AfFabI exists as a tetramer (theoretical molecular weight 136 kDa) (7). Sedimentation velocity analysis confirmed AfFabI exists as a homotetramer in solution with an s 20 value of 6.59 S corresponding to 141 kDa protein (Fig. 3B). AfFabI exhibited high-affinity, cooperative binding to NADH. The K D was estimated by surface plasmon resonance to be 225 nM with a Hill coefficient of 1.6 (Fig. 3C). The AfFabI affinity for its reaction product NAD ϩ was estimated in similar surface plasmon resonance experiments to be orders of magnitude lower (K D ϭ 1.623 Ϯ 0.008 mM) than for NADH (not shown). NADH increased the stability of AfFabI to thermal denaturation by 7°C (Fig. 3D). These data indicated that NADH binding results in a more stable protein structure.
AfFabI enzymatic activity of NADH oxidation to NAD ϩ was monitored by spectrophotometry at 340 nm. AfFabI catalyzed the oxidation of NADH to NAD ϩ in the presence of crotonyl-CoA (Fig. 4A). The NADH apparent K 0.5 was 22.3 Ϯ 4.6 M at 1.25 mM crotonyl-CoA and NADH exhibited positive cooperative behavior with a Hill number of 2.4 Ϯ 0.2 (Fig. 4B). NADPH did not support the reaction (Fig. 4A, inset). Cooperative bind-

Bacteroidetes FabI structure
ing of the reduced nicotinamide substrate is characteristic of FabIs and has been studied in detail with S. aureus FabI (34). The crotonyl-CoA K 0.5 was 457 Ϯ 45 M and exhibited some cooperative behavior with a Hill number of 1.5 (Fig. 4C). Mass spectrometry analysis confirmed that butyryl-CoA was the product in the enzymatic reaction (Fig. 4C, inset). These data demonstrate AfFabI catalyzes the FASII ENR reaction and that AfFabI exhibits kinetic properties that are like other characterized FabI enzymes. Although triclosan inhibited AfFabI with an IC 50 of 1.76 Ϯ 0.57 M, which was similar to triclosan inhibition of EcFabI (35), AfFabI was refractory to AFN-1252 inhibition (Fig. 4D), suggesting its active site was significantly different from the active sites in the clade 1 and clade 3 FabIs that are sensitive to this drug.

The AfFabI crystal structure
The 1.72 Å AfFabI structure was refined to R/R free 20.6/24.5 ( Table 1). Two protomers were in the asymmetric unit and are similar to each other with a C␣ RMSD of 0.34 Å across 260 residues. The protomers interact with two symmetry-related protomers around a crystallographic 2-fold axis to create the tetramer. The presence of tetramers in the crystal is consistent with the gel filtration and sedimentation experiments. The unique AfFabI protomer feature is ␣9, a carboxyl-terminal helix that protrudes from the ␣/␤ core structure that is absent in the clade 1-3 enzymes (Fig. 5A). Structural alignment of AfFabI protomer (lacking ␣9) with EcFabI (PDB ID: 5CFZ) (36,81) showed the two structures are near identical with a C␣ RMSD of 1.48 Å across 448 residues. Like the clade 1-3 FabIs (7), the AfFabI protomer consists of a seven-stranded parallel ␤-sheet flanked by three ␣ helices on either side that adopts a Rossmann fold for dinucleotide substrate binding (Fig. 5A). The signature Tyr-Xaa 6 -Lys FabI catalytic dyad places the active site in the same location as other FabIs (8,9). FabI is a dimer of dimers with three molecular 2-fold noncrystallographic symmetry axes (P, Q, and R) along the subunit interfaces (27,38,39) (Fig. 5B). Each protomer has a total surface area of ϳ13,400 Å 2 , and ϳ4800 Å 2 (ϳ36%) is buried upon tetramerization. These values are similar to other FabIs (28, 36, 40 -42, 81). There are two disordered regions in the structure. The residues constituting the putative ␣6 active site lid (Thr-197 to Gly-209) were expected to be absent from the model because the flexible lid is only resolved in FabI⅐NAD(P)H binary complexes in other FabI clades (27,43,44). The second unstructured region is the final 13 amino acids of the carboxyl terminus.
FabI tetramers are stabilized by extensive intermolecular interactions along the protomer interfaces with conserved dimerization and tetramerization interfaces (27,38,45). Like clade 1-3 FabIs, AfFabI dimerization occurs along the P axis interface and consists of interacting helices ␣8 and strands ␤7 linking the A/B or C/D protomers (Fig. 6A). The AfFabI tetramerization domain links protomers along the Q axis interface and consists of an antiparallel four-helical bundle composed of interacting helices ␣4 (kinked) and ␣5 linking the A/C or B/D protomers (Fig. 6B). AfFabI ␣9 -␣9 interaction domain is a unique feature of clade 4 FabIs and links the A/D or B/C protomers along the R axis (Fig. 6C). Helix ␣9 extends from Ser-261 to Glu-275 and is amphipathic with the hydrophobic

Bacteroidetes FabI structure
side chains forming the ␣9 -␣9 contact interface and the polar side chains extending into solvent.

The AfFabI⅐NADH binary complex
The 1.86 Å AfFabI⅐NADH complex structure was refined to R/R free 15.8/19.7 ( Table 1). The asymmetric unit contained one homotetramer. The protomers are nearly identical with a C␣ RMSD of 0.13 Å across 282 residues and the NADH molecules were well-resolved in all four active sites. The planarity of the cofactor carboxamide relative to its attached pyridine ring is indicative of the NADH redox state (46). The C2N-C3N-C7N-O7N torsion angles in the NADH molecules were between 150.8°and 162.5°and are consistent with reduced NADH being present in the crystals. NADH binding increases the overall protomer surface area to ϳ15,000 Å 2 , and the buried surface area of the subunit interfaces to ϳ5400 Å 2 (ϳ36%). The active site lid containing ␣6 was not resolved in the AfFabI structure but was clearly seen in the AfFabI⅐NADH electron density map. The lid forms several hydrogen bond interactions with the pyrophosphate and nicotinamide of NADH (Fig. 7A). Nucleo-

Bacteroidetes FabI structure
tide binding proteins often stabilize pyrophosphate binding using a positive helix dipole (47). Helix ␣1 creates this helix dipole in AfFabI and hydrogen bonds between NADH and lid residues Thr-199 side chain and Ala-201 backbone amide fix the cofactor in space. Lys-167 on ␣5 is part of the Tyr-Xaa 6 -Lys catalytic dyad and donates two hydrogen bonds to the nicotinamide ribose to position and stabilize the nicotinamide (Fig.  7A). In addition, the nicotinamide moiety is also stabilized by hydrogen bonds that form between lid residue Thr-197 and the nicotinamide ring. The adenine moiety sits in a solvent-ex-posed hydrophobic pocket in the active site cavity. Hydrogen bonds between the adenine nitrogen N-1 and Ala-67 backbone carbonyl, and the adenine ribose and Asn-41 and Ser-45 side chains orient the adenine moiety in this cavity. The locations and interactions between AfFabI and NADH are very similar to other FabI proteins (48 -51).
The NADH-induced conformational change creates four water-filled channels that connect the active sites to a central water well located in the core of the tetramer (Fig. 7B). These water channels enable efficient proton transfer to the active

Bacteroidetes FabI structure
sites for catalysis (52). The resolution of the AfFabI⅐NADH crystal structure enabled the visualization of ordered water molecules within the individual channels making hydrogen bond interactions with the peptide backbone (Fig. 7C). The individual water channels exit the active sites, narrow into tunnels that are formed by residues along the R axis of the promoter interfaces and empty into the water well at the center of the protein (Fig. 7C). The active site cavity is hydrated in the AfFabI crystal structure, and few ordered water molecules are detected because the water channels are only partially formed and the active site lid is open to solvent. As with other FabIs (52), the lid isolates the active site from bulk solvent to create an environment that promotes hydride transfer from NADH.

Helix ␣9 and the carboxyl terminus
A unique feature of AfFabI is how the clade 4 -specific ␣9 helix and disordered carboxyl-terminal tail participate in the conformational change required for NADH binding (Fig. 8). In the example shown, helix ␣9 C and ␣9 B intertwine to connect the C/B partner protomers in AfFabI and the 13 residues of the protein attached to carboxyl terminus of ␣9 are disordered (Fig.  8A). In the AfFabI⅐NADH structure, the interacting ␣9 helices unwind, and along with 10 of the 13 disordered carboxyl-terminal amino acids (Val-277 to Glu-286), form an ordered loop feature that extends to the active site lid of the partner protomer (Fig. 8B). The N terminus of ␣7 lies at the base of the ␣9 -␣9 interaction domain and forms the corner of the active site cavity. The conformational change relocates the C␣ chain of ␣9 up to 20 Å toward the opposite active site and drives the formation of a new protomer-protomer interaction interface that closes the active site and creates the water channels.
The active site is a large solvent-exposed crevice in the AfFabI structure (Fig. 9A). In this conformation, Asn-213 C and Asn-217 C from the last two turns of ␣7 C form hydrogen bonds with Tyr-268 B from ␣9 B . Upon NADH binding, the lid consisting of ␣6 and adjacent residues closes over NADH to convert the solvent-exposed crevice into an enclosed active site (Fig.  9B). Asp-210 C from ␣7 C forms a hydrogen bond with Thr-267 B from ␣9 B (Fig. 9B) as the unwound ␣9 B residues slide along ␣7 C . Lid closure leaves the adenine moiety surface-exposed and buries the pyrophosphate and nicotinamide regions of NADH. The solvent-accessible surface area of the active site crevice decreases from ϳ500 Å 2 to ϳ300 Å 2 . Every residue in the first turn of ␣9 in AfFabI forms a new hydrogen bond interaction in the AfFabI⅐NADH complex (Fig. 9C). The Arg-262 B backbone amide nitrogen forms a hydrogen bond interaction with the Tyr-179 A side chain from ␣5 A in the tetramerization domain. The Arg-263 B side chain forms hydrogen bond interactions with backbone carbonyls from Ser-219 B , Glu-216 B , and Asn-217 B from ␣7 B in the corner of the protomer B active site. The Ala-264 B backbone amide forms a hydrogen bond interaction with the Ser-261 B side chain at the base of ␣9 B , and the backbone carbonyls from Arg-263 B and Met-265 B form hydrogen bond interactions with the Lys-266 C side chain from ␣9 C . Two  A, the AfFabI monomer contains a Rossmann fold. The polypeptide chain is colored from blue (N terminus) to red (carboxyl terminus). The active site lid (␣6) and terminal 13 residues are depicted as disordered (broken dashes) because these residues were not resolved in the electron density map. B, AfFabI is a tetramer with three perpendicular 2-fold noncrystallographic symmetry axes P, Q, and R.

Bacteroidetes FabI structure
residues in the second turn of ␣9 form new hydrogen bond interactions in AfFabI⅐NADH. In addition to the Thr-267 B -Asp-210 C interaction shown in Fig. 9B, the side chain of Lys-266 B forms hydrogen bonds with Arg-263 C and Met-265 C from ␣9 C (Fig. 9D). There are no protein-protein interactions formed between the loop residues derived from the third turn of ␣9 in AfFabI and the protomers in the AfFabI⅐NADH structure. This loop extends away from the surface as seen in Fig. 8B. Ten carboxyl-terminal-disordered residues after ␣9 B in the AfFabI structure become ordered to create a channel connecting the active site to the water well in the AfFabI⅐NADH structure (Fig. 9E). These interactions form "latches" along the R axis that create the surface enclosing a water channel. The first latch is formed by Glu-275 B backbone amide, and Asp-276 B side chain hydrogen bonds to Arg-183 A and Tyr-158 C side chains (Fig. 9E). Asp-276 B is oriented to coordinate a water molecule in the channel. Arg-262 B in the first turn of ␣9 B forms a planar stacking interaction (53) between the guanidinium group in the side chain and the aromatic ring of Tyr-158 C . A van der Waals contact distance is formed between lid residue Leu-206 C side chain and Glu-275 B backbone carbonyl. The second latch is formed by His-278 B backbone amide hydrogen bonding to the Thr-106 C and Asp-108 C side chains, and Gln-279 B backbone carbonyl and side chain hydrogen bonds to the Arg-134 A backbone carbonyl and side chain (Fig. 9E). A van der Waals contact is formed between the Val-277 C and Ile-180 A side chains (not shown in Fig. 9). The molecular latches convert a region of the tetramer that freely exchanges with bulk solvent in AfFabI into an enclosed channel that connects the active site with the central water well in AfFabI⅐NADH. These data indicate that the conformational changes associated with ␣9 and the carboxyl terminus are important for catalysis because they participate in the formation of the catalytic center and the water channel that connects the active site to the water well.

Function of the carboxyl terminus
We constructed a series of truncation mutants to test the role of ␣9 and the disordered carboxyl terminus in catalysis. All the truncated mutant enzymes had similar turnover numbers (V max /[E T ]) but had reduced catalytic efficiencies (V max /[E T ]/ K 0.5 ) that were driven by defects in the apparent K 0.5 for NADH ( Table 2). AfFabI  lacked the disordered carboxyl-terminal residues in the AfFabI structure and eliminated the interactions contributing to sealing the water channels in the AfFabI⅐NADH structure illustrated in Fig. 9E. AfFabI(1-273) was a tetramer that was stabilized by NADH, but the catalytic efficiency of the enzyme was reduced by 3-fold driven by an increase in the apparent K 0.5 for NADH (Table 2). Although AfFabI(1-273) was catalytically compromised, it was able to complement the growth of the fabI(Ts) mutant when overex-

Bacteroidetes FabI structure
pressed. AfFabI(1-269) removed four resides (1 turn) from ␣9 in the AfFabI structure. This region did not form any specific protein contacts in the AfFabI⅐NADH structure because it loops away from the protein and the side chains point toward solvent (Fig. 8B, arrows). Thus, the biochemical properties of AfFabI  were similar to the properties of AfFabI(1-273). The AfFabI(1-265) truncation removed the second turn of the ␣9 helix and the important interactions depicted in Fig. 9, B and D. These two interactions are key to closing one side of the lid that lays over the active site in the AfFabI⅐NADH structure and would be predicted to compromise NADH binding. AfFabI(1-265) was a thermally stable tetramer but was not stabilized by the addition of NADH ( Table 2). The apparent K 0.5 for NADH increased 6-fold and the catalytic efficiency dropped by 5.5-fold leading to a protein that could not complement the fabI(Ts) mutant even when overexpressed ( Table 2). AfFabI  removed the remaining turn of the ␣9 helix and the multiple interactions depicted in Fig. 9C. These interactions connect protomers A and C with protomer B and form a junction with the NADH lid. AfFabI(1-261) reduces AfFabI to the clade 1 structure. Like clade 1 enzymes it is stable and cooperative but had a 23-fold increase in the NADH apparent K 0.5 and was 22.5-fold less catalytically efficient ( Table 2). These data show the importance of ␣9 and the carboxyl terminus in creating the substrate binding lid and the water channels required for efficient catalysis.

Discussion
This work defines the function of the structural features that are unique to clade 4 FabIs. The four clades of bacterial FabIs diverged from their common ancestor in the distant past and each have a long and unique evolutionary history. All FabIs have common features like a Rossmann fold, cofactor-induced conformational change, Tyr-Xaa 6 -Lys catalytic dyad, and dimerization and tetramerization interfaces. These basic structural features are embodied in the prototypical clade 1 EcFabI structure depicted in Fig. 10, upper panel. Using the clade 1 enzymes as the prototypical comparator, each of the other three clades possesses a unique structural feature. Clade 2 FabIs are the Mycobacterium InhA enzymes that have a deeper active site crevice and longer ␣6 lid than the other FabI clades. These features are thought to enable InhA to bind very long-chain enoyl-ACP substrates for the synthesis of cell wall mycolic acids (28,54). Clade 3 FabIs have an insertion between ␤3 and ␣3 that creates peripheral loops that are related by the Q axis interface. A functional or structural role is yet to be described for this feature in the plastid FabIs. Clade 4 FabIs have an extension of the carboxyl terminus that provides a new protomerprotomer interaction surface that links protomers along the R axis. This study shows that the carboxyl-terminal extension is critical for catalysis in clade 4 FabIs by participating in the conformational change that creates the NADH-binding pockets and water channels that feed the active site. This unique feature is present in all the Bacteroidetes FabIs and has evolved into a domain that is essential for catalysis in clade 4 enzymes.
Further, our data explain why a potent antimicrobial drug that targets the FabI of clade 1 and three pathogens does not impact a major constituent of the microbiome. AFN-1252 is a S. aureus-selective (Fig. 10, lower panel) FabI inhibitor that was developed as a pathogen-specific drug (16,17). Treatment of mice with AFN-1252 does not perturb the Bacteroidetes phylum in the gut microbiome, whereas broad-spectrum antibiotics devastate the microbiome (24). Bioinformatic analysis of the microbiome enoyl-ACP reductase distribution indicates the presence of FabK in the Clostridia and commensal Bacteroidetes, and the expression of FabK would account for the resistance of organisms to AFN-1252. However, many Bacteroidetes genera only express a FabI (Fig. 1B). Biochemical analysis shows that clade 4 FabIs are resistant to AFN-1252, explaining why Bacteroidetes genera are not impacted by AFN-1252 therapy. AFN-1252 is a rigid small molecule that contains Figure 7. Active site and water channels in the AfFabI⅐NADH complex upon NADH binding. A, schematic diagram shows hydrogen bond interactions between the active site and NADH designated by purple lines. Lid residues that interact with NADH are shown by yellow highlight. B, molecular surface rendering of the central water well that connects to the four active sites was visualized in the AfFabI⅐NADH complex crystal structure along with the water channels that connect each active site with the water well using PyMOL. The molecular surface of the water channels and central water well are colored according to the contribution from each protomer. Segments of each water channel have more than one color because they are formed by the protomer interfaces along the R axis. The active sites in each protomer are designated with the letter A and the protomer superscript, and the letter W indicates the location of the central water well. C, each water channel contains ordered water molecules (red spheres) that form hydrogen bond interactions with each other and the peptide backbone. The line traces the chain of structured waters that begin at the carboxyamide of NADH in the active site (A C ), exit into a narrow tunnel formed by the A/B/C protomer interfaces along the R axis, and empty into the central water well (W).

Figure 8. Conformational change in helix ␣9 and the carboxyl terminus.
A, in the AfFabI crystal structure, the ␣9 C and ␣9 B helices are wound together, connecting the C and B protomers, and the last 13 residues of the protein are disordered. Broken dashes indicate the relative location of missing carboxyl terminus residues. B, in the AfFabI⅐NADH crystal structure, the ␣9 C -␣9 B interaction domain unwinds, and 10 of the 13 carboxyl-terminal residues form a structured loop feature that connects to the active site lid of the partner protomer. The arrows denote the location of residues (purple) in the ␣9 turn that are removed in the AfFabI(1-269) truncation mutant.

Bacteroidetes FabI structure
an oxotetrahydronaphthyridine and 3-methylbenzofuran moiety connected by a cis-amide. All three components of the drug form interactions with the protein in AFN-1252-sensitive FabIs in clades 1 and 3 (19,29). The AfFabI⅐NADH crystal structure shows that clade 4 FabIs have a differently shaped active site cavity than clade 1 and 3 enzymes (Fig 10, lower panel). In AfFabI, the lid residue Val-205 side chain occupies the space where the 3-methylbenzofuran moiety should reside and the Arg-102 side chain sterically clashes with the oxotetrahydronaphthyridine moiety. Similarly, modeling AFN-1252 in the clade 2 active site reveals a steric clash with Ile-202. These considerations suggest that FabI-targeted drugs against clade 1 or 3 pathogens would have the benefit of having little to no impact on the gut microbiome, sparing patients the complications that arise from current therapies that destroy the microbiome (55,56).

Materials
All chemicals and reagents were obtained from Sigma-Aldrich or Fisher unless otherwise indicated. Strains and plasmids used in this work are listed in Table 3.

Phylogeny
Predicted FabI sequences were collected from the NCBI Reference Sequence database to include characterized FabI sequences and their homologs. Maximum likelihood phylogenetic trees were constructed using the DECIPHER and phangorn packages in R (57,58). Briefly, sequences were aligned using the AlignSeqs and StaggerAlignment functions in DECIPHER. The LG ϩ ⌫ (4) ϩ I model (59) was best fitting by the Bayesian information criteria from the model. Test function in phangorn evaluating the WAG, JTT, LG, and Dayhoff amino acid replacement matrix models with and without gamma distributed rate variation among sites (⌫) and invariant sites (I). The initial neighbor-joining tree was constructed using the distance matrix, with the maximum likelihood tree generated from the neighbor-joining tree using the LG ϩ ⌫ (4) ϩ I model with stochastic branch rearrangement. Bootstrap method of 1000 replicate trees was used to determine the confidence of the tree topology. The maximum likelihood tree was visualized using ggtree (60).

Molecular biology
The predicted A. finegoldii FabI and FabK have been previously identified (61). Representative predicted A. finegoldii genes AffabI (UniProt ID: A0A174E195) and AffabK (UniProt ID: I3YI65) were optimized for gene expression in E. coli using GeneArt Gene Synthesis Technology (Life Technologies). An NdeI restriction site was engineered at the 5Ј-end of the gene with the start codon in the NdeI site, whereas a stop codon and an EcoRI restriction site were sequentially engineered at the 3Ј-end of the gene. The genes were cloned into the pPJ131 plasmid (a modified version of the pBluescript plasmid with the multiple cloning site from pET28a) via the NdeI and EcoRI (New England Biolabs) restriction sites (29,80). To create the AffabI carboxyl-terminal truncation mutant constructs, premature stop codons were engineered into the gene using the QuikChange II Site-Directed Mutagenesis Kit (Agilent) and primers 5Ј-CAAGCATGGGTATGTCATGATGACGTCGT-GCAATGAAAAC, 5Ј-TGTCACGTCGTGCAATGTGATG-AAAAACCTATGAAAAAGG, 5Ј-CAATGAAAACCTATG-AATGATGAAAAGGTATGCGCTTTGA, and 5Ј-ATGAAA-AAGGTATGCGCTGATGATTTGAAGATGTGCACCA and their reverse complements (The bold letters emphasize the stop codons that were engineered into the gene sequence). The amino acid numbering convention used assumes the initiating methionine of the native protein as residue 1. For protein purification, the AffabI constructs were cloned into the pET28b plasmid via the NcoI and EcoRI (New England Biolabs) restriction sites.

Minimum inhibitory concentration analysis
The minimum inhibitory concentrations for AFN-1252 and triclosan against E. coli strain ANS1 (⌬tolC) were determined using a broth microdilution method as described previously (62). ANS1 is a tolC knockout mutant that is used to eliminate the contribution of type 1 secretion systems to drug resistance (63). Briefly, pPJ131 expression plasmids containing nothing, EcFabI, CtFabI, AfFabI, CaFabK, or AfFabK were transformed into ANS1, and these strains were grown to A 600 ϭ 0.6 in LB before being backdiluted 1:30,000 in 1% DMSO in 1% tryptone media. Diluted cells (100 l) were added to each well of a U-bot-

AfFabI protein expression and purification
BL21 (DE3) cells harboring the pET-AfFabI plasmid were grown in LB medium with 100 g/ml carbenicillin at 37°C and 200 rpm shaking to A 600 ϭ 0.6. The culture was cooled to 16°C then induced with 1 mM isopropyl 1-thio-␤-D-galactopyranoside overnight. Cells were centrifuged and pellets were resuspended in 20 mM Tris-HCl, pH 8.0, 200 mM NaCl, 10 mM imidazole (30 ml/liter culture). Cells were lysed via a cell disruptor and the amino-terminal His 6 -tagged AfFabI was purified by standard nickel chelation chromatography (29). The protein was then gel filtered into 20 mM Tris-HCl, pH 8.0, 200 mM NaCl, 10 mM EDTA using a preparative Superdex 200 column with dimensions 16 mm ϫ 60 cm. The AfFabI molecular weight was estimated using an analytical XBridge BEH SEC 200 Å 3.5-m column with dimensions 7.8 mm ϫ 150 mm. Approximately 26 mg of purified AfFabI was obtained per liter of culture. The same methods were used to overexpress and purify AfFabI mutant proteins.

Analytical ultracentrifugation
Sedimentation velocity experiments were conducted in a ProteomeLab XL-I analytical ultracentrifuge (Beckman Coulter) following standard protocols unless mentioned otherwise (64,65). Samples in buffer containing 20 mM Tris-HCl, pH 8, 200 mM NaCl, 10 mM EDTA were loaded into cell assemblies comprised of double sector charcoal-filled centerpieces with a 12 mm path length and sapphire windows. The cell assemblies, containing identical sample and reference buffer volumes of 390 l, were placed in a rotor and the samples and cells were incubated at 20°C for 2 h before being accelerated from 0 to 50,000 rpm. Rayleigh interference optical data as well as absorbance data at 280 nm were collected at 1-min intervals for 12 h. The velocity data were modeled with diffusion-deconvoluted sedimentation coefficient distributions c(s) in SEDFIT (RRID: SCR_018365) using algebraic noise decomposition and with signal-average frictional ratio and meniscus position refined with nonlinear regression (66). The s-value was corrected for time and finite acceleration of the rotor was accounted for in the evaluation of Lamm equation solutions (67). Maximum entropy regularization was applied at a confidence level of p ϭ 0.68. The partial specific volumes of the proteins, based on its amino acid composition, was calculated in SEDFIT.

Enzymology
The AfFabI enzymatic activity was determined by measuring the conversion of NADH to NAD ϩ at 340 nm. The enzyme reactions were 100 l in volume and monitored in Costar UV half-area 96-well plates with a SPECTRAmax 340PC instru-ment taking 340-nm readings at 10-s intervals at 30°C. In these experiments, enzyme concentrations are reported relative to monomer concentration. The velocity of the AfFabI enzyme (100 nM) was measured by adding 1.25 mM crotonyl-CoA and 100 M NADH in 20 mM Tris-HCl, pH 8.0. The apparent K 0.5 of crotonyl-CoA was determined by adding 100 nM AfFabI to 200 M NADH and 0, 0.05, 0.1, 0.25, 0.5, 1, 1.5, 2, or 2.5 mM crotonyl-CoA. The apparent K 0.5 of NADH was determined by adding 100 nM AfFabI to 1.25 mM crotonyl-CoA and 0, 5, 10, 15, 20, 30, 40, 50, or 70 M NADH. The reactions were mixed for 10 s by the mix function on the plate reader, and data were acquired at 10-s intervals for 30 min. Initial velocity was calculated from the linear phase of the progress curve and fit using the Hill equation to determine the apparent K 0.5 (68

Surface plasmon resonance experiments
Surface plasmon resonance experiments were conducted at 20°C using a ForteBio Pioneer optical biosensor (ForteBio). Poly-His-tagged AfFabI constructs were immobilized on polycarboxylate hydrogel-coated gold chips preimmobilized with nitrilotriacetic acid (HisCap chips, ForteBio). The chip was primed in chelating buffer (10 mM HEPES, pH 7.4, 150 mM NaCl, 50 M EDTA, 0.005% Tween 20) and was preconditioned at 10 l/min with three 60-s injections of wash buffer (10 mM HEPES, pH 8.3, 150 mM NaCl, 350 mM EDTA, 0.05% Tween 20) and one 60-s injection of chelating buffer before being charged with a 60-s injection of 500 M NiCl 2 in chelating buffer. After priming into binding buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl, 0.01% Brij-35, 5% DMSO), FabIs were injected until ϳ1500 -1700 resonance units of protein were captured. One flow cell on the chip was charged with Ni 2ϩ without adding protein to be used as a reference cell.
NADH was prepared in binding buffer as a 2-fold dilution series with maximum concentration of 4 M for binding to AfFabI and was injected in triplicate for each concentration at a flow rate of 30 l/min. A series of buffer-only (blank) injections was included throughout the experiment to account for instrumental noise. NADH fully dissociated from the protein surfaces, eliminating the need for a regeneration step. The data were processed, double-referenced, microcalibrated, and ana-Bacteroidetes FabI structure lyzed using the software package Qdat (version 4.3.1.2, Forte-Bio). Equilibrium binding levels were determined and exported to GraphPad Prism for fitting to the Hill equation.

Protein thermal shift assays
Protein thermal shift analysis was conducted to investigate if NADH enhanced the thermal stability of the AfFabI constructs. Solutions (30 l) of AfFabI (10 M) and AfFabI (10 M) ϩ NADH (100 M) in 20 M HEPES, pH 7.5, 150 mM NaCl, 20 mM MgCl 2 , and 2.5ϫ Sypro Orange Dye were added to wells of ThermoGrid optically clear PCR plates (Denville Scientific). The plates were centrifuged at 1000 ϫ g for 5 min and then analyzed by the ABI 7300 real-time PCR system as described previously (29). The temperature was ramped from 25°C to 95°C at 1°C/min with the fluorescence read six times at each temperature ramp. The resulting data were fit to a Boltzmann sigmoidal equation to determine the melting point of each AfFabI construct with and without substrate. Each enzyme or enzyme and substrate condition was replicated six times, and the thermal melting point of each replicate was determined independently. The melting points from each replicate were averaged to determine the reported thermal melting point. A representative thermal shift experiment is shown in Figure 3D and average thermal melting points from triplicate experiments are shown in Table 2.

Mass spectrometry analysis of the AfFabI enzyme reaction
Samples were diluted with an equal volume of 80% acetonitrile ϩ 15 mM ammonium hydroxide and were analyzed by direct injection to a QTrap 4500 equipped with a Turbo V ion source (Sciex). The QTrap 4500 was operated in the positive mode using neutral loss scanning, and the ion source parameters were ion spray voltage, 5000 V; curtain gas, 15 psi; temperature, 275°C; collision gas, medium; ion source gas 1, 15 psi; ion source gas 2, 20 psi; neutral loss, 507.0 m/z; declustering potential, 50 V; and collision energy, 40 V. The system was controlled and analyzed by the Analyst® software (Sciex).

Crystallization and structure determination
The AfFabI protein was concentrated to 12 mg/ml for crystallization. Initial screening was performed at 20°C against the Protein Complex Suite (Qiagen) by hanging drop vapor diffusion method combining 300 nl protein and 200 nl precipitant. Diffraction quality crystals were obtained by combining 1.5 l protein and 1 l 1.2 M sodium potassium tartrate and 100 mM Tris-HCl, pH 8.0. Crystals were cryo-protected with 1.3 M sodium potassium tartrate, 100 mM Tris-HCl, pH 8.0, and 20% v/v glycerol, and then flash-frozen in liquid nitrogen for X-ray diffraction experiments. The diffraction datasets were collected at the SER-CAT beam line 22-ID at the Advanced Photon Source and processed using HKL2000 (69). The AfFabI structure was solved by the molecular replacement method using the program Phaser (70) and the coordinates of FabI from Aquifex aeolicus (PDB ID: 2P91) (71) as the search model. The AfFabI⅐NADH complex was achieved by incubating 5 mg/ml (150 M) AfFabI with 500 M NADH for 2 h at room temperature prior to crystallization. Crystals were grown by hanging drop vapor diffusion by combining 1.5 l of the protein⅐substrate mixture and 1.5 l 12.5% PEG 1000, 200 mM NaCl, and 100 mM MES, pH 6.0. Crystals were cryo-protected with 12.5% PEG 1000, 200 mM NaCl, 100 mM MES, pH 6.0, and 25% glycerol, and then flash-frozen in liquid nitrogen for X-ray analysis. Diffraction datasets were collected at the NSLS-II NYX beamline (19-ID) and AMX beamline (17-ID-1) at the Brookhaven National Laboratory and processed using XDS (72). The structure of the AfFabI⅐NADH complex was solved by molecular replacement using the AfFabI structure lacking ␣9 as a search model. The structures were completed by iterative rounds of refinement using REFMAC (73) and manual rebuilding using Coot (74). The refinement was monitored by following the R free value calculated from a random subset (5%) of omitted reflections. A tight turn resolved in the electron density map led to 0.2% of the atoms to be in the Ramachandran outliers (Asn-161 A in the AfFabI structure, and Asn-161 A , Asn-161 C in the AfFabI⅐NADH complex structure). A summary of the data processing and structure refinement statistics is provided in Table 1. The coordinates have been deposited in the Protein Data Bank (accession code 6VLX AfFabI; 6VLY AfFabI⅐NADH complex). The figures related to the protein structure were generated with PyMOL (75).

In silico analyses
C␣ RMSD calculations were performed using SSM Superpose in Coot (74). Surface areas were calculated in PyMOL as described previously (76). The buried surface area by dimerization was calculated by subtracting the surface area of two protomers linked along the P axis from the combined surface area of the two individual protomers. The buried surface by tetramerization was calculated by subtracting the surface area of two protomers linked along the Q axis from the combined surface area of the two individual protomers. The buried surface area from the helix-helix interaction domain was calculated by subtracting the surface area of two protomers linked along the R axis from the combined surface area of the two individual protomers. The water channels and water well were visualized using the PyMOL surface view in Cavities and Pockets (Culled) mode. AFN-1252 binding visualization across FabI clades was done by structural alignment of the catalytic dyads from clade I EcFabI⅐NAD ϩ ⅐AFN-1252 (PDB ID: 4JQC) (77), clade II MtInhA⅐NAD ϩ (PDB ID: 4DRE) (78,82), clade III CtFabI⅐NADH⅐AFN-1252 (PDB ID: 4Q9N) (29,83), and clade IV AfFabI⅐NAD ϩ (PDB ID: 6VLY).

Statistical analysis
Statistical analysis (i.e. standard error, K 0.5 ) and mathematical modeling (i.e. Hill equation, variable slope nonlinear regression) were performed using GraphPad Prism software version 8.2.1.

Data availability
Coordinates and structure factors for AfFabI and AfFabI⅐NADH crystal structures have been deposited in the Protein Data Bank (PDB) under accession codes 6VLX and 6VLY, respectively. All remaining data are contained within the article.