Three enzymes and one substrate; regulation of flux through the glyoxylate shunt in the opportunistic pathogen, Pseudomonas aeruginosa

The glyoxylate shunt bypasses the oxidative decarboxylation steps of the tricarboxylic acid (TCA) cycle, thereby conserving carbon skeletons for biosynthesis. The branchpoint between the TCA cycle and the glyoxylate shunt is therefore widely considered to be one of the most important junctions in the whole of microbial metabolism. In Escherichia coli, AceK-mediated phosphorylation and inactivation of the TCA cycle enzyme, isocitrate dehydrogenase (ICD), is necessary to redirect flux through the first enzyme of the glyoxylate shunt, isocitrate lyase (ICL). In contrast, Mycobacterial species lack AceK and employ a phosphorylation-insensitive isocitrate dehydrogenase (IDH) at the branchpoint. Flux partitioning here is controlled “rheostatically” through cross-activation of IDH by the product of ICL activity, glyoxylate. However, the opportunistic human pathogen, Pseudomonas aeruginosa, expresses IDH, ICD, ICL and AceK. Here, we present the structure, kinetics and regulation of each branchpoint enzyme. We show that flux partitioning is coordinated through reciprocal regulation of the enzymes involved, beautifully linking carbon flux with the availability of key gluconeogenic precursors in a way that cannot be extrapolated from an understanding of the branchpoint enzymes in other organisms.

Structure and regulation of P. aeruginosa ICL. At 531 amino acids, P. aeruginosa ICL is longer than E. coli ICL (434 amino acids) and shares only 27% amino acid sequence identity with the latter. It also lies on a distinct branch of the evolutionary tree for this enzyme (which also includes ICL from pathogens such as Burkholderia cepacia and Acinetobacter baumanii), with no structurally-characterised homologues to date ( Figure S2). To investigate this further, we solved the x-ray crystal structure of P. aeruginosa ICL to 1.9Å resolution, with Ca 2+ and glyoxylate embedded in the active site. Crystallographic statistics are given in Table S1. The presence of a Ca 2+ (which was coordinated with the glyoxylate ligand and three water molecules) was confirmed using Checkmymetal. In the crystal structure, there is one ICL polypeptide per asymmetric portion of the unit cell, which belonged to the I222 space group ( Figure 1A). These polypeptides are related by a crystallographic two-fold axis to yield a tetrameric structure ( Figure 1B) with extensive contacts made between each protomer ( Figure  S3). This is consistent with gel filtration/multi-angle light scattering (GFC/MALS) and analytical ultracentrifugation (AUC) data indicating that ICL behaves as a 231 kDa tetramer in solution ( Figure  S4). Each protomer in P. aeruginosa ICL is comprised of 17 α-helices and 14 β-strands arranged around a central TIM barrel-like core (secondary structure assignments are shown in Figure S5). However, the overall fold of each protomer is distinct when compared with ICL from E. coli and Mycobacterium tuberculosis ( Figure 1C). In particular, helices α1 and α2 of P. aeruginosa ICL are flipped as a unit by almost 180° relative to the axis of α3, and α6 is extended by an additional 4 helical turns. In addition, α13 and α14 extend away from the globular core of the protein, generating structural projections which give the tetramer a distinctly rugged, star-like profile. The most obvious difference though, is the presence of a relatively unstructured "head-domain" spanning the region between Ile272 and Ile306. The head domains of adjacent protomers form some of the more intimate contacts underpinning tetramer formation ( Figure 1B and Figure S3). Interestingly, ICL from the fungus Aspergillus nidulans also has a head-like domain, although this is rich in α-helices and does not appear to play a significant role in inter-protomer interactions 11 .
The catalytic core of the enzyme is conserved. The active site of ICL is comprised of two distinct parts; those residues associated with the β-strands and loops around the rim of the TIM barrel, and a flexible loop of structure between β4 and β5 12 . The latter contains the catalytic general base, Cys222 (P. aeruginosa numbering). In our structure, this β4-β5 loop points away from the active site. Presumably, in the presence of isocitrate, this loop swings inwards to cap the active site, as it does in ICL from M. tuberculosis 13 . The glyoxylate moiety in our structure is displaced by 6Å towards the β4-β5 loop, leaving it poised ready to exit the active site, although it remains anchored on the enzyme through coordination with the Ca 2+ ( Figure 1D).
Given its position at the TGB, it is unsurprising that the activity of P. aeruginosa ICL is modulated by certain metabolites. P. aeruginosa ICL activity was inhibited by oxaloacetate, pyruvate, succinate, phosphoenolpyruvate and CoA ( Figure 2). No activators were identified. Oxaloacetate (Ki 1.9 mM), pyruvate (Ki 2 mM) and CoA (Ki 1.2 mM) all inhibited the enzyme non-competitively ( Figure 2A) whereas the reaction product, succinate, inhibited uncompetitively (Ki' 0.6 mM) and phosphoenolpyruvate (PEP) displayed mixed inhibition ( Figure 2B). In common with other isocitrate lyases, the enzyme was also susceptible to uncompetitive inhibition by itaconic acid (Ki' 0.5 mM) and nitropropionic acid (Ki' 0.6 mM) 14,15 , and potent mixed inhibition by maleic acid ( Figure 2C). The latter may reflect the partial structural similarity between PEP and maleic acid.
Structure and regulation of P. aeruginosa IDH. In P. aeruginosa, two isocitrate dehydrogenases, IDH and ICD, compete with ICL for isocitrate. In solution, IDH (monomeric molecular mass 82 kDa) had an apparent molecular mass of 235 kDa (GFC/MALS) or 273 kDa (AUC), suggesting that it adopts a higherorder structure ( Figure S6). To investigate the enzyme further, we solved its x-ray crystal structure to 2.7Å resolution ( Table S1). The asymmetric unit comprised two molecules of IDH; one (designated chain A) contained bound α-ketoglutarate and NADP, and the other (chain B) contained no bound substrate/product ( Figure 3A). However, the interface between these monomers was small and analysis using PISA 16 suggested that the two molecules cannot be considered to be protomers of a functional dimer. Comparison of the three-dimensional structure of the P. aeruginosa A and B chains revealed large conformational differences, especially in the smaller domain of each chain ( Figure S7), suggesting that catalysis is accompanied by structural rearrangements.
The α-ketoglutarate binding cleft is sandwiched between the two IDH domains ( Figure S7A), with side chains from both domains involved in binding. The active site residues are remarkably well-conserved between IDH and ICD ( Figure 3B). This is consistent with the proposed common evolutionary origins of both proteins 17 , and raises the question of why IDH is not a substrate for AceK ( Figure 3C). Comparison of the P. aeruginosa IDH active site architecture with that of ICD in the AceK-ICD complex from E. coli (PDB; 3LCB) reveals that in spite of the conserved constellation of catalytic residues, the active sites are differently-structured, with the two ICD motifs known to be critical for recognition by AceK (designated the P-loop and AceK recognition segment (ARS)) absent in IDH ( Figure 3D).
Glyoxylate was recently reported to be the principal regulator of IDH in M. smegmatis, enabling "rheostatic" control of flux through the glyoxylate shunt 7 . However, and although glyoxylate did stimulate P. aeruginosa IDH activity, pyruvate and oxaloacetate were far more potent regulators. All three compounds changed the isocitrate dependency of IDH kinetics from hyperbolic to sigmoidal ( Figure 4A), and all had a pronounced effect on kcat with only a small impact on KM ( Table I), indicating that the enzyme is of the rarer V-type allosteric class. The shift to sigmoidal isocitrate kinetics in the presence of these activators suggests cooperativity in substrate binding, consistent with IDH adopting a higher-order structure in solution. However, these effectors had much less impact on the NADPdependency of IDH kinetics, which remained essentially hyperbolic ( Figure 4B). In common with previously characterised isocitrate dehydrogenases, IDH (and also, ICD) was strongly inhibited by mixtures containing oxaloacetate and glyoxylate, presumably due to the non-enzymatic formation of oxalomalate 18 .
Structure and regulation of ICD. The second isocitrate dehydrogenase encoded by P. aeruginosa is ICD. P. aeruginosa ICD crystallised as a dimer ( Figure S8, Table S1) and also behaved as a dimer in solution ( Figure S9). The structure, solved to 2.7Å resolution, was very similar to that reported for the E. coli enzyme ( Figure S8B), with a clasp-like structure mediating the dimerization 19 . The residues lining the substrate binding site were also conserved between the E. coli and P. aeruginosa enzymes ( Figure  S8C). ICD was competitively inhibited by the reaction product, α-ketoglutarate (Ki 666 µM, Figure  S8D). However, the main mode of regulation was through AceK-dependent phosphorylation ( Figure  5A,B). Interestingly, in all of our experiments with AceK and ICD from P. aeruginosa (hereafter, AceKPA and ICDPA), the maximal inhibition attributable to AceK-mediated phosphorylation was only around 75% ( Figure 5B). This is not due to a lower intrinsic activity of AceKPA, because AceKPA was able to almost completely inactivate ICD from E. coli (ICDEC) ( Figure 5B). Comparison of the ICDPA and ICDEC active site region revealed slight differences in the spatial disposition of the P-loop required for AceK recognition ( Figure S8E), making it formally possible that ICDPA is less sensitive to AceK-mediated phosphorylation. However, AceKEC was able to completely inactivate ICDPA, indicating that these local differences in P-loop conformation are unlikely to be functionally significant ( Figure 5B). Taken together, these data suggest that in the absence of external factors to stimulate its kinase activity, AceKPA cannot fully inhibit ICDPA.
Although the P. aeruginosa and E. coli ICD enzymes are 79% identical at the amino acid sequence level, and the site of phosphorylation (Ser115 in P. aeruginosa ICD) is conserved, the corresponding AceK orthologues show substantial differences. In particular, the N-terminal regulatory domain is just 35% identical between the two species. In contrast, the C-terminal catalytic domain is 60% identical and retains all the sequence motifs thought to be important for kinase and phosphatase activity. These data suggest that P. aeruginosa AceK is likely to be regulated differently compared with the E. coli orthologue. To investigate this further, we examined how a panel of potential regulators affected AceKPA-dependent inactivation of ICDPA ( Figure S10). ICD was first phosphorylated by AceK for 60 min (enough to maximally inactivate the dehydrogenase ( Figure 5B)). Following this, the indicated potential regulators were added and restoration of isocitrate dehydrogenase activity was monitored after 30 min ( Figure S10A). As a control, and to confirm the species specificity of each regulator, we also examined (i) whether the same compounds affected AceKEC-dependent restoration of ICDEC activity ( Figure S10B) and (ii) whether the regulators had any intrinsic impact on ICD activity ( Figure  S11). With the exception of citrate (which presumably competitively blocks the ICD active site) and to a lesser extent, glyceraldehyde-3-phosphate and phosphoenolpyruvate, none of the tested potential regulators affected ICD activity directly. The phosphatase activity of AceKEC was stimulated by αketoglutarate, glyceraldehyde 3-phosphate, 2-keto-3-deoxyphosphogluconate (KDPG), pyruvate and oxaloacetate, and to a much lesser extent by fumarate, AMP and ADP ( Figure S10B). The phosphatase activity of AceKPA was stimulated by an overlapping, but larger number of regulators, including αketoglutarate, fructose 1,6-bisphosphate, glyceraldehyde 3-phosphate, glyoxylate, AMP, ADP, glycolate, pyruvate and oxaloacetate. Intriguingly, fumarate was the most potent activator of the phosphatase in AceKPA, yet this compound only weakly stimulated the phosphatase activity of AceKEC. Similarly, succinate had no apparent effect on AceKEC but strongly stimulated AceKPA phosphatase activity. In contrast, the Entner-Doudoroff pathway intermediate, KDPG, only marginally stimulated AceKPA phosphatase activity, yet this compound was one of the more potent activators of the AceKEC phosphatase. Perhaps the most noticeable regulator was acetyl-CoA. This compound weakly stimulated the phosphatase activity of AceKEC, whereas in in AceKPA, it appeared to stimulate the kinase activity. To investigate this further, we conducted a time-course analysis of ICDPA activity following AceKPA treatment in the presence of acetyl-CoA, and fumarate ( Figure 5C). What is immediately apparent is that, whereas fumarate maximally stimulates AceKPA phosphatase activity, acetyl-CoA maximally activates the AceK kinase activity, allowing the enzyme to inactivate >75% of the ICD, and at a faster rate than AceK alone.
Discussion. The driver behind this work is that P. aeruginosa exhibits a particular predilection for catabolising fatty acids, especially during infection scenarios 20 . Under these conditions, the glyoxylate shunt becomes centrally-important for growth. Indeed, a mutant of P. aeruginosa in which ICL and malate synthase (MS 21 ), are absent is cleared from a mouse pulmonary infection model within 48 hours 22 , indicating that the shunt is an excellent target for the development of adjuvant interventions. The physiological importance of the shunt in infection may not solely be due to the metabolic defect either; aceA mutants also show pronounced defects in virulence [23][24][25] . Consequently, therapeutic agents which inhibit flux through the glyoxylate shunt (or redirect flux away from the shunt) could potentially deliver a powerful "double whammy" by eliciting both metabolic insufficiency and reduced virulence. However, a better understanding of the enzymology of the P. aeruginosa TGB will be essential if we are to realise its full potential as a therapeutic target. Indeed, in the current work, we show that this understanding cannot be gleaned by extrapolation from previously-characterised species. For example, in P. aeruginosa, three enzymes compete for isocitrate, and ICL has a higher affinity for the substrate than the competing dehydrogenases. Also, the enzymology appears to be more complex in P. aeruginosa compared with other species, involving both AceK (absent in e.g., M. smegmatis and M. tuberculosis) and IDH (absent in many enterics, such as E. coli).
The key regulators in the P. aeruginosa system are oxaloacetate and pyruvate, which reciprocally regulate ICL and IDH, thereby elegantly coordinating metabolic flux partitioning between the TCA cycle and glyoxylate shunt. When these compounds are abundant (signalling to the cell that there are sufficient gluconeogenic precursors for biomass production), IDH becomes activated and ICL inhibited, leading to greater flux around the TCA cycle. In a similar vein, oxaloacetate and pyruvate also stimulate the phosphatase activity of AceK, leading to increased ICD activity. In contrast, when these gluconeogenic precursors are in short supply, IDH becomes deactivated and ICL becomes disinhibited, thereby restoring flux through the glyoxylate shunt. It is worth noting that ICD activity also becomes depressed via a different mechanism when demand for gluconeogenesis is high. Acetyl-CoA strongly stimulates the kinase activity of AceK, leading to inactivation of ICD (again, promoting flux through the glyoxylate shunt). This is significant because acetyl-CoA would be expected to accumulate during growth on fatty acids or acetate, or when its condensation partner in the TCA cycle, oxaloacetate, is in short supply (signalling that anaplerosis is necessary). As might be expected, uncharged CoA was an inhibitor of ICL, suggesting that flux through the bifurcation point is also partially dictated by enzymatic surveillance of the acetyl-CoA : CoA ratio.
In addition to gluconeogenic precursors, we also found that the products of ICL activity per se play an important role in coordinating flux partitioning; succinate feeds back to inhibit ICL and also stimulates the phosphatase activity of AceK (thereby re-activating ICD). Similarly, glyoxylate also stimulates the AceK phosphatase (reactivating ICD), and directly activates IDH (although not as strongly as pyruvate or oxaloacetate). In M. smegmatis, this type of mechanism has been termed "rheostatic" 7 , although in P. aeruginosa, this appears to be secondary compared with the more dominant role(s) played by oxaloacetate, pyruvate and acetyl-CoA.
Unlike most other organisms characterised to date, the ICL in P. aeruginosa has a lower KM for isocitrate than the isocitrate dehydrogenases. This may be to always ensure adequate carbon flux through TCA cycle, especially when demand for NADPH is high e.g., during oxidative stress or anabolism. The isocitrate dehydrogenase reaction is one of the main sources of NADPH in the cell 26 , and it may be that P. aeruginosa has evolved a "belt-and-braces" solution to ensure that it never runs out of this important anti-oxidant, even when the demand for flux through the glyoxylate shunt is high. This may also explain why it encodes two independently (albeit, similarly) regulated isocitrate dehydrogenases, ICD and IDH. The same logic may also explain why, even after prolonged exposure to AceK, ICD retains some activity (although this was not the case in E. coli, where 100% inhibition was achieved after treatment with either AceKPA or AceKEC). Complete ICDPA inhibition was only observed in the presence of acetyl-CoA, presumably because an accumulation of this metabolite indicates to the cell that the "relief valve" of the glyoxylate shunt needs to be fully opened.
Although ICD and IDH were expressed at similar levels during growth on glucose and acetate, ICL showed strong induction on the latter. Presumably, this reflects induction of the aceA gene, and this has been noted by earlier workers 10 . However, and unlike the situation in E. coli, aceA is expressed at appreciable levels on glucose, and there is significant flux through the glyoxylate shunt on this carbon source 27 . This may reflect the fact that (for reasons that are not yet clear), under some conditions, optimal virulence is dependent upon flux through the glyoxylate shunt 23 .
One of the more remarkable observations made in this work relates to the structure of ICL. This large protein is an established drug target 28 , and clearly responds to multiple regulators. We found that much of the significant additional sequence present in this sub-class of ICL enzymes manifests itself as structural features projecting away from the globular core of the enzyme. Although we do not yet know what these projections do, one possibility (given their surface exposure) is that they are involved in protein-protein interactions, and we are currently investigating this. Another area of ongoing interest is to understand at an atomic level of resolution how the regulator molecules exert their effects, especially on ICL and IDH. V-type allosteric regulators remain relatively poorly characterised, and a better understanding of these may open the way towards the design of drug-like "antiactivators" as well as inhibitors.
In conclusion, we have shown that the TGB in P. aeruginosa has a complex enzymology that is profoundly different to that in all other organisms for which the TGB has been characterised to date. This notwithstanding, common themes are discernable among the identified regulatory molecules, and the structural data lay a solid groundwork for downstream drug design efforts.

Acknowledgements
This work was supported by BBSRC grant BB/M019411/1.

Cloning, Overexpression, and Purification
The ORFs encoding the ICL (PA2634), IDH (PA2624) and ICD (PA2623) enzymes were PCR-amplified from the genomic DNA of P. aeruginosa strain PAO1 and cloned into a derivative of the NEB vector pMAL-c2X which had been previously modified to introduce a hexa-histidine tag onto the N-terminus of the MBP. The constructs were confirmed by DNA sequencing and introduced into E. coli DH5α. For over-expression, the cultures were grown with good aeration in LB medium at 37°C to OD600 = 0.6. IPTG was then added to a final concentration of 0.3 mM and growth was allowed to continue for a further 2 h. The cells were harvested by sedimentation (3430 × g, 20 min, 4°C) and the pellets were resuspended in 60 mL of buffer A (200 mM NaCl, 20 mM Tris-HCl, 1 mM EDTA, 1 mM DTT, pH 7.4) containing a Complete EDTA-free protease inhibitor cocktail tablet (Roche). The cells were lysed by sonication on ice and the cell debris was removed by sedimentation (15000 × g, 30 min, 4°C). The supernatant was applied to an amylose resin column and washed (overnight) with 500 mL of buffer A at 4°C. The protein was then eluded in 10 mL of buffer B (200 mM NaCl, 20 mM Tris-HCl, 1 mM EDTA, 10 mM maltose, pH 7.4). The eluted sample was dialysed against 2 × 1L of buffer C (50 mM NaCl, 25 mM Tris-HCl, 10% (v/v) glycerol, pH 7.4). Following this, the sample was concentrated and cleaved with Factor Xa protease (1:100 ratio of protease:sample) at 6°C for 24h with constant gentle mixing. The protease was removed using p-aminobenzamidine agarose and the cleaved protein mixture was applied to a column (2 mL packed resin volume) of Ni-NTA resin equilibrated with buffer D (100 mM NaCl, 50 mM Tris-HCl, 5% (v/v) glycerol, 5 mM β-mercaptoethanol, pH 7.5) to remove the His6-MBP tag. The flow through (consisting of cleaved purified target protein) was collected and dialysed against buffer E (100 mM NaCl, 25 mM Tris-HCl, 10% (v/v) glycerol, 1 mM EDTA, 1 mM DTT, pH 7.5). The dialysed sample was concentrated by ultrafiltration and then snap-frozen in liquid nitrogen for storage at -80°C. Protein concentration was determined spectrophotometrically using the calculated molar extinction (εcalc = 54320 M -1 .cm -1 (ICL), 82280 M -1 .cm -1 (IDH) and 57870 M -1 .cm -1 (ICD)).
The AceK (PA1376), ICDPA and ICDEC (from E. coli strain DH5α) used for phosphotransfer/AceKinhibition assays were purified slightly differently. The PCR-amplified ORFs were cloned into the expression vector pET-19m, which introduces a TEV-cleavable N-terminal hexa-histidine tag onto each protein. For purification of the His6-tagged proteins, the cells were grown in LB medium at 37°C with good aeration to OD600 = 0.5. The temperature was then lowered to 20°C and IPTG was added to 0.5 mM final concentration to induce expression of the cloned genes. The induced cultures were grown for a further 16 h and then harvested by sedimentation (6000 × g, 4°C, 15 min). The cell pellet was resuspended in 20 mL of buffer F (50 mM sodium phosphate, 200 mM NaCl, 10% (v/v) glycerol, pH 8.0) and the cells were ruptured by sonication (3 × 10 sec, Soniprep 150, maximum power output). The cell lysate was clarified by centrifugation (11,000 × g, 4°C, 30 min), and the supernatant was filtered through a 0.45 μm filter. The filtered lysate was then loaded onto an Ni-NTA column (2 mL packed resin bed volume) and the column was washed overnight at 4°C with buffer F containing 10 mM imidazole. The His6-tagged proteins were eluted with buffer F containing 250 mM imidazole. The purified protein was mixed with His6-tagged TEV-protease and dialyzed overnight at 4°C against 2 L buffer G (20 mM Tris-HCl, 50 mM NaCl, 5% (v/v) glycerol, pH 7.5). The proteins thus released were cleaned up by batch extraction in a slurry of Ni-NTA resin equilibrated in buffer G. AceKEC was expressed and purified as above, but from a modified pET-15b vector (a generous gift from Prof. Jia) 29 .

Gel filtration and AUC
GFC. Analytical gel filtration with multi-angle light scattering was carried out using a 300 × 7.8 mm TSK-Gel G3000 SWXL column (Toso Haas) equilibrated with 150 mM NaCl, 20 mM Tris-HCl (pH 8.0) operating at a flow rate of 0.5 mL min -1 . The column eluate was monitored in-line with a Mini-DAWN light scattering detector (λ = 690 nm), a quasi-elastic light scattering detector, differential refractometry and an absorption detector (280 nm). Molecular masses were calculated using Astra Software (Wyatt Technologies) and the Debye fit method 30 . A gel filtration marker kit (29 -700 kDa range) from Sigma-Aldrich was used to confirm accuracy of the measured masses.
AUC. Protein samples were first dialysed against 100 mM NaCl, 25 mM Tris-HCl (pH 7.5) to remove glycerol and DTT. AUC was performed using a Beckman Optima XL-I (AN-60 Ti rotor) fitted with absorbance and interference optics. Epon double-sector centrepieces were filled with 400 μL of sample solution or blank buffer and sedimented at 29160 × g for 24 h at 20°C. Absorbance data were acquired at λ = 280nm with time intervals of 2 min; interference scans were taken with time intervals of 1 min. Buffer viscosity, protein partial specific volumes and frictional ratios were calculated using SEDNTERP 31 . Data were analysed using SEDFIT 32 .
The enzyme activity of IDH and ICD were measured using the same assay. The enzyme was incubated in 50 mM Tris-HCl, 5 mM MgCl2 (pH 7.5). To measure the KM value for isocitrate, initial rates were measured across a range of (+)-potassium D-threo-isocitrate concentrations (0 to 600 µM) with a fixed concentration of NADP + (200 µM). To measure the KM value for NADP + , initial rates were measured across a range of NADP + concentrations (0 to 600 µM) at a fixed concentration of isocitrate (200 µM). NADPH formation was recorded spectrophotometrically at λ = 340 nm (εcalc = 6220 M -1 cm -1 ). All assays were carried out at 37°C.
AceKEC and AceKPA kinase/phosphatase activity was assayed by coupling AceK activity to ICD activity as described previously 33 with the following modifications. In a 96-well plate, 180 μL of phosphorylation reaction mixture (100 mM Tris-HCl, 1 mM ATP, 2 mM MgCl2, (pH 7.0) containing 10 μg ICD and 5 μg AceK) was incubated in each well for 1 h at 37°C. Following this, putative allosteric regulators were added to each reaction mixture to 5 mM final concentration. Each assay was carried out in triplicate. The reactions were then re-incubated for a further 30 min at 37°C to allow dephosphorylation of ICD and reactivation of ICD kinase activity. Aliquots (20 μL) of the reaction mixture were then transferred to a new UV-clear 96-well plate and 180 μL of reduction solution consisting of 100 mM Tris-HCl, 1 mM threo-D,L-isocitrate, 0.5 mM NADP + and 2 mM MnCl2 (pH 7.0) was added to each well. The activity of ICD was detected spectrophotometrically by monitoring the rate of reduction of NADP + at 340 nm using a FLUOstar Omega plate reader (BMG Labtech). As a control, ICD alone was assayed for its ability to reduce NADP + in the presence of the putative allosteric regulators using the same method. Because phosphorylation of ICD by AceK inhibits the activity of ICD, higher ICD activity corresponds to increased AceK phosphatase activity. Confirmation of AceKPA/EC phosphotransfer to ICDPA/EC was also demonstrated using a [ 32 P]-based phosphotransfer assay. The reaction conditions were identical to those described above, however, γ[ 32 P]ATP (PerkinElmer) was used in place of unlabelled ATP. The proteins were resolved by SDS-PAGE and radiolabelling was revealed using a phosphor screen (GE Healthcare). Radioactivity was determined with a Typhoon PhosphorImager and quantitated with ImageQuant.

X-ray Diffraction, Structure Determination, and Refinement
Diffraction data were collected remotely on the I02 or I04-1 beamline (as indicated in Table S1) at the Diamond Light Source Synchrotron (Didcot, UK). Diffraction data were processed using FastDP 27 , and the structures were determined by molecular replacement using Phaser 34 . ICL, IDH and ICD were solved using the structural templates 3I4E (to be published), 4ZDA (to be published) and 1BL5 35 from Burkholderia pseudomallei, Mycobacterium smegmatis and Escherichia coli, respectively. Automated refinement was performed using Refmac5 36 . Manual modelling and refinement were performed in COOT 37 . Data collection and refinement statistics are listed in Table S1.

Western blotting
Polyclonal antibodies were raised in rabbits against each of the purified proteins (Biogenes.De). The antisera were pre-absorbed against an acetone extract of a mutant strain defective in the protein of interest (e.g., the anti-ICD antisera were pre-absorbed against an acetone extract of a confirmed icd mutant). The cleaned-up antisera, appropriately diluted, were then used directly in Western assays. Western blots were developed using goat anti-rabbit secondary antibodies.  1F8I) and E. coli ICL (red, PDB 1IGW) revealing that the catalytic core of the enzyme is highly conserved. Consistent with this, the root mean square deviation (RMSD) between P. aeruginosa and E. coli ICL is 1.35 Å, and the RMSD between P. aeruginosa and M. tuberculosis ICL is 1.30 Å. However, note how ⍺1 and α2 are flipped by 180° in P. aeruginosa ICL, the extended helices ⍺13 and α14, and the presence of a head-domain (absent from M. tuberculosis and E. coli ICL. (D) Conserved active site residues in P. aeruginosa ICL. The β4-β5 loop containing the general base, Cys222, is highlighted. The electron density map for glyoxylate and Ca 2+ is contoured at 1.7σ.  Close-up view of the conserved active site residues in P. aeruginosa IDH (light pink, PDB 6G3U) and P. aeruginosa ICD (yellow, PDB 5M2E). Residue numbering is based on the IDH sequence. (C) P. aeruginosa ICD, but not IDH, is inactivated by AceKdependent phosphorylation. The figure shows the loss of isocitrate dehydrogenase activity over time following treatment with AceK/ATP. Reaction mixtures (200 µL) contained 100 mM Tris-HCl (pH 7.0), 1 mM ATP, 2 mM MgCl2, 5 µg purified P. aeruginosa AceK, and 10 µg P. aeruginosa ICD or IDH (as indicated). Reactions were allowed to proceed at 37°C, and at the indicated times, aliquots were withdrawn and assayed immediately for isocitrate dehydrogenase activity as described in Materials and Methods. Activity was considered to be 100% at T0. (D) The architecture of the active site is different in ICD and IDH. In spite of the conserved constellation of active site residues (panel (B)) in P. aeruginosa ICD and IDH, the P-loop containing the phosphoserine in ICD (yellow) and the AceK recognition sequence (ARS) are replaced in IDH by two helices ⍺10-α14 and ⍺4-α6, respectively. This altered arrangement prevents AceK from accessing the active site serine in IDH.  aeruginosa AceK and P. aeruginosa ICD (AceKPA and ICDPA) or between E. coli AceK and E. coli ICD (AceKEC and ICDEC), as indicated. Aliquots were removed for sampling after 1, 2 and 3 min incubation, and resolved by SDS-PAGE. Radioactivity was monitored using a Phosphoimager. (B) P. aeruginosa AceK is intrinsically less efficient at phosphorylating P. aeruginosa ICD than E. coli AceK. The figure shows a "mix n' match" experiment in which the efficiencies of P. aeruginosa and E. coli AceK at inactivating the ICD homologues from each species are measured. ICDPA and ICDEC can each be phosphorylated by either AceKPA or AceKEC. However, the ICDPA/AceKPA pair never achieves full inactivation, in spite of the fact that AceKPA fully inactivates ICDEC (indicating that the kinase is competent to do this) and that AceKEC fully inactivates ICDPA (indicating that the ICD is potentially fully inactivatable). Reaction conditions as in Figure 3. (C) Fumarate stimulates the phosphatase activity of AceKPA (thereby reactivating ICD), whereas acetyl-CoA stimulates the kinase activity of AceKPA (thereby inactivating ICD). Fumarate and acetyl-CoA were present at 5 mM concentration. . In all the cases, ሾ‫ܫ‬ሿ = ‫.ܯ݉1‬   Figure S2. Phylogenetic tree of ICL from a subset of pathogens. Note how the P. aeruginosa ICL forms a distinct cluster (boxed in red) with ICL from Acinetobacter baumanii, Azotobacter vinelandii and Burkholderia cepacia. The tree was generated after alignment of all amino acid sequences using ClustalOmega and neighbour joining was calculated from the percentage sequence identity in JalView.    The overall molecular mass of IDH inferred from GFC-MALS analysis was 236 kDa. AUC analysis yielded a higher estimated molecular mass of 273 kDa. Given that the calculated molecular mass of one IDH polypeptide is 82 kDa, these data suggest that IDH could be a trimer or elongated dimer in solution. However, we also note that the protein has a rather low frictional coefficient (f = 1.15), consistent with it adopting a compact, globular configuration.  Given that the calculated molecular mass of one ICD polypeptide is 45 kDa, the ICD isocitrate dehydrogenase is likely to be a dimer (Mw = 90 kDa) in solution. Figure S10. Impact of small molecule regulators of AceK kinase/phosphatase activity. (A) ICDPA was first inactivated by incubation for 60 min with AceKPA and ATP. Following this, the indicated small molecule regulators were added (5 mM final concentration in all cases) and the reaction was allowed to continue for a further 30 min. The isocitrate dehydrogenase activity of each sample was then measured. Note how ICD retains about 25% of its activity even after extensive treatment with AceK, unless acetyl-CoA or citrate are present (see main text for details). The blue line represents the activity of ICD that has not been treated with AceK and the red line indicates the activity of ICD that has been treated with AceK (but not further treated with potential small molecule regulators). (B) A similar experiment to that just described was also carried out with ICDEC with AceKEC. The E. coli AceK more profoundly inhibits E. coli ICD activity compared with the P. aeruginosa homologue AceK/ICD pair. All plots were generated using GraphPad Prism 6, and the errors bars correspond to ± 1 standard deviation. All experiments were performed in triplicate. Figure S11. ICD isocitrate dehydrogenase activity is not intrinsically affected by most of the regulators that impact on AceK.

Supplementary Information Figure Legends
(A) The metabolites tested as putative AceK regulators in Figure S10 were also evaluated for their ability to directly activate/inhibit P. aeruginosa ICD. Note that ICD retains almost full activity in the presence of most of these compounds, although PEP, glyceraldehyde 3-phosphate, 2-keto-3-deoxyphosphogluconate, and ribose 5-phosphate slightly depress ICD activity when present at 5 mM, and citrate blocks ICD activity completely. (B) The same experiment as in (A) carried out with ICDEC.