The Structure of the PanD/PanZ Protein Complex Reveals Negative Feedback Regulation of Pantothenate Biosynthesis by Coenzyme A

Summary Coenzyme A (CoA) is an ubiquitous and essential cofactor, synthesized from the precursor pantothenate. Vitamin biosynthetic pathways are normally tightly regulated, including the pathway from pantothenate to CoA. However, no regulation of pantothenate biosynthesis has been identified. We have recently described an additional component in the pantothenate biosynthetic pathway, PanZ, which promotes the activation of the zymogen, PanD, to form aspartate α-decarboxylase (ADC) in a CoA-dependent manner. Here we report the structure of PanZ in complex with PanD, which reveals the structural basis for the CoA dependence of this interaction and activation. In addition, we show that PanZ acts as a CoA-dependent inhibitor of ADC catalysis. This inhibitory effect can effectively regulate the biosynthetic pathway to pantothenate, and thereby also regulate CoA biosynthesis. This represents a previously unobserved mode of metabolic regulation whereby a cofactor-utilizing protein negatively regulates the biosynthesis of the same cofactor.


INTRODUCTION
Coenzyme A (CoA) 1 is essential for the growth of all organisms and is derived from pantothenate 2 ( Figure 1A). Pantothenate 2 is an essential nutrient for humans and is synthesized via a common pathway in plants, yeast, and bacteria (Webb et al., 2004). The majority of vitamin and amino acid biosynthetic pathways in bacteria are highly regulated either by metabolite-responsive transcription factors (Grose et al., 2005), product inhibition of key enzymes (Farrar et al., 2010), or metabolite-binding riboswitches (Winkler and Breaker, 2005). The pathway from pantothenate onward to CoA is regulated by feedback inhibition of pantothenate kinase (Rock et al., 2003;Yun et al., 2000) in all organisms, including bacteria, but no evidence for regulation of the pathway to pantothenate has been described.
In all organisms, the final step in pantothenate biosynthesis is the condensation of pantoate 3 and b-alanine 4 to form pantothenate 2, but the source of b-alanine is different in bacteria, yeast, and plants. In bacteria, b-alanine is derived from L-aspartate 5 by the action of the enzyme aspartate a-decarboxylase (ADC) encoded by panD (Williamson and Brown, 1979). ADC is one of a small subset of enzymes containing a protein-derived, covalently linked pyruvoyl cofactor (van Poelje and Snell, 1990). This cofactor was first observed in histidine decarboxylase (Snell, 1986) and is present in a set of highly conserved enzymes, including S-adenosylmethionine decarboxylase (Pegg, 2009) and the membrane-bound phosphatidylserine decarboxylase (Schuiki and Daum, 2009). In all cases, the zymogens undergo an N/O acyl shift in their peptide backbones to form an ester ( Figure 1B, III) that is then cleaved by elimination to generate the pyruvoyl cofactor ( Figure 1B, V). The structural basis for this rearrangement in these enzymes has previously been investigated using a combination of site-directed mutagenesis (Gelfman et al., 1991;Schmitzberger et al., 2003;Webb et al., 2012Webb et al., , 2014, the structure of the zymogen (Schmitzberger et al., 2003), and the structure of inactive site-directed mutants (Schmitzberger et al., 2003;Tolbert et al., 2003aTolbert et al., , 2003bWebb et al., 2014). In some of these cases there is evidence for regulated activity. In the case of S-adenosylmethionine decarboxylase, both formation of the cofactor and subsequent catalytic activity are positively allosterically regulated in vitro by binding of the substrate for the succeeding step in the biosynthetic pathway (Bale et al., 2008;Stanley et al., 1994). In contrast, the catalytic activity of histidine decarboxylase is regulated by pH-dependent unfolding at neutral pH (Schelp et al., 2001). Similar peptide backbone modifications, leading to either autoproteolysis or backbone rearrangement, have been observed in a diverse range of systems; including the post-translational processing of inteins (Paulus, 2000), post-translational activation of the N-terminal nucleophile hydrolases (Buller et al., 2012;Kim et al., 2006), recA-mediated cleavage of the DNA-binding protein umuD (McDonald et al., 1998), and autoproteolysis of SEA domains (Johannson et al., 2009;Levitin et al., 2005). In most cases these post-translational modifications are autocatalytic, but the reaction rates of a subset of enzymes have been shown to be enhanced by accessory proteins (McDonald et al., 1998), although the structural basis for such enhancement has not been determined.
Since the first identification of pyruvoyl-dependent enzymes (van Poelje and Snell, 1990;Williamson and Brown, 1979), it has been widely presumed that their post-translational rearrangement is also autocatalytic (Ramjee et al., 1997). Recently, however, accessory proteins essential for the activation of some members of this class of proteins have been identified Stuecker et al., 2012a;Trip et al., 2011). HdcB, the essential accessory protein for histidine decarboxylase activation (HdcA) from Streptococcus thermophilus, was identified by Trip et al. (2011). Subsequently, the accessory protein for maturation of PanD, PanZ, was described independently in Escherichia coli  and the closely related Salmonella typhi (Stuecker et al., 2012a). PanZ is essential for activation of the zymogen PanD to form ADC in vivo, and its deletion leads to b-alanine auxotrophy . In this article, we report the structure of the protein complex formed between PanZ and PanD, which has allowed us to determine both the basis for CoA-dependent interaction and how the tight interaction of PanD and PanZ leads to activation of PanD to form ADC. Secondly, we report our investigation of the role of PanZ in global regulation of the pantothenate, and thereby the CoA, biosynthetic pathway, leading us to propose a novel mode of metabolic regulation via ligand-dependent protein-protein interaction.

RESULTS
Architecture of the PanD-PanZ Complex Reveals Origin of CoA-Dependent Interaction PanD (encoded by panD) is post-translationally modified to form its catalytically active form, ADC, by cleavage of the peptide backbone between residues Gly24 and Ser25, leading to the formation of a pyruvoyl cofactor from Ser25 ( Figure 1B). The residues in PanD required for activation have been previously explored by mutagenesis. As expected, mutation of Ser25 to alanine leads to loss of activation (Schmitzberger et al., 2003) but the only proximal residue found to be required for activation is Thr57, where mutation to the isosteric valine leads to complete loss of activation (Webb et al., 2014). We have previously used this inactivatable PanD(T57V) mutant to characterize the interaction of PanD with PanZ (Monteiro et al., 2012), and demonstrated that interaction is dependent on the presence of CoA. We observed substoichiometric binding between ADC and PanZ, probably due to the formation of disulfide-linked CoA dimers in solution. As PanZ binds acetyl CoA (AcCoA) with equal affinity Figure 1. Biosynthetic Pathway to CoA and Mechanism for Cofactor Formation in PanD (A) Biosynthetic pathway to pantothenate 2 and coenzyme A (CoA) 1 in bacteria. D-Pantoate 3 is generated from a-ketoisovalerate 6, an intermediate in valine biosynthesis, via hydroxymethylation, to generate ketopantoate 7 and subsequent NADPH-dependent reduction. D-Pantoate is ligated to b-alanine 4, which is generated via decarboxylation of L-aspartate 5 by aspartate a-decarboxylase (ADC). The principal characterized regulatory step is phosphorylation of pantothenate 2 to generate phosphopantothenate by pantothenate kinase (PanK), which is feedback-regulated by the terminal metabolite of the pathway CoA (Yun et al., 2000). (B) Previously proposed mechanism of PanD activation. Thr57 hydrogenbonds to Gly24, polarizing the carbonyl bond (Schmitzberger et al., 2003;Webb et al., 2014). Ser25 attacks Gly24 to a ring intermediate II which opens to the ester intermediate III (Albert et al., 1998). This is cleaved by elimination to a dehydroalanine IV, which can then hydrolyze to the pyruvoyl cofactor V.
Following this, we characterized the structural basis of protein complex formation. Bipyrimidal crystals were reproducibly obtained using a 1.1-fold excess of PanZ over PanD(T57V) and a 2-fold excess of AcCoA, with respect to PanZ. X-Ray diffraction data were collected from these crystals to 1.7 Å resolution at room temperature using an in-house X-ray source. The crystals contained a single PanD protomer and one PanZ molecule per asymmetric unit (Table S1). The structure of the protein complex revealed a highly symmetric heterooctameric complex. The four active sites of the PanD heterotetramer and the loops that are cleaved to generate them lie at the interface of each pair of PanD protomers. The PanD-PanZ protein complex is a crossshaped heterooctamer, with one PanZ molecule bound to each of the PanD protomer-protomer interfaces ( Figure 2A).
We used small-angle X-ray scattering (SAXS) to confirm that the architecture observed in the crystal corresponds to that of the complex in solution ( Figure 2B). A solution sample of the two proteins at a 1:1 ratio in the presence of two equivalents of AcCoA was used for data collection at varying concentrations. The calculated atomic scattering factors for the crystallographic model fitted well to the observed scattering at low protein concentrations, with small deviations at high and low scattering angles ( Figure S2). These deviations could be accounted for by inclusion of both the crystallographically unresolved affinity purification tags (shown modeled in Figure 2B) and a small population of a dimer of the PanD-PanZ complex (two copies of the heterooctamer), consistent with the packing observed in the crystal structure. The population of this dimer species increased in scattering data recorded at higher concentrations, suggesting a concentration-dependent aggregation phenomenon.
Nuclear Magnetic Resonance Demonstrates that CoA Is an Absolute Requirement for Complex Formation As expected from our preliminary biophysical characterization (Monteiro et al., 2012), each molecule of PanZ binds one molecule of AcCoA. The binding site for AcCoA is very close to the protein-protein interface, accounting for the previously observed CoA dependence of the interaction (Monteiro et al., 2012). The structure of PanZ, when in complex with PanD(T57V), is largely isostructural with the monomeric structure for PanZ previously elucidated by nuclear magnetic resonance (NMR) in the presence of CoA (PDB 2K5T, Figure 3A), suggesting that binding of CoA causes PanZ to adopt an ADC-binding conformation. In particular, binding of AcCoA appears to stabilize the PanZ Leu66-Gly76 loop, which forms contacts with PanD ( Figure 3B and Figure S3). At the same time, the acetyl portion of the cofactor is held away from PanD on the distal face of PanZ, suggesting that acetylation is not required for activation, consistent with the findings of Stuecker et al. (2012b). Binding of AcCoA is dependent upon hydrogen-bonding interactions to the PanZ Leu66-Gly76 loop, which also forms hydrogen bonds with PanD. The acetyl group of AcCoA is distal to the PanZ-PanD interface. AcCoA, PanD, and PanZ are colored gray, green, and cyan, respectively. (C) 1 H-15 N HSQC spectrum of PanZ after purification from E. coli shows a complex spectrum due to the presence of co-purified CoA derivatives (Monteiro et al., 2012), leading to a mixture of free PanZ together with PanZ.RCoA. (D) 1 H-15 N HSQC spectrum after addition of CoA is simplified to yield a well-dispersed spectrum of the PanZ.RCoA complex.
(E) Addition of excess PanD to PanZ yields a well-dispersed 1 H-15 N HSQC spectrum of apo-PanZ, which is distinct from that observed after addition of CoA. PanD-PanZ.AcCoA is not observed.
(F) Addition of both excess CoA and PanD to PanZ leads to loss of all signals in the 1 H-15 N HSQC due to formation of the PanD-PanZ.AcCoA complex ($124 kDa) in which presumed fast relaxation leads to signal broadening and consequent loss of signal (see also Figure S3).
While we crystallized the complex of PanZ and PanD with bound AcCoA and observed CoA concentration effects on the interaction by isothermal titration calorimetry (ITC) (Monteiro et al., 2012), this does not demonstrate unambiguously that the proteins cannot interact in the absence of CoA. We therefore generated 15 N-labeled PanZ and used a 1 H-15 N heteronuclear single quantum coherence (HSQC) experiment to assess the homogeneity of the protein preparation. As previously reported by Cort et al. (2009), we observed more peaks than expected for the size of the protein ( Figure 3C), presumably due to the presence of both CoA-bound-and apo-PanZ in the sample. When PanZ is overexpressed and purified, it is isolated with a substoichiometric quantity of what are presumed to be a mixture of CoA and its thioesters (annotated as RCoA), as shown by our previous observation of substoichiometric binding of AcCoA and CoA to the purified protein (Monteiro et al., 2012). Addition of excess Ac-CoA led to simplification of the spectrum ( Figure 3D) due to formation of a single species in the sample tube. Finally, we added excess PanD(T57V) to AcCoA-saturated PanZ. This led to loss of almost all signals from the NMR spectrum ( Figure 3F): the PanD-PanZ.AcCoA complex is approximately 124 kDa, and the slow tumbling rate of this species in solution leads to extensive broadening of the signals such that they are no longer observed.
This loss of signal upon formation of the complex provides an effective assay for complex formation. If AcCoA is not essential for PanD-PanZ complex formation, addition of PanD(T57V) to the mixed population of PanZ and PanZ.RCoA obtained upon protein purification should lead to loss of all signals. Addition of PanD(T57V) to the mixed population of 15 N-labeled PanZ also led to simplification of the spectrum ( Figure 3E), similar to that observed by addition of AcCoA; however, a different set of peaks were resolved in the spectrum, presumably corresponding to apo-PanZ. To confirm that this simplification was due to insufficient CoA rather than substoichiometric PanD(T57V), we then added an excess of AcCoA. Once again, this led to loss of the observed spectrum (not shown), demonstrating unambiguously that CoA or a derivative is required for formation of the PanD-PanZ complex.

Structural Basis for Activation of ADC by PanZ
The PanD-PanZ interaction is mediated by the PanZ N terminus, Arg43-Leu48 loop, and Gly66-Leu76 loop (which also interacts with AcCoA). Together, these elements interact with the surface of PanD to either side of the PanD C-terminal tail (residues Lys119 to Ala126, Figures 4A and 4B). This portion of ADC is normally completely unstructured and has only previously been observed in the structure of the ADC(N72A) mutant, in a different, extended conformation that was stabilized by crystal packing contacts . The C-terminal sequence of PanD is hydrophobic ( 119 KAIPVQVA 126 ) and is fully conserved among organisms that encode PanZ ( Figure 4D) , despite lying outside what can be regarded as the core fold of the protein. The PanD C-terminal region is at the core of the protein-protein interaction and becomes completely buried as a result of complex formation. We used calorimetry to investigate the importance of the C terminus of PanD for complex formation. We generated the site-directed mutant PanD(T57V/ K119Stop) in which the seven C-terminal amino acids are deleted, and repeated the titration of PanZ.AcCoA into the purified PanD(T57V/K119Stop). No interaction between the proteins could be observed, suggesting that the C terminus of PanD is essential for binding ( Figure 4F). The remainder of the protein-protein interface is maintained by numerous hydrogen-bonding interactions between conserved and semi-conserved residues in PanD and PanZ. One key hydrogen-bonding interaction is between the side chain of PanZ-Asn45 and the backbone amide proton of PanD-Glu23 ( Figure 4C). Both residues are conserved in organisms that express PanZ. Mutation of PanZ-Asn45 has previously been shown to result in cells unable to generate active ADC, suggesting that this contact might be key to the activation of PanD . To test this hypothesis, we investigated the interaction of a PanZ(N45A) mutant with PanD(T57V) using ITC ( Figure 4G). This revealed an 80-fold decrease in affinity, to $4 mM. The most recent consensus measurements of protein abundance derived from proteomic studies (Wang et al., 2012) for PanD and PanZ are 240 and 72 ppm, respectively, i.e. concentrations of approximately 500 and 150 nM. The lower affinity of the mutant PanZ(N45A) for PanD is therefore sufficient to prevent formation of the complex, and explains the inability of PanZ(N45A) to effectively complement the b-alanine auxotrophy of DpanZ strains .
Previous studies of the ADC activation mechanism have focused on the identification of residues that are required to catalyze the rearrangement of the peptide chain. It was originally hypothesized that at least two catalytic residues would be necessary for general acid-base catalyzed ester formation (Schmitzberger et al., 2003;Webb et al., 2012Webb et al., , 2014. However, only mutagenesis of Thr57 leads to loss of activation. To confirm that the observed PanD(T57V) conformation when in complex with PanZ is solely caused by the protein-protein interaction rather than as a result of the site-directed mutation, we determined the structure of a second non-activatable mutant, PanD(S25A), in complex with PanZ.AcCoA. Diffraction data were collected at room temperature in-house to 2.1-Å resolution. The overall architecture and major protein-protein interactions were completely consistent between the two complexes (PanD(S25A)-PanZ and PanD(T57V)-PanZ). However, owing to the change in the available hydrogenbonding interactions, slightly different conformations of the active-site loop were seen ( Figure S4). Nevertheless, in both structures one of the two residues at the site of cleavage adopted a Ramachandran-disallowed conformation (Ramachandran et al., 1963) (Figure S5), suggesting that the PanD activation region is forced into an unfavored and, therefore, high-energy conformation upon binding of PanZ. This destabilization may lower the energy barrier for activation of PanD to ADC.
To further investigate how the activation reaction is catalyzed, we compared the structure of the PanD(T57V)-AcCoA.PanZ complex with that of the previously reported structure of the wild-type PanD zymogen in the absence of PanZ (Schmitzberger et al., 2003). The location of the peptide backbone in the immediate region of eventual peptide cleavage (Glu23-Cys26) in the PanD zymogen is largely similar to that observed in the PanD-AcCoA.PanZ complex, although the exact conformations of individual residues, particularly the carbonyl of Gly24 ( Figures  5A and 5B), have changed. In contrast, the position of the The PanD processing loop (residues Glu23 to Cys26) from the crystal structure of the wild-type zymogen determined by Schmitzberger et al. (2003). (B) The PanD processing loop (residues Glu23 to Cys26) from the PanD(T57V)-PanZ protein complex (green). PanZ-Asn45 (cyan) forms a single hydrogen bond to the PanD-Glu23 backbone (see Figure 4C). The processing loop is compacted relative to the wild-type structure and the carbonyl of Gly24 now interacts with the side chain of Tyr58 rather than Thr57. (C) Comparison between PanD(T57V)-PanZ (green) and the high-occupancy conformation adopted by the zymogen (1PPY, yellow). The non-processing-prone conformation is incompatible with the binding of PanZ (cyan), sterically clashing with the PanZ-Asn45 loop. In the activatable conformation, the loop moves upward to form a new b sheet, 6.4 Å away from its original position (see also Figures S4-S6).
PanD loop between Thr16 and Tyr22 is wholly different. In the published PanD zymogen structure, this region adopts two distinct conformations (Schmitzberger et al., 2003). One is well resolved and consists of a loop without defined secondary structure ( Figure S6A), whereas the second, low-occupancy, conformation is poorly defined in the electron density (Figure S6B). This second low-occupancy conformation corresponds to the conformation observed, at full occupancy, in the structures of fully activated ADC (Albert et al., 1998) and our newly determined structures of the PanD-PanZ.AcCoA complex. The N terminus of PanZ and its Arg43-Leu48 loop are positioned in the space that would be occupied by the high-occupancy conformation of the uncleaved peptide chain of free PanD ( Figure 5C). We hypothesize that, in the absence of PanZ, this portion of inactivated PanD can explore a large number of alternative conformations. Binding of PanZ results in restriction of the PanD Thr16-Tyr22 loop to conformations that resemble the activated form. This, in turn, places the Glu23-Ser25 backbone into a conformation that favors the activation reaction. In the two uncleavable PanD site-directed mutants used in this study, the unfavorable nature of this adopted conformation is reflected by the formally disallowed peptide backbone angles observed for either PanD-Gly24 or Ser25. Since the formation of the PanD-PanZ complex depends on the presence of CoA, this suggests that activation of PanD will in turn be controlled by the intracellular CoA concentration.

Inhibition of ADC Catalysis Reveals a Second Global Role for PanZ in Regulation of Pantothenate Biosynthesis
The CoA dependence of the PanD-PanZ interaction and, therefore, PanD activation suggests the presence of a positive feedback mechanism in pantothenate biosynthesis. To further investigate the cellular function of PanZ, we first re-examined the effect of PanZ overexpression in vivo. Deletion of panZ leads to cells auxotrophic for b-alanine, and we have previously shown that leaky, uninduced expression of PanZ is sufficient to complement the b-alanine auxotrophy of E. coli DpanZ cells . In this case, weak expression of His-tagged PanZ under the control of the arabinose promoter in glucose minimal media leads to complementation of the panZphenotype ( Figure 6A). However, when PanZ is overexpressed following growth on arabinose, complementation is not observed and, in fact, overexpression of this essential protein leads to retention of b-alanine auxotrophy. Since the conformation of the inactivated zymogen in the PanD-PanZ.AcCoA complex is similar to that of the free activated protein, we hypothesized that PanZ.AcCoA might also interact with activated ADC to regulate catalysis.
We investigated the effect of PanZ.AcCoA on catalysis of L-aspartate decarboxylation by ADC. This activity can be characterized using calorimetry to directly measure the enthalpy of decarboxylation and protonation (Todd and Gomez, 2001), yielding kinetic parameters similar to those obtained by stopped assays of product formation. Addition of 0.5 equivalents of PanZ.AcCoA to fully activated ADC led to $50% reduction in the maximal ADC activity, whereas addition of 1 or 2 equivalents of PanZ.AcCoA led to a further drop in ADC activity to $5%-10% maximal activity ( Figure 6B). ADC is a relatively inefficient enzyme (k cat $ 0.2 s À1 ) (Ramjee et al., 1997;Williamson and Brown, 1979), and as a result the enzyme concentration required for signal detection in the ITC assay is high (2.5 mM) relative to the affinity of PanZ.AcCoA for PanD ($100 nM). We therefore used 1 H NMR to characterize the effect of PanZ.AcCoA on ADC activity at lower enzyme concentrations (100 nM ADC). Using this assay we observed inhibition of ADC-catalyzed turnover of L-aspartate by PanZ at low L-aspartate concentration (500 mM,  (B) Effect of PanZ.AcCoA on catalysis by activated ADC using an ITC-based assay. Addition of PanZ.AcCoA to ADC leads to inhibition of catalytic formation of b-alanine. Data are derived from the steady-state heat production following sequential addition of fixed volumes of L-aspartate to the enzyme in the instrument. Individual curves are fitted to the Michaelis-Menten curve with a shared value of K M (see also Figure S7). 3K M ) but not at high L-aspartate concentration (10 mM, 60K M ), suggesting a competitive mode of inhibition with respect to Laspartate ( Figure S7).

DISCUSSION
The structure of the PanD-PanZ.AcCoA complex resolves two major questions in pantothenate biosynthesis: how does a CoA-dependent protein-protein interaction promote activation of PanD to form catalytically active ADC, and how is b-alanine biosynthesis regulated in enteric bacteria? We have identified an additional level of regulation mediated by this small-molecule-dependent protein-protein interaction, and this leads us to propose the first model for global regulation of this biosynthetic pathway.

How and Why Does CoA Binding Mediate Protein-Protein Interaction?
PanZ is a member of the GNAT superfamily of acetyl transferases (Vetting et al., 2005). It retains many of the conserved active-site residues of this superfamily, but Stuecker et al. (2012b) have previously shown, in the Salmonella typhimurium system, that mutation of potential substrate lysine residues in PanD does not affect the PanD activation reaction and that acetylation of PanD is not required for activation. In accordance with this observation, in the PanD-PanZ.AcCoA complex the acetyl group of AcCoA is sited away from the PanD-PanZ interface. The proximity of CoA to the PanD-PanZ interface and the similarity between the structure of PanZ in the PanD-PanZ.AcCoA complex and that of the PanZ.CoA complex previously determined by NMR suggest that binding of either AcCoA or CoA places PanZ in a PanD binding-competent conformation. Major changes are observed in the 1 H-15 N HSQC spectrum of PanZ upon binding CoA, corresponding to a substantial conformational change. Such a change would be consistent with data from other members of the GNAT family for which the structure of both apoenzyme and binary complexes have been determined (Vetting et al., 2005). In many of these cases, binding of AcCoA or CoA leads to substantial rearrangements in the P loop (which binds the pyrophosphate moiety of CoA). In the case of the PanD-PanZ.AcCoA complex, this loop forms key interactions with the C terminus of PanD, suggesting that CoA mediates this protein-protein interaction by structuring this loop.

How Does Interaction of PanZ with PanD Promote
Activation of PanD to Form Catalytically Active ADC? Direct interaction of PanD with the PanZ Arg43-Leu46 loop promotes PanD to adopt a reactive conformation, which leads to activation. In the absence of PanZ.AcCoA, this conformation is still accessible to PanD and is observed as a minor population in the structure of the uncleaved zymogen (Schmitzberger et al., 2003) in comparison with the major, unrestrained conformation ( Figure S6B). This conformation places the PanD Glu23-Cys26 loop in the correct spatial arrangement for posttranslational cleavage, and destabilizes the PanD Gly24-Ser25 peptide bond for facilitated nucleophilic attack and subsequent cleavage.
Observation of multiple zymogen backbone conformations is consistent with other autoprocessing systems. The active-site peptides of N-terminal nucleophile hydrolases have been shown to explore multiple conformations, only one of which is competent for autoproteolysis (Buller et al., 2012). In the case of PanD, sampling of this conformation probably accounts for the observed slow thermal activation to ADC after purification in vitro (Ramjee et al., 1997). The rate of thermal activation is, however, insufficient to support pantothenate biosynthesis in vivo in the absence of PanZ . Nonetheless, PanD from other organisms can autoactivate, suggesting that subtle differences in the side chain interactions within the PanD activation loop may induce formation of the matureenzyme-like b sheet between Thr16-Asp19 and Ile69-Asn72, leading to autocatalytic activation.
The structure of the PanD-PanZ.AcCoA complex reported here also allows us to propose a new mechanism for the activation reaction ( Figure 7A). In the structure of the zymogen (PDB 1PPY), the carbonyl of Gly24 forms a hydrogen bond with the side chain of Thr57, which was previously proposed to polarize the carbonyl of Gly24 to favor nucleophilic attack by the hydroxyl of Ser25 ( Figures 7A I and 7B) (Schmitzberger et al., 2003;Webb et al., 2014). However, the Ser25 hydroxyl is poorly positioned in the zymogen structure. It is 4.3 Å away from the Gly24 carbonyl carbon and in the plane of the carbonyl bond approximately 4 Å from where significant orbital overlap leading to reaction could occur. Substantial further backbone rearrangement would be required for the N/O acyl shift to occur. In contrast, in the PanD-PanZ.AcCoA structure the hydrogen bond from Gly24 to Thr57 is no longer present, due to a 120 rotation of the Gly24 carbonyl ( Figures 7A II and 7C). Instead, a new hydrogen bond is formed between the Gly24 carbonyl and Tyr58. The Gly24 rotation places the Ser25 hydroxyl 3 Å from the Gly24 carbonyl carbon, and only approximately 1.5 Å from a position with significant orbital overlap between the Ser25 lone pair sp 3 -orbital and the carbonyl p* orbital. Ser25 is also hydrogen bonded to the carbonyl of Glu23, which may act as a proton-shuttling residue to transfer the proton from Ser25 during the rearrangement reaction.
What, then, is the role of Thr57 in the first stage of the activation reaction? There is no evidence for ester formation in the structure of PanD(T57V), either alone (Webb et al., 2014) or in complex with PanZ.AcCoA. Furthermore, incubation of the PanD(T57V)-PanZ.AcCoA complex with hydroxylamine does not lead to chemically induced cleavage of PanD (data not shown), demonstrating that an ester intermediate is not formed. If Thr57 is not required to polarize the carbonyl of Gly24 toward nucleophilic attack, it must function in the next step of the reaction and deliver a proton to the amide nitrogen as the intermediate oxyoxazolidine ring ( Figure 7A III) is opened. The pK a of the amide anion formed is such that it could readily deprotonate a threonine side chain. The resulting deprotonated Thr57 residue ( Figure 7A IV) could then act immediately to catalyze elimination of the ester to generate a dehydroalanine residue ( Figure 7A V), which subsequently hydrolyzes non-enzymatically to generate the pyruvoyl group. While unusual, this mechanism effectively couples the ester formation and elimination steps, avoiding the formation of the hydrolyzed products observed from thermal cleavage (Ramjee et al., 1997).

How Does Interaction of PanZ and PanD Regulate the Pantothenate Biosynthetic Pathway?
The biosynthesis of CoA has previously been shown to be regulated at the level of pantothenate phosphorylation to form phosphopantothenate (Rock et al., 2003;Yun et al., 2000). Pantothenate kinase catalyzes this step and is allosterically regulated by CoA. Until panZ was identified there were no known regulatory mechanisms for the steps upstream of pantothenate. The pantothenate biosynthetic pathway is relatively simple (Figures  1 and 7D). a-Ketoisovalerate, the oxoacid of valine, is hydroxymethylated and reduced to form pantoate, which is condensed with b-alanine to form pantothenate. The two reactions that form pantoate are reversible and so, in the absence of b-alanine, these metabolites exist in equilibrium with those in the pathway to valine ( Figure 7D). Since the cellular pools of L-aspartate and D-pantoate are in equilibrium with other primary metabolites, formation of b-alanine is the only committed step in the pathway. How, then, can PanZ regulate this step? We have characterized two distinct functions for PanZ: activation of PanD to form ADC and subsequent inhibition of ADC. We therefore propose a new regulatory model whereby ADC activity is limited and regulated by the concentration of CoA in the cell ( Figure 7D).
We hypothesize that such inhibitory activity is actually the primary metabolic role of PanZ (although the activation is also clearly essential). To activate PanD, the PanZ.CoA complex needs to interact with PanD only once; the activation is irreversible and thus can still occur, even at low CoA concentrations. In contrast, inhibition of catalysis requires accumulation of Pan-Z.RCoA to form a substantial concentration of the inhibited ADC-PanZ.RCoA complex. Full inhibition of ADC will therefore only occur at sufficiently high CoA concentrations. Given that CoA concentrations in the cell can be as high as 4 mM (Bennett et al., 2009) compared with an affinity of 2 mM for the PanZ.CoA complex, we anticipate that activated ADC actually exists predominantly as the inhibited ADC-PanZ.RCoA complex in the cell. We suggest that this is the principal regulatory point for pantothenate biosynthesis and, by extension, de novo CoA   (Rock et al., 2003). PanZ controls both the activation (ii) and catalytic activity (iii) of PanD in a CoAdependent fashion. In contrast to b-alanine, formation of D-pantoate is reversible and b-alanine is the sole committed step in the biosynthetic pathway. CoA-dependent regulation by PanZ can therefore control the flux through the whole pantothenate biosynthetic pathway. Finally, the active form of PanD, ADC, undergoes substrate-mediated inhibition (iv) after approximately 300 turnovers, limiting the total flux through the pathway (Konst et al., 2009). biosynthesis, and is sufficient to globally negatively regulate pantothenate biosynthesis ( Figure 7D). Regulation of biosynthesis by terminal metabolites via allosteric inhibition is well characterized; the involvement of a second protein is, however, unusual. PanZ was first identified as a putative N-acetyltransferase and, while there is no indication that it does not also carry out this enzymatic activity in vivo, ADC is not its substrate (Stuecker et al., 2012b). Acetylation is an ubiquitous post-translational modification in Gram-negative bacteria, and numerous such uncharacterized acetyltransferases exist (Jones and O'Connor, 2011). The limited phylogenetic distribution of PanZ (and the absolute requirement for it in only a small subset of bacteria) suggests to us that PanZ was first recruited to regulate catalysis by ADC in response to CoA levels, independent of its acetylation activity, before becoming essential for the activation of ADC.
The PanD-PanZ interaction provides the first regulatory mechanism for the pantothenate biosynthetic pathway in E. coli. The biosynthetic pathways for many of the other B vitamins are tightly regulated by mechanisms such as conserved riboswitches, despite the apparent lack of a significant fitness cost for vitamin overproduction. No equivalent regulatory mechanism for pantothenate has previously been identified. Here we have shown that in the case of pantothenate biosynthesis, this regulation is instead provided by this pantothenate metabolite-binding protein. We further suggest that regulation of catalysis by such metabolite-binding proteins could be a widespread phenomenon in other biosynthetic pathways and that many other such regulatory systems await discovery.

SIGNIFICANCE
Biosynthesis of many vitamins in bacteria is tightly regulated, yet no evidence for regulation of pantothenate biosynthesis has been reported. Here, we report the structure of the complex formed between the zymogen of aspartate decarboxylase, PanD, and its activating factor PanZ. Formation of this complex is dependent on the presence of CoA, the cofactor derived from pantothenate. The reported structure reveals the structural basis for the CoA dependence of interaction and provides a model for how formation of a PanD-PanZ.RCoA complex stimulates activation of the zymogen to form the activated enzyme. The involvement of CoA in this process suggested a paradoxical situation, in which an increase in concentration of a cofactor would stimulate its own biosynthesis. Further investigation, however, revealed that PanZ.RCoA can, in fact, act to inhibit catalysis by the activated enzyme. This suggests that the physiological role of PanZ is 2-fold: to catalyze the generation of active enzyme and to regulate subsequent activity by this enzyme. We suggest that this activity may have evolved by recruitment of an existing CoA-utilizing enzyme into a regulatory role, and that such regulatory protein-protein interactions may operate in other biosynthetic pathways.

EXPERIMENTAL PROCEDURES
Full details of protein expression, X-ray crystallography, NMR, SAXS analysis complementation assays, and ITC can be found in Supplemental Information.

Protein Crystallization and Structure Solution
Protein complexes were prepared with a 10:11 ratio of PanD to PanZ at a total protein concentration of 9-11 mg ml À1 , and a 2-fold molar excess (with respect to PanZ) of AcCoA added. Crystals were obtained in 20% (w/v) polyethylene glycol 3350, 0.1 M bis-Tris propane (pH 7.4), and 0.2 M potassium thiocyanate. Following data collection, X-ray data were indexed and integrated in space group I4 using iMosflm (Leslie and Powell, 2007), and scaled and merged using Aimless (Evans and Murshudov, 2013). Phasing was carried out by molecular replacement using Molrep (Vagin and Teplyakov, 1997) and the coordinates from PDB 4AZD (ADC) and 2K5T (PanZ). The solutions were subjected to iterative rounds of manual rebuilding and refinement using Coot (Emsley et al., 2010) and Refmac5 (Murshudov et al., 2011).

SAXS
SAXS data were collected on beamline 4-2 of the Stanford Synchrotron Radiation Light Source (SSRL). PanD(T57V) and PanZ were mixed together with AcCoA in a 1:1:2 ratio, respectively. The data were integrated with SASTool and examined with PRIMUS (Konarev et al., 2003). Higher-order dimeric species were identified using OLIGOMER (Petoukhov et al., 2012), and a simulated scattering profile computed by FoXS (Schneidman-Duhovny et al., 2013) was subtracted from the SAXS data to yield the scattering for the isolated complex. Ab initio shape reconstructions were generated by DAMMIF (Franke and Svergun, 2009) using P4 symmetry and averaged with DAMAVER . CORAL (Petoukhov et al., 2012) was used determine the positions of the crystallographically disordered residues. SUPCOMB (Kozin and Svergun, 2001) was used to align the high-resolution model with the envelope reconstruction.
ITC ITC experiments were performed using a Microcal iTC200 (GE) or Microcal VP-ITC (GE) thermostated at 25 C as described in the Supplemental Information. For global fitting, data were integrated using NITPIC (Keller et al., 2012) before global fitting to a one-site binding model in SEDPHAT (Houtman et al., 2007).

NMR
All protein NMR experiments were run at 25 C in 50 mM Tris-HCl (pH 7.5), 0.1 M NaCl, and 0.1 mM DTT. 1 H-15 N HSQC spectra were obtained using a 500 MHz Varian Inova spectrometer. A spectral window of 8,000 Hz for 1 H (2,048 complex points) and 1,800 Hz for 15 N (92 increments) was used with 96 scans per increment. A protein concentration of 0.2 mM PanZ was used with a total acquisition time of 6 hr. All data were processed with NMRPipe (Delaglio et al., 1995) and analyzed using NMRView (Johnson and Blevins, 1994).