Crystal Structures of Human Pantothenate Kinases

Pantothenate kinase (PanK) catalyzes the first step in CoA biosynthesis and there are three human genes that express four isoforms with highly conserved catalytic core domains. Here we report the homodimeric structures of the catalytic cores of PanK1α and PanK3 in complex with acetyl-CoA, a feedback inhibitor. Each monomer adopts a fold of the actin kinase superfamily and the inhibitor-bound structures explain the basis for the allosteric regulation by CoA thioesters. These structures also provide an opportunity to investigate the structural effects of the PanK2 mutations that have been implicated in neurodegeneration. Biochemical and thermodynamic analyses of the PanK3 mutant proteins corresponding to PanK2 mutations show that mutant proteins with compromised activities and/or stabilities correlate with a higher incidence of the early onset of disease.

Pantothenate kinase (PanK) catalyzes the first step in CoA biosynthesis and there are three human genes that express four isoforms with highly conserved catalytic core domains. Here we report the homodimeric structures of the catalytic cores of PanK1␣ and PanK3 in complex with acetyl-CoA, a feedback inhibitor. Each monomer adopts a fold of the actin kinase superfamily and the inhibitor-bound structures explain the basis for the allosteric regulation by CoA thioesters. These structures also provide an opportunity to investigate the structural effects of the PanK2 mutations that have been implicated in neurodegeneration. Biochemical and thermodynamic analyses of the PanK3 mutant proteins corresponding to PanK2 mutations show that mutant proteins with compromised activities and/or stabilities correlate with a higher incidence of the early onset of disease.
Pantothenate kinase (PanK) 2 is the first and rate-limiting enzyme in the CoA biosynthetic pathway, which catalyzes the phosphorylation of vitamin B5, pantothenate (1). In human, there are three genes that express four catalytically active PanK isoforms. PanK1␣ and PanK1␤ arise from the PANK1 gene through the use of alternate initiating exons (2). PANK2 encodes a protein that is processed twice and localizes to the mitochondria (3,4) and PANK3 encodes a single cytosolic isoform (5). All of these PanK isoforms share a common, highly homologous catalytic core that is composed of about 355 residues (e.g. the identity of ϳ83%). PanK1␤ and PanK3 consist of the catalytic core alone, whereas PanK1␣ has a feedback regulatory domain in an amino-terminal extension (2). The NH 2terminal domain of PanK2 includes two mitochondrial targeting signals, which are removed proteolytically concomitant with its mitochondrial localization (3,6). In addition, PANK4 encodes a protein with sequence similarity to the other PanKs, but it lacks an essential catalytic glutamate for enzyme activity and has a long carboxyl-terminal extension of unknown function (3).
Mutations in PANK2 have been linked to the inherited pantothenate kinase-associated neurodegeneration (PKAN) (7), which constitutes a subset of neurodegeneration with brain iron accumulation, formerly known as Hallervorden-Spatz syndrome (8). Patients with PKAN exhibit motor symptoms, such as dystonia, dysarthria, intellectual impairment, and gait disturbance (9) and early onset patients have rapidly progressive disease, whereas later onset patients have a slowly progressive, atypical disease (9,10). The reason for brain iron accumulation in PKAN is still unknown. A current hypothesis of PKAN pathogenesis states that PANK2 mutations lead to a loss of PanK2 function and reduced phosphopantothenate, resulting in the accumulation of cysteine-containing molecules and CoA deficiency (6,7). These cysteine-containing substrates may undergo rapid autoxidation in the presence of iron, leading to free radical generation and cell damage (6,7). PANK2 mutations are clustered in the catalytic core domain, and insertion, deletion, and frameshift mutations in the PANK2 gene are linked clinically to early onset disease (10), indicating that null mutants of PanK2 lead to rapidly progressive PKAN. In contrast, the missense mutations have mixed early and late onset patterns (10), and a set of missense mutant proteins expressed in transiently transfected cells are active (4,11), suggesting that defects unrelated to catalysis contribute to PKAN disease.
A common feature of the PanK isoforms is that they are feedback inhibited by CoA and its thioesters. This regulation is a primary mechanism that controls PanK activity and hence the intracellular CoA level (2,5,12). Recent biochemical data show that acetyl-CoA is a more potent feedback inhibitor than free CoA for mouse PanKs (5,11). Structural studies on PanK have been reported only in bacterial orthologs (13)(14)(15)(16)(17). The crystal structures of Escherichia coli PanK reveal why its activity, unlike mouse PanKs, is more potently inhibited by free CoA, (13,14). Staphylococcus aureus PanK (SaPanK) is the most similar in sequence to the mammalian PanK catalytic core (e.g. the identity of 23%), but is refractory to feedback regulation by CoA (17,18). Due to the low sequence homology, the structural interpretation of the disease-causing human PanK2 mutations using the SaPanK model is limited to three missense mutations that directly affect the active site (17).
We have determined the homodimeric structures of the catalytic cores of human PanK1␣ and PanK3 in complex with acetyl-CoA. These crystallographic data provide the first look at the structures of metazoan PanKs and their allosteric regulation. The high degree of similarity between human PanK3 and PanK2 allows an analysis of the mutations associated with PKAN. PanK2 missense mutations mapped onto the active site, dimerization interface, and protein interior tend to be associated with the early onset of PKAN phenotypes and their corresponding PanK3 mutant proteins show PanK inactivation and destabilization. On the other hand, mutations of surface residues have less severe effects on enzyme function and correlate with the mixed early and late onsets of disease.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The genes encoding the catalytic core domains of human PanK1␣ (residues 231-597, abbreviated PanK1 below) and PanK3 (residues 12-368) were amplified using PCR and cloned into the expression vector pET28a to obtain NH 2 -terminal His 6 tag fusion proteins. E. coli BL21(DE3) (Stratagene) was transformed with the resulting plasmids and both recombinant PanK1 and PanK3 proteins were overexpressed upon 1.0 mM isopropyl 1-thio-␤-D-galactopyranoside induction for 18 h at 18°C. The proteins were purified using nickel-nitrilotriacetic acid affinity and Superdex 200 (GE Healthcare) gel-filtration columns. Purified PanK1 and PanK3 samples were concentrated to 25 and 30 mg/ml, respectively, and stored at Ϫ80°C in the gel filtration buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl, 10 mM dithiothreitol).
Crystallization-PanK proteins were crystallized using the sitting drop vapor diffusion method at 18°C. To achieve full occupancy of bound ligands, the proteins were mixed with 5-fold molar excess of an ATP analogue AMP-PNP or acetyl-CoA. The mixtures were incubated overnight at 4°C. Equal volumes of protein solution and reservoir solution were combined for crystallization. Crystals of PanK1 were grown at 0.1 M Na-Hepes (pH 7.5), 28% PEG400, 0.2 M CaCl 2 , and 0.02 M tris(2carboxyethyl) phosphine hydrochloride-HCl, and PanK3 was crystallized in 0.1 M Tris-HCl (pH 8.5), 16% PEG3350, 0.2 M ammonium citrate, and 0.02 M octaethyleneglycol dodecyl ether (C12E8). The mercury derivatives were obtained by soaking PanK1 crystals in the reservoir solution with 5 mM ethylmercury thiosalicylate overnight. All crystals were cryoprotected in the 50:50 mixture of Paratone-N and mineral oil, and flash-frozen in liquid nitrogen. The diffraction quality crystals of PanK1 were obtained in the presence of acetyl-CoA. Although AMP-PNP was added for the crystallization of PanK3, we found acetyl-CoA molecules in the structure that co-purified with PanK3.
Data Collection and Structure Determination-Diffraction data were collected on beamline 19-BM at the Advanced Pho-ton Source, running at 100 K and processed using the HKL2000 program package (19). The program SOLVE (20) was used to locate four mercury sites and to calculate initial phases of PanK1 and the program RESOLVE (21) to improve the initial phases. The model of PanK1 was completed in several repetitions of manual building with the program O (22). The structure of PanK3 was solved by the molecular replacement method using the program MOLREP in the CCP4 package (23) and the PanK1 structure as a search model, and built using the program O (22). Finally, both PanK structures were refined with the program REFMAC5 (24). The crystals of PanK1 contained a dimer in the asymmetric unit and the two monomers were related by noncrystallographic 2-fold symmetry. In PanK3, two dimers were in the asymmetric unit. A dimer state is the most common functional organization for PanK proteins (4,5,25). Data collection and refinement statistics are given in Table 1.
Site-directed Mutagenesis of PanK3-Based on the PanK2 mutations identified in PKAN patients (10), site-directed mutagenesis in PanK3 was accomplished using the Stratagene QuikChange kit (Stratagene). Prior to protein expression, the sequences of all mutations were verified by sequencing, and each mutant protein was purified as described for the wild-type protein, except the concentration of dithiothreitol was 1.0 mM.
PanK Activity Assay-The effect of acetyl-CoA on the PanK1␣ and PanK3 K m values for ATP was investigated by using the assay previously described (2,12). The standard reaction mixture contained 180 M D-[1-14 C]pantothenate (Amersham Biosciences), 10 mM MgCl 2 , 0.1 M Tris-HCl (pH 7.5), 200 ng of PanK1␣ or 100 ng of PanK3 and increasing concentrations of ATP and acetyl-CoA, as indicated. The mixture was incubated at 37°C for 10 min and the radioactive product was quantitated by scintillation counting as described previously (11). Activities of PanK3 mutant proteins were compared with the wild-type using the standard pantothenate kinase assay (11). Briefly, the reaction mixture contained 90 M D-[1-14 C]pantothenate, 2.5 mM ATP, 10 mM MgCl 2 , 0.1 M Tris-HCl (pH 7.5), and 100 ng of purified protein, and assayed as described above. Relative activities of the mutant proteins were checked three times and these data were combined.
UV-visible Spectra and Mass Spectrometry-UV-visible spectra of PanK3 wild-type and mutant proteins (1 mg/ml in 10 mM Tris-HCl (pH 8.0), 50 mM NaCl, 0.5 mM EDTA, 0.5 mM dithiothreitol, and 50% (v/v) glycerol) were recorded as described previously (27). The supernatant (100 l) from a heat-denatured sample of PanK3 (1.8 mg/ml in 50 mM ammonium acetate, pH 7.0) was desalted on a C-8 MicroTip column (Harvard Apparatus), washed with water, and the proteinbound small molecules were eluted with methanol (100 l), dried under N 2 flow, and resuspended in methanol:water (1:1, v/v, 10 l). Mass spectrometry analysis was performed using a Finnigan TM TSQ Quantum TM (Thermo Electron Corporation, San Jose, CA) triple quadrupole mass spectrometer equipped with the nanospray ion source.
Stability Studies by Differential Static Light Scattering-The thermostabilities of PanK3 and its mutant proteins were studied using differential static light scattering (28,29), monitoring protein stability by its aggregation properties. Protein samples at 0.2 mg/ml were heated from 25 to 80°C at a rate of 1°C/min in clear bottom 384-well plates (Nunc) in 50 l of buffer (100 mM Hepes, 150 mM NaCl, pH 7.5). Protein aggregation was measured by recording the scattered light using a CCD camera and taking images of the plate every 0.5°C. The pixels intensities in a preselected region of each well were integrated using proprietary software to generate a value representative of the total amount of scattered light in that region. These intensities were plotted against temperature for each sample well and fitted to Equation 1 as described by Matulis et al. (30), where I o and I f are the intensities of light scattered by the protein before and after denaturation, ⌬H is the enthalpy of unfolding, R is the general gas constant, T agg is the temperature of aggregation that represents the melting temperature, and ⌬C p is the heat capacity difference between the unfolded and folded states of the protein. ⌬C p of 4.93 kcal/mol was calculated based on the protein sequence and was used in all subsequence calculations (31). The free energy difference ⌬⌬G between the wildtype and each mutant protein at the temperature of aggregation of the wild-type was calculated using Equation 2, where ⌬⌬G m(Twt) is the denaturation free energy difference between the wild-type and mutant PanK3, T wt and T m are the aggregation temperatures of the wild-type and mutant proteins, respectively. Stability Studies by Isothermal Aggregation (ITA)-Isothermal aggregation was measured at 37°C. Protein samples at 0.2 mg/ml were incubated at 37°C in clear bottom 384-well plates (Nunc) in 50 l of buffer (100 mM Hepes, 150 mM NaCl, pH 7.5). Protein aggregation was monitored by detecting the scattered light by taking images every 30 s using a CCD camera for 2 h. Images were analyzed and the intensities were plotted versus time. All curves were normalized to compensate for small differences in initial intensities. The time it takes to have 50% of the protein in the aggregation state, here on, is termed ITA 50 . Each experiment was repeated 5 times and the average has been reported here.

Overall Structure Description
The PanK1 and PanK3 proteins we crystallized represent the catalytic core domain, which is highly conserved in all human PanK isoforms. The identity between the catalytic core domains of the two isoforms is ϳ84% (ClustalW 1.7). The overall monomer fold places the human PanKs in the actin fold protein kinase superfamily (32), and contains two domains (Fig.  1A). The A domain consists of a six-stranded ␤-sheet (strands Ab1-Ab6) flanked by five helices (Ah1-Ah3 on one side and Ah4 and Ah11 on the other side) and a glycine-rich loop, characteristic of many nucleotide-binding sites (33, 34), lies  . S1). As the A domain in the PanK1 model was partly incomplete due to poor electron density (supplemental Fig. S1), the PanK3 structure was used for "Results " and "Discussion, " and only its differences from PanK1 were pointed out.

Dimerization Interface
The major portion of the dimerization interface is composed of two long helices, Bh9 and Bh9Ј from each monomer, which interact with each other in parallel resulting in a V-shaped conformation ( Fig. 1B and supplemental Fig. S2A). Helices Bh9 and Bh9Ј widen at their NH 2 -terminal ends and form a four-helix bundle with two short helices Bh5 and Bh5Ј (supplemental Fig.   S2A). The center of the four-helix bundle is lined by non-polar residues forming a compact hydrophobic core. The narrow COOH-terminal ends of helices Bh9 and Bh9Ј are held together by van der Waals interactions between Val 304 and Met 307 of each monomer, and are capped by Asn 311 and Asn 311 Ј which face each other. The middle of helices Bh9 and Bh9Ј is attached by both direct and water-mediated hydrogen bonds between Asn 300 and Asn 300 Ј. Additional van der Waals interactions exist between Bh5 of one monomer and the extended loop of the second monomer contributing to the stabilization of the PanK dimer (supplemental Fig.  S2B).

The Acetyl-CoA Binding Site
In both PanK1 and PanK3 structures, the acetyl pantetheine diphosphate moiety of acetyl-CoA is bound parallel to the ␤ strands of the B domain of one monomer and covered on the other side by the extended loop of the second monomer ( Fig. 2A). The main interactions of acetyl-CoA with PanKs occur via the acetyldiphosphopantetheine moiety (Fig. 2B). The 3Ј-phosphoadenonsine moiety is poorly defined in the electron density map, suggesting that it is disordered and thus not included in the final PanK structures. Residues lining the binding site of the acetyldiphosphopantetheine are identical in both the PanK structures, with the exception of residues 266 and 340 involved in hydrophobic interactions (Fig. 2B). The acetyldiphosphopantetheine moiety forms direct or water-mediated hydrogen bonds and van der Waals interactions with the protein core. The polar residues involved in hydrogen bonds are from the ␤ strands of one monomer (Bb9 -Bb11) and the non-polar residues involved in van der Waals interactions are contributed by the extended loop and Bh10Ј of the second monomer ( Fig. 2A). Acetyl-CoA was very tightly bound to PanK3 in the structure, consistent with the low IC 50 value for acetyl-CoA inhibition of the enzyme (5).

Predicted Binding Sites for ATP and Pantothenate
The ATP binding site of PanK3 was modeled on the basis of its structural similarity to the SaPanK⅐AMP-PNP complex (Protein Data Bank code 2EWS) (17) and the location of the conserved glycine-rich loop (Fig. 3A). Superposition of the B domain of PanK3 and the corresponding domain of the SaPanK⅐AMP-PNP complex placed the ATP molecule between the two domains of the human PanKs. The two domains of PanK3 and also PanK1 were in an "open" conformation com- pared with the "closed " conformation SaPanK⅐AMP-PNP (Fig.  3A). The binding of the acetyl-CoA kept PanK in the open conformation, characteristic of the ligand-free structures of actinfold proteins (32). Acetyl-CoA was a competitive inhibitor of both PanK1 and PanK3 (Fig. 3, B and C), suggesting that the binding of acetyl-CoA interfered with the ATP binding by stabilizing the open conformation. Based on the superimposed structures of PanK3⅐acetyl-CoA and SaPanK⅐AMP-PNP, the ␣,␤-phosphates of acetyl-CoA and the ␤,␥-phosphates of ATP occupied the same space (Fig. 3A), explaining the competitive nature of acetyl-CoA and ATP for PanK. Also, we found that a high proportion of purified PanK3 was bound to acetyl-CoA. Heat denaturation of the protein released the bound compound into the supernatant, and analysis of this supernatant by mass spectrometry identified it as acetyl-CoA (Fig. 3D). This indicates that acetyl-CoA remained bound to PanK3 during purification.
The Pan binding site was predicted on the basis of the location of bound acetyl-CoA in PanK3. The same residues lining the binding pocket for the Pan moiety of acetyl-CoA (Fig. 2B) were considered for the binding of Pan. A key residue was Arg 207 that formed a salt bridge to the carboxamide group of the Pan moiety of acetyl-CoA (the carboxyl group of Pan) (Fig. 2B). This salt bridge interaction would anchor one end of Pan. A similar salt bridge interaction was observed in the Pseudomonas aeruginosa PanK structure (17). However, Glu 138 , another key residue, predicted to act as a general base for the nucleophilic activation of the hydroxyl group of Pan in SaPanK (17), was not located adjacent to the 4Ј-hydroxyl group of Pan (data not shown), suggesting that the enzyme exists in a catalytically inactive state. This finding is consistent with the idea that the acetyl-CoA-bound PanK3 structure adopts the open conformation (see above). Therefore, a switch to the closed conformation by the binding to ATP and/or Pan is required for the nucleophilic activation of the OH group of Pan by Glu 138 .

Crystallographic Mapping of Missense Mutations Associated with PKAN Disease
The sequence similarity in the catalytic core domains of PanK3 and PanK2 allowed mapping of the PanK2 missense mutations (10) onto the PanK3 structure (Fig. 4A) and we correlated the locations of the missense mutations with their biochemical and disease phenotypes ( Table 2). The mutations were grouped into three categories.
Active Site Mutations-Two PanK2 mutations (G219V and G521R) mapped to the ATP binding site and of these, G521R is most frequently observed in PKAN patients (10). Gly 321 (Gly 521 of PanK2) was located in the loop between Bb14 and Bh10 on one side of the ATP binding site, whereas Gly 19 (Gly 219 of PanK2) formed part of the glycine-rich loop (supplemental Fig. S3A). These glycine residues create space for the docking of the ATP phosphates, and the introduction of a side chain by mutation, especially a bulky one, was predicted to prevent proper ATP binding. A previous study of both mutations in PanK2 has shown that these mutant proteins were catalytically inactive, and the PanK2-(G521R) failed to properly fold (11).
Mutations at the Dimer Interface and Protein Interior-The PanK3 structure placed eight PanK2 missense mutations (L413P, D447N, S471N, I497T, N500I, I501T, A509V, and N511D) at the dimer interface (Fig. 4A). Of these, Leu 213 (Leu 413 of PanK2) made van der Waals interactions with nonpolar residues (supplemental Fig. S3B), and Asp 247 , Ser 271 , Asn 300 , and Asn 311 (Asp 447 , Ser 471 , Asn 500 , and Asn 511 of PanK2) formed direct or water-mediated hydrogen bonds to helix Bb9Ј of the second monomer (supplemental Figs. S2A and  S3, B and C). On the other hand, Ile 297 , Ile 301 , and Ala 309 (Ile 497 , Ile 501 , and Ala 509 of PanK2) were indirectly involved in PanK3 dimerization. Many water molecules were trapped at the dimer interface and formed a network through the interaction with amino acid residues along the dimer interface (supplemental Fig. S3, B and C). This network, conserved in both PanK1 and PanK3 structures, is considered to tighten the dimerization of two monomers and the mutations at Asp 247 , Ser 271 , and Asn 300 of PanK3 were predicted to perturb the ordered water molecules. Three PanK2 mutations, L282V, A398T, and L563P, were located in the protein interior (Fig.  4A) and the corresponding residues (Leu 82 , Ala 198 , and Leu 363 ) in PanK3 were involved in the hydrophobic interactions with surrounding nonpolar residues.
Mutations on the Surface of PanK-Fifteen PanK2 mutations were located on the surface of the PanK3 structure with their side chains exposed to solvent (Fig. 4B). Most notable was PanK2-(T528M), the second most common mutation in PKAN disease (10,35). Thr 528 of PanK2 was equivalent to Thr 328 of PanK3 and Met 553 of PanK1, which was located at the NH 2 terminus of helix Bh10 (supplemental Fig. S3D). Based on our sequence alignment (data not shown) and both PanK structures, the side chain hydroxyl group of Thr 528 in PanK2 was predicted to form a hydrogen bond to Arg 532 and PanK2-(T528M) was predicted to break this interaction as shown in wild-type PanK1. The side chain of Lys 332 in PanK3 formed water-mediated hydrogen bonds to residues in the loop between Bb7 and Bb8 (supplemental Fig. S3D), which could be achieved by the corresponding Arg 532 in PanK2. Interestingly, this arginine mutation, PanK2-(R532W), was found in PKAN patients, supporting the idea that the Thr 528 -Arg 532 interaction is important in positioning the Bb7-Bb8 loop for some physiological function of PanK2.

Biochemical and Thermodynamic Characterization of PanK3 Mutant Proteins
We cloned, expressed, and purified 27 PanK3 mutant proteins that correspond to the PanK2 missense mutations described above. The effects of these mutations on activity and stability of the protein were investigated using in vitro PanK activity and thermostability assays (Fig. 4C, Table 2). The mutations located at the ATP binding site severely affected the catalytic activity and stability of the protein. PanK3-(G19V) (PanK2-(G219V)) had negligible activity (Ϸ1%) and its T agg (36.7°C) was lower than the wild-type (45.4°C) with a significant destabilization energy of Ϫ2.46 Kcal/mol ( Table 2). PanK3-(G321R)(PanK2-(G521R)) was insoluble and could not be studied with either technique.
Most of the surface mutations exhibited moderate or no significant reduction in their thermostability (T agg ) or activity (Fig. 4C). Of these, PanK3-(R64W) (PanK2-(R264W)) was stable with a T agg value similar to the wild-type and low ⌬⌬G (0.16 Kcal/mol), but the activity was reduced to 36% of the wild-type. On the other hand, the T agg values of the three other surface mutant proteins, R86C (PanK2-(R286C)), S151P (PanK2-(S351P)), and K332W (PanK2-(R532W)), decreased to ϳ40°C with ⌬⌬G between Ϫ0.9 and Ϫ1.57 Kcal/ mol; nevertheless, their activities were similar to that of the wild-type (Table 2). These findings suggest that the effects of these surface mutations are not always the same on both pro- Top histogram (blue) shows the kinase activities of PanK3 mutant proteins relative to the wild-type. Each bar represents the mean of at least three independent measurements. Error bars show S.D. Bottom histogram (red) indicates denaturation free energy differences (⌬⌬G) between wild-type PanK3 and its mutant proteins. SDS-PAGE is shown between the two histograms. Insoluble mutant proteins are marked with an asterisk. tein thermostability and activity. The other 11 surface mutations including T328M (PanK2-(T528M)) displayed moderate or no protein destabilization and their activities were similar to the wild-type.
Some of the surface mutations such as PanK3-(T34A), PanK3-(R49P), PanK3-(R64W), PanK3-(N155S), PanK3-(C228Y), and PanK3-(T328M) had only a negligible effect on the stability of the protein (⌬T agg Ͻ 2°C) ( Table 2), which is within the experimental error of differential static light scattering. We performed further tests with these mutant proteins by ITA, a more sensitive technique than differential static light scattering that monitors the aggregation of the proteins at constant temperature under a defined set of buffer conditions (36,37). At 37°C, the time required to reach 50% aggregation (ITA 50 ) for wild-type PanK3 was 33.5 min, whereas PanK3-(T328M) corresponding to PanK2-(T528M) was 25 min (supplemental Table S1). Most of the PanK3 surface mutant proteins aggregated more quickly than the wild-type protein (supplemental Table S1), suggesting that the mutations had reduced protein stability.

DISCUSSION
PanK Regulation by Feedback Inhibition-Crystallographic identification of the acetyl-CoA binding sites of PanK1 and PanK3 provides an explanation for the weaker inhibitory effects of CoA. CoA and its thioesters (acetyl-CoA, malonyl-CoA, and palmitoyl-CoA) inhibit PanKs, but they show different potencies for mammalian PanK isoforms (2,5). CoA inhibits mouse PanK3 with a significantly higher IC 50 than acetyl-CoA (5). Similarly, CoA inhibits mouse PanK1␣ with the IC 50 of 80 M, whereas acetyl-CoA has an IC 50 of 5 M (5). The structures of catalytic core domains of two PanK isoforms suggest that the lower potency of CoA, compared with its thioesters, is attributed to the lack of the thioester carbonyl oxygen that forms a hydrogen bond with the main chain amide of a valine (supplemental Fig. S4). Additionally, the binding pocket for the acetyl moiety of acetyl-CoA is partially exposed to solvent, providing the opening for longer chain acyl moieties to interact with the same site.
The monomer folding and dimer assembly of human PanK catalytic core domains are similar to that of SaPanK (17). Unlike human PanKs, however, SaPanK is refractory to inhibition by CoA and its thioesters (18). Compared with the human PanK3 structure, SaPanK has two residues that line the acetyl-CoA pocket that would hinder the binding of feedback inhibitors (Ala 337 of PanK3 versus Tyr 240 of SaPanK and Trp 341 of PanK3 versus Arg 244 of SaPanK) (supplemental Fig. S4). The substitution of alanine to a bulky residue, tyrosine, may cause steric hindrance for the binding of CoA and the substitution of tryptophan to arginine introduces a positive charge, disrupting the hydrophobic binding pocket for the sulfhydryl group of CoA or the acyl groups of the thioesters. These two residues in PanK3 are also conserved in the other PanK isoforms (supplemental Fig. S4). PanK2 Instability as a Determinant of PKAN-Structural analysis of the catalytic cores of PanK1 and PanK3 provides an opportunity to understand why many of the mutations that give rise to PKAN disease abolish the catalytic activity of PanK. Of the two mutations at the ATP-binding site, PanK3-(G19V) exhibits large reductions in both protein stability and enzymatic activity, and PanK3-(G321R) is insoluble (Fig. 4C). Low enzymatic activity, structural instability, and abnormal processing are observed with the corresponding human PanK2-(G219V) and PanK2-(G521R) mutations (4,11). Many of the inactivating mutations are located in the dimerization interface and require between Ϫ0.77 and Ϫ4.36 Kcal/mol less free energy than the wild-type to unfold. Five of eight dimerization mutations give rise to catalytically inactive proteins in vitro, although three dimerization mutations have a less severe effect on activity (Fig. 4C). Similarly, one of the three mutations in the inner protein core, PanK3-(L363P), is insoluble (data not shown). The corresponding PanK2-(L563P) mutant protein is catalytically inactive (11). Our findings on the activities of the PanK3 mutant proteins reflect data obtained with the authentic PanK2 protein (4,11), supporting the close relationship between the structures of these proteins. These findings suggest that proper monomer folding and dimerization are critical for the stability of PanK3 and enzymatic activity. More than 80% of patients who have the PanK2 active site or dimerization mutations are linked to the clinically early onset phenotype (10) ( Table 2), thus structural destabilization of monomers that cannot form dimers or improperly formed dimers appears as a major contributor to PKAN.
The group of mutations that has confounded the field maps predominantly to the protein surface. Hydrophilic or charged residues are frequently substituted with hydrophobic residues (11 of 15 surface mutations), and these mutations do not result in the loss of PanK2 (11) or PanK3 catalytic activity and have ⌬⌬G values Ͻ2 Kcal/mol in the PanK3 model (Table 2). Surface residues of globular proteins are mainly hydrophilic, and the stabilities of globular proteins are altered by changing surface residues involved in salt bridge or hydrogen bond interactions (39 -44). Substitutions of solvent-exposed hydrophilic residues to hydrophobic residues may locally disrupt the hydrophilicity of the protein (45), destabilizing protein conformation. Of 15 surface mutant proteins of PanK3, eight have moderate reductions (⌬T agg Ͼ 2°C) in the T agg values and 14 are rapidly aggregated in vitro at 37°C compared with wild-type, but retain activity ( Table 2 and supplemental Table S1). It will be important to extend these experiments from the PanK3 model to actual PanK2 mutant proteins because the degree of destabilization by elimination of surface salt bridges is highly dependent on the context in which they occur, and there are 56 amino acid differences between PanK3 and PanK2 plus an additional 60-residue amino-terminal extension in PanK2 that interacts with the catalytic core in an unknown way and may either positively or negatively affect the degree of protein destabilization by surface mutation changes.
Kotzbauer et al. (4) examined the stability and processing of PanK2 mutant proteins in cultured cells. PanK2-(G521R) has a shorter half-life and is abnormally processed compared with wild-type, consistent with the compromised structural integrity of the corresponding PanK3 mutant protein. These data indicate that the other PanK2 mutant proteins that correspond to PanK3 mutant proteins that alter protein stability may have a shorter half-life in vivo than their wild-type counterpart, and perhaps explains why these mutations are associated with early onset of the disease. However, the surface mutations have modest effects on PanK3 stability and similar in vivo experiments with the PanK2-(T528M) and PanK2-(T234A) did not reveal any differences between these two mutants and the wild-type with respect to activity, stability, and processing (4). Our biophysical studies on the analogous PanK3 isoform may suggest that there may still be more subtle effects on stability than noted in these experiments, or that alternatively, the surface mutations may relate to compromising the binding of PanK2 to as yet unidentified protein partners or regulatory ligands. On the other hand, PanK2 is translocated into the mitochondria (3,4) and mitochondria produce most of the reactive oxygen species (46), suggesting the damage of mitochondrial PanK2 by reactive oxygen species is possible. Hydrogen peroxide is one of the major reactive oxygen species in mitochondria and modifies sulfur-containing amino acids (26,38). Some PanK2 surface mutant proteins such as PanK2-(R278C), -(R286C), and -(T528M) may be more susceptible to hydrogen peroxide modification than wild-type and when modified, enhance protein destabilization in mitochondria.
In summary, we have presented the first structure of human PanK1 and PanK3 isoforms in complex with acetyl-CoA. The structures show a common architecture of the catalytic cores of human PanK isoforms and explain the allosteric regulation during CoA biosynthesis. Mapping of PanK2 mutations to the PanK3 structure and mutational studies on PanK3 isoform reveal that mutations at the ATP binding site, dimer interface and protein interior compromise their enzyme activities and stabilities and are mostly associated with the early onset of the PKAN phenotype, whereas surface mutations are catalytically active but show slight instabilities relative to the wild-type and are linked to the mixed onset of disease phenotype. These findings contribute to our understanding of the dysfunctional effects of polymorphic PanK2 mutations and help decipher the manner in which enzymatically active mutant proteins are linked to disease.