Structural evidence that brain cyclic nucleotide phosphodiesterase is a member of the 2H phosphodiesterase superfamily.

2',3'-Cyclic-nucleotide 3'-phosphodiesterase (CNP) is an enzyme abundantly present in the central nervous system of mammals and some vertebrates. In vitro, CNP specifically catalyzes the hydrolysis of 2',3'-cyclic nucleotides to produce 2'-nucleotides, but the physiologically relevant in vivo substrate remains obscure. Here, we report the medium resolution NMR structure of the catalytic domain of rat CNP with phosphate bound and describe its binding to CNP inhibitors. The structure has a bilobal arrangement of two modules, each consisting of a four-stranded beta-sheet and two alpha-helices. The beta-sheets form a large cavity containing a number of positively charged and aromatic residues. The structure is similar to those of the cyclic phosphodiesterase from Arabidopsis thaliana and the 2'-5' RNA ligase from Thermus thermophilus, placing CNP in the superfamily of 2H phosphodiesterases that contain two tetrapeptide HX(T/S)X motifs. NMR titrations of the CNP catalytic domain with inhibitors and kinetic studies of site-directed mutants reveal a protein conformational change that occurs upon binding.

The abundance of the enzyme 2Ј,3Ј-cyclic nucleotide 3Ј-phosphodiesterase (CNP 1 ; EC 3. 1.4.37) in the central nervous system of all mammals and some other vertebrates such as amphibians and birds has long been an enigma. This derives from the continuing failure to identify a physiological substrate for this enzyme. CNP has an apparent specificity for nucleoside 2Ј,3Ј-cyclic phosphate, which it cleaves to 2Ј-nucleotide end products, none of which (with the exception of NADP/NADPH) are found in metabolite pools. The last 4 decades of research have failed to attribute a function to this protein, although many possibilities have been considered (extensively reviewed in Refs. [1][2][3]. More recently, RICH, a neuronally associated homolog of CNP, has been discovered in fish (4,5), and the catalytic active site of CNP has been investigated (6).
Members of this enzyme superfamily occur across a vast range of organisms ranging from bacteria to mammals. It has been suggested (16) that all four classes of enzymes originated from a common ancestor because they all have two similarly spaced histidine-containing tetrapeptides; their catalytic domains have a similar size of ϳ200 residues with similar pattern of predicted secondary structural elements; and they all catalyze hydrolysis of either 2Ј,3Ј-cyclic phosphates to 2Ј-phosphates or 1Љ,2Љ-cyclic phosphate to 1Љ-phosphate. Recently, new members of this superfamily have been identified (17).
Investigations of CNP have provided a variety of observations concerning the relationship of CNP to the cytoskeleton and its localization to discrete regions of oligodendrocytes and paranodal compartments of the myelin sheath, adjacent to the axon (18 -25). CNP comprises ϳ4% of the central nervous system total myelin protein and is most abundant in oligodendrocytes. Recently, it has been reported that CNP binds to tubulin and that it may play a role in anchoring microtubules to the plasma membrane as well as in regulating tubulin polymerization (26). A second isoform (CNP2) has also been identified, which contains a unique 20-amino acid N-terminal domain that targets the protein to mitochondria (27). Also, recent studies on CNP-null mutant mice revealed that the absence of CNP causes axonal swelling and neuronal degeneration (28). These observations underline the importance of this enzyme in brain and point to a multifaceted role for CNP in myelinogenesis and the maintenance of the myelin-axonal interface.
Here, we describe the structure of the brain CNP catalytic domain as determined by NMR and show that it is highly similar to the plant CPDase and the archaebacterial RNA ligase despite low overall similarity in amino acid sequence. This work brings us a step closer to understanding the function of CNP and its evolutionarily conserved enzymatic activity.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The catalytic fragment of CNP (CNP-CF, residues 164 -378) was subcloned into pET15b (Novagen, Madison, WI) and expressed in the Escherichia coli expression host BL21(DE3) (Stratagene) as a His-tagged fusion protein. The protein was purified by immobilized metal affinity chromatography on a Ni 2ϩloaded chelating Sepharose column (Amersham Biosciences). Isotopically labeled CNP-CF was prepared from cells grown on minimal M9 medium containing [ 15 N]ammonium chloride and/or [ 13 C]glucose (Cambridge Isotopes Laboratory, Andover, MA). For the backbone assignments, partially deuterated triple-labeled ( 2 H, 15 N, 13 C) CNP-CF was produced by expressing the protein in 90% D 2 O-and 10% H 2 O-containing minimal M9 medium. The N-terminal His tag was cleaved from CNP-CF by overnight dialysis with thrombin (Amersham Biosciences) at 1 unit/mg of fusion protein at room temperature. Benzamidine-Sepharose and Ni 2ϩ -loaded chelating Sepharose were used to remove thrombin and the His tag peptide from CNP-CF. The resulting 219amino acid protein contained four extraneous residues from the His tag. The sequence composition of purified CNP-CF was confirmed by mass spectrometry.
CNP catalytic fragment mutants were created by overlap extension PCR using the Expand High Fidelity PCR system (Roche Diagnostics) (29). The mismatched oligonucleotide sequences used to generate the mutants were as follows (only the sense oligonucleotides are The boldface underlined sequences correspond to the mutated codons. The authenticity of the substitutions and the absence of any undesired mutations were confirmed by sequence analysis. The CNP-CF histidine mutants (H230L and H309L) were generated as previously described (6).
NMR Spectroscopy-NMR resonance assignments of the catalytic fragment of CNP were determined previously (30). All NMR experiments were recorded at 310 K. NMR samples were 1-2 mM protein in 50 mM sodium phosphate buffer, 0.15 M NaCl, 1 mM dithiothreitol, and 0.1 mM sodium azide at pH 6.0. Attempts to use Pf1 phage or compressed polyacrylamide to obtain residual dipolar coupling constraints were not successful. Nuclear Overhauser effect correlation spectroscopy (NOESY) constraints for the structure determination were obtained from 15 N-edited NOESY (mixing time of 100 ms) and 13  (mixing time of 100 ms) three-dimensional experiments using the Varian Inova 800-MHz spectrometer at the Canadian National High Field NMR Centre (NANUC). NMR spectra were processed with GIFA (31) and XWINNMR Version 2.5 (Bruker Biospin) and analyzed with XEASY (32).
Structure Calculation-For the structure determination, a set of 1925 nuclear Overhauser effects (NOEs) were collected from 15 N-and 13 C-edited NOESY spectra of CNP-CF (amino acids 164 -378) acquired at 800 MHz. Automated NOE assignments were made using ARIA (33), and the structure was refined using standard protocols in CNS Version 1.1 (34). The starting structure for ARIA was generated using MOD-ELLER according to the CPDase fold and was in agreement with a set of manually assigned NOEs. PROCHECK NMR was used to check the protein stereochemical geometry (35). The coordinates have been deposited in the Protein Data Bank (code 1N4T), and the NMR assignments have been deposited in the BioMagResBank Database (accession number 5202).
CNP-CF Titrations with Inhibitors-2Ј-AMP, 3Ј-AMP, 5Ј-AMP, NAD, and pyrophosphate (Na 2 H 2 P 2 O 7 ) were purchased from Sigma and used without any additional purification. An RNA oligoadenylate hexanucleotide (A 6 ) was chemically synthesized and purified by University Core DNA & Protein Services (University of Calgary, Calgary, Canada). The purity and composition of the oligonucleotide was verified by onedimensional NMR spectroscopy.
Titrations were monitored by 15 N-1 H heteronuclear single quantum correlation spectroscopy (HSQC) following addition of inhibitors to 15 Nlabeled CNP-CF (amino acids 164 -378) on a Bruker Avance 600-MHz spectrometer. The sample contained 50 mM MES, 0.15 M NaCl, 1 mM dithiothreitol, and 0.1 mM sodium azide at pH 6.0 and ϳ0.4 -0.5 mM CNP-CF at 310 K. Inhibitor concentrations varied from 0.1-0.2 to 7-60 mM depending on the affinity and solubility of the inhibitor. The A 6 hexanucleotide concentration varied from 0.06 to 0.9 mM. HSQC spectra were assigned by monitoring chemical shift changes upon addition of the substrate because the binding takes place in fast exchange. The pH of the NMR samples was verified during the titrations and adjusted as needed. Chemical shift changes for individual residues were fitted to a one-site binding equation using the computer program GraFit (Version 3.0, Leatherbarrow) to determine the K d of binding.
CNP-CF Inhibition Assays-CNP-CF activity assays were performed using the spectrophotometric coupled enzyme assay described previously (36). CNP-CF activity was determined by monitoring the formation of NADPH at 340 nm (⑀ ϭ 6.22 mM Ϫ1 cm Ϫ1 ) using a Cary UV-visible spectrophotometer (Varian). The assay was initiated by adding 25 ng of CNP-CF to 1 ml of assay buffer containing 50 mM MES (pH 6.0), 30 mM MgCl 2 , 2Ј,3Ј-cyclic NADP (Sigma), 5 mM D-glucose 6-phosphate, and 5 g D-glucose-6-phosphate dehydrogenase (Roche Diagnostics). The concentration of cyclic NADP was varied from 0.05 to 2.0 mM. The initial velocity values were obtained from the Cary WinUV enzyme kinetics application and were fitted to the Michaelis-Menten equation. To test the inhibition of CNP-CF by AMP analogs, 1.5, 0.75, and 0.5 mM 5Ј-AMP, 2Ј-AMP, and 3Ј-AMP, respectively, were included in the 1 ml of assay buffer. CNP-CF activity data in the presence and absence of inhibitors were replicated at least twice for each inhibitor.

CNP Belongs to the Superfamily of 2H
Phosphodiesterases-We determined the structure of the catalytic fragment of rat brain CNP (CNP-CF), the first of a vertebrate-specific 2Ј,3Јcyclic nucleotide 3Ј-phosphodiesterase (Fig. 1). The previously reported resonance assignments (30) were used to assign NOEs from 15 N-and 13 C-edited three-dimensional NOESY experiments. The 20 lowest energy structures of 60 calculated were chosen to represent the final ensemble. The structural statistics are shown in Table I. On average, 10.9 constraints per residue were used to calculate the CNP-CF structure. This is below the typical number of 15-20 constraints per residue in high resolution NMR structures and results from the number of unresolved overlapping NOEs and a lower sensitivity of NOESY experiments because of the relatively large protein molecular mass (24.3 kDa). The tendency of CNP-CF to aggregate also limited the protein concentration in NMR samples.
The structure shows a bilobal arrangement of two modules, each consisting of a four-stranded antiparallel ␤-sheet and two antiparallel ␣-helices located on the outer part of the modules (Fig. 1b). The first lobe consists of strands ␤1, ␤2, ␤6, and ␤7 and helices ␣2 and ␣3, whereas the second one consists of strands ␤3, ␤4, ␤5, and ␤8 and helices ␣1 and ␣4. The internal face of the modules forms a large cavity. Intense peaks in the 15 N-1 H HSQC spectrum of CNP-CF indicate a flexible backbone for residues from Gly 208 to Lys 214 . Coupled with low NOE density for this region, this loop between helix ␣1 and strand ␤2 appears to be mobile.
The structure has a striking similarity to CPDase from A. thaliana (12) and to 2Ј-5Ј RNA ligase from T. thermophilus (15). Topologically, CNP-CF differs only in an extra C-terminal strand ␤8 that extends the antiparallel ␤-sheet containing strand ␤3 (Fig. 1c). One possible role of this ␤-strand is to place the N and C termini on the opposite sides of the CNP domain and to position the CNP C-terminal isoprenylation site at the membrane. The structural similarity of CNP-CF provides direct evidence that CNP belongs to the superfamily of phosphodiesterases containing dual catalytic tetrapeptide HX(T/ S)X motifs.
CNP-CF has a more open cavity than CPDase. This difference suggests that a larger natural ligand could exist for CNP. Sequence comparison (Fig. 2) provides possible explanations for the more closed CPDase structure. The turns between strands ␤3-␤4 and ␤6 -␤7 in CPDase make hydrophobic contacts with each other via side chains of Phe 84 and Leu 168 and limit the size of the catalytic cavity. The ␤3-␤4 turn in CPDase is very hydrophobic, with a triplet of phenylalanine residues, Phe 82 , Phe 83 , and Phe 84 , whereas both turns in CNP-CF are hydrophilic and positively charged, suggesting that they are solvent-exposed and may be involved in interactions with a negatively charged ligand. Interestingly, the 2Ј-5Ј RNA ligase has hydrophilic turns and an open conformation similar to that of CNP. This conformation likely allows RNA to access the catalytic site (15).
Binding of CNP Inhibitors-To obtain more information about the active site of CNP, we titrated 15 N-labeled CNP-CF

Solution Structure of CNP
with several compounds previously shown to inhibit CNP activity (reviewed in Ref. 1). These included orthophosphate, pyrophosphate, 2Ј-AMP, 3Ј-AMP, 5Ј-AMP, ␤-NAD, and NADP. The titrations were monitored by 1 H-15 N correlation spectroscopy, and shifts of amide signals as a function of ligand addition were recorded. These shifts act as a fingerprint and identify amino acid residues affected by binding (Fig. 3a). Titration of the catalytic fragment of CNP with orthophosphate resulted in chemical shift changes, indicating that it binds to CNP and was present in the structure determined by NMR. The biggest 1 H and 15 N amide chemical shift changes were observed for Thr 232 (0.56), Thr 311 (0.53), Gly 324 (0.24), Val 228 (0.21), Ala 308 (0.17), His 230 (0.16), Gly 305 (0.15), and Thr 323 (0.15). Thr 232 , Thr 311 , and His 230 are part of the tetrapeptide motifs, which are important for the catalytic activity. This shows that the phosphate group binds in the active site. The catalytic threonines, Thr 232 and Thr 323 , likely coordinate the phosphate moiety through hydrogen bonds. The chemical shift changes correlate with the regions of highest sequence conservation in the catalytic domains of CNP from different species (Fig. 4b).
Interestingly, CNP-inhibitor interactions were pH-dependent. The chemical shift changes upon phosphate binding were much smaller at pH 6.5 (and above) than at pH 6.0 (data not shown). The likely reason for this is deprotonation of the catalytic histidines or the phosphoryl group. This would change the electrostatic charges and interfere with hydrogen bonding to the phosphate ion. In support of this, the CNP enzymatic activity was optimal at pH 5.5-6.5 and decreased at higher pH (data not shown). Whether this reflects protonation of activesite histidine(s) or the substrate phosphoryl group remains to be determined.
The binding of AMPs resulted in a pattern of chemical shift changes very similar to that observed upon binding of phosphate (Fig. 3a). The relatively larger shifts observed for AMPs reflect the stronger binding to CNP-CF. The AMP titrations also allowed us to identify additional residues affected by binding. Located in the loop between strand ␤2 and helix ␣2, Cys 236 and Asp 237 showed minor chemical shift changes. These residues are relatively close to the tetrapeptide motifs and could participate in substrate recognition by interacting with the mobile loop Gly 208 -Lys 214 . Speculatively, Asp 237 could interact with either Lys 212 or Lys 214 to close the mobile loop upon substrate binding. However, this loop is poorly conserved among mammalian CNPs (Fig. 4b) and shows relatively small changes upon inhibitor binding (Fig. 3a), suggesting that either the loop does not function as a flap or that it needs the proper substrate for specific loop-ligand interactions.
The titration experiments also allowed us to compare binding affinities of CNP inhibitors. In the weak binding (fast exchange) regime, the signals in HSQC spectra gradually move from the unbound position to the fully bound position depending on the amount of inhibitor added. These changes can be fitted using the binding equation to estimate the dissociation constant (K d ). The dissociation constants obtained from NMR titrations are shown in Table II. The results show that 3Ј-AMP has the highest affinity for the CNP catalytic domain, followed by 2Ј-AMP and 5Ј-AMP. This might reflect differences in the pK a of the phosphoryl group in the different AMPs or may be an inherent property of the CNP catalytic site. Studies at a second pH are needed to resolve the issue. Comparison of NAD and NADP shows that a terminal phosphate significantly improves binding affinity.
Kinetic Properties of Active-site Mutants-To define the enzymatic role of the conserved tetrapeptide motif residues, we mutated each residue individually and measured the kinetic A. thaliana (A. th.) and RNA ligase from T. thermophilus (T. th.). The secondary structural elements refer to rat CNP-CF. The conserved catalytic residues are indicated in boldface. parameters of the mutant enzymes using cyclic NADP as substrate (Table III). Consistent with our previous results (6), mutation of His 230 and His 309 resulted in a 15,000-fold reduction in k cat without any effect on K m , suggesting that both conserved histidines are critical for catalysis, but not for substrate binding. Mutation of Thr 232 or Thr 311 significantly decreased k cat by 100-and 700-fold, respectively. The Thr 232 mutant also exhibited an 8-fold increase in K m , whereas the Thr 311 mutant showed a minimal 3-fold increase. These results suggest that Thr 232 is more critical for substrate binding, whereas Thr 311 is more important for catalysis. Parallel studies with yeast CPDase showed that mutation of Ser/Thr residues in the tetrapeptide motifs has minor effects on enzymatic activity and suggested that the residues play a larger role in substrate recognition (16). Interestingly, mutation of the non-active site residue Gly 324 (helix ␣4) in CNP resulted in a 5-fold increase in K m and 37-fold decrease in k cat . Mutations of other proximal residues on helix ␣4 such as Gln 322 and Leu 327 did not affect the kinetic parameters. These results agree with the NMR titration data (see below) that suggest that Gly 324 is important for the conformation of helix ␣4. The other remaining mutants (D237V, G276A, A308G, and Y376A) displayed wild-type kinetic parameters, indicating that these residues are not particularly important for CNP activity.

FIG. 2. The catalytic domains of rat and human CNPs and the phosphodiesterase domain of the RICH70 protein from goldfish (g) show low sequence homology to CPDase from
Inhibitor Binding Causes a Conformational Change in the CNP Catalytic Domain-NMR titrations showed ligand-induced chemical shift changes in the residues around Gly 324 (Fig. 3a). These residues are located in helix ␣4 and relatively far from the active site. This could represent a second binding site or a minor conformational change that affects these residues through contacts with strand ␤5 in the active site.
To distinguish between these possibilities, we compared the dissociation constants obtained by NMR for two residues, Thr 232 and Gly 324 , using both weak (orthophosphate) and high affinity (2Ј-AMP) inhibitors (Table II). Identical K d values were observed for both residues, indicating a concerted conformational change that arises from a single binding site. It is very unlikely that two distinct binding sites would demonstrate identical K d values for both substrates. In helix ␣4, the most affected residues are Thr 323 and Gly 324 . These amino acid types are less common in ␣-helices, and we speculate that the conformational change observed is elongation of the N-terminal part of helix ␣4 in the presence of bound substrate.
The CNP Catalytic Fragment Weakly Binds Hexanucleotide RNA-An interesting feature of the CNP catalytic fragment is the presence of several aromatic (Tyr 168 , Phe 172 , Phe 235 , and Tyr 352 ) and positively charged (Lys 214 , Lys 234 , and Arg 307 ) residues in the vicinity of the active site (Fig. 5). The majority of these residues are located on the N-terminal lobe (strands ␤1, ␤2, and ␤7). ␤-Sheets with an abundance of positively charged and aromatic residues are common RNA-binding surfaces (for a review, see Ref. 37). The presence of these residues, coupled with a large binding cavity, hints that RNA may be a CNP substrate. Furthermore, many proteins from the 2H phosphodiesterase superfamily are involved in RNA-processing pathways, and some of them bind RNA (17). To test this hypothesis, we titrated CNP-CF with an RNA hexanucleotide (A 6 ). The   fingerprint of A 6 binding was very similar to that of phosphate and other inhibitors (Fig. 3a), but its affinity could not be determined from the NMR data because the data did not fit a simple one-site binding (data not shown). The chemical shift changes with A 6 RNA were smaller than with 3Ј-AMP or 5Ј-AMP, suggesting weaker binding. Enzyme assays were used to determine the inhibition constants (K i ) for the AMPs and the A 6 oligonucleotide (Table II). All the inhibitors showed competitive inhibition (increased K m and unchanged V max ). Because the N-terminal fragment of CNP has some homology to ATP-binding domains (38), we also tested ATP and observed very weak inhibition, with a K i of ϳ13 mM. It was previously reported that poly(A) RNA is a potent CNP inhibitor (39). Surprisingly, our data show that A 6 oligonucleotide has a rather weak ability to inhibit the CNP activity, with a K i of ϳ0.8 mM. One possible explanation for the previous result is the presence of RNase activity in the CNP-CF preparation, which would produce mononucleotides from added RNA and so inhibit CNP activity. Based on our results, we conclude that single-stranded RNA does not produce any specific interactions with CNP-CF above those identified for mononucleotides. The possibility that CNP binds double-stranded RNA is currently under study. DISCUSSION Compared with the C terminus, little is known about the function of the N-terminal half of CNP. Sequence analysis revealed similarity to numerous proteins containing P-loop ATPase domains. This led to the suggestion that CNP may be a polynucleotide kinase (38), although experimental evidence for this has been unsuccessfully sought. 2 Comparison of CNP with RICH shows divergence at the N terminus, suggesting the existence of shared (C-terminal) and divergent (N-terminal) functions. Structure determination of the N-terminal domain and/or full-length CNP may help to define the global role of CNP in oligodendrocytes, where it is most abundant.
Given that other members of this protein superfamily are involved in RNA-processing pathways, could RNA be a physiological substrate for CNP? The structural features of the catalytic domain such as the RNA recognition motif-like fold of each lobe, the abundance of positively charged and aromatic residues on the ␤-sheet surface, and the large curvature of the binding cavity are consistent with this hypothesis. Although our data demonstrate that single-stranded oligo(A) possesses weak, millimolar affinity for the CNP catalytic domain, we cannot rule out 1) that CNP might bind only to double-stranded nucleic acids or to a specific RNA sequence, 2) that a 3Ј-or 5Ј-terminal phosphate might be important for binding, or 3) that the N-terminal portion of CNP may regulate binding or affect binding activity.
In considering the possible cellular functions of CNP, it is important to take into account other information such as the cellular localization of this enzyme. Both CNP isoforms are membrane-associated via isoprenylation at their C termini (21). The larger isoform of CNP, CNP2, contains a mitochondrial targeting sequence at the N terminus (27). CNP is also known to interact with tubulin (26), leading to the speculation that CNP might have a role in mRNA transport as observed in oligodendrocytes (40 -42). Proteins such as myelin basic protein and carbonic anhydrase (43,44) and tau (45) are specifically synthesized at the periphery of the myelin-forming processes; some type of specific mRNA transport and localization machineries must exist. In addition, there are signaling molecules such as nicotinic acid-adenine dinucleotide phosphate and cADP-ribose (reviewed in Ref. 46) upon which CNP might potentially act. Future structural and enzyme studies of CNP will hopefully clarify the cellular function of this highly conserved yet enigmatic protein.