Effect of N-terminal and Met23 Mutations on the Structure and Dynamics of Onconase*

Onconase (rONC), otherwise known as ranpirnase or P-30 protein, which was initially purified from extracts of Rana pipiens oocytes and early embryos, exhibits anticancer activity both in vitro and in vivo and is in phase III clinical trials for tumor therapy. We have determined the solution NMR structure of a recombinant onconase with Met-1, Gln1, and Leu23 residues (M-1, Q1, M23L)rONC. The 20 best solution structures had a backbone root mean square deviation of 0.41 ± 0.09 Å with respect to the average structure. The energy-minimized average NMR structure had a backbone root mean square deviation of 0.72 Å from the x-ray crystallographic structure of native onconase; however, the orientation of the N-terminal residue in the two structures was very different. Comparison of the 15N HSQC spectrum of (M-1, Q1, M23L)rONC with that of a mutant E1S-rONC, which is identical to the nONC except with the N-terminal pyroglutamyl residue replaced by Ser, showed that N-terminal and residue 23 mutations induced structural changes in regions beyond the mutation sites. Model-free analysis of the backbone amide 15N-T1, 15N-T2, and 15N-1H NOE relaxation data for (M-1, Q1, M23L)rONC and E1S-rONC revealed that the E1S-rONC molecule showed very little flexibility, whereas (M-1, Q1, M23L)rONC exhibited substantial flexibility, which may account for the previously observed reduced stability and increased protease susceptibility. The α1 helix and β-sheets of (M-1, Q1, M23L)rONC displayed bending motions. These data provided strong evidence for the presence of an N-terminal hydrogen bond network in E1S-rONC, but not in (M-1, Q1, M23L)rONC.

Onconase (also known as ranpirnase or P-30), which was initially purified from extracts of Rana pipiens (Northern leopard frog) oocytes and early embryos, exhibits anticancer activity both in vitro and in vivo (1). Sequence analysis and structural comparison have shown that the native natural protein, nONC, 1 is highly homologous to pancreatic RNase A and two lectins found in the eggs of Rana catesbeiana and Rana japonica (2)(3)(4)(5)(6)(7)(8). nONC displays cytotoxic and cytostatic activity against numerous mammalian cell lines in vitro. Although it has a specific activity approximately one-hundredth of that of RNase A, its toxicity in animals is about 5000 times greater than that of RNase A, and it is resistant to RNase inhibitors, placental ribonuclease inhibitor, and Inhibit-Ace TM (6, 9 -11). nONC also displays antitumor activity in vivo (12,13) and specifically inhibits human immunodeficiency virus type 1 replication in infected H9 leukemia cells at noncytotoxic concentrations (14,15). It is currently undergoing phase III clinical trials for the treatment of cancer.
nONC is a 104-amino acid protein with a molecular mass of 11.8 kDa and an isoelectric point of 9.7 (6). Its x-ray crystal structure has been determined (16) and found to be very similar to that of RNase A, RC-RNase (4), and RNase 4 (17). The active site is located at the junction of the N-terminal ␣-helix and two ␤-sheets. A unique structural feature of the molecule is the orientation of the N-terminal pyroglutamyl residue (Pyr 1 ), which folds back against the N-terminal ␣-helix and forms hydrogen bonds with the side chain nitrogen of Lys 9 and the carbonyl group of Val 96 in the C-terminal ␤-sheet and is proposed to play an important role in catalysis. A similar arrangement of Pyr 1 is seen in RC-RNase (18). A mutant of the recombinant enzyme (rONC), (MϪ1, Q1, M23L)rONC, in which a methionine was added to the N terminus, Pyr 1 was replaced with Gln, and Met 23 was replaced with Leu, shows considerably reduced ribonuclease activity and cytotoxicity; however, cleavage of the N-terminal Met, followed by cyclization of the new N-terminal Gln residue to the native pyroglutamyl, results in the M23L-rONC mutant, which has an IC 50 33.3% of that of native ONC and a catalytic activity 10-fold higher than that of (MϪ1, Q1, M23L)rONC and about 2-fold higher than that of nONC. Using a different approach, Notomista et al. (19) used Aeromonus proteolytica aminopeptidase to cleave the N-terminal Met from (MϪ1, Q1, M32L)rONC, and their refolded M23L-rONC mutant was 5-fold more active than nONC and retained all of the antitumor activity. Furthermore, M23L-rONC was shown to be significantly more susceptible to thermal and GuHCl-induced denaturation and less resistant to proteolysis (20). Newton et al. (21) further showed that modification of the 5Ј-region of the nONC gene to encode Ser 1 (MϪ1, E1S)rONC or Tyr 1 (MϪ1, E1Y)rONC results in an active enzyme, and the activity of these two mutants was proposed to be due to the presence of the hydroxyl oxygen atoms in the Ser and Tyr residues, which assume the role of the carbonyl oxygen of the pyroglutamate residue. More recently Liao and co-workers (22) have shown that an endogenous methionine aminopeptidase in Escherichia coli is capable of cleaving the Met Ϫ1 residue preceding a small amino acid such as Ala and Ser. Thus, the restored activity in (MϪ1, E1S)rONC as observed by Newton et al. (21) may be due to the activity of E1S-rONC with the Met Ϫ1 removed by methionine aminopeptidase, and the presence of a hydrogen bond network involving Ser 1 and the active site residues was also proposed but not proven. It is therefore important to explore the structural basis of mutation-induced functional changes. At present, no solution structure of onconase and no information on its dynamics have been reported. The structures of the onconase mutants are also not available. We have used NMR techniques to study the structure-function relationship of the cytotoxic RNases (4,23,24), and the 1 H, 13 C, and 15 N chemical shifts and J HNH␣ coupling constants of (MϪ1, Q1, M23L)rONC have been deposited in BioMagResBank (available on the World Wide Web at www.bmrb.wisc.edu) under BMRB accession numbers 4371 and 5835. In this paper, we describe the solution structure and dynamics of the onconase mutant, (MϪ1, Q1, M23L)rONC, and the dynamics of the E1S-rONC mutant. The structure coordinates of the solution structure of (MϪ1, Q1, M23L)rONC have been deposited in the Protein Data Bank (accession numbers PDB ID 1PU3 and RCSB ID RCSB019565).

EXPERIMENTAL PROCEDURES
Protein Sample Preparation and Characterization-A synthetic onconase gene coding for (MϪ1, Q1, M23L)rONC, kindly provided by Dr. R. J. Youle (NINDS, National Institutes of Health), was cloned into the pET-11d plasmid and expressed in the E. coil BL21 (DE3) strain using isopropyl-1-thio-␤-D-galactopyranoside as inducing agent (25). The expressed protein was isolated as inclusion bodies. After vigorous washing, the protein was dissolved in 6 M guanidine HCl containing 100 mM reduced glutathione and incubated at room temperature under nitrogen for 2 h. The protein was then renatured by rapid dilution into 20 mM Tris acetate buffer, pH 7.5, containing 0.5 M L-arginine and 8 mM oxidized glutathione, followed by incubation at 4°C for 24 h prior to purification, as described previously (25). The purity of the protein was confirmed by electrophoresis on 10 and 27% SDS-polyacrylamide gels and was found to be greater than 95%. The plasmid encoding the (MϪ1, E1S)rONC gene was a generous gift from Dr. S. M. Rybak (NCI-Frederick Cancer Research and Development Center); the protein was expressed and purified as described previously (21). Two fractions of protein were found in our final purification step, and the enzymatic activities and cell cytotoxicities of these fractions were checked as reported previously (21). Fraction 2 was found to retain the enzymatic activity and cell cytotoxicity as reported previously, whereas fraction one had significantly reduced activities. The molecular mass of fraction 2 was checked with electrospray ionization mass spectrometry (ESI-MS), and the first 10-amino acid sequence was determined by liquid chromatography-MS-MS. The molecule was found to correspond to E1S-rONC mutant with no N-terminal methionine instead of the expected (MϪ1, E1S)rONC. This fraction (named E1S-rONC hereafter) was used for further NMR structural characterization. We suspect that Met Ϫ1 in fraction 2 was removed by methionine aminopeptidase as reported by Liao et al. (22). Uniformly 13 C-and/or 15 N-labeled protein samples were purified from E. coli grown in M9 medium, supplemented with 2 g/liter [u-13 C]glucose and/or 1 g/liter 15 NH 4 Cl (Cambridge Isotope Laboratories, Andover, MA).
Briefly, 2500 cells in 0.1 ml of Dulbecco's modified Eagle's medium plus 10% fetal bovine serum were plated in each well of a 96-well plate, and then, after 24 h, the test samples (10 l) were added, and the cells were incubated for 3 days at 37°C in a humidified CO 2 incubator. To measure protein synthesis, the serum-containing medium was replaced with 100 l of serum-and leucine-free RPMI, and then 0.1 mCi of [ 14 C]leucine (10 l) was added, and incubation continued for 2-4 h at 37°C. The cells were then harvested on glass fiber filters using a PHD cell harvester; the filters washed with H 2 O and dried with ethanol; and the bound radioactivity was measured in a scintillation counter. Each point was carried out in triplicate. As expected from a previous study (21), nONC showed the highest cytotoxicity (IC 50 4 g/ml), followed by E1S-rONC (IC 50 [45][46][47][48][49][50] g/ml) and then (MϪ1, Q1, M23L)rONC (IC 50 Ͼ 100 g/ml).
Mass Spectrometry-ESI-MS experiments were carried out on a Finnigan LCQ TM Deca ion trap mass spectrometer (ThermoFinnigan, San Jose, CA) with an electrospray ionization interface. The ESI source was operated in positive ion mode. The sample solution was applied using a high pressure liquid chromatography pump with loop injection. The ESI was operated at a spray voltage of 4.5 kV, a heated capillary temperature of 160°C, and a flow rate of 5 l/min. The mass spectrometer was operated in full-scan profile mode with a scan range from m/z 150 to 2000. Xcalibur (version 1.3; ThermoFinnigan) and BioWorks TM 3.0 software packages were used for data acquisition and analysis and in the reconstruction of the molecular weight from the spectra. A molecular weight of 12,095 was obtained for 15 N-labeled (MϪ1, Q1, M23L)rONC, compared with the expected value of 12,102. For 15 Nlabeled E1S-rONC, the measured molecular weight was 11,935, compared with a calculated value of 11,946.
NMR Experiments-Samples for NMR experiments (about 0.35 ml of 2 mM protein in 50 mM potassium phosphate, 10% D 2 O at pH 7.0) were sealed in Shigemi tubes. The pH was measured using a JENCO microelectronic pH-vision model 6071 pH meter equipped with a 4-mm electrode. All reported pH values are direct readings from the pH meter without correction for the isotope effect. To monitor the exchange rates of labile protons, a concentrated sample in H 2 O was lyophilized and redissolved in D 2 O (99.99% D), and NMR spectra were acquired immediately and thereafter at appropriate time intervals. All NMR experiments were performed on a Bruker AVANCE600 NMR spectrometer with the probe temperature set at 310 K. Spectra were processed using XWINNMR (Bruker AG, Karlsruhe, Germany) or NMRPipe (26) and analyzed using AURELIA (27) or SPARKY (T. D. Goddard and D. G. Kneller, SPARKY 3, University of California, San Francisco, CA). Linear predictions (28) were used in the indirect dimensions to improve the digital resolution. Chemical shifts were referenced to 2,2-dimethyl-2silapentane-5-sulfonate at 0 ppm (29). The 15 N and 13 C chemical shifts were indirectly referenced using consensus ⌶ ratios of the zero-point frequencies at 310 K of 0.101329118 for 15 N/ 1 H and 0.251449530 for 13 C/ 1 H (30). Experimental Restraints and Structure Calculation and Analysis-An ensemble of NMR structures of (MϪ1, Q1, M23L)rONC was calculated based on NOE data, backbone dihedral angles, and hydrogen bond constraints. Hydrogen bond constraints were derived from the long range NOE data and H-D exchange experiments. At the initial stage of the structure calculation and NOE assignment, only protons that show slow exchange (longer than 8 h), display the characteristic NOE pattern, and show chemical shift indices (CSI) that indicate the presence of secondary structure (31) were assumed to be involved in hydrogen bonds. In some cases, hydrogen bond constraints were set to have multiple partners. In total, 49 hydrogen bonds were included in the final calculation, giving 98 hydrogen bond constraints.
Backbone dihedral angles were obtained using the 3 J HNH␣ coupling constants derived from HNHA experiments (32) and the empirical Karplus equation (33). Backbone dihedral angles predicted using TALOS (34) were included in the structure calculation if they were found to be consistent with the secondary structure obtained from the CSI and NOE patterns. These gave rise to a total of 94 and 56 dihedral angle constraints.
NOE distance restraints were derived from three-dimensional 15 N NOESY-HSQC and 13 C NOESY-HSQC and two-dimensional 13 C-filtered NOESY spectra, all collected with a mixing time of 100 ms. The NOE spectra were peak-picked and quantified using the automatic picking routines of AURELIA and Sparky. Initially, the three-dimensional 15 N-separated NOESY spectrum was partially assigned manually, and then the manually assigned and unassigned NOEs from the 15 N-separated spectrum and unassigned 13 C NOESY spectra were used as input for the ARIA (35) software package for further NOE assignment. The NOE cross-peaks assigned by ARIA were interactively reexamined manually using Sparky.
For the structure calculation, we used CNS (36) interfaced with ARIA 1.2. We used the parallhdg5.3 force field with ARIA default SA parameters during iterations it0 -it8. Stereospecific assignment of prochiral groups was achieved using a floating assignment approach (37) as implemented in the ARIA package. The 20 lowest energy structures, chosen from 100 calculated structures, were subjected to further refinement after hydration with a shell of water molecules. This ensemble of structures was analyzed using the programs InsightII (Accelrys, San Diego), GRASP (38), MOLMOL (39), and PROCHECK-NMR (40).
Relaxation Data Analysis-The longitudinal and transverse relaxation rate constants, R 1 (equal to 1/T 1 ) and R 2 (equal to 1/T 2 ), were obtained from a Marquardt-based nonlinear least-squares fit (using SigmaPlot (Jandel Scientific) and subroutines in A. G. Palmer's Modelfree package (New York)) of the measured cross-peak intensities to a single exponential function. The reported R i values are the mean values of two independent data sets. The heteronuclear NOE value (XNOE) was determined from the peak intensity ratio in the spectra recorded, respectively, with or without proton saturation. The reported XNOE value is the average value of three pairs of XNOE experiments. The diffusion tensor analysis followed previously described procedures (41).
Model-free Formalism-The heteronuclear 15 N relaxation rate constants, R 1 and R 2 , and the 1 H-15 N steady state XNOE values were further analyzed by model-free formalism, pioneered by Lipari and Szabo (42,43) and further extended by Clore et al. (44). In this approach, the overall and internal molecular motions are assumed to be independent, and the spectral density function for a molecule undergoing isotropic tumbling is given by the equation, where m is the overall rotational correlation time of the molecule, is the effective correlation time for internal motions on the fast (or slow) time scale ( f Ͻ s Ͻ m ), and S 2 (equal to S f 2 S s 2 ) is the generalized order parameter measuring the degree of spatial restriction of the bond vector. S f 2 (or S s 2 ) is the order parameter for internal motions (which ranges from 0 for completely isotropic motion to 1 for completely restricted motion) on the fast (or slow) time scale. The six dynamic parameters in Equation 1 were extracted using five simple models proposed by Palmer and co-workers (45). These five models contain the following parameters: 1) S 2 (assuming S 2 (assuming S f , for model 1, 2, 3, 4, and 5, respectively. 15 N-1 H vectors with relaxation data that can be fitted to model 1 usually have large S 2 values and are more rigid. 15 N-1 H vectors with relaxation data that must be fitted to models 3 and 4 are those displaying conformational exchange processes on the s/ms time scale. The most flexible 15 N-1 H sites, such as those in the termini and those in the unfolded state, are often be fitted to model 5. 15 N-1 H vectors with relaxation data fitted to model 2, 4, or 5 display internal motions on the ps time scale. Model-free parameters were obtained using the Model-free program, version 4.1. Model fitting and selection followed the procedures detailed in Mandel et al. (45), as described in our previous paper (41).

RESULTS
Secondary Structure Determination-The 15 N HSQC spectrum of (MϪ1, Q1, M23L)rONC has an excellent chemical shift dispersion, all resonances are well resolved, and the complete assignments of backbone and side chain resonances have been reported (24) and deposited in BMRB 4371 and 5835. We used standard two-and three-dimensional heteronuclear NMR techniques to deduce the secondary and tertiary structures. Fig. 1 summarizes the NMR parameters used to determine the secondary structure of (MϪ1, Q1, M23L)rONC; these were the H N -H ␣ coupling constant ( 3 J HNH␣ ), the medium range backbone NOEs (d NN , and d␣ N (i, i ϩ 4)), and the consensus CSI derived from the CSI of the H␣, C␣, C␤, and CЈ resonances (31). On the basis of these data, (MϪ1, Q1, M23L)rONC was found to consist of three helices (␣ 1 , Trp 3 -His 10 ; ␣ 2 , Cys 19 -Ile 22 ; and ␣ 3 , Pro 41 -Cys 48 ) and seven ␤-strands (␤ 1 , Ile 11 -Thr 12 ; ␤ 2 , Lys 33 -Tyr 38 ;  , these seven ␤-strands were found to form two ␤-sheets, with the first sheet, S1, formed by strands ␤ 1 , ␤ 2 , ␤ 4 , and ␤ 5 , and the second sheet, S2, composed of strands ␤ 3 , ␤ 6 , and ␤ 7 . Tertiary Structure of (MϪ1, Q1, M23L)rONC-A total of 1550 NOE restraints (789 long range, 212 medium, and 549 sequential range NOEs), four disulfide bonds, 98 hydrogen bond restraints, and 150 dihedral angles were used for structure calculation. Fig. 2a shows an overlay of the 20 best structures, and Fig. 2b shows a ribbon representation of the energyminimized average structure. The coordinates of these structures have been deposited in the Protein Data Bank (PDB ID 1PU3 and RCSB ID RCSB019565). Table I summarizes the structural statistics. The 20 best structures showed no NOE violation greater than 0.3 Å. The r.m.s. deviation of the backbone atoms of the ensemble of NMR structures was 0.30 Ϯ 0.06 Å for the secondary structure region, 0.41 Ϯ 0.09 Å for the backbone atoms of all residues, and 0.80 Ϯ 0.06 Å for all heavy atoms. PROCHECK analysis (data not shown) showed that, in the Ramachandran plot, 82.5% of the dihedral angles fell within the most favorable region, 17.1% in the additionally allowed region, 0.4% in the generally allowed region, and none in the disallowed region. The G-factors were Ϫ0.248 for dihedral angles and 0.613 for covalent bonds. The r.m.s. deviation of the backbone atoms between this NMR structure and the x-ray crystal structure of nONC, excluding residues Met Ϫ1 and Gln 1 (Pyr 1 ), was 0.72 Å. Thus, within experimental error, the overall solution NMR structure of (MϪ1, Q1, M23L)rONC was indistinguishable from that of the x-ray crystal structure of nONC. One major difference between the two structures was the orientation of the N terminus. In native onconase, the N-terminal cyclized pyroglutamyl residue folds back to the binding pocket in such a way that it is hydrogen-bonded to the -amino group of the Lys 9 side chain via O ⑀ and to the CO group of Val 96 via the NH group. In contrast, the N terminus of the averaged NMR structure for (MϪ1, Q1, M23L)rONC pointed away from the active site (Fig. 2a, red trace), and the distance between Gln 1 -O⑀ 1 and Lys 9 -N was 5.37 Å, and that between Gln 1 -N⑀ and Val 96 -CO 5.06 Å, both too far to form hydrogen bonds. Our H-D exchange experiments also showed that the ⑀2-amino protons of Gln 1 were in fast exchange and, thus, unlikely to form hydrogen bonds. In the x-ray crystal structure of recombinant onconase with Met Ϫ1 and Glu 1 residues, the N terminus is also in an unfavorable orientation for Glu 1 to interact with Lys 9 (25).
Chemical Shift Perturbation in Amide Resonances Due to MϪ1 and M23L Mutation- Fig. 3a shows the overlay of the 15 N-HSQC spectra of u-15 N-labeled (MϪ1, Q1, M23L)rONC (red) and u-15 N-labeled E1S-rONC (blue). The sequence variation of the measured chemical shift differences between the two recombinant proteins is shown in Fig. 3b. The chemical shift differences were calculated using the formula, ⌬␦ ϭ ((0.17⌬␦ N ) 2 ϩ ⌬␦ H 2 ). Although there appeared to be a surprisingly large number of resonances displaying different chemical shifts between the two mutants, the largest shifts (⌬␦ Ͼ 0.1 ppm) formed clusters. Fig. 4 shows that residues with ⌬␦ Ͼ 0.10 ppm were clustered primarily near the N terminus and residue 23. The large shifts in Trp 3 , Thr 5 , Phe 6 , and His 97 were probably due to the E1S mutation. Most of other residues showing large changes in chemical shift cluster around residue 23, and these were probably due to the perturbation introduced by the M23L mutation. However, a few residues further from residue 23 also showed substantial shifts. Of particular interest were the large shifts observed for the catalytically crucial residues (i.e. Lys 31 and Thr 35 ). Thus, N-terminal and M23L mutations caused structural changes not only to residues near the mutation site but also to active site residues. The observation of resonance shifts for many residues on the C-terminal ␤ 7 strand suggests that the relative positioning of the two ␤-sheets that form the binding pocket may differ in the two mutants.
Relaxation Parameters of Backbone Amide Nitrogens of (MϪ1, Q1, M23L)rONC and E1S-rONC- Fig. 5 shows the sequence variation of the values for 15 N-R 1 (a), 15 N-R 2 (b), and 1 H-15 N NOE (c) for u-15 N-labeled (MϪ1, Q1, M23L)rONC (empty circles) and E1S-rONC (filled circles), obtained at 600 MHz. A total of 91 R 1 , 90 R 2 , and 94 1 H-15 N NOE data points were obtained for (MϪ1, Q1, M23L)rONC, and 90, 90, and 93 data points, respectively, were obtained for E1S-rONC. The backbone nitrogen relaxation rates of all four proline residues (amino acids 41, 43, 74, and 95) could not be measured due to the absence of a proton. The amide resonances of Q1/S1 and Phe 98 were too weak to be measured. In addition, the resonances for residues 1   Model-free Analysis-To gain a deeper insight into the dynamic behaviors, we applied model-free formalism to extract molecular dynamic information from the NMR relaxation data (42)(43)(44)(45). Model-free formalism uses the magnitude of order parameters to describe the degree of angular restriction of the motion of the NH bond vector. The motional time scales are expressed as effective correlation times (models 2, 4, and 5) or as conformational exchange rates (models 3 and 4). Thus, a mobile region displaying large amplitude angular fluctuations on the ps/ns time scale has a lower order parameter, whereas a rigid region should have a higher order parameter. In general, residues that can be fitted to model 1 are rigid residues with no flexibility, whereas residues that must be fitted to other models display a certain flexibility. Residues that must be fitted to models 2, 4, and 5 display fast (ps/ns time scale) motion, whereas residues that must be fitted to models 3 or 4 have a nonvanishing R ex term and display slow exchange motion.
The average overall rotational correlation time, m , estimated from the 10% trimmed mean value of T 1 /T 2 for the backbone amide 15 N spins, gave values of 5.80 and 6.51 ns for (MϪ1, Q1, M23L)rONC and E1S-rONC, respectively. These values are consistent with the size of these two proteins, suggesting that both onconase molecules exist as monomers in solution. The diffusion tensor obtained was axially symmetric, with D // D Ќ ϭ 1.22. The numbers of residues fitted to the different models for (MϪ1, Q1, M23L)rONC and E1S-rONC (in parenthesis) were 61 (77), 5 (1), 4 (5), 2 (0), and 3 (0) for models 1, 2, 3, 4, and 5, respectively. Nine (MϪ1, Q1, M23L)rONC relaxation data (amino acids 11-13, 29, 59, 77, 80, 87, and 101) and six E1S-rONC resonances (amino acids 27, 28, 42, 66, 67, and 73) could not be fitted satisfactorily to any model. The relaxation parameters and the residue-specific model-free dynamic parameters, including the generalized order parameters (S 2 ), the effective correlation times for internal motions ( e ), and the conformational exchange terms (R ex ), are given in the supplementary materials. Fig. 6a shows the sequence variation of the generalized order parameters for (MϪ1, Q1, M23L)rONC (empty symbols) and E1S-rONC (filled symbols). Of those that could be fitted satisfactorily, none had an order parameter smaller than 0.75. No long stretch of segments showed unusually small order parameters. The mean value for the generalized order parameters was 0.891 Ϯ 0.051 and 0.937 Ϯ 0.047 for (MϪ1, Q1, M23L)rONC and E1S-rONC, respectively. These values indi-cate that the bulk of both mutants exhibited limited mobility on the ps/ns time scale. Fig. 6b shows the exchange rates extracted from models 3 and 4. In total, 6 and 5 residues in (MϪ1, Q1, M23L)rONC and E1S-rONC, respectively, exhibited a slow conformational exchange motion on the s/ms time scale. The exchange rates in E1S-rONC were generally larger than those in (MϪ1, Q1, M23L)rONC.

DISCUSSION
Comparison of the NMR Solution Structure of (MϪ1, Q1, M23L)rONC and the X-ray Crystal Structure of nONC As described above, the overall folding of (MϪ1, Q1, M23L)rONC was indistinguishable from that of the x-ray crystal structure of nONC. To explore the structural basis for the reduced activity of (MϪ1, Q1, M23L)rONC, we compared the conformations of the side chains of the active site residues in these two proteins (catalytic residues, Pyr 1 , Lys 9 , His 10 , Lys 31 , and His 97 ; B1 site, Thr 35 and Asp 67 ; B2 site, Glu 91 ). Fig. 7 shows the side chain conformations of these active site residues and of residue 23 for an ensemble of (MϪ1, Q1, M23L)rONC NMR structures (gold), the energy-minimized average structure for the mutant (red), and the crystal structure of nONC (blue). The side chains of His 10 , Thr 35  structures. Unfortunately, due to the large r.m.s. deviation of the side chain of these residues in the solution NMR structure, it was not possible to detect conformational differences of these residues between the solution NMR structure of (MϪ1, Q1, M23L)rONC and the nONC x-ray crystal structure. Comparing the amide resonance chemical shift difference between the two mutants ((MϪ1, Q1, M23L)rONC and E1S-rONC) offers a sensitive alternative. Our results clearly showed that there were substantial differences in the conformation of several residues located in the active site groove, including Trp 3 , Phe 6 , Lys 31 , Thr 35 , His 97 , and Val 99 . At present, we cannot quantify these changes in structural terms or pinpoint the precise functional effect of these changes. However, enzyme activity is highly sensitive to minor changes in the distances between critical functional groups (46). It has been estimated that, in the hydride transfer reaction catalyzed by dihydrofolate reductase, a 0.1-or 0.3-Å increase in the distance between the crucial carbon-carbon bond from the optimal distance of 2.6 Å causes an increase in the energy of activation of 0.7 or 5.0 kcal/mol, respectively, translating into a reduction in the rate of the hydride transfer step by a factor of 3 or 5000, respectively. Thus, a small change in the relative positioning of the active site residues might account for the change in enzymatic activity. Subtle changes in the active conformation that cannot currently be easily detected by structural studies may affect the catalytic activity and cytotoxicity. Thus, the observed large chemical shift changes for the active site residues between these two mutants may be significant.

Dynamics of (MϪ1, Q1, M23L)rONC and E1S-rONC
The dynamic behavior and the spatial distributions of 15 N-1 H vectors fitted to different models can be much more clearly seen in the color-coded sausage model (Fig. 8), generated using the program MOLMOL (39). The radius of the sausage is inversely related to the magnitude of the order parameter. Thus, mobile regions with smaller order parameters have larger diameters. The main features that can be seen from the figure are the following.
(MϪ1, Q1, M23L)rONC Shows the Presence of Substantial Dynamics-The evidence for this is as follows: 1) 14 residues had to be simulated with models other than model 1; 2) 12 residues showed weak resonances or nonexponential relaxation decay; and 3) 9 residues had relaxation parameters that could not be modeled satisfactorily with any of the five models.
(MϪ1, Q1, M23L)rONC Exhibits Domain Motion-In (MϪ1, Q1, M23L)rONC, in addition to the presence of flexibility in the loop regions, the linker between strands ␤ 5 and ␤ 6 appeared to be highly dynamic, as indicated by the presence of four contiguous residues displaying flexibility. These were residues 85 (model 2), 86 and 88 (model 3), and 87 (which could not be fitted to any model). Residue 59, the linker residue between strands ␤ 3 and ␤ 4 , also could not be fitted to any model. The C terminus also exhibited high flexibility, as indicated by the presence of two residues fitted to model 2 (residues 102 and 103), one that could not be fitted to any model (residue 101), and one with an intensity too weak to be measured (residue 98). The presence of mobile linker regions between strands ␤ 3 and ␤ 4 and between ␤ 5 and ␤ 6 suggests domain movement, possibly a bending motion involving the opening and closing of ␤-sheets S1 and S2. A similar domain motion has been proposed in RNase A (47-49). The diameter of the sausage is inversely related to the order parameter of the corresponding residue. The color-coding is as follows. Gray, model 1; cyan, model 2; green, model 3; orange, model 4; red, model 5; blue, proline residues; black, resonances that were not used to deduce the dynamic information due to weak intensities or multiple decay or inability to fit them satisfactorily to any model. These images were generated using the MOLMOL program (39). The presence of domain motion was further supported by the presence of flexible residues in ␤ 4 (residue 64 displayed multiple exponential decay and is colored black) and ␤ 5 (residue 76 was fitted to model 4, whereas residues 77 and 80 could not be fitted to any model).
Helix ␣ 1 of (MϪ1, Q1, M23L)rONC Is Not Rigid-Residues showing flexibility in the ␣ 1 helix included residue 4 (fitted to model 5), residue 8 (which showed multiple exponential decay), and residue 9 (fitted to model 4). Furthermore, residues 11-13 could not be fitted to any model, and the presence of flexibility in these residues suggests the possible presence of a large scale bending motion for the ␣ 1 helix, clear evidence for the lack of hydrogen bonds involving the N-terminal residue and the bulk of the protein in (MϪ1, Q1, M23L)rONC. This contrasts with the x-ray crystal structure of nONC, in which hydrogen bonds involving Pyr 1 , Lys 9 , and Val 96 are seen.
The Mutation Site at Leu 23 Displays Considerable Flexibility-Residue 22 was fitted to model 2, and residue 23 was fitted to model 3. Furthermore, residues 26 and 27 show nonexponential relaxation decay. This observation explains the increased susceptibility of M23L-rONC to protease digestion at this site (20). It is also consistent with our observation of the disruption of the hydrogen bonds between Lys 31 -NH and Phe 28 -CO and between Phe 28 -NH and Thr 25 -O␥ 1 in the solution structure of (MϪ1, Q1, M23L)rONC.
E1S-rONC Is Much More Rigid Than (MϪ1, Q1, M23L) rONC-Seventy-seven resonances in E1S-rONC could be fitted to model 1 compared with 61 in (MϪ1, Q1, M23L)rONC. Only six resonances in E1S-rONC needed to be fitted to other models (one to model 2 and five to model 3) compared with 14 in (MϪ1, Q1, M23L)rONC. In addition, 11 resonances displayed multiple exponential decay, and six resonances could not be fitted to any model, the corresponding numbers in (MϪ1, Q1, M23L)rONC being 12 and 9. The ␣ 1 helix in E1S-rONC is not flexible, since all resonances in this helix could be fitted to model 1. The first residue showing some flexibility was Arg 15 , the amide resonance of which relaxed nonexponentially. This lack of flexibility of the N-terminal ␣ 1 helix suggests the possibility of the formation of a hydrogen bond network involving the first serine residue. The importance of structural integrity due to the presence of a hydrogen bond network involving the N terminus has been emphasized (22), and the formation of a hydrogen bond network between the N-terminal residue and the active site Lys 9 has been suggested (22,50). Our data provided strong support for the formation of hydrogen bonds between the Nterminal serine and the bulk of the protein. Additional regions that showed reduced flexibility in E1S-rONC compared with (MϪ1, Q1, M23L)rONC included the linker region between strands ␤ 5 and ␤ 6 and residues at the M23L mutation site. Thus, the two onconase variants displayed substantially different dynamic behavior.

Structural Basis for the Mutation-induced Functional Changes
E1S-rONC and (MϪ1, Q1, M23L)rONC differ in only three residues, the presence or absence of Met at position Ϫ1 and the Gln 1 to Ser 1 and Met 23 to Leu 23 mutations. Whereas (MϪ1, Q1, M23L)rONC loses enzymatic activity and has a reduced stability, E1S-rONC retains both catalytic activity and cytotoxicity, and its melting temperature (89°C) is close to that of nONC (22). This raises the question of the structural basis for these drastic functional differences. The present study indicated that the structure and dynamics of (MϪ1, Q1, M23L)rONC and E1S-rONC differed significantly. The structural differences included the orientation of the N-terminal residue, the retention of a hydrogen bond network involving the N-terminal residue in E1S-rONC, but not in (MϪ1, Q1, M23L)rONC, and certain, as yet undefined, structural changes in the conformation of some of the active site residues. These changes may be responsible for most of the observed differences in enzyme activity and cytotoxicity.
The observation of a significant difference in dynamics between (MϪ1, Q1, M23L)rONC and E1S-rONC was surprising but not totally unexpected. Notomista et al. (20) have shown that M23L-rONC is cleaved by pepsin and chymotrypsin under conditions in which nONC is largely resistant to proteolysis and have demonstrated that the Phe 28 -His 29 peptide bond is the primary site of proteolysis. Thus, the Phe 28 -His 29 peptide bond is more accessible for proteolysis in the M23L mutant than in wild-type nONC. They also showed that the M23L mutation lowers the denaturation temperature from that of 88.7°C for the wild-type protein to 82.8°C and makes M23L-rONC more susceptible to GuHCl-induced denaturation, lowering the half-denaturation GuHCl concentration from 4.5 to 3.4 M and the ⌬G from 58.0 to 47.0 kJ mol Ϫ1 for rONC and M23L-rONC, respectively; this mutation also makes the recombinant protein more susceptible to proteolysis. Our observation of the significantly increased flexibility of the (MϪ1, Q1, M23L)rONC molecule is consistent with the observed reduced stability. As discussed above, the dynamics extracted from relaxation data were consistent with (MϪ1, Q1, M23L)rONC displaying two major motions that are absent in E1S-rONC, these being the domain motion of the ␤-sheets and the bending motion of the ␣ 1 helix. The increased motion in the ␣ 1 helix in (MϪ1, Q1, M23L)rONC can be explained by the loss of hydrogen bonds between the N-terminal residue and the active site residues, Lys 9 and Val 96 . The lack of flexibility in the ␣ 1 helix of E1S-rONC confirms the speculation that Ser 1 assumed the role of Pyr 1 and is capable of forming a hydrogen bond with the active site residues (21,22,50). The presence of domain motion of the ␤-sheets in (MϪ1, Q1, M23L)rONC may be due to the weakening of domain interactions. In the structure of onconase, the ␣ 2 helix is in contact with the ␤ 5 strand. As shown in the present study, the M23L mutation causes a significant change in chemical shift and dynamics for numerous residues in the vicinity of residue 23. This is probably due to the presence of a small void in the interface between the ␣ 2 helix and the ␤ 5 strand, created by the M23L substitution, which loosens the interaction between the ␣ 2 helix and the ␤ 5 strand and makes the ␤ 5 strand more mobile, which, in turn, causes the domain motion of the two ␤-sheets. In addition, we found a change in hydrogen bond pattern near the M23L mutation site.
In the x-ray structure of nONC, Lys 31 NH forms a hydrogen bond with Phe 28 CO, and Phe 28 NH forms a hydrogen bond with the Thr 25 side chain O ␥1 . In the NMR structure of (MϪ1, Q1, M23L)rONC, these groups were located at an unfavorable distance and angular orientation for hydrogen bond formation. The H-D exchange experiments showed that the Lys 31 NH proton displayed a medium exchange rate and that the Phe 28 NH was in fast exchange, in agreement with the absence of hydrogen bond formation for these two protons, rendering the Phe 28 -His 29 bond more accessible for proteolysis. The increased flexibility of the Leu 23 -Asp 32 loop region, the change in Nterminal orientation that disrupts the hydrogen bond network involving the N-terminal residue and the active site residues, the loosening of the contact between the ␣ 2 helix and the ␤ 5 strand, and the disruption of hydrogen bonds involving residues near Leu 23 may all contribute to the reduced stability and increased flexibility, which probably cause the change in the enzymatic properties of (MϪ1, Q1, M23L)rONC compared with nONC.