A Vibrational Structure of 7,8-Dihydrobiopterin Bound to Dihydroneopterin Aldolase*

Dihydroneopterin aldolase (DHNA) catalyzes the conversion of 7,8-dihydroneopterin to 6-hydroxymethyl-7,8-dihydropterin and glycolaldehyde. An inhibitor of the enzyme, 7,8-dihydrobiopterin, free in solution and bound in its complex with the enzyme has been studied by Raman difference spectroscopy. By using isotopically labeled 7,8-dihydrobiopterin and normal mode analyses based on ab initio quantum mechanic methods, we have positively identified some of the Raman bands in the enzyme-bound inhibitor, particularly the important N5=C6 stretch mode. The spectrum of the enzyme-bound inhibitor shows that the pK a of N5 is not significantly increased in the complex. This result suggests that N5 of 7,8-dihydroneopterin is not protonated before the bond cleavage of 7,8-dihydroneopterin during the DHNA-catalyzed reaction as has been suggested. Our results also show that the N5=C6 stretch mode of 7,8-dihydrobiopterin shifts 19 cm− 1 upon binding to DHNA. Various possibilities on how the enzyme can bring about such large frequency change of the N5=C6 stretch mode are discussed.

The vibrational spectrum of a molecule reports on the bonds orders of molecular bonds and is very sensitive to the small changes in the distribution of electrons within bonds that come about when an enzyme binds its substrate. The changes are a fingerprint of the enzyme-substrate interactions that exist in the ground state. Raman difference spectroscopy is well suited for the measurement of the Raman spectrum of a ligand bound to an enzyme (2). In this study, we have applied the technique to the DHNA⅐7,8-dihydrobiopterin (Scheme II) enzyme-substrate complex. Previous studies have shown that 7,8-dihydrobiopterin is an inhibitor to the DHNA-catalyzed conversion of 7,8-dihydroneopterin, and it also promotes the formation of enzymic tetramer, an inactive form of the enzyme, from the active octamer. 2 In addition, our current studies show that 7,8-dihydrobiopterin is a poor substrate of DHNA. It is converted to a not as yet fully characterized species but at a reaction rate slow enough so that Raman difference spectroscopy is feasible. Since 7,8-dihydrobiopterin and 7,8-dihydroneopterin (the normal substrate of dihydroneopterin aldolase) are very similar structurally, particularly with regard to the pterin ring and close to the bond breaking point in the catalyzed reaction and also undergo enzymic catalysis, we use bound 7,8-dihydrobiopterin as a mimic for the Michaelis complex of this enzyme.
The binding site structure of DHNA around the pterin ring of the substrate is very much like that of dihydrofolate reductase (3)(4)(5). For example, the N2-C2-N3 moiety of the dihydropterin ring forms hydrogen bonds to Glu-74, similar to those found for the structurally conserved Asp in dihydrofolate reductases (DHFR). This, other structural similarities, and chemical considerations prompted the suggestion that the reaction mechanism of DHNA may be similar in some respects to DHFR (1). In particular, it has been postulated that the electron flow out of the C-C bond that is broken by DHNA (Scheme I) could be stabilized by protonation of N5 of the pterin ring. A recent study using Raman difference spectroscopy of the electronic structure of substrates bound to Escherichia coli dihydrofolate reductase showed that the pK a of N5 is raised from 2.6 in solution to 6.5. This can be determined in titration studies using vibrational spectroscopy since the N5ϭC6 stretch "marker" band is located near 1650 or 1675 cm Ϫ1 for unprotonated or protonated N5, respectively, exhibiting then easily measured and well separated bands (6). Hence, a particular focus of this investigation is the protonation state of N5 in the pterin ring for substrate bound to DHNA. We found little, or no, protonation of N5 when 7,8-dihydrobiopterin binds to DHNA. However, other unusual binding-induced changes to the elec-tronic structure of the substrate within the pterin ring are observed. MATERIALS  Spectroscopy-The Raman spectra were measured using an optical multichannel analyzer system. The optical multichannel analyzer system uses a Triplemate spectrometer (Spex Industries, Metuchen, NJ) with a model DIDA-1000 reticon detector connected to an ST-100 detector (Princeton Instruments, Trenton, NJ). Details of the system can be found elsewhere (2). The 514.5 nm line from an argon ion laser (model 165, Spectra Physics, Mountain View, CA) was used to irradiate the sample (ϳ100 milliwatts). Separate spectra for enzyme and enzyme-inhibitor complexes in solution, approximate concentration of 4 mM, were measured using a special split cell (the volume of each side being about 30 l) and a sample holder with a linear translator as described previously (2). The spectrum from one side of the split cuvette is taken, the split cell is translated, and the spectrum from the other side is taken. This sequence is repeated until sufficient signal to noise is obtained. A difference spectrum is generated by numerically subtracting the sum of the spectra obtained from each side. In general, the two summed spectra do not subtract to zero, as judged by the subtraction of well know protein marker bands (for example, the amide-I, amide-III, and the 1450 cm Ϫ1 bands, the latter band being especially useful since it is generally not affected by protein conformational changes). These protein marker bands are determined from their bandwidths (generally much broader than those from spectra of bound substrates) and their characteristic positions. Hence, one summed spectrum is scaled by a small numerical factor, generally between 1.05 and 0.95, which is adjusted until the protein bands are nulled (see e.g. Ref. 8). The same control procedures were performed on all the difference spectra results herein. Resolution of the spectrometer is 8 cm Ϫ1 for the present results. A spectral calibration is done for each measurement using the known Raman lines of toluene, and absolute band positions are accurate to within Ϯ2 cm Ϫ1 . None of the spectra presented here have been smoothed.
Calculations-The ab initio calculations were carried out on the models by Hatree-Fock method with the 6 -31 g** basis set, as implemented in Gaussian 98. The geometries of the model compound of 7,8-dihydrobiopterin were first optimized energetically, and then the vibrational normal modes were calculated using the same basis set. True local minimum on the potential surface of the complexes for the geometry-optimized complexes was verified from the vibrational frequency calculations in which no imaginary frequency was found. In all cases, a stable structure of the model compound is achieved without any geometry constraint.

RESULTS
When 7,8-dihydrobiopterin binds to DHNA, it slowly undergoes product formation. Fig. 1 shows the UV-visible absorption measurements of the DHNA⅐7,8-dihydrobiopterin complex at 1-, 10-, 25-, 55-, 80-, 120-, 150-, 180-, and 360-min time intervals, respectively, after mixing. The sample pH was 7.8, and the temperature was set to 32°C. The initial spectrum has a max of 332 nm, slightly red-shifted from the 328 nm max of 7,8-dihydrobiopterin in solution at the same pH. DHNA then slowly converts 7,8-dihydrobiopterin into a new species, which has an absorption band at 423 nm. Separate kinetic experiments of this complex at various pH values indicate that the reaction rate is lowered by about 2.5-fold for pH 6.5. The intensity increase of the 423 nm absorption band can be fitted with a first order exponential function. Based on this result and the observation of a clear isosbestic point found in the absorption spectra of complex at various stages of the reaction (Fig. 1), it is reasonable to conclude that only one product is formed. A similar spectral change occurs for 7,8-dihydroneopterin, the normal substrate of DHNA. However, no isosbestic point was found, and the kinetics cannot be fitted with a single exponential function. This product of the reaction of 7,8-dihydrobiopterin is not due to protonation of N5 of the pterin ring; the red shift is much too large. Titration studies of 7,8-dihydrobiopterin in solution show that when N5 of 7,8-dihydrobiopterin is protonated, a new absorption band at ϳ365 nm appears. If protonated 7,8-dihydrobiopterin has a similar absorption in DHNA, we can estimate the percentage of this form in the DHNA⅐7,8-dihydrobiopterin complex from its absorption spectrum taken just after mixing. On this basis, less than 10% of N5 is of the protonated form in the enzyme⅐7,8-dihydrobiopterin complex. Furthermore, there is no detectable increase of this SCHEME I. SCHEME II.
form when the sample pH is changed to 6.5. Fig. 2 shows the Raman spectra of 7,8-dihydrobiopterin in solution at pH 6.5. The pK a of N5 is 2.6 (6) so that N5 of pterin ring is unprotonated in the spectra of Fig. 2. There are four major peaks above 1400 cm Ϫ1 evident in Fig. 2a, at 1480, 1559, 1605, and 1636 cm Ϫ1 , respectively. The band at 1636 cm Ϫ1 can be assigned to the C6ϭN5 stretch based on its 17 cm Ϫ1 down shift in the 15 N5-labeled compound spectrum (Fig. 2b) (6). The other three bands have not been assigned before. However, with the aid of ab initio frequency calculations at the HF/6-31 g** level performed on a model compound for 7,8-dihydrobiopterin, preliminary assignments of these bands can be made.
Here we need to point out that the chemical bond lengths, especially the double bonds which contain oxygen or nitrogen, are underestimated by ab initio methods at the Hatree-Fock level compared with those observed values. This is due to the neglect of electron correlation forces in the calculations, the limited basis set, and also that the calculations are "gas phase" which do not treat the interactions with solvent. Consequently, the calculated stretch frequencies of these bonds are typically overestimated (15-20%) compared with the observed values. Nevertheless, the overestimation of the calculated frequency differences compared with experimentally determined values tends to remain constant. Hence, by applying a uniform scaling to the calculated frequencies, reasonably accurate normal mode assignments can be made based on the calculations. Our previous experience on such calculations suggests that, in many cases, the calculations provide a reasonable description of the normal modes and their response to isotopic labeling or to environmental change around the molecule (9 -11).
It is well known that the Raman bands in the spectral region between 1400 and 1700 cm Ϫ1 consist of normal modes mainly due to CϭC, CϭO, and CϭN stretch motions and CH bending motions. For 7,8-dihydrobiopterin, there are four double bonds, and thus we expect to observe at least four Raman bands in these region, although some of the modes may contain more than one double bond stretch characters. According to the ab initio frequency calculations on an isolated 7,8-dihydrobiopterin molecule, there are five vibrational modes that have relatively strong Raman intensities in this region. The calculations predict that the band with lowest frequency in this region, at 1480 cm Ϫ1 in Fig. 2a, which is not sensitive to the 15 N5 labeling but shifts down by 6 to 1474 cm Ϫ1 when the exchangeable protons of 7,8-dihydrobiopterin are deuterated as occurs for solution samples in D 2 O (panels b and c of Fig. 2, respectively), is likely due to the in-plane bending mode of the C7H2 group of the dihydropterin ring. There are two ways to make assignments of the two bands at 1559 and 1605 cm Ϫ1 . The first is to assign the 1559 cm Ϫ1 band to a CϭC ring stretch which also contains contribution from N8H bend. This mode is not sensitive to the 15 N5 labeling (Fig. 2b) but shifts down 38 cm Ϫ1 for deuterated samples (Fig. 2c) which removes coupling to NH bending motions because of the large down shift of a ND bend. In this case, the 1605 cm Ϫ1 band in Fig. 2a is assigned to the C1ϭN2 stretch with contributions from N2H and N3H bending motions. In D 2 O, the frequency of this mode would down shift to 1577 cm Ϫ1 as observed (Fig. 2c) because its coupling with the NH bending is removed. A second way to assign these two modes that is compatible with the calculations is that the C1ϭN2 stretch and CϭC stretch of the dihydropterin ring are coupled to form an in-phase and an out-of-phase pair. The in-phase mode has a higher Raman intensity and lower frequency, the 1559 cm Ϫ1 band in Fig. 2a can be assigned to this mode. The out-of-phase mode has a lower Raman intensity and higher frequency, and the 1605 cm Ϫ1 band in Fig. 2a can be assigned to this mode. Since both of these modes have contributions from the ring NH bending motions, significant frequency shifts are predicted for deuterated samples, which is consistent with experimental observations. FIG. 1. UV-visible absorption spectra of DHNA⅐7,8-dihydrobiopterin complex taken at 1, 10, 25, 55, 80, 120, 150, 180,  The concentration of the sample was 10 mM. and the sample pH was 6.5. The spectrum of water was subtracted from these spectra.
The ab initio calculations also predict that the two highest frequency modes in this region are due to the C6ϭN5 and C4ϭO stretch modes. The C6ϭN5 stretch mode is predicted to have the highest Raman intensity and its frequency is insensitive to NH deuteration (unless N5 is protonated, which is not the case here). Such predictions are in very good agreement with the experimental observations (see Fig. 2, a and c). The calculations also predict that the frequency of the C4ϭO stretch mode, predicted to have significantly lower (0.2-0.1 times) Raman intensity than that of C6ϭN5 mode, is very sensitive to the strength of hydrogen bonding to the C4ϭO oxygen. In vacuum, its frequency is higher than the C6ϭN5 stretch but, if a hydrogen-bonded water molecule is included in the calculations, its frequency is comparable to the C6ϭN5 stretch mode. This prediction is supported by the observation of the Raman and IR spectra of 7,8-dihydrobiopterin in Me 2 SO (data not shown). The C6ϭN5 stretch shifts down by 6 only cm Ϫ1 for 7,8-dihydrobiopterin in Me 2 SO compared with its value in water, whereas a new Raman band is observed at 1658 cm Ϫ1 that is insensitive to 15 N5 labeling (data not shown). Based on this and the fact that the 1658 cm Ϫ1 band is very strong in IR, the 1658 cm Ϫ1 band can be easily assigned to CϭO stretch for 7,8-dihydrobiopterin in Me 2 SO. Thus, it is reasonable to assume that the C4ϭO stretch mode of 7,8-dihydrobiopterin in water is hidden somewhere under the intense C6ϭN5 stretch mode at 1636 cm Ϫ1 in Fig. 2a.
Our calculations also predict that, under certain conditions, the C4ϭO stretch and C6ϭN5 stretch can be mixed. It this case, the Raman intensity of the C4ϭO becomes significantly higher by borrowing intensity from the C6ϭN5 stretch. Such results provide a rationale for the observation of the doublet at 1623/1638 cm Ϫ1 in 7,8-dihydrobiopterin in D 2 O (Fig. 2c). Apparently, this doublet is due to the coincidental mixing of the two stretch modes so that the intensity of the C4ϭO mode becomes much stronger. This interpretation is further supported by the observation that only a single band is observed at 1614 cm Ϫ1 (data not shown) for the 15 N5-labeled compound in D 2 O. The band at 1623 cm Ϫ1 can be assigned to the C4ϭO stretch, whereas that at 1638 cm Ϫ1 can be assigned to the C6ϭN5 stretch since the former is strong in the IR spectrum compared with the latter (data not shown).
Since 7,8-dihydrobiopterin is slowly converted to product during the Raman measurements, the difference spectrum between DHNA⅐7,8-dihydrobiopterin complex and DHNA obtained in the usual way (see "Materials and Methods") will contain Raman bands from the bound enzyme product of 7,8-dihydrobiopterin. Thus, an extra Raman spectrum of DHNA⅐7,8-dihydrobiopterin complex was taken after the reaction completed, as monitored by UV-visible absorption. Since it has been shown that there is just one product (see above and Fig. 1) and the product has characteristic Raman bands that are distinct from those of 7,8-dihydrobiopterin (data not shown), it is possible to remove all of the peaks of the enzyme product from the Raman difference spectrum. This was achieved by obtaining two difference spectra between DHNA⅐7,8-dihydrobiopterin complex and DHNA as follows: the first was from the freshly prepared DHNA⅐7,8-dihydrobiopterin, and the was second from the DHNA⅐7,8-dihydrobiopterin complex 3 h later after the reaction was completed. An appropriate amount of the second difference spectrum is subtracted from the first to remove the Raman peaks due to the enzyme product.
The results of this procedure are presented in the difference spectra of Fig. 3. In these spectra the major bands are from bound 7,8-dihydrobiopterin, and the minor bands are either from unbound 7,8-dihydrobiopterin or from DHNA itself since protein Raman bands are slightly affected by binding ligands (2). The major bands at 1482, 1559, 1605, and 1655 cm Ϫ1 in the Fig. 3a spectrum can be assigned to the bound 7,8-dihydrobiopterin since corresponding bands in the solution spectrum of solution 7,8-dihydrobiopterin can be readily found (Fig. 2a). The spectral features of the small band at 1679 cm Ϫ1 in Fig. 3a and the dip at its low frequency side are likely due to incompletely nulled protein peaks in the difference spectrum. The 1679 cm Ϫ1 peak does not shift upon 15 N5 labeling of 7,8-dihydrobiopterin (Fig. 3b) or deuteration of the sample (Fig. 3c); thus it does not arise from the protonated C6ϭN5 stretch. It is interesting to note that the bands at 1482, 1559, and 1605 cm Ϫ1 are all within 2 cm Ϫ1 of their corresponding position in aqueous solution, thus they can be assigned to the C7H2 bending, CϭC stretching, and C2ϭN1 stretching modes, respectively. The 1655 cm Ϫ1 band in Fig. 3a, assigned to the C6ϭN5 stretch based on its 18 cm Ϫ1 down shift upon 15 N5 labeling (Fig. 3b), shifts up by 19 cm Ϫ1 relative to its solution value (Fig. 2a). Since this band remains at 1655 cm Ϫ1 when the sample is deuterated (Fig. 3c), N5 of 7,8-dihydrobiopterin is not protonated when bound to DHNA. Thus, on the basis of the Raman results, no significant amount of protonated N5 of bound 7,8dihydrobiopterin can be detected under our experimental conditions. Similar experiments performed at pH 6.5 yield the same results (data not shown). We tentatively assign the 1615 cm Ϫ1 band observed in Fig. 3c to the C4ϭO stretch because it is near its position at 1623 cm Ϫ1 found in the solution spectrum ( Fig. 2c) and has about the right Raman intensity for this band. The apparent small 7 cm Ϫ1 down shift of the C4ϭO stretch upon binding is consistent with a small amount of bond polarization that is often observed for keto groups bound to proteins. Such a bond polarization comes about from increased hydrogen bonding in the protein relative to water (2).
In an attempt to understand the influence of the binding pocket on the C6ϭN5 stretch frequency, we have conducted additional ab initio calculations on a series of model systems of 7,8-dihydrobiopterin, including its complex with a carboxyl group in contact with N2H2 and N3H, a water molecule hydrogen-bonded to the CϭO or OH groups in various configurations, and different orientations of the OH group of the chain relative to the N5. The results show that the addition of a carboxyl group has a relatively large effect on the C6ϭN5 stretch mode, causing a shift of this mode by about 10 cm Ϫ1 but toward lower frequency. The hydrogen bonding on the OH or CϭO by a water molecule does not have significant effect on the C6ϭN5 stretch, causing less than 5 cm Ϫ1 shift. On the other hand, the orientation of the first OH group of the arm on C6 has a significant effect on the C6ϭN5 stretch. When the orientation of this OH changes from anti (relative to N5) to a syn conformation, the C6ϭN5 stretch is predicted to shift by about 10 cm Ϫ1 toward higher frequency.

DISCUSSION
The substrate-binding site of DHNA near the pterin ring bears a strong similarity to the binding of folate substrates to dihydrofolate reductases. For example, the carboxyl of Glu-74 in the DHNA binding pocket is positioned very similarly to that of the strictly conserved Asp group (Asp-27 in E. coli DHFR) in the reductases. A structural water molecule is also found in E. coli DHFR (12) in some of the crystallographic structures in a place similar to that in the DHNA-binding site (see Scheme I).
In the dihydrofolate to tetrahydrofolate reaction, N5 of the pterin ring is protonated. It is often supposed that the conserved Asp groups of DHFRs either donate the proton to N5 via active site structural water molecules or stabilize the positive charge on protonated N5 (4,13,14). There is some evidence that the pterin ring CϭO bond is polarized (5), and this is consistent with a proton relay system from Asp-27 to N5 via binding site water molecule(s). It has been shown that the pK a of N5 of dihydrofolate bound to E. coli DHFR is raised from its solution value of 2.6 to 6.5, and this structural attribute can be a major factor in the activity of this enzyme. Asp-27 is essential for enzymic activity in DHFR (17) as well as structurally necessary for raising the pK a of N5 in situ (6). Given these structural similarities, it is reasonable to compare the structures of the bound pterin ring in DHFR to that in DHNA.
Our UV-visible and Raman studies indicate that, in the Michaelis complex of DHNA⅐7,8-dihydrobiopterin, more than 90% of N5 is unprotonated. Furthermore, there is no detectable change in its UV-visible and Raman spectra when the sample pH is changed from 7.8 to 6.5, suggesting that no significant amount of protonated N5 is present. The results place the pK a of N5 of bound 7,8-dihydrobiopterin to be less than around 5. The kinetic data on the DHNA-catalyzed reaction on 7,8-dihydrobiopterin shows that the optimal pH for the DHNA-catalyzed reaction is about 9, similar to that for the 7,8-dihydroneopterin, consistent with the argument that protonated N5 is not required in the ground state or in the transition state of the reaction. We also find that the CϭO group of the pterin ring of bound 7,8-dihydrobiopterin is apparently not particularly polarized from its solution value. All of this is in contrast to DHFR where the pK a of N5 is 6.5 and the pH profile for hydride transfer has its half-maximum also near this value (15,16). Thus the effect on the pterin ring by binding to DHNA is quite different than that found for dihydrofolates bound to (E. coli) DHFR despite the similarities in the structure of the binding pocket. The pterin ring binding pocket of DHNA would appear to be the design for binding of the ring and not for promotion of the catalyzed reaction. This notion is strongly reinforced in studies of mutant DHNA whereby only a small decrease (to 25% of the wild type) in activity is found for enzymes that lack the active site Glu-74. 2 It needs to be pointed out here that our preliminary results suggest that the product from DHNA-catalyzed conversion of 7,8-dihydrobiopterin is not the same as that from 7,8-dihydroneopterin, the natural substrate of DHNA. Thus, the terminal OH group in 7,8-dihydroneopterin plays a crucial role in the reaction, since its presence does not only dictate the reaction rate but also the reaction pathway. On the other hand, the unprotonated state of N5 in 7,8-dihydrobiopterin is unlikely the reason for the slow turnover rate compared with 7,8-dihydroneopterin since both of these molecules have similar pH reaction profiles, as mentioned above.
The only significant observed change in the electronic structure of pterin ring of 7,8-dihydrobiopterin bound to DHNA relative to its solution form observed by our study is within the C6ϭN5 bond. The C6ϭN5 stretch increases 19 cm Ϫ1 upon binding. This is an unusual change and is difficult to bring about. As outlined under "Results," our calculations indicate that none, i.e. Glu-74, the active site water molecule, Lys-100 (see Scheme I), of the active site polar groups can bring about the increased electron density in the C6ϭN5 bond through electrostatic interactions (i.e. hydrogen bonding) that would result in the increased frequency of the C6ϭN5 stretch. This is particularly surprising with regard to the hydrogen bonding of the NH 2 group of Lys-100 to substrate that certainly participates in enzymic catalyzed C-C bond breaking. This leaves steric interactions at the active site that change the conformation and electron distribution of the pterin ring through geometry changes and/or imposition of strain on the bound substrate. For example, there is a substantial upward shift in frequency (although only about half that observed) for a C6ϭN5 bond that is syn to the adjacent N6 -C(OH)group compared with the anti conformation (see Table 2 in Ref. 10). It is certainly possible that the enzyme imposes strain on the alcoholic alkane "arm" of the bound biopterin which brings about sufficient electron delocalization to increase the electron density in the C6ϭN5 bond, hence increasing the C6ϭN5 stretch.