Kinetics of CO ligation with nitric-oxide synthase by flash photolysis and stopped-flow spectrophotometry.

Interaction of CO with hemeproteins has physiological importance. This is especially true for nitric-oxide synthases (NOS), heme/flavoenzymes that produce .NO and citrulline from L-arginine (Arg) and are inhibited by CO in vitro. The kinetics of CO ligation with both neuronal NOS and its heme domain module were determined in the presence and absence of tetrahydrobiopterin and Arg to allow comparison with other hemeproteins. Geminate recombination in the nanosecond time domain is followed by bimolecular association in the millisecond time domain. Complex association kinetics imply considerable heterogeneity but can be approximated with two forms, one fast (2-3 x 10(6) M-1 s-1) and another slow (2-4 x 10(4) M-1 s-1). The relative proportions of the two forms vary with conditions. For the heme domain, fast forms dominate except in the presence of both tetrahydrobiopterin and Arg. In the holoenzyme, slow forms dominate except when both reagents are absent. Geminate recombination is substantial, approximately 50%, only when fast forms predominate. Stopped-flow mixing found dissociation constants near 0.3 s-1. These data imply an equilibrium constant such that very little CO should bind at physiological conditions unless large CO concentrations are present locally.

Nitric-oxide synthase is a hemoprotein that catalyzes conversion of the amino acid L-arginine to citrulline and nitric oxide (1,2). The NO produced activates another heme enzyme, guanylate cyclase (see Reaction 1), that catalyzes formation of cyclic guanosine monophosphate, a second messenger that mediates numerous biochemical events including vascular smooth muscle relaxation, platelet disaggregation, photoreceptor cell signaling, ion transport in gastrointestinal cells, and myeloid cell differentiation.
At least three isoforms of NOS 1 have been purified and characterized. In each, a heme prosthetic group, iron protoporphyrin IX, is attached to the protein by a proximal cysteine. Visible absorption spectroscopy and other physicochemical measurements reveal similarities to cytochromes P450 (3)(4)(5)(6)(7)(8).
Like that family of proteins, reduced (ferro) NOS forms carboxy derivatives with an absorption maximum at ϳ446 nm (4). Marletta (9) and Stuehr et al. (10) have proposed mechanisms for the oxidation of Arg catalyzed by NOS, based largely upon the similarity of NOS and P450, and Masters (11) has proposed a scheme involving redox interactions with the flavoprotein module of NOS. A key observation was that enzyme activity is inhibited by CO (4,6,9) as well as by NO (12), the physiological significance of which remains uncertain. Enzyme inhibition in vitro required large concentrations of either CO or NO. If those reagents play a role in vivo, the mechanism must involve locally enhanced concentrations. For NO, fast turnover and slow diffusion away from the site of production in unstirred cytoplasm might possibly lead to large concentrations locally and, consequently, to autoregulation of NO production. Recent evidence suggests a role for CO produced by heme oxygenase as a messenger molecule in the same cell types as nNOS (13)(14)(15). Could the activity of heme oxygenase produce high concentrations? The CO concentration needed depends upon the equilibrium constant for CO binding. Unfortunately, equilibrium constants for binding CO and NO are not known for NOS. Matsuoka et al. (16) reported CO combination rate constants for neuronal NOS but not CO dissociation data, which could have been combined with the combination rate constant to calculate an equilibrium constant. In hemeproteins, CO dissociation rate constants vary from 6.5 s Ϫ1 in P450 to 7.2 ϫ 10 Ϫ5 s Ϫ1 in horseradish peroxidase. Therefore, knowledge of CO association and dissociation is important for determining NOS reactivity relative to myoglobin, guanylate cyclase, and other heme proteins, which potentially could compete with NOS for reactive ligands.
In addition to the interest generated by its physiology, this sensitivity of CO dissociation rate constants to protein structure makes it an excellent parameter for providing qualitative information about the local heme environment. A review of CO dissociation constants reveals two trends. 1) CO off rates are markedly affected by the nature of any base proximal to the heme-CO bond, whereas for a given base, any weakening of the base-heme bond, or its total rupture, dramatically increases the CO dissociation rate. 2) Steric hindrance on the distal side reduces both association and dissociation rates for CO in all cases known (17).
These same issues arise for carboxy guanylate cyclase activation. In vitro activity is significantly affected by 1 atm of CO, but equilibrium constants indicate that in vivo, little, if any, guanylate cyclase activation by CO should be expected. Nevertheless, several biological phenomena seem to be affected by small amounts of CO (15). We have no explanation, but the example points out the difficulty in extrapolating from in vitro to in vivo conditions. Nevertheless, a knowledge of equilibrium and rate parameters is the essential starting point of all discussion.
The studies reported here determined the kinetics of ligation of CO with neuronal nitric-oxide synthase stably expressed in human kidney 293 cells (4) and a holoenzyme (18) and heme module (residues 1-714) (19) expressed in Escherichia coli. Flash photolysis and stopped-flow spectrophotometry were used to determine association and dissociation rate constants. We investigated the effects of tetrahydrobiopterin (BH 4 ) and L-arginine (Arg). For the latter, advantage was taken of the ability in E. coli to express nitric-oxide synthases in the absence of BH 4 and with diminished Arg concentrations (18,20) in the case of the heme domain. In addition to bimolecular association, we also investigated geminate recombination of CO-iron pairs confined within the protein. These studies allowed us to estimate the equilibrium binding constant of CO with NOS and provided insights into the nature of the heme pocket in the enzyme.

MATERIALS AND METHODS
Carbon monoxide (99.8%), nitric oxide (99.9%), high purity argon, and premixed CO in argon were all from Matheson. NO was further purified by passage through a column of fresh KOH. Buffers were prepared (20 mM Tris-HCl, usually pH 7.8, 100 M EDTA, 100 mM NaCl) in a gas-tight syringe and deoxygenated by bubbling with argon for Ͼ40 min. When indicated, buffers also contained 250 M BH 4 and/or 0.5 or 10 mM Arg. Microperoxidase solutions were prepared as described elsewhere (21).
Rat nNOS holoenzyme was expressed in two different systems. Unless otherwise indicated, it was in E. coli, purified and reconstituted as described previously (18). Nitric oxide formation, measured in E. coli-expressed nNOS after Arg and BH 4 repletion using a hemoglobin capture assay described previously (22), was found to be 250 nmol min Ϫ1 mg Ϫ1 (25°C) or 540 nmol min Ϫ1 mg Ϫ1 (25°C). Expression of rat nNOS in human embryonic kidney cell was performed as reported previously (4); nitric oxide formation was measured using an L-[ 3 H]citrulline assay (23) and found to be 420 nmol min Ϫ1 mg Ϫ1 (25°C). The amino-terminal heme-binding domain (residues 1-714) of rat nNOS was expressed in E. coli and purified as described previously (19). All proteins were Ͼ90% homogeneous as determined by sodium dodecyl sulfate-polyacrylamide gel electrophoresis. Protein solutions were degassed and then reduced using degassed 1% sodium dithionite solution. Finally, CO was added in a premixed ratio with argon by gentle gas flow over the liquid using a large surface-tovolume ratio. Spectra for reduced holoenzyme and heme domain were consistent with previously published data as were the spectra of liganded preparations (7). Spectra were taken before and after flash photolysis measurements to detect any possible degradation.
Flash photolysis used an improved version of an instrument described previously (24). Photolysis laser pulses were 4 ns in duration at 545 nm with up to 4 mJ of energy over an area of 0.1 cm 2 . The probe had an 8-nm bandwidth selected from a stable tungsten lamp, except that measurements of geminate recombination required a pulsed xenon flash. The photomultiplier was wired with both a large standing current and large interdynode capacitors to optimize linearity. The digitizer had been upgraded to a Lecroy Model 9361 digitizing oscilloscope.
A quantitative treatment of signal-to-noise ratio (S/N) is essential whenever there is a question of complicated kinetics. Heme concentrations were near 1 M with 1-cm optical path length using a volume of 0.5 cm 3 in a 1-cm square cell. (This gave a large surface area for gas exchange after we learned that even mild stirring caused degradation.) Initial absorbance changes ⌬A (t ϭ 0) were typically about 0.03-0.04. By accumulating 80 -200 recordings, the root mean square uncertainty at each time point was reduced below 0.00008. As a result, we achieved S/N Ϸ 500 or better. Nanosecond geminate recombination had to be measured with a wider electronic bandwidth, resulting in about a 5-fold loss in S/N despite the brighter, pulsed lamp. Good S/N for random scatter exposes a measurement to systematic distortions in the data. These have been reduced through several generations of the instrument. The most convincing short argument that systematic distortion is not responsible for any major feature described below is that, by coincidence, while these measurements were being made we were also measuring kinetics of recombination in HbA 2 . The latter system gave beautiful single exponential decay curves.
Stopped-flow measurements were carried out at 20°C using a Durrum instrument following protocols described earlier (25). The CO dissociation rates were determined both by the microperoxidase method (21) and by replacing CO with NO (26). For the microperoxidase method, after mixing, the sample was 0.5 M protein, 200 M sodium dithionite, and 5 M CO. The microperoxidase concentrations varied between 3 and 10 M. For the NO substitution method, the sample was 0.5 M protein, 25 M sodium dithionite, and 150 M CO. The NO concentrations varied between 175 and 275 M. Spectra before and after reactions were as expected (27). For both methods, absorption changes were monitored at 443 nm. Measurements were done in duplicate and were proved to be independent of concentration of either microperoxidase or NO.

RESULTS
The kinetics of the reaction of CO with both the complete nNOS holoenzyme and the heme domain fraction of nNOS were characterized, and the rates in the presence and absence of Arg and BH 4 were measured. Thus, there are eight cases to distinguish. The effect of Ca 2ϩ /calmodulin on rates of CO binding has been previously shown to be negligible (16).
Association-Flash photolysis measurements were made at 23 Ϯ 0.5°C. They were carried out at both pH 7.0 and pH 7.8, with identical results. There was also no discernible difference between nNOS from kidney cells and that from E. coli. There was no irreversible photochemistry. Transient recovery was Ͼ99.5% after each flash, and spectra did not change during a series of flashes. Association rates were determined by monitoring recombination after photodissociation in the presence of excess CO. Kinetics were quite complex. In general, we observed three phases on different time scales. There was a very fast process with a half-life on the order of 100 ns that was unaffected by changing CO concentration and was assigned as geminate recombination of protein-caged heme-ligand pairs. All slower processes varied systematically with [CO]. Solutions were equilibrated with premixed gas mixtures at 1 atm (10 5 pascals) total pressure. We used CO in argon at molar ratios of 5, 10, 20, and 50% as well as 100% CO. Molar concentrations were calculated assuming that the solubility of CO in aqueous solution at 25°C is 930 M atm Ϫ1 .
We must distinguish a kinetic "phase," as the term was used above, from an exponential component of a fit. The two words are frequently used as synonyms. Here, however, a phase may be composed of two or even three exponentials that are quite close in characteristic time constant. The group as a whole differs from other rates by at least a factor of 10, and the components behave similarly as CO concentration is changed.
Best fits to the data were calculated using our own implementation of the classic Marquardt algorithm for a nonlinear least squares fit to a sum of exponentials, where s is an overall scale factor, a i are amplitudes, which sum to unity, and k i is the rate constant. The nanosecond geminate phase involved only a single exponential, characterized by a gem and k gem . The bimolecular association required as many as four additional components, but they could be grouped into fast and slow phases characterized by k f or k s , each defined as a sum over one, two, or, occasionally, three exponentials as follows. 2) mimics what a single exponential fit to the same data yields.
To determine the bimolecular association constants k a,f or k a,s , measurements were carried out for several [CO] to prove the linear dependence on [CO]. The phases were distinguished according to Equations 1 and 2, and the rates k a,f and k a,s were plotted versus [CO]. The association rate constants are the slopes of those graphs.
There are three reasons for simplifying to two bimolecular phases. First, it permits easier comparison with other experiments that do not have the S/N to resolve all components. Second, fitting exponentials with similar rates is a difficult statistical problem. Parameters are highly correlated, and even though heterogeneity is clear, little significance should be attached to a particular expression of that heterogeneity. It is probable that there is a continuous distribution of rates. Third, since so little is known at present about NOS, it is preferable to simplify the discussion for the present while documenting whatever detail is measured for its potential future significance.
To confirm that the complex kinetics were not an instrumental artifact, we carried out a number of control experiments. 1) We investigated pulse energies from 0.4 to 4 mJ to confirm that there was no effect on the measured decay curves. 2) In pre-liminary measurements, we recorded kinetic traces at 5-nm intervals from 380 to 450 nm and obtained a transient spectrum very similar to that reported previously (16). In particular, we confirmed the extremes of the difference spectrum, the location of the isosbestic point, and the near equality of the magnitude of peak transient absorption and bleach. 3) We confirmed that we had a well defined isosbestic point at 430 -432 nm (slightly shifted between the heme domain and the holoenzyme), which persisted throughout the entire time range. Subsequently, in each measurement traces were collected both at the maximum transient bleach, 445 nm, and the maximum transient absorption, 405 nm. Kinetics reported are averages of the two. The heme domain decay curves were often indistinguishable at the two wavelengths. There may be a small difference in some cases for the holoprotein, presumably due to very small, time-dependent shifts in the transient spectrum related to conformational relaxation of the holoprotein.
Principal data for all measurements are collected in Table I, but the following sections add information about certain details, mainly the multiple exponentials included in certain phases.
Heme Domain with neither L-Arginine nor BH 4 -The fast phase could be fit fairly well using two exponentials of equal amplitude with rates differing by a factor of 3, but at very high S/N it became clear that two exponentials were not completely adequate. The steady-state absorption spectrum showed a small amount of extra absorption near 420 nm, consistent with that reported and explained by Wang et al. (28). We also noticed additional complexity in the transient spectra, particularly around the usual isosbestic point. This could be pursued, if one were interested in the so-called P420 form of NOS, but for our present purpose we believe that by choosing wavelengths characteristic of the P450-like spectrum, we can include this case in the comparison study, with minimal contamination by the P420 form.
Heme Domain with L-Arginine but Not BH 4 -The fast phase required at least two exponentials with rates differing by about a factor of 3.
Heme Domain with BH 4 but Not L-Arginine-The fast phase required two exponentials differing by a factor of 4 -5, slightly more than in the cases above.
Heme Domain with both L-Arginine and BH 4 -Adding both reagents had a striking effect, not observed with either alone. Geminate recombination was greatly reduced so that the slow phase dominated bimolecular association. The slow phase could be well fit using two exponentials differing in rate by a factor of 3, but instead of the amplitudes being approximately equal, the slower contributed 90% to the phase.
Holoenzyme with neither BH 4 nor L-Arginine-The bimolecular combination could be described very well using three exponentials, of which two could be grouped into a fast phase. The fastest is 5-6 times faster than the middle rate, which is about 15 times the slowest. The fastest also provides most of the amplitude, being about 10 times larger than the middle rate, which is somewhat larger than the slowest. The primary data were obtained using a preparation that had been prepared entirely in the absence of Arg or BH 4 . This was possible by expressing nNOS holoenzyme in E. coli in which BH 4 is not biosynthesized (18). Similar results were obtained for a sample originally prepared with BH 4 and Arg but then dialyzed 48 h against buffer lacking both. That, however, is rather harsh treatment, required because BH 4 is strongly bound.
Holoenzyme with L-Arginine but Not BH 4 -The slow phase was well fit using two exponentials differing by a factor of only 2, with the slower carrying the majority of the amplitude.
Holoenzyme with BH 4 but Not L-Arginine-The slow phase presence of tetrahydrobiopterin and arginine and 20% atm CO, 80% atm argon. a, the best single exponential fit (⌬A ϭ 0.0539 e Ϫ5.11t ) might be thought to fit several half-lives, but it clearly deviates at long times and, although difficult to see in the semilog plot, also at short times. b, a triple exponential fit of the major slow phase plus a small single exponential for a fast phase (⌬A(t) ϭ 0.0566 (0.0177 e Ϫ763t ϩ 0.0864 e Ϫ14.8t ϩ 0.5883 e Ϫ5.75t ϩ 0.3074 e Ϫ3.74t )) appears coincident with the data. The weighted mean rate of the dominant, slower phase is 5.17, only 1% different from the best single exponential. Bottom panel, a linear display of the residuals between fits and data shows clearly that the single exponential (a) fails miserably (reduced 2 ϭ 18 for root mean square noise ␦⌬A ϭ 9.1 ϫ 10 Ϫ5 ), whereas the triple exponential (b) fits well (reduced 2 ϭ 1.1). required two exponentials differing in rate by a factor of about 3, with the slower carrying most (about 70%) of the weight.
Holoenzyme with L-Arginine and BH 4 -Protein with 0.25 mM BH 4 and either 0.5 or 10 mM Arg showed no detectable geminate recombination (Ͻ3%), and the bimolecular association was almost exclusively (Ͼ95%) a slow phase. That phase, however, was not at all a single exponential (see Fig. 1). It required at least two and often three exponentials, distributed in rate over a range of 5-10. The small, fast phase was at least as fast as the fast phase observed in the absence of Arg. Although a small fast phase at the beginning of a decay can be quite apparent, it is difficult to characterize since it is affected by the random noise in the much larger slow phase.
Dissociation-Plots of association rates versus [CO] were linear and could be extrapolated to the y intercept, but the plots were not very useful for estimating dissociation constants. Only upper limits could be assigned, typically Ͻ20 s Ϫ1 for the fast phase and Ͻ2 s Ϫ1 for the slow phase. Consequently, we resorted to stopped-flow methods carried out at 20°C. With one exception, all stopped-flow measurements were well fit by single exponentials over the time scale that could be monitored. A typical trace is shown in Fig. 2. The absorbance change of 0.07 is what was expected for the change from reactants to products, so that any additional component cannot be very large. Results are included in Table I. The holoenzyme lacking Arg showed about 15-20% faster phase with rate constant 15 Ϯ 2. Perhaps surprisingly, the heme domain did not show any fast process under any conditions.

DISCUSSION
Heterogeneous Kinetics-Bimolecular association required three or four exponentials to fit, but the exponentials tend to group into two distinct time regimes. Fig. 1 displays a large amplitude slow phase requiring three exponentials varying in rate by only a factor of 4 between the extreme values together, with a minority fast phase that is 150 times faster. We attribute the different rates to different protein conformations, each reacting with its own rate constant. This is based on two considerations. 1) Protein expression should produce proteins with a common sequence identity. 2) Lack of spectral evidence for intermediates appears to rule out sequential mechanisms. (Additional spectral features can be detected under certain circumstances but not the conditions discussed here.) We postulate a quasi-continuous distribution of conformations, which is bimodal, clustering about the two most common forms. The effect of cofactors, substrates, and other reagents is to bias the probability distribution in ways that can be described approximately (but not completely) as shifting the protein between fast reacting forms and slow reacting forms. The one previous report of combination kinetics in NOS distinguished the two major phases, but did not recognize additional complexity (16).
Tentatively assuming that kinetic heterogeneity is due to multiple conformational states, the question remains as to how much is intrinsic and how much may be an artifact of in vitro preparations. The much-studied myoglobins exhibit no such heterogeneity. The hemoglobins do have distinct R and T quaternary conformations with different kinetics (29,30). There are also ␣ and ␤ chain differences in Hb (30 -32), but those reflect different primary sequences. On the other hand, NOS is compared not with Mb or Hb but with P450, and a similar multiplicity of kinetic phases is observed in human cytochrome P450 3A4 (33). Three phases were recognized over the range 10 5 -10 7 M Ϫ1 s Ϫ1 , and it was suggested there may be further sub-phases. Heterogeneity may be common for enzymes, even in vivo, although rare for the simpler oxygen transport proteins. On the other hand, the globins purified by simpler methods are less likely to introduce artifacts. Bimolecular Association-Dissociation-Despite the complexity of association kinetics, Table I shows that the behavior can be modeled approximately using a fast phase with k a,f ϳ 2-3 ϫ 10 6 M Ϫ1 s Ϫ1 and a slow phase with k a,f ϳ 2-4 ϫ 10 4 M Ϫ1 s Ϫ1 in various ratios. Table I also shows that such a two-state model is an approximation. When the fast dominates, it becomes faster; when the slow state dominates, it is at its slowest. The full story apparently involves a complex redistribution among multiple states.
For the holoenzyme in the presence of both BH 4 and Arg, the slow phase dominates but never quite reaches 100%. The opposite extreme occurs in the heme domain in the absence of Symbols are data points, and the continuous line is the best fit to a single exponential, with parameters as in Table I. Note that this is total absorbance, not absorbance change, so it plateaus at a nonzero value at long times.  BH 4 and Arg. All other cases lie in between. The holoenzyme lacking both BH 4 and Arg is similar to the heme domain, but adding either reagent is sufficient to convert it largely to the slow reacting form even though both are needed for the full effect. For the heme domain, either reagent alone has a small effect, but both together have a dramatic effect. The limiting cases are reproducible. For the intermediate cases, different preparations vary slightly in the percentage of each phase, again suggesting heterogeneity. Thus, it is not surprising that there is some slight difference between our numbers and those reported previously, which we believe pertain to two intermediate permutations (16). It is known that both the BH 4 and the Arg binding sites are located in the heme domain (19,34). Our results confirm that model. It has also been reported that addition of Arg to the native enzyme is accompanied by changes in the UV-visible spectrum that are consistent with conversion of a low-spin, six-coordinate iron to a high-spin, five-coordinate iron (35). This implies that the Arg does not bind directly to the iron but perturbs or displaces some as yet unidentified ligand (or protein side chain) coordinated to the iron (35,36). Steric interference by Arg of the ligand binding site was suggested (37). It is not surprising, therefore, that in the presence of both BH 4 and Arg the CO association rate is reduced by a factor of 100 below the fast phase. Lack of any profound effect by Arg alone in the heme domain suggests that Arg can be incorporated into the heme domain, but its spatial orientation is different and directed away from the ligand binding site. In the holoenzyme, Arg alone is only slightly different from Arg ϩ BH 4 , showing that the options for spatial orientation are severely limited. These conformational constraints may also correlate with quaternary changes, that is, the formation of homodimer, which is facilitated by BH 4 and may be more favored in the holoenzyme.
The dissociation rate for CO from either NOS-CO itself or the heme domain subunit is close to 0.3 s Ϫ1 in the presence of Arg and 0.16 s Ϫ1 in its absence. Either number is high compared with the oxygen-binding hemeproteins but substantially less than in P450. Consider: five-coordinate carboxyheme (a protein-free model compound) in cetyltrimethylammonium bromide detergent has a CO dissociation rate constant of 400 s Ϫ1 . Coordination of proximal imidazole trans to CO in either a protein or a model system reduces the CO dissociation constant to between 0.1 and 0.01 s Ϫ1 , a trans effect of about 10 4 . In six-coordinate bacterial cytochrome P450 cam , in the absence of substrate, the CO dissociation rate constant is 6.3 s Ϫ1 (38) upon extrapolation to room temperature, which represents a much smaller trans effect of only 64. In the presence of substrate, CO dissociation from carboxy-P450 decreases to 1.4 s Ϫ1 , but either with or without substrate the proximal cysteine thiolate in P450 exerts a much smaller positive trans effect than does imidazole. The dissociation rate for NOS-CO, which also has a trans-cysteine, is intermediate: lower than in bacterial cytochrome P450 cam but much greater than when histidine (imidazole) is trans. The smaller trans effect of thiolate is probably due to its ionic nature; ions generally have lower affinity for ferroheme. In fact, for Mb mutants in which the proximal histidine was replaced by cysteine, carboxy derivative formation was accompanied by rupture of the proximal bond, a related consequence of a small (or negative) trans effect (39,40).
We noted above the evidence for some steric crowding at the ligand binding site. As mentioned in the introduction, such crowding seems always to decrease rate constants for both association and dissociation of CO. Consequently, we can infer from kinetic measurements that the pocket in NOS is probably more constrained than in P450 cam . The is no significant sequence homology between NOS and P450 and, according to computer modeling, no structural similarity. Consequently, differences in their heme pockets should be expected.
From the rate constants, one may calculate association equilibrium constants for the fast and slow phases of ϳ10 7 and ϳ10 5 . (Conditions that generate some fraction of faster dissociation will reduce those numbers even further.) This affinity is 300 times less than in human Mb. It is unlikely that CO binding will be significant in vivo, where low CO concentrations are expected, unless there is much local enhancement and an absence of nearby traps, such as Hb or Mb. Ingi et al. (15) predicted, on the basis of different evidence, that a 1 mM concentration of CO would be needed to inhibit NOS activity. The equilibrium constant reported here suggests that their value is too high, but we concur with the conclusion that CO inhibition is unlikely. In fact, a recent study suggests just the opposite; far from being an inhibitor, small amounts of CO, 1000 ppm, may increase NO production (41). If that observation is confirmed, it will have important implications for the mechanism of NOS action. One possibility is that small amounts of CO may stabilize the reduced, ferro intermediate of iron in NOS in the reaction scheme proposed by Marletta (9).
Geminate Recombination-After bond dissociation by flash photolysis, CO may remain in the heme pocket for some time. The simplest model for such a geminate pair temporarily trapped in a protein pocket is as follows.
In this model, admittedly oversimplified (24), the rate of disappearance of the caged pair intermediate is k gem ϭ k 2 ϩ k 3 , whereas the geminate yield (the fraction recombining directly from the geminate pair) is Y ϭ k 2 /(k 2 ϩ k 3 ). For NOS, Table I shows that k gem is about 1.5 ϫ 10 7 s Ϫ1 and varies by less than a factor of 2 among our eight conditions, while Y changes by a factor of 10, from Ͻ5% to almost 50%. This implies that k 3 remains approximately constant, while k 2 , the rate of bond formation proper, changes significantly. Changes in k 2 may be due to changes either in the distal environment or in strain on the proximal ligand, both of which are influenced by protein conformation. Hence, geminate recombination provides direct evidence that in addition to other consequences, when the two domains are combined or BH 4 or Arg is added, one effect is a substantial perturbation at the heme. Resonance Raman spectroscopy also probes the heme and its environment. Binding Arg causes a frequency shift of the Fe-NO stretch (37). Furthermore, that mode is broad, suggesting multiple conformations (7). All this echoes what is inferred from bimolecular combination rates.
The rates k gem are similar in all hemeproteins. In P450, two components were detected, one 2-3 times faster than we observe in NOS, the other 2-3 times smaller (42). For three different hemoglobins, k gem varied from approximately equal to about half the rate measured in NOS (43). Part of that variation is due to the fact that k gem ϭ k 2 ϩ k 3 is faster when k 2 is greater. Escape from the protein k 3 is surprisingly constant for different hemeproteins.
The geminate yield Y varies greatly among hemeproteins. For CO, Y is no more than about 5% in Mb, 20 -50% in various Hb R , depending on pH, and somewhat less in Hb under conditions that favor T-like structures (42). In P450 cam , Y is reported to be as large as 90% in the absence of substrate, reduced to only 2% with camphor bound (41). NOS is like P450 in having very small Y with substrate bound. In the absence of substrate, P450 is like NOS without BH 4 or like the heme domain and unlike the competent holoenzyme lacking only substrate. This echoes the conclusion from measurements by magnetic circular dichroism that "the active sites of NOS and P450 may share some common structural features, but significant distinctions exist" (44).
Geminate yield is related to bimolecular association. According to equation 3, where k 4 is a bimolecular rate constant that describes entry into the protein and involves both diffusion to the periphery of the protein and a steric factor describing the fraction of encounters in which the ligand penetrates the protein. After entry, only a fraction Y of the ligands bind to the iron; the rest escape and must try again. From Table I, a correlation is evident between Y and the fraction of the bimolecular fast phase. The fast binding forms have substantial, ϳ50%, geminate yield, whereas slow binding conformations show very little geminate recombination. The 100-fold ratio in association rates between fast and slow phases can be accounted for largely by their different Y.
Geminate recombination strongly suggests that conformational effects on ligand binding are mediated not by changes in ligand entry into the protein but rather by interactions related to binding at the iron, involving strain from the proximal cysteine or polar or steric effects from the distal side.