Entrapment of Carbon Dioxide in the Active Site of Carbonic Anhydrase II*♦

The visualization at near atomic resolution of transient substrates in the active site of enzymes is fundamental to fully understanding their mechanism of action. Here we show the application of using CO2-pressurized, cryo-cooled crystals to capture the first step of CO2 hydration catalyzed by the zinc-metalloenzyme human carbonic anhydrase II, the binding of substrate CO2, for both the holo and the apo (without zinc) enzyme to 1.1Å resolution. Until now, the feasibility of such a study was thought to be technically too challenging because of the low solubility of CO2 and the fast turnover to bicarbonate by the enzyme (Liang, J. Y., and Lipscomb, W. N. (1990) Proc. Natl. Acad. Sci. U. S. A. 87, 3675–3679). These structures provide insight into the long hypothesized binding of CO2 in a hydrophobic pocket at the active site and demonstrate that the zinc does not play a critical role in the binding or orientation of CO2. This method may also have a much broader implication for the study of other enzymes for which CO2 is a substrate or product and for the capturing of transient substrates and revealing hydrophobic pockets in proteins.


REACTION 1
The second stage is the transfer of a proton from the zincbound water to bulk solvent to regenerate the zinc-bound hydroxide (Reaction 2). Here B is a proton acceptor in solution or a residue of the enzyme itself.
For hCAII (␣ class CA), this reaction is facilitated by a solvent molecule with a pK a near 7 that is directly coordinated to the zinc (6). This centrally located zinc exhibits a tetrahedral configuration with three histidines (His-94, His-96, and His-119) and either a water or a hydroxide molecule. The active site cavity can be loosely described as being conical in shape having a 15 Å diameter entrance that tapers into the center of the enzyme. The cavity is partitioned into two very different environments. On one side of the zinc, deep within the active site, lies a cluster of hydrophobic amino acids (namely Val-121, Val-143, Leu-198, Thr-199-CH 3 , Val-207, and Trp-209), whereas on the other side of the zinc, leading out of the active site to the bulk solvent, the surface is lined with hydrophilic amino acids (namely Tyr-7, Asn-62, His-64, Asn-67, Thr-199-O␥ 1 , and Thr-200-O␥ 1 ) (Fig. 1a).
Previously, molecular dynamics studies have implied that the hydrophobic region of the active site sequesters the CO 2 substrate and orients the carbon atom in readiness for nucleophilic attack by the zinc-bound hydroxide (Reaction 1) (1,8). Additionally, crystallographic studies have identified an ordered water molecule, positioned near the hydrophobic pocket, termed "deep water," W DW , that is stabilized by the amide nitrogen of Thr-199 and the zinc-bound hydroxide (Fig. 2a). It has been proposed that this water is likely displaced upon the infusion of CO 2 into the binding pocket (8,9). The hydrophilic wall of the active site has been shown, by x-ray crystallography, to create a well ordered hydrogen-bonded solvent network. It is hypothesized that this network is required to permit the transfer of a proton from the zinc-bound water to the bulk solvent via the experimentally identified proton shuttling residue His-64 (Reaction 2) (10 -13). Taken together, these two very different active site environments permit the sustained and rapid catalytic cycling of CO 2 to bicarbonate.

EXPERIMENTAL PROCEDURES
Protein Purification and Crystallization-hCAII was expressed in a recombinant strain of Eshcherichia coli (BL21(DE3)pLysS) containing a plasmid encoding the hCAII gene (14). Purification was carried out via affinity chromatography as described previously (15). Briefly, cells were enzymatically lysed with hen egg white lysozyme, and the lysate was placed onto an agarose resin coupled with p-(aminomethyl)-benzene-sulfonamide, an hCAII inhibitor. The protein was eluted with 0.4 M sodium azide, 100 mM Tris-HCl, pH 7.0, and the azide was removed by extensive buffer exchanging against 10 mM Tris-HCl, pH 8.0.
Preparation of Apoenzyme-The zinc from the holoenzyme was removed by incubation at 20°C in the chelation buffer (100 mM pyridine 2,6 dicarboxylic acid, 25 mM MOPS; pH 7.0) for 8 h. The enzyme was then buffer-exchanged against 50 mM Tris-HCl, pH 7.8, to remove the chelating agent (16). The complete loss of enzyme activity was verified through kinetic studies. The enzyme activity was revived through the addition of 1 mM ZnCl 2 , attributing the loss of activity to the absence of zinc rather than to the denaturation of the enzyme.
Crystallization and X-ray Data Collection-Crystals of holoand apohCAII were obtained using the hanging drop vapor diffusion method (17). 10-l drops of equal amounts of protein and precipitant were equilibrated against precipitant solution (1.3 M sodium citrate, 100 mM Tris-HCl, pH 7.8) by vapor diffusion at room temperature (ϳ20°C) (13). Crystals grew to ϳ0.2 ϫ 0.2 ϫ 0.5 mm in size after ϳ5 days. CO 2 Binding-To capture CO 2 in the active site of hCAII, it was essential to cryo-cool the crystals under CO 2 pressure. This was achieved using the high pressure cryo-cooling method that was originally developed for crystal cryoprotection (18). The crystals were first soaked in a cryo-solution containing 20% glycerol in precipitant solution. The crystals were then coated with mineral oil to prevent crystal dehydration and loaded into the bottom of high pressure tubes. In the pressure tubes, the crystals were pressurized with CO 2 gas at 15 atm at room temperature. 25 min later, without releasing CO 2 gas, the crystals were slowly frozen over 2 min by dipping the sealed end of the pressure tubes into liquid nitrogen. During the cooling process, it was noticed that the CO 2 gas pressure gradually dropped from 15 atm to below 1 atm due to CO 2 solidification.
X-ray Diffraction Data Collection-Diffraction data were collected at CHESS, beamline A1, at a wavelength of 0.9772 Å. Data were collected using the oscillation method in intervals of 1°step on an ADSC Quantum 210 CCD detector (Area Detector Systems Corp.), with a crystal to detector distance of 65 mm. A total of 624 and 360 images were collected for the holo and apo data, respectively. Indexing, integration, and scaling were performed using HKL2000 (19). The crystals of the CO 2 -bound holo-and apohCAII diffracted to 1.1 Å resolution and were processed to a completeness of 99.9% and an R sym of 8.8% and a  completeness of 93.1% and an R sym of 8.0%, respectively. Complete processing statistics are given in Table 1.
Structure Solution and Model Refinement-The structures of CO 2 -bound holo-and apohCAII were solved in a similar manner using the program SHELXL (20). Prior to refinement, a random 5% of the data were flagged for R free analysis (21). The previously determined 1.54 Å resolution crystal structure of holohCAII (PDB ID: 2CBA) (16) was stripped of all waters, the zinc, and any alternate conformers and used as the initial phasing model in a round of least squares, rigid-body refinement to 2.5 Å resolution to an R factor /R free of 31.3/33.2% for holo and 28.0/28.6% for apo enzyme. The data were then extended to 1.5 Å resolution, and the model was refined using conjugant gradient least squares refinement. After 20 cycles, the model and related -weighted electron density maps were read into the molecular graphics program Coot (22). Improperly built side chains and the zinc (in the holo structure only) were placed into their respective density, and the model was run through another round of conjugant gradient least squares refinement. Waters with positive density in the -weighted difference map were kept until all waters with reasonable density were built. The data sets were then extended to 1.1 Å resolution, and the final waters were built. Disorder was then modeled into the density by modeling all visible alternate conformations for both amino acid side chains and waters. Riding hydrogens were then placed on all residues except the imidazole nitrogens of the histidines. The weighting factor was then changed to 0.2 for one round followed by the use of all data for the final round. The final R final /R free for holo was 10.9/12.9%, and the final R final /R free for apo was 10.4/13.9%. Complete refinement statistics can be found in Table 1. The model geometries and statistics were analyzed by PROCHECK (23).
Additional Structural Features-There were no major changes in any amino acid atomic positions other than that mentioned for the phenyl ring of Phe-226 (due to CO 2 binding) (Fig. 2). There were a large number of dual amino acid conformations in these structures; however, this can be attributed to the near atomic resolution of the structure and not necessarily to the pressurized environment that the crystals experienced (the alternate conformations were Ile-22, Leu-47, Ser-50, Asp-52, His-64, Ser-152, Ser-217, and Val-223 for the holo, and Ile-22, His-64, Gln-103, Asp-162, Lys-172, Glu-214, Ser-217, and Val-223 for the apo structure). Many previous structural studies of hCAII have shown that His-64 occupies two conformations, termed "in" and "out." Generally it has been seen that the "in" conformation is favored, although there may be pH effects on the ratio of "in" to "out" (10,13). In both the holoCO 2 -bound and the apoCO 2 -bound structures, His-64 was seen to have a preference for the "out" conformation. Due to the presence of glycerol as a cryo-protectant, a well ordered glycerol molecule was observed at the mouth of the active site cavity. The oxygens of the glycerol overlap with the positions of the waters of the proton wire, thus indicating that the glycerol might be binding advantageously due to the "out" prevalence of His-64. Interestingly this glycerol molecule is seen only in the holo enzyme. A second molecule of glycerol is seen in the same posi-tion in both the holo and the apo structures on the surface of the protein located near residues 243-245 (Fig. 3).

RESULTS
Here we describe for the first time, to our knowledge, the experimental capture of CO 2 in the hydrophobic cavity of hCAII (Fig. 1). The holo-and apohCAII CO 2 -bound structures were refined to 1.1 Å resolution with final R factors of 10.90 and 10.35, respectively (Table 1) Table 2) and were refined, assuming full occupancy, and had final  a Values in parentheses are for the highest resolution shell. b R sym ϭ (⌺͉I Ϫ ͗I͉͘/⌺͗I͘) ϫ 100. c R factor ϭ (⌺͉F o ͉ Ϫ ͉F c ͉/⌺͉F obs ͉) ϫ 100. d R free is calculated the same as R factor , except it uses 5% of reflection data omitted from refinement. e The first number given is the average B-factor for the active site-bound CO 2 , the second is for the CO 2 bound near Phe-226. f The root mean square deviation of C ␣ positions as compared with the 1.1 Å resolution crystal structure of unbound holo hCAII (PDB ID: 2ILI) (13).

Comparison of holo-and apoCO 2 -bound hCAII-
The holohCAII structure shows, as modeled previously (1,8), that one of the oxygens of the CO 2 , O(2), interacts (3.5 Å) with the amide of Thr-199, and in doing so, causes a displacement of the water molecule W DW , whereas the O(1) is positioned between the zinc and Val-121. This arrangement places both CO 2 oxygens nearly equidistant from the oxygen of the zinc-bound solvent with distances of 3.0 and 3.1 Å, respectively, putting the carbon 2.8 Å from the zinc-bound solvent. This results in a side-on orientation of CO 2 with the zinc-bound solvent, at a distance that is well suited for the nucleophilic attack to take place on the carbon by the lone pair electrons of the oxygen in the zinc-bound hydroxide ( Table 2, Fig. 1b). Additionally, a new (or displaced) water molecule, W I , not previously observed in other holohCAII structures is seen to occupy a space between Thr-200-O␥ 1 and the O(2) oxygen of CO 2 (Figs. 1, b and c, and 4).
Interestingly, the CO 2 molecule in the apo enzyme shares a very similar geometry despite the absence of the zinc (Fig. 1c). A water is positioned near what would have been the zinc-bound solvent in the holohCAII, although it is ϳ0.6 Å closer to the histidine ligands. Both the CO 2 oxygens are positioned ϳ3.1 Å from this water molecule. The small shift of this water allows the CO 2 to pivot about the O(1) atom, shifting O(2) into a slightly tighter interaction with the amide nitrogen of Thr-199 (3.15 Å for apo when compared with 3.5 Å for holohCAII) (Fig.  1, b and c).
Secondary CO 2 Binding Site-In addition to the catalytic binding site, another CO 2 binding site (not believed to be involved in catalysis) was observed in a second hydrophobic pocket, ϳ11 Å away from the active site (Fig. 4a). In this pocket, the CO 2 displaces the phenyl ring of Phe-226, inducing a 30°tilt with respect to the plane of the ring (Fig. 4b).
Furthermore, this pocket lies next to Trp-97, a residue that biophysical analyses have shown acts as an initiator of proper folding of hCAII (24).

DISCUSSION
Catalysis of the hydration of CO 2 by hCAII at 10 8 M Ϫ1 s Ϫ1 approaches the diffusion-controlled limit and follows Michaelis kinetics with a maximal turnover near 10 6 s Ϫ1 and K m near 10 mM. The diffused CO 2 is expected to be loosely bound since it has no dipole moment, and the fact that CO 2 is more soluble in organic solvents is consistent with the observed hydrophobic binding site, which suggests that solvation is a significant contributor to binding. The dissociation constant of CO 2 at the active site of hCAII was estimated by infrared spectroscopy to be 100 mM (25), a value consistent with the kinetic properties of the catalysis. The constant of Henry's Law for the solubility of CO 2 in water under the conditions of these experiments (15 atm CO 2 ) indicates a maximal concentration of CO 2 near 0.45 M (26). These considerations suggest a nearly complete occupancy of CO 2 at the active site. With an energy barrier for catalysis near 10 kcal/mol, an insignificant reaction rate is expected at liquid nitrogen temperature. However, in our procedure, CO 2 was introduced to the crystal at room temperature, a procedure that surely decreased the effective pH of the crystal and surrounding solvent and promoted the forma-  (32). The binding modes of both the CO 2 substrate and the HCO 3 Ϫ product are similar, with the substrate favoring the hydrophobic side (green) and product favoring the hydrophilic side (orange). Note that the deep water (W DW , gray sphere) is displaced and a new water occupies the area between the side chain of Thr-200 and CO 2 (W I , cyan) upon CO 2 binding. The CO 2 is orientated so that the carbon is primed for the nucleophilic attack by the zinc-bound hydroxide (orange sphere). b, a superposition of the V143Y variant of hCAII (9,27). Note that the side chain of Tyr-143 (white) acts as a steric block to the CO 2 binding site. The figure was created using PyMOL.

TABLE 2 Distance (Å) geometry of CO 2 for holo-and apohCAII
The bond distances from the CO 2 molecule are given within a radial shell of 3.9 Å. The atoms of the CO 2 molecule are in bold text. The numbering of the CO 2 oxygens are in accordance to the text and figures.

Holo
Apo tion of the zinc-bound water at the active site. The observation of CO 2 at the active site is consistent with a zinc-bound water in our structures since this form would predominate at acidic pH and is unreactive toward CO 2 . The zinc-bound hydroxide form of the enzyme reacts with CO 2 ; however, the observation of no bound bicarbonate suggests that this form of the enzyme was not prominent. That the binding of CO 2 does not involve first-shell coordination to the zinc is consistent with previous spectroscopic studies (27,28). Moreover, the observed CO 2 binding site confirms previous kinetic and structural analyses of mutations made at Val-143 (9,29). From these studies, it was shown that bulkier substitutions led to significant decreases in activity. For example, a V143Y mutant had less than 0.02% the activity of the wild-type enzyme. A structural least squares superposition of V143Y with that of CO 2 -bound wild-type enzyme (C ␣ r.m.s.d. ϭ 0.26 Å) clearly shows that the tyrosine would directly interfere with CO 2 binding, thus blocking the substrate from binding in an orientation that is optimal for nucleophilic attack by the zinc-bound hydroxide (Fig. 2b).
The binding interactions of CO 2 determined here are very similar to those of the isoelectronic NCO Ϫ ion that is a potent inhibitor of hCAII. Crystallographic analysis of the complex of NCO Ϫ and hCAII shows that cyanate is bound on the hydrophobic surface of the active-site cavity and does not displace the zinc-bound water (30). Moreover, like bound CO 2 , the cyanate ion displaces the deep water and forms a hydrogen bond with the backbone amide of Thr-199; the tetrahedral coordination about the zinc is not disturbed in the complex. The distance between the carbon of bound cyanate and the oxygen of zinc-bound water is 2.4 Å, again similar to the corresponding distance for bound CO 2 . This comparison of the binding of CO 2 and the inhibitor NCO Ϫ supports our hypothesis that the observed binding site of CO 2 (Fig. 1) is a site of productive substrate binding. It is interesting to note that in studies of Co(II)-substituted carbonic anhydrase, cyanate appears to bind directly to the zinc (31).
The observed binding site of CO 2 aids in the interpretation of the next step of catalysis, the formation and subsequent release of the product, bicarbonate. Following the nucleophilic attack, two mechanisms have been proposed for the subsequent release of the HCO 3 Ϫ ion based on the theoretical free energy calculations of CO 2 /HCO 3 Ϫ intercoversion. The Lipscomb mechanism (32) propounds a monodentate Zn-HCO 3 Ϫ intermediate wherein a proton rapidly migrates from the original Zn-OH Ϫ to one of the other two oxygen atoms of the HCO 3 Ϫ ion. The zinc in this mechanism is held in a tetrahedral coordination (Fig. 5a). In contrast, the Lindskog mechanism (33) proposes a bidentate Zn-HCO 3 Ϫ intermediate that requires one of the two oxygen atoms of the original CO 2 molecule to coordinate directly with zinc, resulting in a penta-coordinated metal ion held in a trigonal bipyramidal geometry (Fig. 5b).
In addition, physical evidence is provided by previous structural work. Xue et al. (34) made a T200H mutant that displayed a higher affinity for HCO 3 Ϫ ion than the wild-type enzyme, and as a consequence, were able to capture bicarbonate in the active site and study the complex using x-ray crystallography (Figs. 2a  and 4). Least squares superposition of this structure on the wild-type hCAII CO 2 -bound structure (C ␣ r.m.s.d. ϭ 0.21 Å) shows that the CO 2 substrate molecule exists in the same plane as the Zn-HCO 3 Ϫ product (Fig. 2a). From a strictly structural perspective, the pseudo-bidentate nature of the captured Zn-HCO 3 Ϫ complex seems to favor the Lindskog hypothesis. Nevertheless, in both mechanisms, the release of the HCO 3 Ϫ product from the Zn-HCO 3 Ϫ intermediate is associated with the binding of a water molecule to the metal. The appearance of the previously unseen water, W I , in close proximity to the zinc was observed in the CO 2 complex structures of both the holo and the apo enzymes. This water could be either the displaced W DW water, seen prior to CO 2 binding (Fig. 2a), or a new ordered water, possibly arising due to a change in the local electrostatic environment. The position of this water with respect to the zinc leads us to suggest that this water may be the best candidate in the aforementioned water-associated displacement of product bicarbonate (Figs. 2 and 6).
In summary, this method of using pressurized gases, such as CO 2 , may be applicable to other enzymes to capture weakly bound substrates and/or identify hydrophobic pockets in enzymes that might play important roles in substrate binding or protein folding. The identification of a CO 2 binding site at the active site of hCAII, in combination with the previously known bicarbonate binding site and the location of ordered water molecules, W DW and W I , should now guide molecular dynamics studies to examine the catalytic event itself and decipher whether the Lipscomb (32) or Lindskog (33) mechanism is more feasible.