Active-site solvent replenishment observed during human carbonic anhydrase II catalysis

Ultrahigh-resolution crystallographic structures of human carbonic anhydrase II (hCA II) cryocooled under CO2 pressures of 7.0 and 2.5 atm are presented. The structures reveal new intermediate solvent states of hCA II that provide crystallographic snapshots during restoration of the proton-transfer water network in the active site.


Introduction
The reversible interconversion of carbon dioxide (CO 2 ) and water to bicarbonate and a proton (H + ) occurs at a rate that is limited by the diffusion of substrates in the presence of carbonic anhydrases (CAs) as the catalyst (Davenport, 1984;Christianson & Fierke, 1996;Chegwidden & Carter, 2000;Frost & McKenna, 2013;Supuran & De Simone, 2015). CAs are metalloenzymes that mostly contain zinc, although some are found with iron or cadmium. There are six distinct families of CA (, , , , and the family, which was recently subdivided from the family) that are found throughout the animal, plant and bacterial kingdoms. In animals, CAs primarily function to maintain acid-base balance in the blood and other tissues, and to help the transport of CO 2 out of tissues. In particular, mammalian CAs belong to the family and are expressed as many different isozymes (Hewett-Emmett & Tashian, 1996). For instance, 14 forms of human -CA can be divided into four cytosolic (I, II, III and VII), two mitochondrial (V A and V B ), one secreted (VI) and four membrane-bound (IV, IX, XII and XIV). The remaining three isoforms lack catalytic activity and are referred to as carbonic anhydrase-related proteins (CARPs). Among these isozymes, human CA II (hCA II) is expressed in most cell types, with involvement in many physiological processes (Krishnamurthy et al., 2008;Frost & McKenna, 2013;Supuran & De Simone, 2015).
The first crystal structure of hCA II, known at the time as hCA C, was determined by Liljas and coworkers in 1972 and was further refined in 1988 (Liljas et al., 1972;Eriksson et al., 1988). These studies laid the foundation for understanding the mechanism of CA activity. In hCA II, the active-site zinc is located within an $15 Å deep cleft and is tetrahedrally coordinated by three histidine residues (His94, His96 and His119) and an OH À ion (Fig. 1). Furthermore, the active-site cavity subdivides into two distinct sides, formed by hydrophilic residues (e.g. Tyr7, Asn62, His64, Asn67, Thr199 and Thr200) and hydrophobic residues (e.g. Val121, Val143, Leu198, Thr199-CH 3 , Val207 and Trp209). The 'hydrophobic' side sequesters and positions the CO 2 for nucleophilic attack by OH À (Liang & Lipscomb, 1990).
Mechanistically, the conversion of CO 2 to bicarbonate in the hydration direction takes place via the nucleophilic attack of CO 2 by the zinc-bound hydroxide (OH À ) (1). The subsequently generated bicarbonate is then displaced by a water molecule (W Zn ; Silverman & Lindskog, 1988) (1): The next step of catalysis is the transfer of a proton from W Zn to the bulk solvent, regenerating the zinc-bound OH À (2). Here, the tetrahedral coordination of W Zn to zinc causes polarization of the hydrogen-oxygen bond, making the O atom slightly more positive and thereby weakening the bond (Christianson & Fierke, 1996). The general base (B) for the proton transfer is likely to be mediated by ordered waters and His64 within the enzyme, where the hydrophilic side of the active site forms the hydrogen-bonded water network (W1, W2, W2 0 , W3a and W3b) that connects W Zn to His64. This hydrogen-bonded network is believed to act as a proton wire that reduces the work required to transfer a proton from W Zn to the bulk solvent for the regeneration of the zinc-bound OH À (2) Steiner et al., 1975;Cui & Karplus, 2003;Zheng et al., 2008;Fisher, Maupin et al., 2007;Silverman et al., 1979). Neutron studies have been utilized to observe the protonation states and orientation of water molecules in proteins (Langan et al., 2008). Such experiments have determined that the water network in hCA II is pH-dependent, with an unbranched wire between W Zn and His64 at physiological pH that is broken at high pH owing to a rearrangement of the hydrogen bonds of W1 (Budayova-Spano et al., 2006;Fisher et al., 2011). The side chain of His64 is oriented in two conformations, termed the 'in' (pointing towards the active site) and 'out' (pointing away from the active site) positions, that are suggested to facilitate the proton-shuttling process (Tu et al., 1989;Fisher et al., 2005;Nair & Christianson, 1991;Lindskog, 1997;Avvaru et al., 2010). Neutron structures have revealed that His64 is uncharged when occupying the 'in' position, priming the residue for the acceptance of a proton transferred from W Zn , regenerating the enzyme during catalysis (Fisher et al., 2010). Moreover, the fact that binding of small molecules (activators) in the vicinity of His64 changes the catalytic rate by altering the proton-transfer step adds to the hypothesis that proton shuttling occurs via His64 (Supuran, 2008;Temperini et al., 2005Temperini et al., , 2006aBriganti et al., 1997Briganti et al., , 1998  Surface rendition of hCA II depicted using the ultrahigh-resolution (0.9 Å ) crystal structure of 7.0 atm CO 2 hCA II. The hydrophobic and hydrophilic regions are coloured green and yellow, respectively. (a) The overall hCA II shows the CO 2 -binding active site that is located at a depth of 15 Å from the surface and is open to the outside bulk solvent through an entrance (the entrance diameter is $7-10 Å and it is referred to as the 'entrance conduit' in this study). The substrate and product (CO 2 / bicarbonate) of hCA II as well as water molecules can pass in and out through this open entrance conduit. (b) A closer view of the active site (Zn, CO 2 binding site and His64) is shown through the entrance conduit with a 2F o À F c map (in blue) contoured at 2.5 (for His64) and 5.0 (others). The isolated electron density of the C atom of CO 2 is clearly visible in this ultrahigh-resolution structure. The protein surface around His64 is removed and the proton-transfer water network (W I /W2/W3a/ W3b) is not shown for clarity. The proton transfer during the catalytic activity is thought to occur via His64 through the proton wire rather than through the open entrance conduit.
Previously, the capture of CO 2 in the active site of hCA II was achieved by cryocooling hCA II crystals under a 15 atm (1 atm = 101.325 kPa) CO 2 pressure (Kim et al., 2005;Domsic et al., 2008). More recently, attempts have been made to track the intermediate changes during gradual CO 2 release to the CO 2 -free state by incubating 15 atm CO 2 -pressurized hCA II crystals at room temperature (RT) for different time intervals (50 s, 3 min, 10 min, 25 min and 1 h) to decrease the internal CO 2 pressure (Kim et al., 2016). The resulting so-called intermediate snapshots revealed that two deep waters (W DW and W 0 DW ) immediately replace the vacated space as CO 2 leaves the active site. In addition, W I (intermediate water), which is only observed in fully CO 2 -bound hCA II (Domsic et al., 2008), abruptly disappears, while W1 appears, as the CO 2 is released. Moreover, with CO 2 release W2 0 (an alternate position of W2) in close proximity to residue His64 was observed to gradually disappear, whereas His64 concurrently rotated from the 'out' to the 'in' rotameric conformation. Despite the structural changes observed, the rapid changes taking place with the crystal incubation method left some of the key questions unanswered, such as how the protontransfer water network is restored during hCA II catalytic activity.
In this study, we present ultrahigh-resolution structures of hCA II from crystals cryocooled under CO 2 pressures of 7.0 atm (0.9 Å resolution) and 2.5 atm (1.0 Å resolution), which are hereafter referred to as '7.0 atm CO 2 hCA II' and '2.5 atm CO 2 hCA II', respectively. These two structures were compared with the three previous structures (Kim et al., 2016) of hCA II crystals cryocooled under 15 atm CO 2 pressure and then incubated at room temperature for 0 s (1.2 Å resolution structure from PDB entry 5dsi; hereafter referred to as '15 atm CO 2 hCA II'), 50 s (1.25 Å resolution structure from PDB entry 5dsj; hereafter referred to as '15 atm CO 2 hCA II -50s') and 1 h (1.45 Å resolution structure from PDB entry 5dsn; the 'CO 2 -free state' and hereafter referred to as '15 atm CO 2 hCA II -1h'). The structural comparison reveals that 7.0 atm CO 2 hCA II and 2.5 atm CO 2 hCA II are previously unknown intermediate states between 15 atm CO 2 hCA II and 15 atm CO 2 hCA II -50 s. Together, these studies provide a view of how hCA II utilizes a water reservoir to fill the void in the active site as CO 2 is released.

Protein expression and purification
The zinc-containing hCA II was expressed in a recombinant strain of Escherichia coli BL21 (DE3) pLysS transformed with a plasmid encoding the hCA II gene (Forsman et al., 1988). Purification was carried out using affinity chromatography as described previously (Khalifah et al., 1977). Briefly, bacterial cells were enzymatically lysed with hen egg-white lysozyme and the lysate was loaded onto agarose resin coupled with p-(aminomethyl)benzenesulfonamide, which binds to hCA II. The protein on the resin was eluted with 400 mM sodium azide in 100 mM Tris-HCl pH 7.0. The azide was removed by extensive buffer exchange against 10 mM Tris-HCl pH 8.0.

Protein crystallization
Crystals of hCA II were obtained using hanging-drop vapour diffusion (McPherson, 1982). A 10 ml drop consisting of equal volumes of protein solution (5 ml) and well solution (5 ml) was equilibrated against 1 ml well solution (1.3 M sodium citrate, 100 mM Tris-HCl pH 7.8) at room temperature ($20 C) (Domsic et al., 2008). Crystals grew to approximate dimensions of 0.1 Â 0.1 Â 0.3 mm in a few days.
2.3. CO 2 entrapment using pressure cryocooling CO 2 entrapment was carried out as described in previous reports (Domsic et al., 2008;Kim et al., 2016). The hCA II crystals were first soaked in a cryosolution consisting of the reservoir solution supplemented with 20%(v/v) glycerol. The crystals were then coated with mineral oil to prevent dehydration and loaded into the base of high-pressure tubes. Once in the pressure tubes, the crystals were pressurized with CO 2 gas to two different pressures (7.0 and 2.5 atm) at room temperature. After 10 min, the crystals were cryocooled to liquid-nitrogen temperature (77 K) without releasing the CO 2 gas. Once the CO 2 -bound crystals had been fully cryocooled, the crystal-pressurizing CO 2 gas was released and the crystal samples were stored in a liquid-nitrogen dewar until X-ray data collection. Note that once cryocooled, the CO 2 -bound research papers IUCrJ (2018 hCA II crystals were handled just like normal protein crystals and were flash-cryocooled at ambient pressure.

X-ray diffraction and data collection
Diffraction data were collected on CHESS beamline F1 (wavelength of 0.9180 Å , beam size of 100 mm) under a nitrogen cold stream (100 K). Data were collected using the oscillation method in intervals of 1 on an ADSC Quantum 270 CCD detector (Area Detector Systems Corporation) with a crystal-to-detector distance of 100 mm. For the 7.0 atm CO 2 hCA II data set (0.9 Å resolution), an initial data set consisting of 180 images was collected with 1 s exposures to cover diffraction resolution up to 1.1 Å . The detector was then offset to cover diffraction resolution up to 0.88 Å , and a second data set consisting of 360 images was collected with 10 s exposures. For the 2.5 atm CO 2 hCA II data set (1.0 Å resolution), a single data set consisting of 360 images was collected with 10 s exposure for each image. For each X-ray data set, the estimated absorbed X-ray dose was $2 Â 10 7 Gy. No significant diffraction resolution decay was observed up to this X-ray dose. Indexing, integration, merging and scaling were performed using HKL-2000 (Otwinowski & Minor, 1997). Data-processing statistics are given in Table 1.

Structure determination and model refinement
The structures of hCA II at CO 2 pressures of 7.0 and 2.5 atm were determined using the CCP4 program suite . Prior to refinement, a random 5% of the data were flagged for R free analysis. The previously determined 1.1 Å resolution crystal structure (PDB entry 3d92; Domsic et al., 2008) was used as the initial phasing model. Maximumlikelihood refinement (MLH) was carried out using REFMAC5 (Murshudov et al., 2011) and the water molecules were automatically picked up using ARP/wARP (Perrakis et al., 1999) during the MLH cycles. The refined structures were manually checked using the molecular graphics program Coot (Emsley & Cowtan, 2004). Reiterations of MLH refinement were carried out with anisotropic B factors and riding H atoms. The partial occupancies of W1 in 7.0 and 2.5 atm CO 2 hCA II were estimated such that the electron density in the F o À F c map disappears. The refinement statistics are given in Table 1. We also re-refined the water molecules in the three previously reported structures (PDB entry 5dsi for 15 atm CO 2 hCA II, PDB entry 5dsj for 15 atm CO 2 hCA II -50s and PDB entry 5dsn for CO 2 hCA II -1h; Kim et al., 2016) for accurate comparison of the bound water molecules in the active site and the entrance conduit. The re-refined structures The active site of hCA II at different internal CO 2 pressures. (a) 15 atm CO 2 hCA II, (b) 7.0 atm CO 2 hCA II, (c) 2.5 atm CO 2 hCA II, (d) 15 atm CO 2 hCA II -50s, (e) 15 atm CO 2 hCA II -1h. The electron-density (2F o À F c ) map (in blue) is contoured at 1.3, except for W 0 I in (d), which is contoured at 1.0. The intermediate waters W I and W 0 I are coloured light grey and the entrance-conduit water W 0 EC1 is coloured cyan for clarity. Note that CO 2 is fully bound in the active site in (a) and (b). Concurrent with the decrease in CO 2 pressure, the electron density for CO 2 fades out in (c) and is eventually replaced by two water molecules in the CO 2 binding site (d, e). Note also the dynamic changes reflected by the electron densities of W I , W 0 I and W1 that take place as the internal CO 2 pressure decreases. These events are more explicitly explained in Fig. 4. were updated in the PDB with the new PDB codes 5yui (superseding 5dsi), 5yuj (superseding 5dsj) and 5yuk (superseding 5dsn). Details of the structural analysis of the bound water molecules are given in the Supporting Information. All structural figures were rendered with PyMOL (Schrö dinger).

Results and discussion
3.1. CO 2 binding sites: active site (CO 2 /W Zn /W DW /W 0 0 0 DW ) and secondary CO 2 site near Phe226 Structural examinations show that the five structures are very similar. The all-protein-atom r.m.s.d.s between 15 atm CO 2 hCA II and the other four structures (7.0 atm CO 2 hCA II, 2.5 atm CO 2 hCA II, 15 atm CO 2 hCA II -50s and 15 atm CO 2 hCA II -1h) are 0.14, 0.12, 0.10 and 0.13 Å , respectively. Although changes in the zinc-coordinating histidines (His94, His96 and His119) are negligible between the structures, the electron densities for the CO 2 binding site differ significantly, as expected (Fig. 2). While 15 and 7.0 atm CO 2 hCA II show a clear position for the CO 2 (Figs. 2a and 2b), deterioration of electron density for the CO 2 site occurs in the 2.5 atm CO 2 hCA II structure, represented by sparsely connected lobes (Fig. 2c). When the density is modelled and refined with only CO 2 , the CO 2 occupancy is at most 0.7. This difference suggests that the CO 2 site is occupied by both CO 2 and two waters (deep waters W DW and W 0 DW ) at a pressure of 2.5 atm. The manifestation of W DW at this pressure is supported by the extended electron-density connection from CO 2 to W I (Supplementary Fig. S1). In 15 atm CO 2 hCA II -50s, the electron density for the CO 2 binding site is further shifted towards Zn and W Zn , which correlates with the known positions of W DW and W 0 DW (Fig. 2d). This argues that 2.5 atm CO 2 hCA II has a higher internal CO 2 pressure than 15 atm CO 2 hCA II -50s. Finally, in 15 atm CO 2 hCA II -1h, the electron density of the CO 2 binding site splits into two distinct lobes, indicating that the CO 2 site is completely replaced by W DW and W 0 DW (Fig. 2e). Previously, the binding of a secondary CO 2 molecule which is 15-16 Å away from the active-site CO 2 molecule was reported in a hydrophobic pocket created by Val223 and Phe226 (Domsic et al., 2008). Comparison of the 15 atm CO 2 hCA II and 15 atm CO 2 hCA II -1h structures in this region indicates that the side chain of Phe226 must rotate to accommodate the secondary CO 2 molecule (Supplementary Figs. S2a and S2e). Interestingly, in the lower pressured 7.0 atm CO 2 hCA II, the subdued electron density for the secondary CO 2 results in dual conformations of the Phe226 side chain (Supplementary Fig. S2b). In the cases of 2.5 atm CO 2 hCA II and 15 atm CO 2 hCA II -50s, the secondary CO 2  Rotameric states of His64 and solvent positions at different internal CO 2 pressures. (a) 15 atm CO 2 hCA II, (b) 7.0 atm CO 2 hCA II, (c) 2.5 atm CO 2 hCA II, (d) 15 atm CO 2 hCA II -50s, (e) 15 atm CO 2 hCA II -1h. The electron density (2F o À F c ) for W 0 I in (d) is contoured at 1.0. In all other cases, the electron density (2F o À F c ) for His64 and the electron density (2F o À F c ) for waters are contoured at 1.5 and 1.3, respectively. The intermediate waters W I and W 0 I are coloured light grey and the entrance-conduit water W 0 EC1 is coloured cyan for clarity. As the internal CO 2 pressure decreases, W2 0 gradually dissipates and the His64 side chain shifts from the 'out' to the 'in' position from (a) to (e). The intermediate water, W I , is clearly observed in (a) and the electron density gradually subsides (b, c) and finally disappears (d, e). In accordance to the decrease in W I , electron density for W1, which is not observable in (a), appears in (b) and subsequently increases gradually (c, d, e). When the models are refined with partial occupancy, the W1 occupancies are 0.8 in (b) and 0.9 in (c). Interestingly, the electron density for the newly observed intermediate water W 0 I increases gradually from (a) to (c), but decreases in (d) and disappears in (e). The measured distance between W I and W1 in (b) and (c) is 2.0 Å . The electron density for W3a is well isolated in all cases, but W3b shows an alternate position W3b 0 in (a) which grows in (b) but subsequently disappears (c, d, e).
was not present and the Phe226 side chain sits in the position observed in the CO 2 -free 15 atm CO 2 hCA II -1h (Supplementary Figs. S2c, S2d and S2e). Hence, the observation of the secondary CO 2 and the dual conformations of the Phe226 side chain in 7.0 atm CO 2 hCA II imply that 7.0 atm CO 2 hCA II has a higher internal CO 2 pressure than 15 atm CO 2 hCA II -50s.
3.2. His64 and the water network (W1/W I /W2/W2 0 0 0 /W3/W3a/ W3b) near the active site As described above, structural examinations of the CO 2 binding sites suggest that both 7.0 atm CO 2 hCA II and 2.5 atm CO 2 hCA II have a higher internal CO 2 pressure than 15 atm CO 2 hCA II -50 s. Furthermore, 7.0 atm CO 2 hCA II intuitively has a higher internal CO 2 pressure than 2.5 atm CO 2 hCA II, hence leading to the conclusion that the internal CO 2 pressure decreases in the sequence 15 atm CO 2 hCA II, 7.0 atm CO 2 hCA II, 2.5 atm CO 2 hCA II, 15 atm CO 2 hCA II -50s, 15 atm CO 2 hCA II -1h. Such an interpretation ascertains that 7.0 atm CO 2 hCA II and 2.5 atm CO 2 hCA II are intermediate states that fill the gaps between the 15 atm pressurized CO 2 hCA II and the earliest time point of CO 2 release (15 atm CO 2 hCA II -50s) observed in the previous study (Kim et al., 2016). On this foundation, His64 and the water network near the active site were analyzed in order of decreasing internal CO 2 pressure (Fig. 3).
Although the side chain of His64 lies predominantly in the 'out' position in 15 atm CO 2 hCA II (Fig. 3a), the electron density of His64 infers that it occupies dual 'out' and 'in' positions as the internal CO 2 pressure decreases (Figs. 3b, 3c and 3d). However, in the CO 2 -free 15 atm CO 2 hCA II -1h, His64 is observed to primarily occupy the 'in' position ( Fig. 3e). In concert with His64 moving from the 'out' to the 'in' position, the density for W2 0 (an alternate position of W2) is observed to gradually dissipate.
In the previous studies, it has been recognized that when CO 2 is fully bound in the 15 atm CO 2 hCA II structure, W I but not W1 is observed (as in Figs. 2a and 3a; Kim et al., 2016). However, this W I disappeared in 15 atm CO 2 hCA II and W1 was observed to appear instead in the CO 2 -free 15 atm CO 2 hCA II -1h (as in Figs. 2e and 3e; Kim et al., 2016). Because the measured distance between W I and W2 is $4.8 Å , the hydrogen-bonded water network (via W1, W2 and His64) necessary for the proton-transfer wire was presumed to be broken when CO 2 fully binds to the active site. In this study, we observed the dynamic replacement of W I with W1 as the internal CO 2 pressure decreases, since dually occupied Solvent positions in the entrance conduit. (a) 15 atm CO 2 hCA II, (b) 7.0 atm CO 2 hCA II, (c) 2.5 atm CO 2 hCA II, (d) 15 atm CO 2 hCA II -50s, (e) 15 atm CO 2 hCA II -1h. In all cases the electron-density (2F o À F c ) maps are contoured at 1.3, except for the electron density (2F o À F c ) for W 0 I in (d), which is contoured at 1.0. The intermediate waters W I and W 0 I are coloured light grey and the entrance-conduit waters are coloured cyan for clarity. As the internal CO 2 pressure decreases, alternate positions appear and disappear, especially for W EC1 , W EC2 and W EC5 , suggesting dynamical motions that are correlated with the dynamical changes in W I , W 0 I and W1 (a, b, c). For example, note that W 0 EC1 in (a) and (b) is located next to the electron density for W3b 0 , attesting to the interaction between the two. As W I and W 0 I disappear in (d) and (e) together with appearance of W1, the entrance-conduit water molecules return to the singly ordered positions (d, e).
positions of W1 and W I are seen for 7.0 and 2.5 atm CO 2 hCA II (Figs. 3b and 3c). The reduction of electron density for W I is observed in the order 15, 7.0 and 2.5 atm CO 2 hCA II, with complete disappearance in 15 atm CO 2 hCA II -50s and 15 atm CO 2 hCA II -1h (Fig. 3). In contrast, W1 electron density starts to emerge in the 7.0 atm CO 2 hCA II, is more pronounced in 2.5 atm CO 2 hCA II, and is fully occupied in 15 atm CO 2 hCA II -50s and 15 atm CO 2 hCA II -1h. The close 2.0 Å distance between W1 and W I in 7.0 and 2.5 atm CO 2 hCA II suggests that W1 and W I exhibit partial occupancies rather than being two separate, stable waters. The inverse correlation, with a decrease in W I and increase in W1 electron density upon decreasing internal CO 2 pressures, suggests that W I moves to the W1 position upon CO 2 release.  (Figs. 2, 3b and 3c). When the previously reported structures were compared, W 0 I existed in the 15 atm CO 2 hCA II and 15 atm CO 2 hCA II -50s structures, but was overlooked because of its faint electron density (Figs. 2, 3a and 3d). Compared with W I , W 0 I is located farther away from the active site and more towards the entrance that connects the active site of hCA II to bulk solvent. Because the entrance is near the active site where water, substrate and product (CO 2 / bicarbonate) can interchange/interact with bulk solvent, we will refer to it as the 'entrance conduit' (Fig. 1). The conduit consists of the hydrophobic residues Leu198, Val135, Leu204, Pro202 and Phe131 on one side, and the hydrophilic residues His64, Gln92 and Thr200 on the other. It should also be noted that the proton-shuttling His64 is positioned perpendicularly to this entrance conduit (Fig. 1).
A close inspection of the structures further identified five water molecules (named entrance-conduit waters or W EC s) that are ordered along the surface of the entrance conduit in the CO 2 -free 15 atm CO 2 hCA II -1h (sequentially named counterclockwise as W EC1 , W EC2 , W EC3 , W EC4 and W EC5 starting from the one closest to water W3b; Fig. 4). Unlike for W EC3 and W EC4 , alternate positions for W EC1 (W 0 EC1 and W 00 EC1 ), W EC2 (W 0 EC2 ) and W EC5 (W 0 EC5 ) exist in the different internal CO 2 pressure structures. Tight hydrogen-bonding networks stabilize W EC1 , W EC2 , W EC3 , W EC4 and W EC5 , which are mediated by residues lining the entrance conduit ( Supplementary Fig. S3). For instance, the side-chain amide N atom of Gln92 binds to W EC2 , and the main-chain carbonyl O atom of Pro201 and the side-chain hydroxyl O atom of Thr200 bind to W EC5 , which are conserved throughout all of the internal CO 2 pressure structures. Hydrogen-bonding interactions also exist in all of the structures between the five W EC waters (W EC1 -W EC2 , W EC2 -W EC3 , W EC3 -W EC4 and W EC4 -W EC5 ).
The Entrance-conduit water dynamics. (a) 15 atm CO 2 hCA II, (b) 7.0 atm CO 2 hCA II, (c) 2.5 atm CO 2 hCA II, (d) 15 atm CO 2 hCA II -50s, (e) 15 atm CO 2 hCA II -1h. In all cases, the electron-density (2F o À F c ) maps are contoured at 1.3. The intermediate water W I is coloured light grey and the entranceconduit waters are coloured cyan for clarity. Although all five W EC waters (W EC1 , W EC2 , W EC3 , W EC4 and W EC5 ) exist in all of the structures regardless of the different internal CO 2 pressures, dramatic variations of W EC1 (near to the proton-shuttling His64), W EC2 (near to W I , W 0 I and W1) and W3b 0 are manifested by multiple alternate positions during the internal CO 2 pressure decrease. These observations indicate that the waters of the proton-transfer Although all five W EC waters are present regardless of the different internal CO 2 pressures, some perturbations of W EC1 (near to the proton-shuttling His64) and W EC2 (near to W I , W 0 I and W1) were observed during the internal CO 2 pressure decrease, which are manifested by multiple alternate positions (Fig. 5). The dynamic motions of W EC waters imply their direct interplay with the proton-transfer water network in the active site. Specifically, interactions between W EC1 and W3b were observed. Previously, the positions of W3a and W3b were thought to be invariant and singly occupied regardless of CO 2 binding, leading to the belief that the main role of W3a and W3b was to stabilize the W2 water molecule that is directly located within the proton-transfer wire. However, in this study an alternate position of W3b [named W3b 0 , which is different from the two alternative positions (W3b 0 and W3b 00 ) in CO 2bound apo CA II in Kim et al. (2016)] was observed along with an alternate position of W EC1 (named W 0 EC1 in this study) in 15 and 7.0 atm CO 2 hCA II (Figs. 5a and b). In these structures, the distances between W3b, W3b 0 , W EC1 and W 0 EC1 are so close (1.3-1.7 Å ; Supplementary Fig. S5) that they organize into a continuous tube of electron density (Fig. 5). W3b 0 and W 0 EC1 disappear in lower internal CO 2 pressure structures, with W EC1 recovering to the singly occupied position (Figs. 5c, 5d and 5e). These results suggest that the waters in the protontransfer network (W1/W2/W2 0 /W3a/W3b/W3b 0 ), the intermediate waters W 0 I /W I and the water network of the entrance conduit (W EC1 /W EC2 /W EC3 /W EC4 /W EC5 ) all act interdependently with their motions correlated.

Mechanism of the restoration of the proton-transfer water network
By lowering the CO 2 pressure in hCA II crystals, we captured additional intermediate states, including dual occupied positions of W1 and W I , an active site partially occupied with CO 2 , W DW and W 0 DW , a new intermediate water W 0 I and an alternate position of W3b (W3b 0 ). By realizing that W I and W 0 I are transiently stabilized by several entrance-conduit water molecules and that they rearrange during the restoration of the proton-transfer water network, we propose the sequential events in the formation of the water network that replenishes W Zn and the consequential connection of the His64-mediated proton-transfer wire during the catalytic turnover of hCA II. Although these events have been postulated from the observations during CO 2 release in this study, these mechanisms may also account for restoration of the water network after bicarbonate release, assuming that both CO 2 and bicarbonate molecules do not directly mediate the water-network restoration process.
It is observed that only W I , and not W1, is observed near the active site of hCA II in the fully CO 2 -bound state (Fig. 6a)  Proposed mechanism of water-network restoration. (a) In hCA II with fully bound CO 2 , W Zn exists as OH À and nucleophilic attack occurs to form bicarbonate. In this state, only W I is present and not W1, suggesting that the water network for proton transfer is broken. (b) The sequential events for water-network restoration as the product bicarbonate leaves the active site. After the intermediate water W I fills the places for W1 and W DW , and these W1 and W DW waters or W I can move to the W Zn position. W DW also fills the position for W 0 DW . Note that four waters (W1, W Zn , W DW and W 0 DW ) are eventually filled in from W I during this water-network restoration process. A newly found intermediate water W 0 I is located between W I and the outside bulk solvent, is stabilized by the entranceconduit waters and seems to facilitate the fast charging process of W I . (c) Only after the water network is restored can proton transfer can occur from W Zn to the outside through W1/W2/W2 0 /His64 in /His64 out . Now, with CO 2 binding, hCA II is ready for the next catalytic turnover. of W1 suggests that the hydrogen-bonded water network from W Zn to His64 (charged and in the 'out' position) is disrupted. In fact, when hCA II is fully CO 2 -bound, the proton transfer should have already happened to result in the deprotonation of W Zn to OH À , which is necessary for the nucleophilic CO 2 attack (the first step in equation 1). After bicarbonate formation via this nucleophilic attack, the product bicarbonate subsequently leaves the active site and W Zn is replenished along with restoration of the active-site water network prior to the proton-transfer process from W Zn .
It is likely that W Zn replenishment and water-network restoration are directly mediated by the transient waters W I and W 0 I . After bicarbonate diffuses out of hCA II, W I seems to immediately fill the positions of both W1 and W DW (Fig. 6b). This directional branching movement of W I is predictable since the distance between W I and W1 is 2.0 Å and the distance between W I and W DW is 2.4 Å . This interpretation is further supported by the observation that the electron density of W1 emerges as that of W I disappears (Figs. 3a, 3b, 3c and 3d), as well as by the observation that the electron density of W I is fused to the electron density of W DW ( Supplementary  Fig. S1). Subsequently, W1 can move to W Zn (the distance between W1 and W Zn is 2.6 Å ), and W DW can shift to either W Zn or W 0 DW (the distances from W DW to W Zn and from W 0 DW are 2.4 and 2.2 Å , respectively). Judging by the distance from W I to W Zn (2.6 Å ), W I can also directly flow into the W Zn position (Fig. 6b). As W I replenishes multiple water positions (W1/W DW /W 0 DW / W Zn ), it is important that the bulk solvent supplies the W I position rapidly (acting as a water reservoir), a process that seems to be facilitated by W 0 I . W 0 I is separated from W I by 2.2 Å , is located closer to the bulk solvent and is transiently stabilized by the dynamic motions of water molecules in the entrance conduit, which take place in concert with the changes of solvent in the active site ( Fig. 4 and Supplementary Fig. S4). As the W1/W DW /W 0 DW /W Zn positions are filled, the intermediate water W I is destabilized by steric hindrance with W1 (the distance between W1 and W I is only 2.0 Å ). In addition, the dynamic motions of water molecules in the entrance conduit decrease as the water network is restored in the active site (Fig. 5), which causes destabilization of W 0 I . Therefore, the intermediate waters W I and W 0 I disappear. Finally, the activesite water network is fully restored and proton transfer occurs from W Zn to His64 (uncharged and in the 'in' position) via W1/W2/W2 0 (Fig. 6c).

Conclusions
Structural comparisons between hCA II in complex with CO 2 and during its release reveal intermediate snapshots during the water-network rearrangement in the active site as the waters fill the void following CO 2 liberation. Based on our observations, insight into the water-network restoration prior to proton transfer is proposed. While previous studies of the catalytic activity of hCA II have mainly focusing on the CO 2 binding site (Zn/W Zn /W DW ) and the proton-transfer water network (W1/W2/W3a/W3b), our results indicate that the intermediate and alternate waters (W 0 I , W I , W2 0 and W3b 0 ) and the entrance-conduit waters (W EC1 , W EC2 , W EC3 , W EC4 , W EC5 and their alternative positions) are critically involved in catalysis by hCA II. The substrate CO 2 enters via the hydrophobic half of the active site, while the product HCO 3 À , being a charged molecule, exits by perturbing the ordered waters that fill the hydrophilic half of the active site (Silverman et al., 1979;Koenig et al., 1983). Thus, the ordered waters within the active site and its vicinity are likely to exist in a state of intermittent rearrangement during the forward and reverse reactions of catalysis. Taken collectively, our results provide snapshots of low-energy stages of water rearrangement during catalysis. Future mutation studies to perturb the protein regions that stabilize these waters would provide more evidence of their roles in the reaction. Moreover, our results suggest that the catalytic activity of hCA II can be more thoroughly understood with the 'extended' catalytic waters (W DW /W 0 DW /W Zn / W1/W I /W 0 I /W2/W2 0 /W3a/W3b/W3b 0 /W EC1 -W EC5 ). Molecular dynamics simulations on this extended water network may reveal further insights into the bioenergetic mechanisms utilized by hCA II to generate ordered water networks from the surrounding disordered bulk solvent (Riccardi et al., 2006;Roy & Taraphder, 2007).